Sunteți pe pagina 1din 447

Contents

Cover
Series
Title

Page

Copyright
Dedication
Preface
Acknowledgments
Chapter

1: How

Media,
HFT

Modern

Modern
as

Markets,

Evolution

and

Is High-Frequency

What

Do

How

Many

Those

Past

Methodology

Traders

High-Frequency
in the

Organization

from

Trading?

High-Frequency

Players

Differ

HFT

of Trading

What

Major

Markets

Traders

HFT

of This

Do?
Are

There?

Space

Book

Summary
End-of-Chapter
Chapter

Questions

2: Technological

A Brief

History

Innovations,

Systems,

and

HFT

of Hardware

Messaging
Software
Summary
End-of-Chapter
Chapter

3: Market

Types
Limit

Questions
Microstructure,

Orders,

and

Limit

Order

of Markets
Order

Books

Aggressive
Complex

versus

Passive

Execution

Orders

Trading

Hours

Modern

Microstructure:

Fragmentation

in Equities

Market

Convergence

and

Divergence

Books

Fragmentation

in Futures

Fragmentation

in Options

Fragmentation

in Forex

Fragmentation

in Fixed

Fragmentation

in Swaps

Income

Summary
End-of-Chapter
Chapter
What
How

Questions

4: High-Frequency

Data1

Is High-Frequency

Data?

Is High-Frequency

Properties

Data

Recorded?

of High-Frequency

Data

High-Frequency

Data

Are

Voluminous

High-Frequency

Data

Are

Subject

High-Frequency

Data

Are

Not

High-Frequency

Data

Are

Irregularly

Most

High-Frequency

Data

to the

Normal

Do

Not

Bid-Ask

Bounce

or Lognormal
Spaced
Contain

in Time
Buy-and-Sell

Identifiers

Summary
End-of-Chapter
Chapter

Questions

5: Trading

Overview

of Execution

Costs

Execution

Costs

Transparent
Implicit

Costs

Execution

Background

and

Estimation
Empirical

Costs
Definitions

of Market

Impact

Estimation

of Permanent

Market

Impact

Summary
End-of-Chapter
Chapter

Questions

6:

Performance

and

Capacity

Strategies
Principles
Basic

of Performance

Performance

Comparative

Ratios

Measures

Measurement

of

High-Frequency

Trading

Performance
Capacity

Attribution
Evaluation

Alpha

Decay

Summary
End-of-Chapter

Questions

Chapter

Business

Key

7: The

Processes

Financial

Market

Trading

of HFT

Markets

Economics

of High-Frequency

Suitable

for

HFT

of HFT
Participants

Summary
End-of-Chapter

Questions

Chapter

8: Statistical

Practical

Applications

Arbitrage

Strategies

of Statistical

Arbitrage

Summary
End-of-Chapter
Chapter

Questions

9: Directional

Developing
What

Trading

Directional

Constitutes

Forecasting
Tradable

an

Around

Event-Based

Events
Strategies

Event?

Methodologies
News

Application

of Event

Arbitrage

Summary
End-of-Chapter
Chapter

Questions

10:

Automated

Market

MakingNave

Introduction
Market

Making:

Simulating
Nave
Market
Profitable
Summary

Key

Principles

a Market-Making

Market-Making
Making
Market

as

Strategy

Strategies
a Service
Making

Inventory

Models

End-of-Chapter

Questions

Chapter

11:

Automated

What's

in the

Data?

Modeling

Information

Market

in Order

Making

II

Flow

Summary
End

of Chapter

Chapter

12:

Market

Crashes

Latency

Questions
Additional

HFT

Strategies,

Arbitrage

Spread

Scalping

Rebate

Capture

Quote

Matching

Layering
Ignition
Pinging/Sniping/Sniffing/Phishing
Quote

Stuffing

Spoofing
Pump-and-Dump
Machine

Learning

Summary
End-of-Chapter
Chapter
Key

Questions

13:

Regulation

Initiatives

of Regulators

Worldwide

Summary
End-of-Chapter
Chapter

Questions

14:

Measuring

Risk
HFT

Management

of HFT

Risk

Summary
End-of-Chapter
Chapter
Why

15:

Execution

Order-Routing

Questions
Minimizing
Algorithms?
Algorithms

Market

Impact

Market

Manipulation,

and

Issues

with

Advanced

Basic

Models

Models

Practical

Implementation

of Optimal

Execution

Summary
End-of-Chapter
Chapter
Model

Questions

16:

Implementation

Development

Life

System

Implementation

Testing

Trading

Systems

Summary
End-of-Chapter

Questions

About

the

Author

About

the

Web

References
Index

Site

of HFT
Cycle

Systems

Strategies

Founded

in

publishing

John

company

America,

Europe,

developing
for

1807,

in

the

and

marketing

customers'

&

Sons

United

Australia,

and

our

Wiley

print

is

the

States.

Asia,
and

oldest

With

offices

Wiley

is globally

electronic

products

professional

and

independent

personal

in

North

committed
and

to

services

knowledge

and

understanding.
The

Wiley

survived

Trading
the

series

market's

features
ever

prosperedsome

by

reinventing

basics.

novice

Whether

in-between,
needed
For

these
to prosper

list

www.WileyFinance.com

books
today

of

by

well

titles,

others

professional,

provide
into

traders

temperament

systems,

will

available
.

changing

trader,

and

books

the
the

by

getting
or

advice

and

who

have

and

have
back

to

somewhere
strategies

future.
visit

our

web

site

at

Cover

image:

Cover

design:

Crusitu
John

Copyright

2013

Published

by

The

Edition

First

Algorithmic
Wiley

and

John

of
or

Wiley

Aldridge.

and

Inc.

Publisher,

or

per-copy

fee

the

Web

permission
Wiley

(201)

in

any

form

107

either

author

be

no

Inc.,

completeness
any

implied

purpose.

consult
nor

or
herein

with
author

commercial

damages,

be

written

of

retrieval

electronic,
except

United

States

222

Rosewood

646-8600,

the

or

Publisher

Permissions

for

Department,

Hoboken,

748-6008,

the

appropriate

(978)
to

of

the

Inc.,

Street,

permission

fax

the

River

(201)

Warranty:

efforts

contents
of

NJ

or

While

in preparing
with

07030,

online

of

this

may

be

sales

not

be

book

at

including

for

The
for

your

appropriate.
any
but

loss
not

limited

for

profit
to

particular
by

and

situation.

the

special,

sales

strategies
You

or

or

disclaim

extended

Neither
of

make

accuracy

specifically

advice

and

they

the

fitness
or

publisher

book,

to

and
or

materials.

where

liable

this

created

suitable

the

respect

merchantability

a professional
shall

John

otherwise,

1976

Center,

to

warranties

written
may

the

750-8400,

111

best

warranty

representatives
contained

the

of

in

means,
or

. Requests

of

warranties
No

by

to

or
of

any

payment

addressed

their

representations

108
prior

(978)

fax

used

by

Clearance

Liability/Disclaimer

have

or
the

www.wiley.com/go/permissions
of

Guide

published

stored

scanning,

www.copyright.com

Sons,

or

through

01923,

748-6011,

Limit

Jersey.

Practical

was

reproduced,

recording,

Copyright

MA

&

New
A

Systems

be

Section

should

Hoboken,
Trading:

may

authorization

at

reserved.

in Canada.

without

Danvers,

Inc.,

Trading

publication

to the

rights

in 2010.

under
Act,

All

High-Frequency

photocopying,

Copyright

Inc.

& Sons,

transmitted

permitted

John

Irene

of

this

mechanical,

on

& Sons,

simultaneously

part

Drive,

by

Sons,

system,

as

Wiley

Strategies

Published
No

Robert/iStockphoto

should
publisher

any

other

incidental,

consequential,
For

or other

general

information

technical

support,

within

the

at (317)
Wiley

United

print-on-demand.

this
in

book
the

may

refers

(317)

variety

not

included

media

you

Library

visit

Irene,

pages

trading:

cm.(Wiley

ISBN

978-1-118-34350-0

For

for

Department

the

United

States

electronic
with

standard

and

print

by

versions

or in print-on-demand.

or

may

formats

DVD

that

download

more

is not
this

information

If
included

material
about

at
Wiley

.
Data:

4. Electronic

trading

HG4529.A43

2013

guide

analysis.

to algorithmic

strategies

and

Edition.

series)

(Cloth)ISBN

978-1-118-43401-7

1. Investment

2012048967

you

a practical

trading

index.

332.64dc23

a CD

Aldridge.2nd

Includes

(ebk)

and

Cataloging-in-Publication

systems/Irene

(ebk)ISBN

or

1975

High-frequency
trading

services

Care

outside

in e-books
as

www.wiley.com

and

Customer

included

purchased,

of Congress

Aldridge,

print

such

http://booksupport.wiley.com
products,

our
762-2974,

of

material

be

products

572-4002.

Some

to

version

other

at (800)

or fax
in

our
contact

States

572-3993

book

on

please

publishes

of this

damages.

(ebk)
2. Portfolio

of securities.

I. Title.

978-1-118-42011-9
ISBN
management.

978-1-118-41682-2
3.

Securities.

To

my

family

PREFACE
If

hiring

activity

industries,

then

successful

activity

the

Jobs

is

Journal

there

were

Stanley

in

presidents

were

directors
presidents

were

investment

into

HFT

their

employers

in

for

is

still

comprise

HFTs

in such

Kirilenko

liquid

S&P

average

Source:

its

infancy.

60

to

500

al.,

all

the

instruments

Aldridge

(2012a)

As

(NYSE:
20

associate

To

generate

vice

executive
and

vice

warrant

the

prospective

percent

the

employees

high

returns

high-frequency

market

as

just

trading

development,

of

little

for

Morgan

development,

claim

for

account

Wall
placed

roles.

levels:

all

some

reality.

ETF

related

at

field,

in

2011).

the

for

future.

percent

as

most

example,

of

high-frequency

to

While

70

the

for

in its

levels,

expected

in

Take,

operations.

reached

traded
et

at

foreseeable

account

frequently

HFT

far

advertisements

and

strategy

in

hiring

seldom

still

(see

Figure

in

for

by

expanding

section

five

technology

HFT

clearly

the

All

sought

employees

were

industry

that

2012.

were

is

Investing

candidates

needed

considerable

are

four

rapidly

today.

trading

in HFT

new

(HFT)

and

27,

sought

skills

numbers

hiring

and

sector

Money

required

Despite

traders

the

candidates

were

with

financial

high-frequency

was

HFT

profitable
trading

November

for

operation.

the
in

on

alone

in

high-frequency

classifieds

Street

highest

all

that

percent

as

the

S&P

find

of all

market

activity

500

E-mini

futures

shows,

SPY),

high-frequency
trades.

such

examinations

Figure

of daily

high-frequency

participants,

Scientific

25

trading

even

in

the

traders

very
on

As

shown

remain
17

in Figure

remarkably

percent

same

of

time,

industry

any

computer

that

prepare

industry

science.

required

This

highly

book

is

accurate,

and

companion

many

web

practical

examples,

the

informative

book

site,

the

that

the

yet

with

you,

it

can

data

to

development

strategies

seem
to

the

develop

of

takes

extreme,

it

at

is

not

proficiency

discipline

rests

of

high-frequency

study:

academic

simultaneously
do

computer

scientists

or

have

the

confluence
finance

institutions

offer

competent

in both

not

programs

understand
do

that

academic

not

areas.

computer

have

well
about

as

void:

and

helpful

to

information
to

address

the

subject.

www.hftradingbook.com
updates,

at

a grasp

on

finance.
fill

as

and

toiling

tell

may

graduate-level

trading,
that

will

people

to

up-to-date

At

models.

required

few

academic

written

high-frequency
opinions

most

of

to

reconciling

to entry,

trading

areas

to be

finance-trained

explanation

years

number

as

Very

students

HFTs.

suggests

barriers

15

or specialization.

complex

and

low

in SPY

2012,

to

to HFT

operator

time

through

attributed

A natural

trading

the

complicated

programming,

be

has

HFT

the

other

already

pace.

participation

in 2009

ultra-short-term

from

of HFT

allocation

deploy

While

is particularly

and

the

and

years.

in any

can

HFT

successful

different

Most

SPY

requiring

profitable

three

most

of resource

develop
as

two

in

complex,

consistently

of

days

exists:

proficiently
Indeed,

on

observations

extremely

block

trades

levels

at a rapid

be

really

all

average

is growing

two

least

stable:

evidence

the

HFT

1 , the

teaching
in your

on

the

the

book

, where

future

true,

subject

questions

The

materials.

supply

of
and

has

you

can

find

I hope

you

find

endeavors.

ACKNOWLEDGMENTS
I am
and

extremely
insightful

balanced
to

my

Howarth,
terrific

grateful

to

suggestions,

perspective,
wise

editor
diligent

publisher

to
to

Bill

Pamela

Gaia

Falloon,

senior

my

husband
my

son

Rikhye

Henry
for

ber-patient

production
Van

for

Giessen

editor
for

tireless
for

terrific

helping
front-line

development
Vincent
making

encouragement
me

keep

edits,
editor

Nordhaus,
it all happen.

a
and

Judy
and

CHAPTER
How

Modern

Markets

Structural
fact,

change

it

is

the

leadership
(HFT)

of which

better

market

modern

past

50

HFT

umbrella.

consumer
the

a result,
the

lower

errors

order

has
as

of

the

and

more

execution,

deliver

a business

their

technology
also

are

becoming

The

lower
on

illustrate

the
1.1

the

base

Financial

of

deepening

expensive
squeezes

services
Markets

full-service

in the

the
input,

1970s,

book.

As
and

platforms

savings

from
reliability

computer

firms'

reliance

complexity
capabilities

substantial

1970s

that

this
1.1

and

of

platforms.

and

Figures

before

in
prices

higher

further,

circa

the

through

model.

landscape

under

trading

reporting

margins

the

effective,

financial

without

of

over

hardware

escalating

advanced

overview

2 of this

made

data

traders

technology

cost

business

The

more

traditional

financial

for

in

become

and

stability,

for

falling

Additionally,

systems.

prohibitively

pressure

in Chapter

continuity

requires

cost

in detail

has

market

the

and

markets,

markets

drops

has

years

the

computer

the

trading

costs

significant

powerful.

case

higher

strategies

for

software

If

drive

few
to

securities

demand

transmission
and

past

provides

core

trading
into

helped

execution

book

and

to

the

volatility,

the

HFT

led

has

in the

discussed

in message

regulations

Figure

decades,

investment

accessible

on

two

that

instruments.

High-frequency

lower

precipitated

defines

financial

improvements

and

and

in

over

in lower

technology-enabled

ensuing

code,

that

board,

more

of

limelight

chapter

markets

across

the

operational

This

past

trading

institutions.

transparency,

years,

the

to

Past

innovation

resulted

changes

Those

financial
into

have

investors.

dramatic

new
of

considerable

most

Over

not

stepped

delivered

and

is

from

constancy

of
has

Differ

and

today.

Electronization

puts
1.2

Figure

1.2

Financial

Markets

Today

In

the

and

1970s

and

individuals

shows,

on

earlier,
now

the

the

market

considered

portfolio

participants

were

traditional

management

organizations

players.

or

buy

As

side,

Figure
the

1.1

markets

engaged
Discretionary
funds,

and

Retail

hedge

flow,

others

with

Manual

for
the

market

short-term

makers

facilitating

A single

not-for-profit
wild

transaction
highly
the

low

and

traders

predominantly
to

large

were
share

ultimately,
millions

by

the

the

in
high

the

margin

spoils

proverbial

in

taking

buy

was

members

side,

and

established

and

and

relatively
their

financial

transaction

degree

of

to

lower

error

on

risk

form

of

brokers

to

with
as

intuition

as

Yet,

tune

the

receiving

commissions
the

to

trading,

markets.

large

bonuses

of

and

the

with

leading

associated

experience
bets

landscape

costs,

high

businesses,

fat-cat

the

high

the

the

class

exchange

their

side,

a fee.

asset

high

making

sell

to

labor-intensive

on

bank.

broker-dealers),

for

investors.

proprietary

of their
or

of

in each

by

orders,

intraday

account

trading

therefore

and

a
and,

of

tens

of

of dollars.

entrants

process:

also
of

by

relied

science

the

Fast-forward

hash

of

in

quotations

buy-side

securities;

processing

the

providing

characterized

of

mutual

investors,

middle-men,

exchange

manual

manual

opposed

the

paid

turnover

or for

speculation

was

involved

(representatives

risk,

costs

1970s

1970s

account

inventory

curtail

funds,

capitalization.

facilitation,

generally

of

smaller

transaction

Manual

pension

mom-and-pop

individuals

own

supported

The

individual

comparatively

their

including

funds.

including

markets

to

managers,

speculators,

trading
On

asset

to
successfully

out

precise

today's

markets,

compete
investing

using
models,

illustrated
lean

in

technology

reshaping

the

Figure
and

1.2 :
science

markets

in

new
to
the

Quantitative
funds,

money

are

latest

using

managers,

the

precise

mathematical

securities
Automated
funds,

the

microstructure,

and

market

share

Automated
and

HFT
from

proprietary

such

traders,

high-frequency

deliver

the
of

and

studies

hedge

of

transaction

market

costs,

taking

broker-dealers.
as

use

trading

and

forecasts

broker-dealers

low

traditional

arbitrageurs,

finance,

hedge

investments.

technology,

to

and

close

of their

example,

latest

funds

increasingly

profitability
for

mutual

of economics,

chisel

makers,

harness

over

to

improving

market

as

science

tools

prices,

such

statistical

arbitrage

quantitative

techniques,

hedge

algorithms,

to

deliver

like

new

funds
including

short-term

trading

profits.
Multiple

alternative

pools,

have

quality

financial

These

sprung

and

The

the

largely

of
and

The

trading,

affordable

characteristics

of

the

vastly

democratic

a right

anyone

formerly
club

costs

access:

due

to

can

reserved

trade

in

the

to

members

of

of broker-dealers.

keep

money

in

investors'

pockets;

order

routing,

and

settlement

deliver

a new

low

of error.
competition

proceed

is

done

affected
In

the
as

1970s'

follows:

out
has

the

the
has

squeezing

trading

institutions.

among

however,

broker-dealers,

approaches

often

key

technology,

quotes,

participants,

way

for

dark

later.

extreme

market

demand

and

better:

enjoy

transaction

Automated
degree

the

the

connections-driven

this

market

changed

low-cost

set

Plummeting
on

address

exchanges

services.

for

now

exclusive

more

to

have

markets

markets

venues,

matching

proliferation

for

up

innovations

markets,

The

trading

new
also

entrants

and

resulted

in

relative
marketplace,

over

time

power

of

the

trading

incumbent

reduced

technology-inefficient

changed

old

margins

players.
and

these

consumers
process

newer
and
would

1.

Brokers

clients.

would

The

deliver

ideas

were

based

on

time,

and

were

were

the

customer

broker
2.

If

and

when

assistant.

Such

the

on
to

smaller
similar

the

broker

the

idea

who

on

the

produced

in the

form

calls,

markets
in

to trade

buy-side
phone

to observe

the

verbal

orders

trading

floors

the

size

orders

market

order

to
to

in real

soft-dollar
idea,

the

of

often

trade

the

frequently

he

was

idea,

potentially

on

broker

resulted

impeded

to

the
or

and
higher

idea,

the

in errors:

correct

up

the

broker's
the

noise

understanding

of

for

appropriate

while

lotthe

Smaller

customers
of their

minimum

orders

for

other

order

size

thus

while

be

parcels),

waiting

were

would

would

smaller

desk,

round

steps

orders

in

broker's

execution

next

large

(potentially

the

broker's

often

the

at

favorable

away.

order

acceptable

on

the

order:

away

exchange.

slipped

the

placed

sit

fill

an

order,

right

waiting

price

Once

the

would

on

disadvantage,

of

market

orders

or

several

to the

broker,

sequential

orders

comprised

the

would

route

broker

the

the
order

order
to the

exchange.

Next,

human

specialists,

representatives

would

acknowledgments
specialists

at

rewarded
in

discrimination

match

back
often

connections,

resulted

in

decided

a customer's

the

executable

5.

countless

compensation

decided

customer

phone

receiving

depend

size

for

their

instructions.

After

taken

the

brokers'

customer

4.

pay

to

commissions.

would

3.

through

via

ability

required

customer

would

customer

on

so

ideas

disseminated

generally

do

trading

then-unique

the

to

often

brokers'

arrangementsif
expected

one-off

to

the

in

exclusive

Wall
in-group

the

order

broker.

exchange,
and

It is

well

preferential

terms

of

of

expense

investment

for

the
the

created

of

orders

connections
Street
versus

and

cliques

send

the
that

for

some

of

others.

Such

capable

networks,
of

significant

customers.

as
trade

understood

chummy

out-of-group

known

the
their

behavior
and
price
Even

though

exchanges

operated

peddling

was

anything

resembling

6.

The

common,

broker

and
an

and

power

of the

markets

Figure

1.3

the

client

the

were

were
field

of

bonuses.

and

organizations,

markets

playing

oversized

illustrates

not-for-profit

the

equal

notified

commissions

as

for

a long

execution

and

brokers

compensated

as

investing

way

away

from

all participants.

The

traditional

influence

collected

presided

his

over

the

kings.

process

prevalent

circa

1970s.
Figure

1.3

Broker-centric

Investing

Process

Prevalent

before

Electronization

Fast-forward
has

40-something

shifted.

analysis
Brokers'

Customers
and
area

all-encompassing
more

narrow,

are

years
have

often
of

sell-side
albeit

increased

better

expertise

still

ahead,

and
their

equipped
has

research
important

for

decreased
into
area

the

balance

expertise

in

research

than

in

scope

securities
of

of

algorithmic

power

quantitative
brokers.
from

behavior

the
to

execution

designed
With
the

to help

such

buy-side

process

Figure
Best

1.4

securities
quantitative
The

for

not

3. The

customer

places
and

speed

Selected

of the
algorithm

trading

evolves

waters.

according

to

Scenario

1:

Brokers

Provide

generate
their

an

research
existing

based

on

allocations,

forecasts

of

within

the

all

framework.
order

via

electronic
The

networks,
order

greatly

instantaneously

desktop.

or the

up

flow

misunderstandings.

broker's

designed

observability
4.

brokers,

management

the

market

intraday

Orders

portfolio

on

risk,

Clients'

and

arrives

choppy

Process,

movements

errors

algorithm

the

Investing

customer

reducing

the

1.4 :

Modern

Execution

navigate

investors,

in Figure

1. Customers,

2.

clients

to

broker

selects

the

minimize

the

customer's

execution

whenever

customer's

trading

electronically

broker's

optimal
execution

possible,

execution
costs

and

and

minimize

actions.
parcels

out

the

customer's

order

and

routes

the

order

slices

to

relevant

exchanges

and

other

trading

venues.
5.

Trading

venues

acknowledge
6.

broker

customer,

and

the

broker

lowest
and

Brokers

share

the

steps:

1.

broker

exchange

or

3.

profits.

slices

and

can

watch

and

grants

the

customer

with

the

specific

the

the

or

and

engage

to use
The

now
and
taking

in

direct

consists

to

have

of

the

access

per-volume

customer

broker's

keeping

Customers

privilege

exchange.

broker

markets

process

with

in

technology

time

to

per-trade

allow

use

said

execution

a negotiated

then

the

fee.
the

broker's

customer's

identification

To

order
in

order

exchange.

computer

low-frequency

systems

portfolio

or

human

allocation

analysts

decision

generate

that

involves

high-

one

uses
place

his

his

own

orders

order
directly

splitting
with

and
exchanges

routing
and

algorithms
other

to
trading

venues.
4.

One

or

or

orders.

Customer

optimally

are

their

systems

trading

route

reduction

parties,

behavior.

Merrill

Interactive

away

of

1997,

by

algorithms,

interested

and

with

do

costs

market

offered
Today,

to all

of

may

to

the

(In

100-fold

execution

markets,

broker

was

data

brokers,

to

commission.

Plummeting

back

trade.

prefer

own

of tick

for

per

and

their

forecasts

the

Customer

more

even

just

identification

messaging
2.

order

trades

trade,

further

directly

access,

routing

per

go

distribution

following

own

$70

broker-independent

to

grant

was

to clients.)

short-term

largely

retail

available

the

not

generate

The

customer's

on

$0.70

building

of

fast

access

lower

about

costs

customers,

the

considerably

his

commission

altogether,

enabled

acknowledgment

commission

customers

higher

order

receives

the

transaction

service

the

sends

charges

Some

the

execution.

The

Lynch,

match

several

exchanges

and

trading

venues

match

the

orders,

acknowledg

execution

5. The

broker

for

privilege

the

directly

receives

1.5

summarizes

Figure

1.5

Modern

Best

Access

Modern

Markets,

While

the

market-wide

broker-dealer
changes

side
to the

saved

watching

Clearly,
the

least

and

dollars

steps.

and

bunch

Scenario

have

disturbed

many
have

into

is happy
happens

client

identification.

2:

Clients

the

status

a broker
been

many

and

Decide

on

about
to

such
be

the
the

depositing

phone

The

on

of business,

Gone

taking

screens.

quo

positive,

pockets.

brokers

end

out

mostly

investor

computer

shareholders

everyone

happy

their

access

the

HFT

at large

on

charges

Directly

squeezed

of

and

direct

Process,

Markets

bonuses

to bank
not

these

directly

markets

redirected

broker's

changes

society

multimillion-dollar

information

the

Investing

Execution,

Media,

the

settlement

of using

Figure

to client.

are

the

orders

and

has

been

industry,

and

money

investors.
shifts

brokers

in the
losing

their

income

to

automation.

Stripped

investors'

pockets,

high-frequency

Some

and

technology,

charging

exorbitant

the

investor

poor

take

time

available

of

scare

lambs

compare

said

without

this

to

the
is

the

definitions

and

HFT

umbrella.

HFT

as

up

expense

chapter

HFT

catch

of

devoted

Evolution

to

and

who

analysis

in making

their

decisions

analysis

was

one

popular

with

loudly

many

sophisticated

Technical

analysts

identify

recurring

technical
moving
and

analysis
average

standard

against

traders

to today's

of
and

the

deviation

still

protecting
should
other

book

provide

educated

the

steps

hype

of

innovation,

to

selected

resorting

instead.

the

HFT

the

earliest

is,

in many

in security

measure
of the

outfits

to estimate

self-serving

The

that

to

remainder

evolutionary

current

price,
of

tend

of when

in vogue

patterns

sinister

versus

this

of strategies

econometrics
came

by

nature
fall

under

of
the

Methodology

Brokers

Technical

reduced

Investors

take

explaining

on

of

investors

clients

overview

of Trading

speak

their

allowed

their

guise

of

technical

those

more

a broker

15

and

on

to

the

low-frequency

on

about

even

predators.

and

relying

for

world.

an

through

12,

(2012),

not

in investors

trading

of
of

significantly

with

out

opponents

Saluzzi

were

under

HFT,

vocal

of the

HFT

money

eloquent

been

lure

5,

adverse

wax

cats

the

Chapters

at

and

costs

of

potentially

tactics

fat

to

from

costs

refuse

Arnuk

HFT

transaction

risk,

most

has

seeking

to enable

who

the

demonize

information

managing
brokers

when

lifestyle

are

options.

specific
risk

to

like

easy

the

investors

to

they

been

days
were

attempt

extract

yet

whose

in mind:

have

to

automation,

brokers

brokers,

ability

Brokers

error-prone

exchanges

of

brokers

denounce

manual

the

trading.

example,

goal

of

ways,

early

prices.

price.

HFT
1910s

Many
levels

For

technical
a

position.

that

a direct

other

in the

on

or exit

techniques

or a combination
the

rely

to enter

and

price

to

became
precursor

techniques.
and

sought

techniques

relative
of the
example,

to

used
the

rolling

moving

average

technical

to
in

analysis

technique

divergence

(MACD)

generate

trading

security

prices

market

conditions

moving

next.

in

century,

when

prospered

considerably

information
curtailed

the

and

allowed

The

previous

and

were

future

at

charts

post-WWII

it

the

is

to display

latent

of

the

decades,

when

technical

based

trading

that

analysis

morning's

transmit
hands,
prices,

of securities.
newspaper

successfully

published

infer

information.
began

developed

securities

into

next

to

and

changed

demand

technology

be

twentieth

to

and

on

may

major

inability

analysts

general

the

incorporated

supply
in the

or

at

telegraph
of

The

was

of

to

look

prices

its

of shares

technical

prices

half

in

today.

appeared
for

first

the

complexity

number

information

sufficient

considerably,

trading

than

trades

movement

the

would

events

where

was

which

day's
often

the

limited

pace

through

and

lower

quickly

of

technology

stages

market

idea

averages

analysts

current

a fuller

convergence

moving

technical

with

obtain

average

exponential

Advanced

trading

pneumatic-tube

moving

three

conjunction
to

analysis

as

uses

signals.

Technical

was

known

into

to

In

develop

self-fulfilling

prophecy.
If,

for

example,

pattern

would

instrument,

enough

all

be

followed

the

believers

head-and-shoulders
Subsequently,

modeling

technical

frequencies,
work
very

only
low

for

technical

bands,

have

quants

using

patterns.

in

such

as

the

smallest,

analysis
been

all

by

daily

that

steep

would
thus

indeed

investors

have

powerful

or

weekly

intervals,

least

liquid

securities,

adopted

frequencies.

as
and

It

has

following

to

high-frequency

or even

at

low

marginalized

which

are
per

momentum
extended
long

trading

analysis
is

prediction.

technology,

technical

such

particular

the

moved

day

orders

realizing

computer

per

in

sell

now,

techniques,

head-and-shoulders

sell-off

place

or twice

successfully

investing

By

frequenciesonce

Some

believed

pattern,
institutional

econometric
away

people

been

traded

at

week.
or

by

to

Bollinger

modern-day
shown

that

human

investors

recent

tend

months.

also

likely

that

can

As

indicators,

the

as

simple

bands

standard

deviations

away

arbitrage

extended

Bollinger

band

deviation

of

differences

from

trading
the
of

price

exercise,

and

of

standard

generally,

quants

processes

and

the

language

of statistics.

Another

important

fundamental
noticed

market

price

present

to

Graham

For

example,

implies

present

(1934)

were

their

cash

approach

trading

flows

were

of

the

identify

as

the

popular.
to

on

the

as

affected
back

of

the

to the

security.

purveyors

on

the

when

dividends,

exchange

based

in

1930s

of

the

the

years,

of

economic

of

that

known

pricing

expected

rates

expectation

Over

include

More

mean-reverting

value

earliest

price

prespecified

discounted

market

this

the

stationary

in

then

determination

valuation

traders

technique,

such

is still

based

In

changes.

or

equities

expanded

fundamental

equilibrium

flows

means.

the

the

stable,

flows,

fair

analysis

obvious

cash

cash

the

Dodd

fundamental
no

The

obtain

and

with

future

example,

to pinpoint
with

stay

in

by

difference

ideas

and

originated

levels.

methodology
term

investing

that

and

will

for

when

another

instruments

quantity

analysis,

traders

band

statistical

detect,

trading,

of

of

instruments

price

Similarly,
number

concept

to

pairs

of

are

momentum

tools.

long-running

as

month

prespecified

The

their

of

financial

average

the

principle

price

Bollinger

trade

measured

the

deviations
use

quantitative

financial

exceeds

past

in

moving-average-based

mean.

known

underpriced

instrument

number

the

worked

a tradable

technical

of prices

from

commonly

overpriced
one

deviation

that

in the

forming

complex

Bollinger

strategies

working

month,

more

detect

into

strategies

using

well

money

following

detected
as

pour

a result,

to work
be

to

the

securities
variables.

rates

today

macroeconomic

theories.
Fundamental
century.
expectation

analysis
Today,
that

developed

fundamental
the

prices

through
analysis

will

move

much

of

refers

to

to

level

the

the
trading

twentieth
on

predicted

the
by

supply-and-demand
theory.
still

In

relationships,

equities,

most

often
exchange,

models

specify

inflation,

microeconomic

determined

foreign

expected

variables.

advanced

properties

Various

price

of

high-frequency

the

event

news

the

announced,

price

event

the

instruments.
available

of

at

quants

recomputed

on

refine

up

fly,

an

news

of
earnings

occurrence

of

of

the

be

In
used

to

to

be

variable
process.
models
their

relevant

were

models,

and

values.

Fair

announcement

to

practicing

to

can

of

the

response

announcement.

fundamental
precision

For

content

high-frequency

quants

traders

the

economic

the

the

of the

analysis

the

calculation

enabling

of fundamental

the

some

extended

following

the

of

models,

sped

of

many

response

in

and

fundamental

to further

equities

time

value

who

time

in advance,

the

momentum

security

and

into

microstructure.

the

the

date

known

technical

dramatically

expense

statistical

inputs

market

of

The

arbitrage,

in order

by

values

other

matches

are

of trading

adjustment

fundamental

selected

often

about

fundamentally

underlying

alongside

consists

revealed

high-frequency

adopted

incorporate

analysis

models,

is typically

is usually

Like

fundamental

information.

event

forecast

that

and

In
the

and

traded

analyzes

flows.

information

are

of

are

prevalent;

countries,

models

trading

arbitrage

fundamental
news

different

movements

trading

accompanying
new

of

cash

most

using

prices

demand.

facets

example,

levels

economic

equity

of future
are

Derivatives

commodities

and

values

price

of

apply;

models

econometric

of

Fundamental

present

balances

macroeconomic

supply

as

fundamentals

models

macroeconomic

trade

through

the

reap

the

longhand

were

profits,

at

analysis

the

in Excel

spreadsheets.
Speed,

in

fact,

competition.
the

first

one

to

became

Whoever
to

identify

capture

the

the

was
and

able

trade

biggest

most
to run

on
gain.

obvious
a quant

a market
To

increase

aspect
model

of

the

quant

fastest

inefficiency

and

trading

speed,

was

was

the

traders

began

to

rely

decisions.
new

on

and

computer

executed

advances

buy-and-sell

technology

computer

by

processing

to
HFT

ensure

has

to

traders.

To

trade

into

algorithmic

The

the

assumed
advances

enabled
and

in

fully

Trading
expensive
along
accuracy

trader
with

other
of

these

for
trading

and

made

does

about

determine

market

current

to

by

how
to

market
mentioned

make

portfolio
sell

to the
of

were

made

human

to process

execute

price)

or

traditional

optimization

As

buy

were

refers

or

whether

usually
when

the

process

to the

of

decisions

packets.

not

order
trading

usually

is,

may

the

new

algorithms

strategies

conditions:

from

the

industry.

trading

close

of

to

rudimentary

trading

trading

response

all-inclusive

buy-and-sell

smaller

the

or

passively

price),

in

one

previously,
allocation

which

securities

exogenous.
computer

technology
HFT,

interest

desks

convenient

adoption

trading,

algorithmic

a price

automated

generating

to the

in

original

systematic

market

decisions

the

the

systematic

several

to be

to adapt

data

1990s

execution

removed

trading

decisions;
are

far

the

processthat

once

(on

price

or split

of

term

current

aggressively

trading

systematic

run-time

a billion-dollar

Algorithmic

given

a limit

into

the

the

manager.

order
(on

of

as

state-of-the-art

execution

part

order

From

established

decisions

another

and

execution

best

buy-and-sell

technology

evolved

day,

docking

processed

in

current

mimic
this

automated

portfolio

the

exchanges

known

developed

exchanges.

optimal

designed

an

the

execute

decisions.

trading
in

systems,

that

and

offer

became

systems

High-frequency

and

trading

make

enabled

culture

Computerized
the

to

progress

technology-driven

after

by

computers

Technological

trading.

To

fast

seized
head

in

execution

fueling

cost

count

with

and

the

pushing

upon

advanced

over

lack

of

past

profitability

the

technology

savings

realized

less

computer

the

expensive
technology.

hesitation

offered

decades

have

of trading

desks

even

further.

from
trading

replacing
algorithms

Immediacy
by

machines

and
as

compared

with

banks'

human

decisions

operations.

to switch

Lack

savings

due

particular

traders

to

in

Banks

developed

in turn,
the

been

encouraged

of

capital

following

products

based

traditional

Under

on

the

Dodd-Frank
trading

and

investors

costs,

adding

conditions

or

functionality

in

investors,
trading

daily

by

disclosure

to

found

that

investment

have

little

correlation

trading

strategies,

function,

where

is often

referred

it is now

pure

return,

or

alpha,

care

Traditional

and
1.6

are
be

Long-Term

client

to

close

many

HFT.

In

certain

well

in

the

and

bound
taken,

rather

versus

further
to

take

on

however,

trading,

schematic

traditional

HFT

alive
with

algorithmic
a

not

develops

should

trading,

illustrates

Figure

run

forced

of

banks,

the
the

market-making

than

bank

capital

and

drops

and

prehedging.

systems

Special

were

but
is

technology

high-frequency

systematic,

banks

HFT

to as

computer

electronic

Act,
operations,

prop-trading

1.6

carry

high-frequency

lock-ups

retail

immediate

Institutional

practice

in

to systematic

lending

investors.

role

into

position

intraday

buy-and-hold

formerly

role.

quantitative

trading

high-frequency

shorter
and

significant

translated

tight

to

portfolios.

proprietary

As

has

overnight

buy-side

institutional

played

traditional

adopted

have

Both

to their

and
from

investors.

with

in

demand

influx

from

crisis-driven

environments.

to

also

positions

reduction

high-interest

response

away

of overnight

issue

also

has

and

Algorithmic

Investing

even

to distinguish

more

price,
active

HFT

from

systematic

trading.

between

high-frequency,

difference
long-term

an

in

investing

Figure

styles.

(Systematic)

Trading

and

Systematic
may

be

not

be

trading
held

daily

a month

high

computer

program

buy-and-sell
Another

in

often

to
well

of

order

sequence

is

in

but

than

and

system.

not

synonymous

refers

to

financial
term

places

the

to telephone,

today's

complex

is
accepts

a high-frequency

opposed

the

algorithms

ability
mail,

or

markets

electronic

are

are

trading

designed
have

decisions

not

of sizes

best

about
point

required
in which

to

been

in time
be
the

allocation

to optimize

the

is

best
to

and

execute

should

decisions.

to

route

be

optimally

once

Algorithmic
the

a submitted

immediately,

The

execution

elsewhere.
way

can

processes

trading

made

executed
order

trading

order-execution

portfolio

decisions

the

electronic

spanning

high-frequency

makes

the

as

or may

monthly;

matrices;

trading

that

trading

even

conjunction

networks,

is more

algorithms

exchange,

or

Electronic

may

systematic

is not

orders

therefore

allocation

in

decisions

obsolete.

a variety

the

of

portfolio

electronically

computer

and

weekly,

a system

most

buy-and-sell

execution

daily,

trading.

trading

as

example

mentioned

orders

becoming

execution
the

Such

via

Algorithmic

as

orders.

Since

transmitted

refer

runs

outputs

the

person.

rapidly

that

trading

or a minute

An

is electronic

to transmit

computer-driven

or a day

prices;

term

HFT

to

frequency.

closing

with

refers

and

order

to

order
the

if
best

processed.

Algorithms
and

generating

decisions

example,
to

1 million

100-share
the

of

is

generate

algorithm,

the

high-frequency

decision

that

would

in a particular

best

to

generate
the

1 million

would

then

be

passed

determine

the

optimal

or

The
not

shares

timing

be

IBM.

the

and

high

however,

of

to

of

decision

signals,

on

order

increments

may

HFT

buy

For

a received

price.

may

decisions
security.

using

in the

however,
to

that

handled

run-up

deployed

signals

algorithm

IBM

allocation

determine

a sudden

algorithm

portfolio

a position
may

to prevent

An

make

execution

execution

frequency.

signals

or close

shares

lots

to

would

to enter

algorithmic

buy

fed

HFT

The

execution

routing

of

the

order.
Successful
those

implementation

generating

trading

and

for

those

for

(Chapter

managing

The

intent

survey
full

generation

of

order

14

buy-side

and

algorithms

as

1.7

Reasons

Automated

of
those

through

11)

information

includes

illustrated
in

within

latest

the

trading

by

2012.

survey.
to

Using

Trader

Survey,

reported

decision

and

2012

Figure
In

algorithms

of

1.7

shows
to

portfolio

the

algorithmic

reporting,
their

in Trading

results

addition
of

and

also

Algorithms

the

adoption

management

managers
by

is

1
Source:

execution

to conceal

related

performance

for

algorithms:

of algorithms:

needs.
Figure

of

(Chapters

also

from

factors

driven

groups

Trader

responses

sell-side

to be

types

optimizing

signals

book

Automated

mentioned
such

both

execution

by
of

those

designed
the

both

operations.

algorithmic

spectrum

and

trading

of

of HFT

requires

covers

execution

conducted

previously
trading,

book

risk
of

signals

This

15).Chapter

for

the

HFT

decisions.

designed

of HFT

use

both
of

management

the

True
of

HFT

systems

underpriced

allocation
is

the

execution

examples

and

algorithmic

markets
those

of

financial
are

research

Figure

the

shows,

lag

of

Figure

are

have

been

traded
Adoption

OTC

are

and

as

over

handled

by

systematic

can

As
be

difficult

them,

to synchronize
by

Asset

some
in

According

to

the

most

of the
by

futures.

explained

in

trading

total

2010.

As

Advances

options,

for

Execution

most

percent

illustrated

HFT

3,

are

by

exchange,

visible.

for

algorithms

followed

are

frequency.

(OTC).

50

rapid

and

Chapter

equities

development

of Algorithmic

in

with

foreign

are

their

All
high

HFT

duration,

systematic

not

counter

closely

less

of

trading.

Group,

instruments
trading

fully

inasmuch

the

class,

in

is a prerequisite

HFT,

to be

of

of electronic

1.8

asset

expected

fixed-income

are

Aite

execution

however,

tardiness
of them

by

equities

are

discussed

over

portfolio
of

shorter

Because

however,

for

optimal

or

algorithmic

As

performed

of equities

day

algorithmically

suitable

executed

1.8

systems

identification

characteristic

overnight.

platforms,

conducted

through

one

and

order

yet
is

algorithmic

income,

HFT

from

distinguishing

held

instrument.

not

algorithmically

in

an

of decisions,

times,

systematic

trading

markets

volume

holding

most

to execute

given

The

positions

nature,

range

securities

execution.

no

also

a full

overpriced

position

with

Ability

or

to best
short

usually

make

and
in Figure

by

the

given
as
Class

that

fixed
1.8
relative
many

a result.

Source:

Aite

While

Group

research

relative

to

dedicated

data

evidence

on

naturally
and

itself

precision

volumes

of

of tick

outperform
for

by

designed
performance

from

that
returns

lack

of

the
in

more

than

different
eye

is

seamless

funds
crisis.

inherent

in

gathering
news.
50

data

systematic

points

per

to

most

to

only

to

the
second,

moviegoers.

humans

returns

abstracting
(2009b)

also

funds

attributed

to

in
the

strategies

in such

analysis
human
as

In
24

be

is

as

traders.

industrycinematography.
exposed

of

by

trading

lightning-fast

Physiologically,

Aldridge

nonsystematic
can

to

consistently

Aldridge

finding

superior

and

shown

metric

trading

outperform

of

performance

influences.

human

are

funds

1968),
of

been

metrics.

when

skill

That

emotion-driven

information
and

unique

systematic

computers

key

speed

analysis

has

systematic

(Jensen,

market

of

in turn,

trading
high

high-frequency

operations

alpha
the

times

with

Furthermore,

quotes

trading

are

algorithmic

several

that

broad

emotion

compared

along

shows

measure

as

scarce

anecdotal

demanding

trading,

trading

Jensen's

to

well

is

strategies

and

applications
as

HFT

computer-driven

Systematic

traditional

measured

as

trading

example,

outperform

raw

most

execution,

data.

of

strategies,

Systematic

to

human-led

(2009b),

shows

that

performance

buy-and-hold

strategies.

lends

the

long-term

suggests

high-frequency

to

eye

per

Even

then,

cannot
by

movies,
second,
the

tasks

a multitude

evidenced

modern

frames

of

basic

capture
an

entirely

the

human

which

majority

appear
of

images

of

displayed
In

on

sequential

comparison,

bouncing

modern

quotes,

second

for

just

information
and

processing

the

financial

of information

humans

have

High-frequency

have
This

is

Commodity

that

includes
the

Where

efficient
trading,

succeeding.

HFT

HFT,

opinions

all

of

activity

of HFT,

several

various
what

fast

algorithmic

Subcommittee

of

(CFTC),
back

actually

of HFT:

utilizing

came

market

HFT

definitions

Commission

is a form

decision

of automated

making,

for

comprising

the

tasked

with

U.S.
with

the

following

that

employs:

2012:

trading

execution,

per

assets

to profitable
of

of

Technology

definition

in June

for

multiple

15.

term

common

Trading

a working

Algorithms

for

is key

breadth

divergent

example,

High-frequency

1,000

inter-instrument

data

umbrella

discusses

Futures

definition

Detecting

chance

drastically

exceed

Chapter

little

the

somewhat
section

For

an

Given

of HFT

execution.

in

objects.

Trading?

strategies.

A definition

or

discussed

trading

participants

easily

processing

volumes

Is High-Frequency

draft

instrument.

involves

moving

incorporates

can

of high

What

compiling

which

as

over.

for.

of

classes,

takes

of

continuously

information

number

one

technology-averse

groups

involve

financial

spillovers

asset

stands

frames

each

order

trading

initiation,

individual

generation,

transaction

routing,

without

human

direction;
low-latency
including

technology
proximity

high

speed

high

message

Such

execute
entrusting
algorithmic

and

rates

95

(orders,

captures

his

orders.
broker

manner

for

quotes
many

of

to

response

times,

entry;

traders,

algorithmic

order

high-frequency

retail
in

yet

also

technology

mom-and-pop
his

and

or cancellations).

using

execute

becomes

order

high-frequency

investors
Even

to minimize

services;

to markets

percent

their

is designed

co-location

connections

definition

includes

that

the

most

trader

to

investor
efficient
under

the

definition

proposed

surprisingly,

this

members
1.9

the

definition

of the

Figure

by

CFTC's
generated

subcommittee

HFT

vs.

subcommittee
strong

on
dissent

HFT.

Not

from

many

itself.

Algorithmic

Trading

and

Quant

Portfolio

Management
Source:

The

Gomber,

definition

trading.
HFT

of

Gomber,
as

traders

on

as

in Figure

also

definition

or

A survey

and

of HFT

of hedge-fund

generated

High-frequency

the
trading

Systematic,
Quant-based

models

and

(2011)

this

(2011)

such

drive
definition,

execute
on

managers,
following
comprises

models

in

the

HFTs

holding

generate

in any

one

period

conducted

definition

of HFT:

algorithmic
define

are

to

do

algo

speed

supersonic
not

up
time.

engage

trading

in

signals,

security.

of capital
by

to

HFTs

technology

but

trades

of

proposed

definition,

super-fast

management,

based

subset

Uhle

1.9 . Under

and

construction

Uhle

a latency-sensitive

utilizing

under

models,

and

Lutat,

processes

validating

2009,

HFT

steroids,

Interestingly,
portfolio

Lutat

Arndt,

shown

algorithmic

The

Arndt,

throughput.

FINalternatives

in

With

holding

positions
The

periods

held

survey

based

who

were

prominent

launched

period

retracted.

The

definition

the

as

while

sent

second

to

of

HFT

one

is

also

low

HFT

Intermediaries,

an

day

(no

on

to

noting

that

that
The

characterized

by

HFT.

the
fund

holding

The

fund

market

high

was

activity.

high-frequency
market

researchers

market

at

submicrosecond

identify

generate

10,000

position

observed

(2011)

fund

hedge

from
with

their

Tuzun

other

hedge

(close

average

departure

inventory.

from

worth

in connection

participants

holding

with

far

and

from

Greenwich-based

based

Samadi,

market

to distinguish

It

fund
a

responses

FINalternatives

out).

mentioned

Kyle,

traders

300

to

days,

often

later

of

multibillion-dollar

three

frequencies

fraction

nearly

high-frequency

of

Kirilenko,

on

subscribe

questionnaires
time,

overnight).
was

managers

from

use

volume,
the

all

definition

participants:

low

inventory,

but

not

high

trading

volume.
Fundamental

buyers,

who

are

consistent

net

buyers

intraday.

Fundamental

sellers,

who

are

consistent

net

sellers

within

given

day.
Small

traders,

generating

Opportunistic

traders,

fitting

the

Such

definition

inventory
The

of

their

HFT

of

feasible

generating
an

200
account

or other
on

as

all

other

categories

somewhat

based
A

clients

attribution

rely

defined

traders,

not

above.

arbitrary

cutoffs

of

low

volume.

participants.

human

would

of HFT

high

volume.

loosely

may

definition

segment
for

definition

and

market

low

orders
that

behavior

common

into

trading
and

on

HFT

definition
and

activity

human
per

second

used

non-HFT,
of

infeasible.

consistently

unattainable

each
For
would

succeeds

this

by
by

brokers

definition

specific

deemed

at locking

an

to
calls

account

example,
be

human

into
account

HFT,

as

in a penny

gain

day-in

This

book

While

most

concrete

deployed

by

Do

Despite
market
four

all of the

definition

participants

1.10

What

day-out.

considers

market

Figure

and

HFT,

are

broad

of HFT

Traders

disagreements

participants

has

agree

proven
with

in Figure

Categories

High-Frequency

HFT

discussed

comfortable

summarized

Major

the

of

definitions

to
the

be

range

challenge,

of strategies

Strategies

Do?

about
that

1.10

here.

the
HFT

precise

definition

strategies

fall

of HFT,

into

the

most

following

classes:

1. Arbitrage
2. Directional

event-based

3. Automated

market

4. Liquidity
Arbitrage
equilibria
classes,

trading

making

detection
strategies
or
as

relative
well

as

trade
asset
multiple

away

price

mispricing,
exchanges.

deviations
and

can

Many

from
include
HF

long-running
multiple

arbitrage

asset

strategies

detect

price

Chapter
trading

discrepancies

8.

Several

on

different

strategies,

and

strategies

multiple

strategies

on

as

prices

are

discussed

based

securities,

arbitrage

exchanges,
are

are

in

of

known

in

assumptions

of

the

as

Chapter

discussed
same

in
asset

latency

arbitrage

Most

arbitrage

12.

mean-reversion

of

asset

prices.
Statistical

arbitrage

cross-asset

models,

relationships.

All

economic

the

models

of

plain

against

the

(if

wall

however,

may

fall

in production).

apart

have

been

values,
or

to

a result,

interest
for
via
to

the

other

Chapter

techniques

Directional

strategies

of

in

short-term

price

controversial
arbitrage

news

Chapter

ignition
models

on
used

show

predictable
in

event

filled

stick;

what
and

interest

When

rate

and

selling

of

as

their
bonds
price

may

be

price

instrument

other

models,

futures

trading

a lower-than-expected

the

to

and

with

to

sticks,

long-running

the

with

is likely

prices

arbitrage

trend

includes

other

with

instrument(s).

well

as

the

model

in Chapter

methodology

are

momentum.

in
as

effects
showcased

on
11,

as

Chapter

well

as

of news.
in

This

strategies,

based

discussed

recurrent

or

event-driven

strategies

discussed

arbitrage

Spaghetti

results.

strategies,

and

the

sticking

in
often

spaghetti

and

statistical

strategies

the

deep-rooted

as

their

short-term

movements,

statistical

dependencies,

from

relative

9,

for

deviate

identify

strong

relationships

may

in tandem.

known

including

are

of

bonds

economic

and

plate

move

instrument

models,

known

considerable

reason,

high-frequency

discussed

trading

example,

many

estimation

class

For

price

details

reason

instrument(s),

higher-than-expected

also

something

sound

obvious

buying

throws

any

to

book

statistical

and

statistics,

futures

no

the

mining

possess

of
have

in the

spurious

one

tend

rate

equilibrium
feasible

of

have

shown

as

relative

not

out

range

securities

included

data

Modeling

financial

ruling

using

Principle

comprise

where

theories,

developed

data

models

predictable
well

as

12.

the

Event

performance

of

Various

of

Chapter

types
9,

which

also

includes

specific

references

practical

Automated

broker-dealers,

strategies,

inventory-driven

limits

appropriate.
goal

an

opposite

markets,

the

Perhaps

least

detection

stuffing,
on

controversial
selected
other

venues

to

strategies

have

the

as

no

date.
been

(also

sound

Still,

Pinging

has

been

the

in dark

strategies
of strategy

hypothetical
for

the

deploy

completeness,

a
of

players

in

lumps

the
and

techniques

allow
of their

of liquidity.

sniffing

strategies,

detected

of taking

can

Chapter

the

evidence

included

in

include

ignition

credible

as

with

minimize

investors

and

was

risk

of aggressiveness

known

HFT

built

directionality

or dearth

addressed

well.

like

levels

risk
where

are

To

and

trader's

the

These

low-frequency

pinging

perspective

strategies

produced

to

(pinging

and

and

risk

impending
11.

of
both

hedged

market

forecast

of surplus

explaining

HFT

speculative,

quantities

short-term

in Chapter

category

traders

well-informed

help

spoofing,

a balanced

within

party.

forecast

human

inventory

selection,

that

of

market

Inventory-driven

models

high-frequency

like

and

focuses

draw

as

most

includes

and

better-informed

expectations

strategies,

are

conditions,

help

the

The

of the

positions,

palatable

10.

approaches.

of adverse

covered

the

on

even

of liquidity,

based

well

to

provision

positions

risk

number

and

to choose

Chapter

market-making

to

models

waters,

shortages

to

of

in

minimization

the

the

losing

perhaps

alternative

liquidity

market

position

track

market

book

that

of minimizing

range

quote

joint

given

of such

orders

detail

Information-driven

the

as

automated

accurate

and

on

ensuring

tolerance

comprise

information-driven

to focus

risk,

number

in

making

and

tend

market

studies

encompassing
and

discussed
market

methods

relevant

strategies

cost-effective

automated

latest

examples.

trading

making,

traders

the

market-making

traditional

wide

to

are

liquidity

and

sniping),

12.
the

While

book

this

attempts

methodology
shown

behind
to

pools).

The

have

been

existence

strategies

exist

on

nature

of

mostly
has

like

accompanied

been

ignition
by

brief

analysis

broader

of

Many

The

number

High-Frequency
of

definition

of

definition

proposed

all

Samadi,

HFT

account
(S&P)

just

and

in

that

SPDR

than

Major

Players

Many

HFT

25

20

very

still
like

IV

line

major

easy

to

HFT

on

the

on

the

human

are

most

of

annual

Organization

their

the

&

Poor's

that

HFTs
futures

S&P

average

500

represent

Book

in

of

HFT,

and
on

outfits

Lesser-known

include

trading,

specialist

however,

can

can

specifically
trading

such

and

for

in

all
are

robots

are

than

their

blend

example,

arbitrage

desks

reliable

seamlessly

be

automated

developments

market-making

In Canada,
statistical

HFT

while

HFT

Shaw.

most
of

inexpensive,

the

all the

WorldQuant.

forms

HFT

limelight,

DE
to

automated

arbitrage.

reports.
of This

HFTs

the

well-known

and

advantages

Likewise,

HFT

of the

most

staples

accurate,

of statistical

and

other

earlier,

new

more

that

exchange

fund,

dedicated

Fusion,

becoming

the

Kyle,
produce

find

estimates

on

percent

who

Standard

foreign

out

Capital,

and

counterparts.

activities

(2012a)

The

players

And

the

95

Kirilenko,

and

traders

to stay

profits.

HFT

or

draft

participants.

prefer

DKR

banks.

considerably

in

CFTC

Space

mentioned

see:

volume

high-frequency

between

making,

inventory,

exchange-traded

Capital,

market

traders

liquid

profitable

As

as

most

Renaissance

blurred.

HFTs

Aldridge

of market

20,

HFT.

in EUR/USD

percent

every

the

as

little

of

depends

under

of

percent

considerable

but

banks'

impact

qualify

define

percent

participants

Getco,

out

to 30

in the

include

19

holding

largely

earlier,

would

markets.

the

generating

list

30

(NYSE:SPY),

fewer

2012,

while

E-Mini

comprise

the

and

There?

mentioned

(2011),

about

500

Are
traders

traders

volume

for

As

in June

Tuzun

trading

Traders

used.

and

and

large

The

properties,

high-frequency

investors

firms

feasibility,

markets.

How

of

their

in

banks

category

with
often

in

the

This

book

applied

is written
and

that

high-frequency

are

high-frequency

has

markets.

enabled

through

lay

and

7 delve

Chapter

6 describes

capacity

of HFT

casual

explain

the

data,
the

and

and

enhancing

their

institutions

seeking

activity

against

observers,

seeking

to

present-day

frontiers

in

technological

analysis

evolution

trading.
via

and

of

Chapter

modern

costs.

of high-frequency

for

that

Chapters

description

trading

economics

methodologies

strategies,

yet

markets.

of

into

and

latest,

management

or

high-frequency

foundation

the

trading

describes

high-frequency
6 and

as

book
2

to

starting

clients'

well

5 of the
Chapter

the

providing

individuals

financial

algorithmic

microstructure,
Chapters

as

modern

2 through

in

their

traders,

of

information

and

understand

financial

goal

operations,

their

Chapters

explicit

interested

trading

protect

better

the

ready-to-implement

employees

to

with

trading.

evaluating

performance

7 outlines

the

and

business

case

of HFT.
Chapters

through

implementation
of

of

today's

in

HFT.

and

Chapters
HFT

of

and

externalities.

regulatory
market

thought

focus
Chapter

on

HFT,

manipulation,

market-wide

events

solutions

for

HFT

modern

slicing

HFT
HFT

15

on

information-prying

as
like

flash

low-frequency

techniques

16

markets.
and
HFTs,

are

8 through

strategies.

monitoring

13

through

Chapters

12

15

devoted
dissect

Chapter

of

Chapters

implementation
and

14

management

construction.

running

and

high-frequency

measurement
portfolio

12

14

to

actual

core

focuses

models
on

high-frequency

trading

and

the

nuts-and-bolts

as

best

practices

in

HFT

and

mitigation

of

systems,

16
as

discuss
well

as

risk

well

as

systems.
on

regulation

13

of

presents

summary

discusses

models

well

mathematics

as

crashes.
traders
Chapter
their
and

Chapter

for

15

discusses

respective
may

also

detection
of

15

concerned

of

book

the
the

ability
prove

of

HFT

foreseeing

of the

about

current

impact

latest
to
useful

offers
of
order
avoid
to

high-frequency
trading

traders

seeking

to

further

expand

capacity

of

their

systems.

Summary
High-frequency
The

trading

technological

replace

bank

broker

are

trading

beneficial

End-of-Chapter

markets

created

function

compensation

technology.

with
to

ability

to

cost-effective

end

investors

and

do

2.

What

the

they

play?

are

well

defined,

and

most

of

markets.

groups

How

are

the

do

of

they

core

today's

market

participants.

What

interact?

groups

of

strategies

deployed

by

traders?

do

high-frequency

strategies,
quant

to the

major

high-frequency
How

strategies

Questions

Describe

role

such

as

trading

technical

strategies

analysis,

relate

to

fundamental

other

trading

analysis,

and

strategies?
are

the

major

markets

over

the

past

5. What

is algorithmic

What

6. How

do

1 Most

exchanges

time

of

their

very

much

laid

financial

of trading

shareholders.

them

4.

of

evolution

intermediation

returning

High-frequency

3.

organic

evolution

human-driven

technology,

1.

is an

end

today's

of

benefit

profit.

In fact,

the

standards.

the
no

New
broker

transaction

from

occurred

in

high-frequency

not-for-profit

to

rules:

have

the

financial

years?

became

to

that

trading?

investors

foundation

percent

40

formation
for

profitability

changes

the

1970s,
the

York
was

only

Stock

the

1970s.

to

charge

staggering

From

exchanges

agreement
Exchange,

allowed

volume,

in the

however,

Buttonwood

trading?

of 1792

specified
less

the
were
that

explicit
than

commission

0.25
by

CHAPTER

Technological

Innovations,

Technological

innovation

leaves

financial

markets.

operations

of

financial

instruments,

one-time

disruptions

and

subtle

has

of

analysis,

of technology

over

past

financial

History

Trading

was

(benches

or

Gradual

of

communication.
was

greatly

have

the

years,

the

quality
adoption

aided

by

This

occurred

context

have

the

The

improvements.

that
in

Over

is disseminated,

however,

decades

large-scale

changes

markets.

technological

institutionalized
in

change

of

illustrates,

over

available

to

computing

guided

with
the

traders
has

in

were

revolution
trading

during

currency

bancas

technological

its

speed

new

in

of

the

chapter

technology

enabling

modern

of Hardware

exchange

state

news

the

landscape.

A Brief

first

the

on
of

created

technological

way

developments

mark

introduction

in 1999,

on

services,

several

persistent
the

routines,

the

of

most

EUR/USD

the

costs

HFT

While

impact

and

key

and
the

in market

in financial

the

the

as

improved

ever-plummeting
examines

such

continuous

technology
financial

Systems,

been

of
rapid

the

precursors

operations

of

twentieth

century

past

100

years

has

increased

of
or

so

since

of

the

be

today's

noted

until

enabled

the
As

1980s,

the

current

Figure

computational
while

the

banks).

firms

information.
the

when

can

trading

exponentially,

steadily

Empire,

locations

direct

exchanges

falling

Roman

designated

the
the

the

2.1
speed

the
after

cost

of

reaching

peak.

Figure
Twentieth

2.1

Evolution
Century

of

Speed

and

Costs

of

Technology

over

the

The
the

price

decline

past

(RAM),

20

years.

300

processor
its

found

for

own

as

computer
The

demand

rather

of

players,
the

surge

technology.
and

cost

industry.

fits

little

as

$400

with

drive

dollars
In

only

for

the

cost

scale
the

any

spectacular

2 gigabytes

in 1995,

2012,

in

been

space,

in a standard

by

in

and
and

was

memory

2-gigahertz
big

enough

with

case,

neighborhood

over

of
a

computer
desktop

more

whose
in

mass

Financial
efficiencies

can

of

to

identical

it can

Best

not

by

casual
wallets.

scale

created

of
the

technology
and

desire
and

have
by

largely

also

Buy

or

be
other

cheaper

reaped
the

past

two

activity

has,

technology

be

of

the

latter

video

game

has

prices

benefits

with
the

graphics

plummeting
the

technology
practitioners,

to

lifelike

the
The

decades,

happened

to

overseas.

services

computer

for

traced

chips

financial

users

production
firms

and

the

Over

be

computer

accessible

cost-efficient
sheer

computing

production

driven

thinner
for

of

increasingly

been

considerably
demanders

million

not

in

surprisingly,
but

system
hard

room.

has

store.

decline

efficiency

of

several

specifications

technology

A computer

gigabytes

cost

require

in computer

of
video

fueled
of

fast

innovation
gaming

As

shown

in

multicore

central

arrays

and
use

a shared

given

written,
Figure
Source:

Unlike

schedules
branch

is

CPU

Architecture

Thomas,

where

by

memory

performed

is

in

maximize
parallel

batches

warps.

To

ensure

that

the

Figure
cost

of

of

the

the

and

of

to

chip,

latency,

of

the

Multicore
while
on

architecture
time

this

of
book

a
was

higher.
CPU

(2009)

the

scheduler

size.

brain

computations

Sample

arithmetic

each

so-called

communication,

Multicore

devoted

of identical

threads

Luk

the

the

performs

$100

gate

in memory.

2.2 . At

majority

and

minimize
the

and

so-called

efficiency

and

and
is

inter-CPU

of a Sample

largely

the

fast

thread.

could

Howes

CPUs,

GPU

in

CPU

information

comprise

programmable

(GPUs),

tasks

or

shown

a multicore

occupied
sample

CPU

CPU

2.2

to store
for

field

units

chips.

memory

process

multicore

how

technologies

(CPUs),

processing

architecture

decides

individual

advanced

units

graphics

parallel

computer
CPUs

2.1 , today's

processing

(FPGAs),

massively

each

Figure

space

functions,
the

the

the

units

process
These

batches

however,

care
are

To

are

on

should

be
in

further

executed

of threads

similar

is

operations,

(ALUs).

threads

chip

space

computational

logic

process

on

terms

are

in
called

taken

to

of

the

number

of

loops

and

expertise

is

maximum

efficiency.

GPU.

retail

for

Figure

required

popular
$100

2.3

conditional
to

of

to $700

per

Architecture

Source:

Thomas,

FPGAs

are

entirely

instruction

set

slate

of

functional

any

desired

dedicated
Most

circuit

the

area

infrastructure
units.

Figure

Figure

2.4

Source:

of

2.4

an

Architecture

Thomas,

and

Luk

Howes

GPUs

programming

are

sample

Nvidia

deployed

with

architecture

GTX

series,

of

the

which

can

not

have

any

provides

(2009)

of

chips

Instead,

an

that

be

can

Some

such

FPGA,

GPU

class

units

run-time
shows

words,

card.

units,

the

is

processor.

functional

of

a GPU

architecture.

or

other

illustrates

different

fixed

bitwise

2.3

In

that

of a Sample

Howes

an

ensure

Figure
model

exits.

as

architecture

of a Sample
and

Luk

programmed

is
of

and
dedicated
the

of a sample
FPGA

(2009)

Chip

blank

to

contain

multipliers

however,

do

FPGA

FPGAs

connectivity

the

that

create

a number

memory

blocks.

to

routing

FPGA's
FPGA

the

functional
chip.

of

The

main

written

distinction
directly

programmed
The

are

to

easy
or

chips.

In

the

languages
learn.

VHDL

compiled

and

translated

program

run
into

rather

time,

the

VHDL

variation

in

compilers

commonly
gamers.

and

available,
A blank

Verilog

The

process

and

inexpensive.
of

again
chip

and

produced

simulators

cost

may

be

free

to

CPUs

be

during

an

and

FPGA

is

there

exists

and

Verilog

or

inexpensive

anywhere

FPGA

operations

chips

to satisfy

and

the

need

like

While

quality

Verilog

translates

by

programming

FPGA

as

device

chips

are

language

computer
of

blank

may

such

programs

computer

simulators,

FPGA

trading

is

FPGAs

understood

additional

costs

once

software

the

outset.

programming

language

requiring

latency.

the

code

to C programming

FPGAs,
to

straightforward

significant

The

of

programming

languages,

FPGA

Assembly

absence

the

from

similar

special

into

that

programming
are

is

chip

special

the

eating

FPGAs

onto
using

or VHDL.

Verilog

of

options
demand

from

$4

are

of video
to

(open-source)

$5,000.
or

$20,000.

The

process

software

specific

chips

is

to

often

programs,
state-of-the-art

known

demanding

the

then

the

chip,

Programming

in

undergraduate

electrical

be

easy

However,

to

FPGA

system

to

may

massively

advanced

onto

manufacturer.

tends

as

downloaded

chip

taught

and

format

is

learn.

require

parallel

understanding

of

of

the
FPGA

engineering
achieving

arranging

processor

using

FPGAs

array

in

configuration,

hardware

and

software

optimization.
Performance-wise,
particularly

when

time

series.

key

hardware

input
fed

into

results

becomes

Quad

Source:

The

number

of

the

inputs

Thomas,

be

superior

As

exceeds

vertical

axis

of inputs

figure

the

of
the

number
involving

an

than

of

simultaneously

operation

is less

number

shows

the

illustrates,

threshold,

FPGA

2,000.

speed

When

of

of
the
posts
the

an

FPGA

Single

CPU,

of a GPU.

Performance

and

the

CPUs,

of efficiency

shows

an

2.5

of

Architectures
Howes

of

and

a limited

variables

perform

this

to that

axis

independent

Figure

GPUs

comparison

horizontal

to

to

process

a graphical

of

number

Comparative
CPU

to

The

inputs.

comparable
2.5

shows

required

when

of

2.5

algorithm.

number

Figure

the

cycles

number

and

or

tend

to simultaneously

models.

the

computer

best

used

Figure

size,

given

FPGAs

Luk

(2009)

FPGA,

GPU,

The

choice

of

the

of a chip

computer

determined
and,

by

most

itself

is not

program.

The

the

so-called

important,

the

single

speed

of

oscillator

organization

determinant
each

crystal

of the

of the

computer
within

program's

speed

cycle

each

is

machine

algorithm.

Messaging
Hardware

is

necessary

is

software

in

Figure

the

successful

messaging,

enabling
of

of

various

that

trading.

market

presents

crucial

among

participants.

important

technology

Another

communication

it is also

of messaging

computer

hardware

Just

as

in messaging.
a hurdle

or

speed

In fact,

a bottleneck

it
for

Protocols
messaging
2.6 . The

streaming
is

components

communication.

Messaging
Trading

many

achieving

in hardware,

speed

trading

of

modules

is important
is the

one

for

component
and

just

and
bare

implementation,

is comprised
most

is known
bones
and

basic
as

the

data
utilizing

of

three

levels

level

of

User

Datagram

protocols,

communication

communication
the

of

fewest

Protocol
protocol,
number

enables

data

(UDP).

UDP

lean
of

shown

bytes

in

its
and

messages
UDP
is

to

is very

fast,

the

same

movies

over

does
is

not

identify

the

Figure
Used

The

as

and

the

the
2.6

is

trader's
Three

used
of

therefore,

impact

already

on

from

may
the

packet

the

data

be

way,

games

and

and

there

In trading,

UDP

are

exchange

the

result,
UDP

refreshed

sensitive

deemed

retiring

a
data.

here

that

very
an

As

stream

experience.

not

sent

data.
of sent

to

one

quotes,

quote

of

delivery

viewer's

transmit

resulting

quote

one
loss

the

are,

stream

guarantee
the

impact
to

the

where

If a particular

trader,

not

Internet,

used

information.

hitting

does

significantly

sometimes

revised

but

deliver

technology

continuously,

and

to
fails

to reach

minimal:
lost

lost

quote

new
upon

account.
Levels

of

Complexity

of

Communication

Protocols

in Trading

integrity

of

the

model

development.

quote

stream

quote
A

process,

trading

idiosyncrasies

however,

algorithm
to

generate

can

developer
predictive

matter
may

in
rely

signals

trading
on

the
about

impending

market

stream
the

used

in

quote

of

received

in

designer

should

production

to

the

the
for

The

level

most

all

numbered,

the

used

a
to

packets
total

more

and
Figure

UDP.
FIX

the

Figure

Common

resulting

metrics

of

data

for

not

presently

carry

each

packet

As

information

is

used
the

any
are

is

identifying
sequentially
is

result,
delivery,

counted,
TCP/IP
and
execution

cancellations,

and

important

tends
common

to

similarly
be

three

usage

of

times
UDP,

communication.
of Protocols

in Trading

in

UDP,

acknowledgments,

summarizes

Use

be

order

TCP/IP

of

TCP/IP

Unlike

re-sent.

sets

should

(TCP/IP).

protocol,

within

is

two

protocols

transmission

bytes

back-test.

implementation.

Protocol

TCP/IP

in

example,

the

communication

do

data

received

For
on

and

algorithm

in the

computed

in

orders,

2.7

quotes

and

Care

the

the

used

of

production

to

performed.

design

framework

a trade-off,

in trading
2.7

be

information

corrupt

order
As

be

that

simulation

minimum,
of

quote

working.
in

communication.

number

secure

acknowledgments,

than

of

or

communication

of

cease

data

Protocol/Internet

transmit

information.

can

algorithm

packets

undelivered

provides

also

Web-browsing

individual

monikers,

can

in

compatible

historical

of

Internet

and

the

complexity

Control

e-mail

where

and

of

trader

structurally

in the

metric

from

used

that

tests

significantly

data

frequency

successful

standard

the

that

distribution

comparable

another

is

historical

may

ascertain

data

and

Transmission

that

of the

the

At

complicated

data,

by

environment.

matches

next

ensure

production

autocorrelation

differs

forecasts

algorithms

rolling

structure

development

calculated

taken

back-test

If the

encountered

the

be

More

model

stream

environment,
should

movements.

Communication

slower
TCP/IP

is

Both

the

UDP

and

communication
Protocols
and

to
like

FAST

on

top

of

enumerations,

messages.

Each

data.

field

Each
time

following

USD/CAD

message
of

stamp,
field

2.8

UDP

by

the

binary

illustrates

order,

protocol
various

is then

populated
including

is

rate.

The

FIX

message,

shown

quote

information:
Figure

Version

2.8

Sample

of FIX,

ITCH,

OUCH,

transmit

data

protocol

is

trade

and

the

in

free
related

data

components,

the

information,

process.

comprises

with

of

field
forming

user-generated

version

of

separated

FIX

used,

from

the

a quote

for

1.

a sample

exchange

FIX

and

other

to

layer

trading

(FIX),

TCP

quote,

FIX

message,

and

and

additional

the

eXchange

for

The

an
of

format.

specification

transmission.

definitions,

Figure

Information

used

text

require

messages

machine-readable

XML-based

the

however,

standardize

Financial

are

standardized

message

TCP/IP,

FIX

FIX.4.2

Message

(field

number

8)

transmitting
contains

the

following

The

time

stamp

of the

message,

20070731-15:25:20

(field

number

52)
Security

identifier,

Security

type,

Base

USD/CAD

FOR

currency,

Best

bid

Sizes

USD

and

best

at the

132

number

and

best

ask

of

communication

(field

133,

number

167)

15)
respectively)

(fields

134

and

135)

messages

depends

on

of message
bandwidth

TCP/IP

and

UDP

market

participants

slice.

Once

message,

the

the

specified

and

message

number
The

FIX

ensures

format,

using

messages

In

OUCH

parcel

comprising

and
used,
ITCH
the

is

and

is

fixed

on

among
The

in excess

comparison

The

binary

arrive

in

time

to convert

the

binary

trade-data
a

them
format,

entry

protocol,

number

each
UDP

consists

the

sender

other

FIX

of
and

the

total

variables.

header

can

be

bytes.
with

nature

Nasdaq's
of ITCH

and

machine-readable
from

text

to

and

faster

ITCH

specification.
messages.

binary
OUCH

enabling
while

of

of

and

ITCH

length,

dissemination
limited

the

The

and

of 100

in

out

to

route.

message,

OUCH.

order

their

includes

bytes.

message

attached

number,

slow

per

parcels

destination,

message

the

only

out

20

messages

to

is

is

it
and

addition

support

header

sequence

is often

processing

OUCH

the

header

sent

bytes

sends

individual

header

have

quote

message

at

the

message

as

no

into

TCP/IP

that

transmission.
outbound

many

participant

the

widely

OUCH

sliced

how

receive

market

identifies

elaborate,
is

one

and

typically

TCP/IP

known

back.

is

send

are

parcels

more

protocols

of

specifying

messages

The

standard

While

to

willing

identifications,
of

much

are

length,

header

destination

sizes,

message

the

8 bytes.

window

system

window

parcel,

and

(field

and

speed

Connection

and

bid

55)

exchange

U.S.$

(fields

number

factors:

Size

just

foreign

for

ask

best

Transmission
several

for

(field

Yet

is

the
ITCH

OUCH

provides

the

platform

Order

entry.

Replacement
Receipt

and

is

built

dissemination,

Trade

is able

and

lean

quote

and

past

trade

data

to transmit:

messages.
order

Event

imbalance

such

as

start

of

day,

end

of

day,

and

emergency

halt/resume.

more

complex

participants

are

Core

Message

Tick

information
several

FAST

data.

messages.

controls,

market

just

trading

messaging
in Figure

messages,

often

required

ITCHto use

and

OUCH-enabled

market

participants

FIX.

is transmitted
messaging
a

few

specifications.

message

information.

is built

among

around

FIX,

languages
Despite

strikingly

their
simple

Core

Message

Architecture

ITCH,

enabling

using
OUCH,

architecture,

in Trading

one
and

transmission

complicated

2.9 .
2.9

market

Architecture

quote

are

critical

Figure

fast

data.

Administrative

or

acknowledgements.

for
and

Order-level

For

cancellations.

of execution

ITCH

Net

for:

acronyms,
as

illustrated

of

As

shown

in

Figure

communication
1.

includes

Session

start

message

open

for

Heartbeat

communication

channel.
the

The

session

3. Quote
best
order

and

book

typical

5.

Order

Order

has
The

stopped
resulting

the

online,

down

may

in

state

that

messages
the

then

be

is

participant's

Parties

heartbeat

entity

stream.

the

business.

session

of

fail

to

for

communication
reinstated

using

carrying

quote

Level

II data,

bid

and

ask

information,
like

the

quotes,

such

as

of

the

depth
may

also

be

used

to

transmit

includes

actual

buy

or

order

sell

information.

identifier,

an

size,

limit

or other

specification,

order

and

validity

period

good-till-canceled)

(day,

and

A
order
desired
in

the

of

the

orders.
message
order

include

that

includes
now

and

the

needs

to be

order

confirmations

unique
canceled.

cancellation
of

identifier

acknowledgment

order

placement

or

order

respectively.

Execution

Session

The

that

notifies

shut

sizes.

best

acknowledgment

execution:
8.

is

placed

cancellation,

day.

every

messages.

cancellation

messages

7.

the
quote

price

of limit

and

message

market,

previously
6.

prices

behind

execution
case

is a message

ask

order

is

for

channel

trade

of

a communication

participant

often

and

beginning

participants

that

open

time

the

once

partners'

of

message

typea

and

quote

sequence.

using

Order

the

communication
start

transmitted
4.

health,

message

bid

that

period

in

only

message

communication

preconfigured

sent

market

of

messages:

to establish

recurrent

parties

their

key

message

desires

technological

stream

following

relevant

and
is

every

sometimes

notifies

trading

receive

the

session,

start

good

the

is

communication

2.

2.9 ,

acknowledgment
time

messages

of execution,

end

message

trading
trade

and

obtained
informs

quoting

messaging

price,

parties
for
flow

specify

the

and
that

the

execute

a given

details

of

quantity.
trading

entity

day.

comprises

intuitive

methodology

to

deliver

trading
and

effective,
outfits

fast

log

Neither

traceable

communication.

communication

of

for

potential

algorithm

issues,

errors,

and

the

Most

easy

reconciliation

such

as

network

like.

Security

TCP/IP

most

TCP/IP

plain

text.

nor
and

FIX

ITCH

Nasdaq

and

daily

problems,

and

While

their

identification

connectivity
Speed

reliable,

UDP

UDP

messages

provides
and

OMX

incorporate
are

optional

OUCH

are

sent

via

encryption

send

messages

encryption.

sent

other

Internet

words,

networks

at a considerable

messages
still

In

in

binary

unencrypted

in

latency.

format,

over

most

the

Internet

networks.
What

kind

of risk

messages?
current

To

in

nodes,
have

the

as

the

major

through
service

measures

in

nodes.

down

by

one

needs

At

sending

traffic.

about

Internet

at these
slowing

Internet

face

80

the

core

nodes.

(ISPs)

time,

traffic

consider
most

limiting

same

messaging

to

Today,

providers
place,

unencrypted

These
Verizon,

incidence

nodes

without

Internet

like

the

the

can

of

be

quite

consideration

to

urgency.

If the

core

through
the

nodes

traffic

would
While

network

hop
the

sufficiently

intent,

could

destroy

market
Figure

from

local

one

peer-to-peer

flow.

percent

redundant

to

to

and

prevent

the

markets

to

configuration

can

resulting
and

other's

peer

peer-to-peer

potential

traffic

flow

in which
distributed
allow

the

and

read

communication

party

networks

cause

structures
in a

peer-to-peer

any

would

another,

each
the

traffic

backup

network

full,

Still,

of Internet

user

observe
in

randomized

message

70

networks,

messages

malicious

flow

to fail,

participants

unencrypted

entire

were

peer-to-peer

fashion.

to

of

question,

flows

security

activity

participants

this

flow

world

some

congested,
its

and

such

spying

market

answer

layout

traffic

do

to
may

accumulate
be

hijacking

tremendous

is

of
losses

the

vulnerable
the

order
for

participants.
2.10

describes

three

common

Internet

communication

models

all

prevalent

in trading,

co-location
facility

model,
as

the

secure

ensuring
also

provides

trader
order

the

all

security

due
in

to

of

the

summarizes

latency

Figure

Messaging

and

2.10
Co-location

Figure

2.11

Optical

Fiber

from
risk

of

all

market

17

the

trading

Architecture

in

secure

co-location

access

to

the

comprises

directly

to

the

servers

would

between
savings
connection.
centers

Client-Server,

and

Co-location

milliseconds

co-location

the

intervention

example,

immeasurable

selected

same

participants.

for

trading

In

servers

advantage:

22

model.

the

access

distance

co-located
among

to

the

In

malicious

Nasdaq,

physical
to

the

speed

from

the

addition

as

in

network

network

for

with

anywhere

latency

Chicago,

the

co-located

shave

housed

dedicated

line

benefits

co-location

servers.

dedicated

environment
such

trader
to

The

minimizing

a safe

Chicago

have

communication

exchange,

are

matching

servers

servers.

so-called

servers

exchange's

trading

exchange

the

traders'

the

scenario,

private

including

in a
New

of

allow

the

round-trip
York

and

resulting

from

Figure

2.11

worldwide.
Peer-to-Peer,

Models

Latency
Networks

Incurred
between

by

Electronic

Pairs

Signals

of Locations

Traveling

over

Within

most

various

distances

concerns

result

the
issues,
length

of
of

lengths

are

near

sent

each

the

and

trader

down

servers

the

of

to

traders

little

traders

as

physically

equidistant

facility.

close

be

to

delay

such

exchange

can

feet

address

to

and

100

near

an

the

in a

time

guarantees

millimeter,

natural

co-located
To

Nasdaq's

at

co-located

of a second)

servers

in the

the

raising

as

advantage.

co-located

identical

of

giving

center

from

positioned

of all traders

received,

co-location
cable

itself,

millionth

potential

are

server

difference
(one

Nasdaq

fiber-optic

servers

distance

servers

the

servers

exchange

microsecond

message

exchange

centers,

of connections

physical

in one

every

the

fairness

facility.

may

data

from

about

given

on

co-location

to
The

cable

seen

servers

the

coiled
of

the

to

be

exchange.
Although

some

unaffordably
data

center

the

in

minimum

scanners
A 20-amp

for
and

offers
monthly

the

trader's

real

additional

the

to the

various

data

around

120-volt

ultra-fast
world

per

as
with

private
services

biometric

month.

costs

$350

source

At

follows:

per

power
$175

the

down

source

proximity

$1,500

power

opposite.

example,

equipped

runs

additional

a connection

month.

to

for

broken

hardware

Finally,

centers

point

Jersey,

charges

20-amp
an

co-location

co-location-equivalent

primary

costs

believe

numbers

New

air-conditioning

120-volt

redundancy

the

Secaucus,

Equinix,

A cabinet

An

participants

expensive,

company,
with

market

per

security

month.

designed

for

month.

communication
runs

an

network
additional

$325

linking
per

The

grand

month,

total

a negligible

Network
The

of

the

proximity

cost

for

participants,

architecture
yet

continuously

it is not

of

message

traffic

challenge

their

to

among
exchanges

Huang

(2011),

on

Nasdaq,

one
Saar

placement

minute

from

(2011)

report

and

malicious
is

automated

an

on

per

promptly

matched,

market

moves

away

interests

resubmit

them

explained

limit

cancel

execution

on

ensuring

the

one

makers,

but

eroding

network
for
navigate

on

the

new

12,

the

quotes

bid

on

the

and

trading

hand,

they
many

cancel

issue,

such
10,
market
and

Once

the

it is in the

orders

and

In

time-price

to

addition,
priority

of

place

and

then

to

secure

their

layering.
are

in

as

seem

the

prices.

orders

caught

competing

Even

the

offer

to

order-cancelling

participants.

may

successfully

may

as

resulting

be

with

venues
are

to

maker,

limit

order

Chapters

market

exchanges

known

in

stream.

best

numbers

many

bandwidth

to

brisk

for

revenue

participants

resources,

all market

to

of the

maker

market

other,

order

and

Hasbrouck

activity

close

the

canceled

explanation

quote

a steady

best

into

detail

is

are

grouped

in

biggest

Hautsch

placed.

the

to

the
venues

this

the

uninhibited

Perhaps

are

While

bookin

in a practice

dynamics,

activity

need

excessive

priority,

orders

described

market

at the

the

observer,

the

the

books,

innocuously

time

to

that

allow

orders

percent

as

in Chapter

order

such

of

traders.

runs.

from

to

of all limit

95

similar

thus

enough

market
have

ensure

to

the

of

to

According

makers
top

broad

all

example,

cancellations.

simple:

market

for

order-matching

uninitiated

quite

pricestay

fully

$2,350

to

other

cancellation

to

behavior

delays

just

available

Exchanges,

interested

and

order

circle:

to

investor.

resource

is

all

of

In

up

infrastructure

connections

volume

within

to all.

their

sheer

as

serious

is

free

enhance

bandwidth

best

any

adds

Throughput

messaging

and

setup

the

within
attract

co-location

market

makers

quotes

co-location

vicious

the

market

missed

and
does

space

not

also

are
other
help
faces

capacity

constraints:

co-location

hangar

running

out

there.
the

Nasdaq

As

of

in

network

Mahwah,

space

to

discussed

and

to

the

Chapters
issue,

seen

New

offer

in

bandwidth

developed

has

much

Jersey,

that

parties

3 and
known

implemented

so

in

its

is

reportedly

in

co-locating

a promising

pro-rata

at selected

it

interested

12,

as

demand

solution

execution

, has

to
been

exchanges.

Software
High-frequency
deployed
any
an

trading
over

hardware

software

computer

algorithm

is

high-level

actions,

implement

a given

the

details

of actual

take

into

account

structure

on

Algorithms

Figure

HFT

system

into

properly

defined

developed
task
coding
the

the

algorithm

the

book

to
2.13

are

explain
illustrates

market-making

Figure

2.13

Sample

that

logic,
to

algorithm

be

designs

2.12
of

step-by-step

algorithm:
Market-Making

2.12

will

be

The
of

how

not

delve

but
and

key

in

to
into

may

still

messaging
implemented.

elements

of

the

Algorithm

in Figure

algorithm

coded

system,

The

as

sequence

ultimately

in Figure
of an

known

computer

hardware

in diagrams.

shown

the

the

does

the

will

summarized

with

program.

or

algorithm

of

idea,

software

explain
The

an

As

is subsequently

a full-blown
as

applications
above.

with

or programming

Elements

elements

simple

hand.

visualized

Common

The

begins

idiosyncrasies

often

diagrams

at

to

software

described

lingo,

language

which

are

2.12

Figure

an

ultimately

messaging

in computer-science

a chosen

algorithm

are

and

system,

algorithm

term

systems

Algorithm

common
process

used

throughout

HFT

strategies.

of

the

following

1. Begin
2.

program.

Check

market

market

4. If no,

start
wait

5. Repeat

market
one

step

algorithm

additional

market

conditions

suitable

for

minute.

2.
presented

and

Start

secret

sauce

ones.

Usually,

the

research,

where

given

positive

shown

in Figure
term

the

are

trading,
trading.

significant

enough
to

to be

to

highly

therefore,
fully

The

rigorous
explained

often

used

among

The

in detail

A trader

two

rigid
can

be

the

considered

terms

trading,

and

however,

are

system

process.

frameworks,

tasks

15.

terms,
A

bad

ideas

nested

with

explanation.

the

advanced

competing

electronic

to

be
from

of

in Chapter

the

can

basis

two
market

systems

synonymously

approach
some

HFT
the

Check

tasks

several

testing.

among

following

as

nested

on

distinctions
some

or comprises

only

good

trading,

warrant

automated.

The

systematic

methodical
is

2.13

designed

is selected
of

is

Figure

making.

are

task

2.13

is nested,

distinguishes

tasks

algorithm

low-latency

trading,

that

2.13

in

market

results

high-frequency

refers

in Figure
marked

critical

The

making.

algorithms

conditions

have

Are

making?

3. If yes,

The

conditions:

usually
Systematic

but

does

systematic

not
if

he

manually

pattern.

places

The

term

practicing

methodical

discretion,

and

trader:

trading,

he
the

and

trading

systems

route

implicit

Low-latency

refers

microseconds.

are

often

the

trades.

Once

an

algorithm

is

coded

in

language

coding
computer
possible
the

code.

fastest.
non-FPGA
to

code

is

to

run

The

code

Still,

many

systems.

time

many

may

but
The

electronic

not

are

be

placed

moving

utilizes

on

to

trading

is

not

translate
the
when

written

directly

high-frequency
and

rely

While

on
C++

to

in

happen

to

The

of

algorithm

create

to

chips

FPGA
traders

as

little

read

and

is

the

as

goal

of
into

delay

as

interpret

presently

deploy
such

better

components

computers.

will

remains

recorded

to obtain

the

and

low-latency

into

computer

languages

trader

hold:

down

logic
to

previous

also

traders

by

process,
the

currently

have

broken

the

the
the

often

low-frequency

understood

in

are

connectivity

in

between

systems

it is

fast

described

distance

does

by

time,

architecture
their

algorithmic

electronic

that
As

designed,

and,

during

or

term

trading

accurately

speak

or,

electronic,

case;

traders

measurements

deployed

on

the

and

trading

reverse

prices

and

the

latency

but

the

components.

may

exchanges.

High-frequency

low-latency,

systems

to

measures
Most

of

trading,

that

the

All

electronically

necessarily

be

or

obsolete.

and

latency

exchange.

markets
however,

and

traders

section,

be

most

to

intuition
traders.

nonelectronic

orders

their

orders

are

traders

preferences

high-frequency

have

only

trading

between

place

involve
not

platforms,

becoming

All

may

As

electronic

to

specific

systematic.

execution

trading

does

algorithmically.

the

chooses

using

also

distinguish

discretionary
are

the

telephone.

however,

trading

as

form

to

traders

traders

low-latency

systematic
reverse,

known

algorithmic

indicators

coined

from

describes

whether
over

was

allocations

electronic

perhaps,

certain

systematic

and

term

when

hence

high-frequency
The

trades

C++

fastest

the

standard
and

Java

computer

language

easily

Java

with

understood

workarounds

example,

the

Java

with

Java

describes

best

The

outlining

code
quite

as

few

and

as

sell,

data

practices

lines

as

well

actual

HFT

the

applications

that

trade
can

the

most

lengthy,

system

is its

risk

management

lines

detail

of

in Chapter

code.
14

may

of this

an

for

coded

in

replaced

with

Chapter

the

algorithm

trading

16

number

to

supporting

as

mandatory,
and

lines

of

code,
receipt

5,000

lines

component

of

balances,

management

of

of

is

code.

each

which

HFT

buy
quote

acknowledgment

much

to

comprises

decisions

in 10,000+

and

tends

logic

actual

incorporates

checks

Risk

reportedly

speed.

to the

as

yet

in

Thus,

and

of

cases,

take

Perhaps

50,000+

logic

send-off

also

coded

implementation.

system

that

is

increased

In addition

functionality
as

for

are

components.

disabled

successful

every

systems

system

trading

of code.

however,

retrieval

OMX

of coding

In many

50

slowest

access

the

many

collection

memory

short.

its

Nasdaq

garbage

direct

humans,

of

famed

C++-like

be

by

HFT

can

total

discussed

in

book.

Summary
Algorithmic

execution

necessary
large

function

and

today

small.

can

for

select

few

added

benefits

to

market

you

execution

2. Mr.

Smith

in the

financial

super

computer

Mr.

Smith's

inseparable
delivers

build

from

and

to

investors,
investors

advanced

algos,

including

previously

available

such

as

is

all

most

Services

and

value

It

costs,

use

participants.

markets.

technology

trading,

of security

today's

considerable

plummeting

high-frequency

End-of-Chapter

the

that
With

afford

designed

1. Would

is

algos

only

co-location

to

provide

speed.

Questions
encrypt

your

venue
has

over

read

services
to

the

about

orders

Internet?
the

industry

increase

investment

trading

arms
and

the
justified?

odds

before

transmitting

them

to

Explain.
race
decides
of

fast

Where

of computer
to invest

technology
into

the

order

transmission.

does

most

latest

message

Is

congestion
3. What
4.

On

occur

average,

best

best

offer

exchange

order

today?

much

slower
with

is the

quote

transmission

of

trade

order

messages?

is a heartbeat?

6. The

buy

how

in comparison

5. What

on

cyber-universe

is co-location?

messages

the

in the

550

units

and

algorithm?

offer
on

on

exchange

exchange

C contains
on
send

his

B contains
just

behalf.

them

A contains

to

100
How

300
500

units.

Your

would

you

exchanges

units

units,

of instrument
and

customer
break

under

the

up

the

best

wants
the
minimal

X,
offer
you

customer's
impact

to

CHAPTER
Market

3
Microstructure,

The

study

of

ago,

and

the

participants

market

remain

order

types.

has

changed.

into

the

continuity

the

1980s

In
tick

of Markets

Financial

to

transactions

buy

nonfinancial

his

transaction,

The

grocer's

amount

of

not

customer's

Furthermore,
good
has

takes

produce
grocer

condition
full
financial

rather

grocery
and
rights

to

markets

the
sell

chapter

delves

into

the

matching

issues

defining

other

market

mind

that

settled.

In

this

respect,

his

money

markets

for

grocer.

immediate
for

settlement

customer

bought

nowadays

in cash

form.

expects

that

transaction

all

other

When

execution

of

merchandisefood.

the

incorporate

and
taking

orders,

and

of

expects

the

into

a peace

he

than

were

easily

strategies

from

collected,

data

trading

neighborhood

the

much

peculiarities

captured

investors

and

on

derivative

differences

other

with

customer
that

monthly

store,

register

account
the

as

of

the

allow

different

exchange

the

and

among

that

such

the

be

order

still,

time,

describing

accounted

a grocery

money

can

This

securities

various

the

to build

sell

are

cash

itemizes

data

markets,

properly

an

receipt

centralized

be

enters

tick

that

products,

customer

daily

market

core

principles

to incorporate

venues

and

markets

from

structures,

are

will

financial

grocer

rebate

markets

participants

of

key

decades

most
These

of

one

microstructures.

microstructure
and

At

four

today:

orders.

of these

way

over

explosion

it is possible

market

Types

in

no
that

quickly,

markets.

is

was

Books

true

market
an

dynamics

Today,

today's

the

data

there

account

processes,

and

Despite

initiated

modern

orders
despite

strategy.

trades

limit

Order

remain

used

but

trading

on

Limit

originated

principles

most

differentiating
observed,

and

microstructure

core

rely

types

Orders,

is

customer
similar

function:
and
most

the

food

the
often

he

permanent:
the

total
from

acquires
that

product.

quality

the

control

the
Most
for

financial

products:

floors

of

each

be

ensured

as

via

the

real-time

well-defined

foods

decisions

beginning

markets

have

settings,

stores

have

their

have

been

each

to

common

with

Trading

and

not

by

of

of

investing

and

offering,

type

as
all

well:

used

to

markets

traded

instrument,

or

other

securities.

been

and

gearing

take

venturing
player

is not

always

skills.

cross-asset

iCap

exchange
the

of

and

in the

entering

toward

advantage

frameworks

equity

considering

type

of the

have

exchange

clothes.

financial

options,

now

and

nonfinancial

selling

Similarly,

procurement

foreign

In

makers

the

are

stores

nonfinancial

ice-cream

efficiency

venues

(BATS)

type.

and

by

markets

their

traditional

Systems

be
of

unaware

fragmented

futures,

nonfinancial

trading

to

suitability

risk

financial

from

districts.

equities,

trading

the

fixed-income

The

if

increase

many

highly

butchers,

distribution

Similarly,

specific

tend

are

the

product

differ

organized

however,

well
For

stocks

markets

most

by

be

shops,

only

megastores

trading

to

historically

Recently,

history,

lumber

bakers,
own

can

as

have

evaluate

leaving

organized

carrying

trading

space,

of trading

used

fishmongers,

financial

quality

participants.

for
not

the

themselves.

been

markets

do

on

process

market

offerings

sale

their

exchanges

accreditation,

traders

and

all

on

grocers

person,

or

to

public

like

and

due-diligence

traded

just

purchase

standardized,

distribution

each

profile

to the
the

Food

to

be

Initial

Yet,

risk

to

for

prelaunch

data

structure.

particular

offered

contracts

scrutinized.

Since

tend

a thorough

futures

client's

products

market
via

example,

well

the

launching

Best

Alternative

foreign

exchange

space.
history

York

of the

Stock

for-profit

Exchange

entity

at the

stockbrokers

signed

commissions

and

In subsequent

financial

(NYSE),
end
an

instead

years,

exchanges

as

for

of the

the

U.S.

was

eighteenth

agreement
retain

example,

to
at

economy

first

century,

stop

least

linear.

formed

when

undercutting

0.25

percent

grew,

more

The

two

each
of

each
public

New
as
dozen
other's
trade.
listings

became

available,

public,

stockbroking

NYSE

rose

During

the

the

of

of

and

transfer

brokers

creating
the

away

from

and

U.S.

on

the

lieu

increased

such
similar

to

relatively

Today's

equity

markets

venues,

run

by

Nasdaq,

and

relatively

new

comprise

trade

all

the

Chicago

that

equity

share
as

the

to
their

Nasdaq,

human

traders,
pockets

process.

of

Technology

times,

and,

process

to

NYSE,

the

perhaps
investors.
Exchange,

and

today

entrants,

such

face
as

the

(ICE).
over

arrivals,

dozen
such

such

exchanges

standardized

of

in

Mercantile

exchange

stalwarts

recent

larger

into

of

tags.

appear

such

execution

Exchange

now-industry

Since
and

as

the

transparency

exchange

forces

bonus-collecting

in

the

competition

market

benefits

nonprofit

price

savings

of brokers

on

exchanges,

resulting

other

Commodity

others.

of

the

into

The

seats

retain

New

during

exchange-registered

increasing

competitive

increased

from

1971,

investors.

their

to
U.S.

converted

with

for

witnessed

the

exchanges,

In

1934,

formed

compensation

the

investors

rates,

trajectories

to

In
required

newly

was

excessive

stockbrokers

was

multiple-million-dollar

brokers.

of

then

NYSE

market

these

Intercontinental

oversight

seats

of

NYSE

expectation,

reached

in

error

competition

and

to its

producing

importantly,

followed

the

regulators.

(SEC).

intake

helping

some

lowered

to cap

have

technology

early

up

and
in

investors

Other

and

and

the

the

their

decades

transferring

most

had

succeeded

deployed

from

a secondary

space,

earnings

has

of

to live

two

exchange
have

from

lifestyles

the

Commission

in a bid

1990s

lavish
investors

recession,

some

failed

past

grew

lucrative

Depression,

Exchange

entity

mandate

however,

oversight

War

not-for-profit

The

investing

increasingly

Great

accept

postVietnam

by

in

money-losing

the

and

that

became

scrutiny

Securities

and

interest

recessions,

midst

register

the

in value.

major

drew
in

and

are

products,

various
as

as

the

NYSE

BATS,

subject
exchanges

exchange
and

DirectEdge,
to

U.S.

SEC

presently

compete
In

to

a bid

differentiate

to

attract

exchanges
free

on

In

addition

do

the

entire

not

based

who

Liquidnet

liquidity

in

this

later

fees

exchanges

and

rewarding

whether

they

costs.

chapter),

most

and

offer

are

only

Order

The

cumulative

now

differentiating

large
place

traders

but

market

orders

their

the

orders,

in

for

books,
book

the

information

traditional

of

exchange,

observation,

has

dark

keeping

them

appealed
they

exchange,

dark

matching

an

instead

order

equities

measurable,

size

orders

at any

for

to large

may

lit

reveal
market.

pool.

orders,
so-called

referred

to

as

United

1980s.
a

given

orders

time

the
the

the

given

on

larger

of

the

trading
the

order

has

the

majority

available

a specific

number

markets:

limit

as

in the

limit

of

venue.
number

a finite

venue

order

traders

each

trader's

Liquidity
of

size.

meet

trading

limit

size

to

limit

is

also

orders

Liquidity

was

is
first

(1968).

limit

the

all

and

today's

organized

in

of

larger

liquid

each

Demsetz

account

The

in

and

by

trade

more

finite

centralized
double-sided
States

In a CLOB

recorded

in

sequential

price

increments,

each

price

at

breed
Unlike

available

about

exchange,

the

necessarily

Europe

limit

of an

liquidity.

on

pioneered

is

new

pools.

order

in a dark

example

dark

book
limit

Books

as

available

as

order

orders

market

known

exchanges,

concerned

their

Limit

posted

on

known

limit

are

is an

incoming

regular

Trading

placing

also

this

on

membership

structures,

disclose

dark.

defined

the

not

investors

limit

with
equity

emerged,

the

pools

by

on

addition,

traders

to

has

where

in

(more

mainly

orders.

venues

are

price

offerings

away

In

segmenting

or limit

To

done

registration.

also

is

liquidity

have

themselves

their

book:

table

increment.

and

in

of

contemporary

limit

order

auction.
the

model,
with
rows
Figure

early
all

books
The

1970s

columns

3.1

(CLOBs),

CLOBs
and

incoming

recording

exchanges

limit

were

adopted
orders

in
are

corresponding
sizes

illustrates

of
the

limit
idea.

to
orders
The

limit

order

book

information

participants

as

Figure

3.1

Sample

orders

are

on

right-hand

the

on

In theory,
the

of

side

order

bell-curve
order
As
a

Figure
limit

are
3.2

order

today's

markets

price-based
highest

price

reflected
the

in

best

the

bid

form

the

Best

ask

is

moment

best

size.

price

at each

Figure

3.2

to be

of

buy

limit

also

many

assumed

are

often

a new

limit

order

to
to

its

best

bid,

price,

are

the
with

with

price.

limit

the

price

aggregate

limit

sell

orders

a finite

to

as

best

aggregate

in a Limit

follow

Order

normal
to

hold:

it is placed

into

all

prices

increment,

The

price

mirroring

management

Since

the

respective

about

orders

arrives,

and

referred

Dynamics

orders

tends

buy
of

posted
and
offer.
size

Book

or

in
tick

orders

at

these

orders

size

price.

Sample

buy

asymmetric.

minimum

delineated.

exists

limit

symmetric

to

and

sometimes

4.

sell

risk

normal

ask,

there

assumed

assumptions

bid

market

All

limit

the

Similarly,

best

of time,

posted

the

all

of

clearly

form

other

Book.

Neither

subject

are

book;

all

in Chapter

Order

in

corresponding
are

bins

the

often

are

when

queue

Limit

distribution

books

shows,

in detail

Furthermore,

seldom

to

book.

the

distributions.

books

of

are

orders.

distributed

side

books
with

sell

applications,

of

of the

order

be

discussed

left-hand

price,

limit

II data,
Snapshot

the

limit

market

that

Level

can

the

reported
at

size

the

as
lowest

information.
At

of all

any

given

limit

orders

When

orders,

market
beginning

of the

queue,

the

direction

those

the

price
ask

those

market

buy

market

order

leaves

than

the

instantaneously

the

Similarly,

the
book.

at

the

matched

in

available

limit

buy

through
eating

orders

best

bid

are

low-priced

limit

sell

order

with

are

If the

size

size

of the

best

offer

in limit

usually

orders

The

that
on

equal

end

up

the

bid

or

higher

buy

treated

as

than

the

on

orders.

orders

market

at

and

aggregated

like

in

spread,

buy

prices

queues
available

bid-ask

sell

the

executed

price.

gap

market

buy

the

liquidity

the

market

sell

other

up

for

limit

the

a significant

orders

prevailing

ask

than

subsequent

is similar

with

best

increasing

slippage

Limit

is

sweeps

Sweeping

process

it

is greater

ticks.

with

of

order

price,

order-matching

size

placed

increasing

inducing

matched

arrives,

of

side,

potentially

order

with

incoming

ask

the

buy

orders.

market

sell

orders.
If the
best

size
ask,

of
and

the
the

incoming
aggregate

buy
best

order
ask

is smaller
queue

is composed

size
of

of

the

several

limit

sell

the

limit

differ

orders
sell

placed
orders

from

one

present

practice

(FIFO)

execution

now

match

process
In

market

known

transparency

pro-rata

2005;

Figure

3.3

Price-Time

Reducing

information

limit

book

orders,

buy

most

known

as

limit

order

order

of
may

exchanges

the

several

via

execution,

price

continuous

Jain,

limit

FIFO,

of that

of trading

market

of which

at

first-in-first-out

other

exchanges

at a given

price

in

matching.

the

1996;

Enhancing

orders,

of each

decision

While

for

3.3

as

another.

limit

the

the

also

Figure

Roell,

order

or

price,

priority,

proportion

first

ask
against

to

schedule

priority,

order.

FIFO,

exchange

as

the

best

matched

price-time

known

is also

is

a fixed

time-price

first

at the

bin

to be

illustrates

following

Harris,

Priority

limit

order

matched

the

auction,

the

and

the

with

FIFO
has

that
the

matching

been

measures

shown
(see

arrived
incoming
process.

to enhance
Pagano

and

2003):

Execution

asymmetryall

traders

have

access

to

the

traders

to

add

information.

liquiditya
thereby

CLOB's

organization

providing

a fast

and

CLOB

structure

increasing

market

supports
objective

incentivizes
liquidity.

efficient
order-matching

price

determination

mechanism.

by

Uniform

rules

and

equal

While

most

like

for

market

execution

venues

Mercantile

Exchange
Intercontinental

are

ensure

execution

incoming

market

operational

fairness

the

best

ask.
of

with

equal

fractions

The

larger

the

the

larger

limit

the

fill

with

order
of

of

an

the

at the
that

pro-rata

a fixed

Similarly,
all

Chicago

(ICE)

CME's

best

order.

exchanges,

Board

proportion

bid

and

Figure

the
3.4

at

best

is

order

matched

the

best

ask,

shows

an

limit

order

posted

to

matches

of each
sell

and

switched

schedule

orders

Options

(PHLX),

have

incoming

limit

some

Exchange

Exchange

order

FIFO,

Stock

schedules.
buy

on

(CME),

Philadelphia

Commodity

pro-rata

based

Exchange

(CBOE),

at

participants

access.

Chicago

posted

all

bid.

therefore,

the

pro-rata

process.
Figure

3.4

The

main

point

of

orders,
pro-rata
without
increasing
eliminates

Pro-Rata

Execution

advantage
view

is the

and,

of

the

built-in

special

to

bring

encourages
compensation

exchange
incentives

place

for

traders
like

and

rebates

from

traders

liquidity

profitability.
to

matching

incentives

therefore,

matching

pro-rata

to
to

to
the

post

addition,

then

cancel

exchange

place

large

limit

exchange.
large

discussed

In

the

limit
below,

pro-rata
limit

orders

The
orders
thereby
matching
with

the

intent

to

to and
In

secure

from

the

will

filled

the

to

order

be

of

placed

higher

the
the

order

his

at

the

can

of

than

limit

order

in
be

will

To

trader

intended

size

of

his

of

the

of

The

bin,

in response
his

the

is of the

the
to

an

chances

to place

with

limit

traders.
price

is likely

filled

trader

size

increase

order,

get

exact

order

filled

fraction

a specific

sign.
the

cumulative
limit

will

opposite

that

only

the

orders

follows:

The

other

that

his

that

on
by

therefore,

do

not

explicit
exact

limit
hope

size

as

that

order

the

concept.

price

3.5

A limit

order

order,

or

closer

the

are
or

to

far

order.

destined

the

away

from

A market
to
side

be

is to
order

matched
of

the

orders,

and

passive

be

taken

orders

market

the
sell
the

order

most
the

An

is

to

best-priced
Limit

not

Instead,

proximity
3.5

of

illustrates

Illustration

(a

the

aggressive

do

of.

the

low-priced

considered

price,

book.

orders

Figure

price

market

is the

refers

Orders:

order)

Aggressive

advantage

price.

market

with
limit

aggressive.

Passive

limit

order

or

of

prevailing

and

passive

to

passivity

high-priced
limit

as

malicious

Aggressive

Execution

described
mean

orders

Figure

Passive

be

the

opposite

traffic

order.

aggressiveness

spread

depend

of the

of the

the

the

message

as

schedule.

price

size

size

versus

indicate

will

order,

a larger

Aggressive
Orders

knows

of all orders

entire

intended

reducing

works

pro-rata

same

order

fraction

order

the

order
the

market

with

orders

the

aggregate

the

execution,

incentive

limit

under

percentage

incoming
of filling

filled

orders

of

pro-rata

execute

portion

lower

priority

exchange.

nutshell,

desiring

that

time

more
order,
limit
orders

limit

passive.

The

aggressive
crossing
order
crossing

buy

on

is
the
the
the

spread
are

are

also

While

treated

considered
market

market
the

like

today's
associated
market

order

market

price

placed

price

market

between

the

time

matching

liquidity
Such

pieces
Additional

By

until

the

price

order

prevailing
against

spread,

orders.

Highly
the

cost

of

spread,

specified

is added

to

Limit

orders

limit

transaction

price.
costs,

and

in the
to

may

same

the

limit

times

the

and

limit

orders

executed

as

better

execution

which

vary

are

also
from

subject
one

also

it

market

avoid

the

order

where

incurred

orders

may

when
book,

of

levels.

disruptions,

spread

or

news

fractional

price

market

good

the
market

depth

the

as

is

direction.

fixed

it

it

of

executing

order

reaches

time

the

available

market

is

executed

simultaneously

at different

order
the

at

generally

Limit

The

deplete

moving

book,

as

be
the

also

obtain

the

placed.

algorithms

the

such

price

but

the

than

and

orders

their

orders

limit

aggressive

the

recorded,

was

common

relative

paying

time

execution

adversely

orders

market
it.

is

is placed

market

limit

of

the

exchange

order

through

slippage.

In

component

From

order

at the

and

large

against

costliest

worse

of

uncertainty.

execution

particularly

sweep

the

market

book,

place

in significant

the

negative

order

conditions,

executed

their

the

is

order

A limit

cross

arriving

market

contrast,

placed.

that

price

fees

factors:

the

and

be

in

market

traders

face

the

arrive

many

may

of the

result

a given

order

time

the

is

information

book

and

fills,

transaction

execution.

resulting

situation

process

order

the

may

in

when

market

queue,

guaranteed

incur

can

to several

of

releases,

execution

nearly

spread,
and

time

orders

Each

price.

to

the

due

executed.

the

order

slip,

at

be

Several

pay

market

may

may

the

and

uncertainty

the

is

in

immediate

broker-dealers,
price

with

slippage

and

and

markets,

prevailing

enjoy

cross

exchange

orders

aggressive.

orders

orders

market

sits

order

crossing
by
market

trading

is
the

market
orders

price
to

is

than

positive

or

venue

to

another.
For

all their

riskthe
is

price

risk

matched

price

with

may

quickly

when

to eliminate

to open

engage

in the

Complex

to
trading

example,

in

to an

is executed

the

opposite

from

order

may

direction.

limit

trade.

yet,

unexecuted

also

incur

a cost,

from

new

that

market
it

particular

the

opportunity
limit

of the

it

leaving

present

of the

when
A

order,

misses

And

important

only

and

orders

opportunity

to

strategy.

competition
venues
to

expanded

so-called

iceberg

display

only

rest

orders

their

iceberg

orders

against

a smaller

placed

back

orders

in a dark

at

the

is

matched

more

than

Other

specialized

potential
risk .

range
include
liquidated

of

from
needs

Most

hard
when

for
and

In

a time

part
as

and

have

orders,

creating
traders

book,

part

of the

iceberg

in

full:

trade

tick.

market

is

is

matched

As

being

Unlike

iceberg
the

books,
matched

queue.

an

keep

order
when

of

to

and

basis:

rule,

the

revealed
size

is

iceberg

orders.

sprung

higher

selected

limit

book

For

pools,

priority

or

offerings.

order

FIFO

limit-order

in

matching

limit-order

limit

information

limit

up

as

transaction

costs

well,
and

to
to

generate
serve

the

of customers:

trading

orders

dark.

their

traders
do

allow

nonexecuted

size

orders

revenues

following

of

pool,

other

cost

on

the

end

to

orders

additional

executed

dark

available

in the

the

order

from

orders

the

in

their

of

order

in

order,

iceberg

disseminated

. Iceberg

of their

the

diversified

number

liquidity

are

entrants

competition

the

a portion

of

the

have

response

exchanges

Limit

order

subject

Orders

business,

after

are

a position,

a position
trading

response

the

limit

to close
risk

of

away

unexecuted

market

placed

order

move

placed

orders

A limit

a market

An

problem

limit

of nonexecution.

unexecuted.

In

advantages,

venues

containing
trailing

a price

move

and

broker-dealers

market
stop

risk.

orders,

exceeds

the

The

where

now
order
the

predetermined

offer

examples
position
threshold

is

in

the

adverse

direction

(see

Chapter

14

for

more

information

on

stops).
Speed

of

fastest

execution

market

order,

catch
bid

execution.

the
limit

possible
a

closing
order

spread,

the

book

order

may

market
gradually
Price
that

over

a midpoint

attempting
the

the

often

name

iceberg

by

orders

price

and

the

vanilla

guarantees

that

bests

to

the

only

order.

market

The

half

the

best
of

the

clears
sweep-to-fill

order,

sweeping

by

include

obtaining

orders

hidden

suggests,

order

since

limit

order

block

order

in

large-volume

displays

discussed

in the

Time

to

market.

Orders

orders

liquidity.

for

A fill-or-kill

immediately
in
it is

the

Advanced
triggers,
Algorithmic
order-slicing

order

same

canceled

venue

large
book

options

discounts

such

as

category

another

in the

limit

of the

order

on

if the

order

order,

books.

in the

The

limit

group
the

most

matching

good-till-canceled

is

kept

maximum

hidden

include

order

section.

desiring

is canceled

and

The

time-to-market

orders

order

in

period

the
of

include
immediate

liquidity

limit

is

not

limit

order

order

book

time

set

by

the

a quarter).
These

implied

trading.

orders

include

volatility

Orders

algorithms,
in detail

the

liquidity,

others.

of this

Conversely,

or

(e.g.,

trading.

beginning

market

available.

portion

in

dark

among

displayed

a limited

as

fill-or-kill

deliver

orders,

is not

book,

described

the

the

simultaneously

in the

than

executed

Such
the

orders

trading

order

requested

faster

often

order

enable

to

crossing

sweep-to-fill

size

addition

that

negotiate

to

costs.

iceberg

until

to

Privacy-providing

falling

in

try

time.

improves

its

include,

match

improvement.

Privacy.

as

category

price,

and

is

transaction

and

executed

order

this

order

of

be

in

market-on-close

by

prevailing
order

Orders

in
such

in Chapter

quantitative

in options.
this
as

17.

additional

category
percentage

offer
of

execution
volume

via
(POV),

Trading

Hours

Traditionally,
p.m.

many

Eastern

placed

time.

on

for

selected

that

allow

use

market-on-close

of

result,

4:00

normal

many

hours
a.m.

is

fill

to

in
8:00

considerably

trading
to

is

trading,

Extended

from

trading

effort

afterhours

volume

the

Convergence

markets

has

streamlining

others.

markets

eXchange

Among

today
(FIX)

XML-like

and

left

an

some

the

can

trends

quotes,

hours.

their

Still,

customers'

Divergence

indelible

footprint

aspects

of

in market

accessed

on

trading

all
and

convergence

orders,
among

efficient

The

trading.

via

protocol.

that

acknowledgments,

independent

be

messaging

specification

receive

are

the

allows

order

other
FIX

not-for-profit

The

market

and

FIX

Information
protocol

participants

to

cancellation

messages
is

further

is
send

and

necessary

protocol

body,

Financial

and

execution
for

fast

administered

facilitating

an

and

by

an

proliferation

of

protocol.

Most

markets

worldwide

(LOBs).

The

Singaporean

entities

to use

a different

past

Among

the

Equities

are

now

designed

Stock

as

Exchange

market

limit

was

structure,

but

order

one

of

converted

books
the

last

to LOB

in

decade.

fragmentation

rule,

access

4:00

developments:

Most

the

and

to

more

As

trading.

after-hours

Market

markets,

fragmenting

the

trading.

trading

during

a.m.

orders.

electronization

following

9:30

markets,

extended-hours

the

Microstructure:

modern

of

market

The

from

globalized

extended-hours

observed

brokers

Modern

as

time.

than

operated

premarket

example,

Eastern

thinner

today's

offer

known

equities,

venues

accessibility

today

cumulatively

The

In

expanding

exchanges

p.m.

trading

key

trends

in

of markets
are

whereas

subject
all

market

among
to

equities

the
are

the
National
to be

divergence
asset
Best
executed

is

the

progressing

classes:
Bid

and

at the

Offer
aggregated

(NBBO)
and

disseminated
execute

NBBO
an

obligated

incoming

to route

Futures

or

the
do

exchange

exchanges

All

(OTC),

electronic.
access

to an

interdealer

Option

markets

are

Following
fixed

the

income

electronic
peculiarities,

sections

equities

of

are

can

(2011),

dark

revealing
not

include

the

quotes.

but

are

subject

quotes

negotiated

may

the

now
be

fully

granted

entity.
on

average.
classes

online

asset

or

over

are

asset

coming

further

within

be

each

or

such

as

expanding

classes

has

fragmentation

asset

within

traded

exchange-like

Securities

class

selected

any

pools

about

dark

of

as

in
distinct

the

well.

asset

ordersthe

market

participants.
remaining

78

in the

their

The
of

that

ability

dark

percent

not

with

overall

The

following

classes.

of

the

The

order

related

U.S.

order

equity

darknot
Pragma

aggregate

equity

singular

advantage

orders

size,

as

without
the

of dark

issues.

book

Dark

to

large

differentiated

limit

is

disadvantages

and

pool

book

U.S.

match
the

lit exchanges.

According

the

cited

offer

order

pools.
to

transparency

on

pool.
of

dark

frequently

do

and
the

percent

associated

pools

market

22

in

observable.
lack

of

traded

lies

pools

where

participant

information

the

in dark

entities

is presently

exchanges,

The

Each

activity
new

be

is

in Equities

to

are

will

NBBO

platforms

exchange-like

swaps

in

to be

little

to

exchange

centralized

an

and

peculiarities

displayed

of

have

OTC

with

unable

requirements.

participants

network,

the

pricing,

market

regulation,

discuss

volume

with

continued

Dodd-Frank

exists

Fragmentation

pools

NBBO,

is

frontier.

Fragmentation

U.S.

not

many

large

resulting

securities

are

numerous,

forms.

exchange

mark-to-market

though

selected

the

centralized

do

trades

even

Yet

daily

at

the

exchange

have

markets

at all.

counter

to an

not

and

If

order

order

margining

Foreign

price.

market

exchanges

to unique

better

Unlike

pools
lit

pricing

for

is

disclosed

to

is executed

on

volume

not

limit

orders

and

lit

exchangesvenues

where

and

can

in full

even

be

disseminated

within

the

fragmented
In the
the

as

the

quotes

or

better

available

on

Securities

quotes

are

then
The

with

NBBO

rule,

explicit

or

participants.

an

mercy

of

orders

to

NBBO

rule,

the

dearth

of liquidity

than

are

can

are

the
execute

rule,

at

the

orders

are

recorded

orders

can

be

1.

The

liquiditylimit
2.

The

achieved

using

can

compete
as

large
market
at

the

the

of

due

to

orders

to

a result,

available

top
Under

NBBO

market

As

limit

time.)

the

prices

order

to

at NBBO

at

guaranteed

trades

when

distinct

compete

priced

serving

two

all
were

are

match

is,
limit

to

traders
that

their

nationwide.

requirement.

can

that
their

can

orders

exchange

simultaneously

exchanges

within

exchange

NBBO

by

the

investors'

incoming

available.

possible

the

NBBO,

is

Securities

under

at the
at

NBBO

field:

were

execute

best

Regulation

route

available

exchanges

best

from

orders

to route

NBBO
lit

at the

limit

by

the

investors
to

cannot

required

on

exempt

NBBO

that

the

displayed

failed

quote

by

placed

be

ask

as

playing

individual

said

best

the

orders

executed

pools

when

the

are

where

market

Under

often

exchanges

exchanges

Dark

who
even

the

orders

rule,

broker-dealers,

marketbetter

placing

NBBO

to

and

aggregated

under

the

at

disseminated

forth

whether

has

bid

exchanges

2005

order,

investor,

to the

exchanges,

in

executed

best

The

put

of leveling

limit

individual

the

other

(SEC)

best

(Prior

was

quite

fees.

and

to

But

is

to be

the

(SIP).

rule

purpose

every

required

back

execution

participants.

optimal

exchanges

disseminated

Commission

the

institution

member

transparent

landscape

from

Processor

NBBO

Exchange

NMS,

the

are

compiled

fully

market

to set

equities

Information

references.
and

U.S.

is

the

compete

quotes,
all

book

category,

exchanges
the

order

to interested

markets

lit exchanges,

NBBO

the

lit

the

the

only

when

they

NBBO-priced

books.

Such

limit

NBBO

limit

approaches:
attract

the

top-of-the-book

or better.
to

a proprietary

attract
market

market
maker

orders,
posting

while
NBBO

limit

orders.

The

two

fee

business

structures

now

offer

liquidity.

2,

orders,

on

the
may

of

amounting

to

suppliers

whether

takers

readily

addition

for

exchange

pay

payments,

In

pricing

the

or

of exchanges

exchanges.

divergent

1 or

Such
as

of

Depending

model
limit

models

translate

clearing
of

liquidity

to negative

in

transaction

new

takers

the

providers

bringing

the

exchanges

and

follows

liquidity
for

fees,

liquidity

exchange

pay

into

of

business
for

posting

market

costs,

orders.

are

known

rebates.

The

two

two

distinct

their

different

fee

normal
away

camps,
structure.

liquidity,

market

order,

100-share

limit

determined

by

higher

the

displayed
3.6

and

pays

the

Fee

on

Source:

NYSE

web

The

higher

May

14,

site

of

exact

placing

2012,
Orders

volume

is shown
Placed

$0.13

The

the

on

following

fees:

orders,

taking

bringing

for

to

$0.20

NYSE
of

NYSE

Routed

in

100-share
for

rebate

the

in Figure
and

into

based

orders,

$0.21

of

rebate.

the

limit

from
value

exchanges

market

charges

monthly
the

offer

placing

anywhere

the

NYSE

for

example,

aggregate

Structure

for

the

exchanges,

exchanges

rebates

for

driven

inverted

traders

offer

order.

have

and

normal

NYSE,

volume,
online

normal

charge
and

The

models

The

exchanges

liquidity.

Figure

business

is

traderthe
fee

structure

3.6 .
to

and

from

In

contrast,

rebates

the

to remove

place

limit

example
for
while

traders
are

under

share.

Yet,

25,000
large

25,000

shares

$0.0014

per

BX

Figure
Source:

as
3.7

per
share.

of October
Fees

Nasdaq

on
web

per

orders

per

day

order

are

for

charged

The

snapshot

of

2012,

is shown

site

million

moreare

BX,

charge

an

an

orders

or more

shares

to

trading
3.7 .

Exchange

orders
the

per
limit

$0.0018

at

of

share,

displayed

rebates

Inverted

is

market

limit

in Figure

to

BX)

placing

distribution

fees

per

$0.0015

paid

small

$0.0005

adding

traderstraders

OMX

and

with

paid

Traders

or

Nasdaq

are

3.5

day

10,

traders

OMX

traders

day

share.

pay

(Nasdaq

There,

shares

per

limit

orders),

Exchange

market

$0.0014

exchanges
market

exchange.

million

placing

orders

the

3.5

paid

(place

Boston

inverted

than

inverted

liquidity

orders.

of an

fewer

day

so-called

rate
costs

per
for
of
on

As

result

among

of

equity

show

as

are

the

rates

different

exchanges

regulatory

exchanges

different

well

the

of

to

able

to

the

relationship

market

share

rates.

average
share
Figure
Source:

have

the

United
of

shares.

provide

displays

the

availability

market

able
ones

in

framework

As

the

secure
between

As
highest

shown

States,

NBBO
the

divergence

various

exchanges

their

Availability

Pragma

of NBBO

Securities

(2011)

versus

of

of NBBO

and

Share

Volume

quotes

Figure
rates

and
obtain

as
the

NBBO

share.

3.8 , Nasdaq

books

suggest,

availability

of trades.
3.8

order

would

market

NBBO

in Figure

availability

pricing

occurrence

highest
the

on

intuition

highest
the

and

and

NYSE
the

3.8
the
on

highest

Figure
all

3.9

or

illustrates

only

some

NBBO

are

volume

is

attracts

market

exchange

happens

exchanges.

When

available

offers

inverted

on

and

BYX,

EDGA,

NBBO

is available

the

trading

volume

with

asymmetric

order

advantage

for

Figure

3.9

When

NBBO

Source:

High

inverted

the
of

Is Present

the

exchange,

contributions

to the

bottom

normal

inverted

and

flow.

parent

company

Trades

Executed

high

BYX,

like

smaller

fees

and

available

solely

to

Direct

structure.
BYX,

NYSE,

BX

every

(BYX).

fee

When

and

EDGA,

although

Nasdaq

economic

structures

trade

structure

moves

BX,

The

of

at

incentives

clearly
creates

work:
the

the

first-taker

of BX.
at

Each

Trading

Venue

Venues

(2011)

of

of

the

by

at Certain

Securities

for

to

the

enjoyed

share

profitability

largely

low

ask

When

is not
share

but

on

a similarly

much

except

best

fee

orders,

NBBO

normal

the

liquidity.

market

price

flow

Percentage

market

the

percent

on

with

available

percent

Y-Exchange

but

When

and

inverted

limit

BATS

from

Nasdaq,

Pragma

the

is

30

NBBO

NBBO

everywhere

benefit

bid

the

with

moves

best

where

the

exchange

the

associated

30+

NBBO

BX,

structure,

the

the

over

liquidity.

an

also

to

the

seeking

has

fee

removing

BATS

BX

is routed

for

and

the

when

exchanges,

traders

for

rebates

Edge's

all

at

order

trades

BX

on

executed

except

number
BX,

what

trades

does
as

line

exchanges

not

other

of the

fees

alone
can

exchanges.

charge

traders'

contribute
provide
For

to
significant

example,

fees

the

for

both
routing

orders

to

the

other

current

trading
exchange.

proportion

of

orders

and

can

charging

fees

exchanhes,

but

time

the

rebate

this

been

even

remain
for

the

NBBO
an

not

available

exchange

profitable

routing

is

by

with

lowest

attracting

said

orders

elsewhere.

manner

similar

on

enough

in Futures
exchanges

the

when

Thus,

NBBO

Fragmentation
Futures

venues

operated

without

book

rebate

was

structure

operational

draw

the

written.

exchanges

risk

management

circuit

breakers

and

so

liquidity,

futures

matching,

sprung

up

unlike
have

earlier.

decisions,
described

at

and

Other

predominantly

on,

equity

exchanges

discussed
are

to

or foreseeable

to attract

equities,

futures
and

implemented

In a bid
in

pro-rata

among

their

pricing

prevalent

deploying

differences

in

based

such
in

as

detail

on

when

in

to

Chapter

13.
Fragmentation

in Options

Various

options

exchanges

majority

of the

exchanges,

are

driven

Due

to

and

strike

price

10

or so

same

the

options
The

by

the

with
lack

for

option

market

price

special

knowledge,

customers.

underlying
not

yet

different
are

A
and

options

trading

large

price

usually

on

away

from

the

a bet

rest

most

enables
in

represents

dates

traded.

trade

far

to the

expiration

liquidity

common.

revealed

decade.

principles

barely

being

strike

past

of the

equity

market
long-dated

the
on

The

current

someone's

markets.

in Forex
interdealer

access

Currenex

similar

to provide

information.
with

of the

in

on

options

spreads

activity

insider

or

Fragmentation

large

of

with

most
tend

the

makers.

options

combinations,

option

direct

of

over

operate

of market

broker-dealers

relative

In Forex,

group

number

markets,

an

however,

same

sheer

surveillance

have

to

and

brokers
end

iCAP,
In

begin

traders.
for

breaking

example,
with

competing

The

with

traditionally
now

other

accept
exchanges,

broker-dealers
interdealer

selected
iCAP

for
brokers

institutional
offers

250-microsecond
exchange

validity

(forex)

Fragmentation
While

fixed

to a potential
Thus,
using

the

target

income

has
Nasdaq

in Swaps
also

trading

venues,

(SEFs),

was

book

the

and

launch

by

jointly

The

traded

Act,

shaped

by

the

offering

with

would

industry

and

further

validity

of

all

negotiated

required

to

be

A new

class

of

exchanges.
swaps

engine

privately
are

as

market.

matching

swaps

on

point

fixed-income

iCAP

OTC,

known

signs

a one-second

electronically

collectively

was

foreign

some

a fixed-income

Dodd-Frank

traded

OTC,

of the

providing

traditionally

Under

standardized

top-of-the-book

traded

technology.

investors

Fragmentation

contracts.

to

OMX

quotes.

have

its

exchange-ization

top-of-the-book

this

traditionally

planned

institutional

Swaps

of

Income

near-term

iCAP

all

quotes.
in Fixed

most

on

execution
regulators

facilities
at the

time

written.

Summary
Modern

markets

streamlining

their

immediate,

and

systems.

has

and

delivering

to the

innovation

and

of

and

the

driver

offerings,

key

the

solutions

trading

products

with

the

The

evolution

exchanges

remains

concerned

most

competition
in methods,

alternative

trading

in development

are

becoming

customized

of
more

to

clients'

needs.

End-of-Chapter
What

of

customers.

producing

execution

goal

to their

models

leaner

the

ever

service

technology

customer-centric,

1.

with

led

service

While

businesses

operations

venues

pricing,

unique

complex

cost-effective

of trading

faster

are

Questions
is

the

difference

between

the

normal

to buy

a large

and

inverted

exchanges?
2. What

type

3.

are

You

liquidation

of orders
given

(selling)

the
of

would
task
a large

you
of

use

developing

equity

position.

an

quantity?

algorithm

Your

aim

Why?
optimizing

is to

develop

an

algorithm

of

liquidation

giving

that

maximizes

process.

normal

execution

How

and

would

inverted

costs
you

price

while

minimizing

develop

such

structures

of

speed

an

algorithm

modern

equity

a certain

futures

exchanges?
4.

You

are

contract

receiving

from

liquidity

contracts.

The

reported

next

13:45:00:01:075
following

tick
with

market

from

5.

current

The

stock

on

Ask:
What

as

35.67?

6. Is an
7.

The

The
a

At

at the

bid

data

given

size

last

of 250

of

per

share

order

passive

best
to

buy

bid

limit

the

best

bid

happened

in

the

Best

book

are

for

likely

shares

particular

Bid

Best
at

to receive

Best

is 35.65

information:
100

National

Bid

GMT?

following

you

of

at 12.7962.

citing

has

order

at 35.68,

if the

35.69.
on

your

is advertised

instead?

or aggressive?
price

is 2.65
10

quote

325

a timestamp

contracts

What

the

shares

National

trade

the

200

shares

with

of 900

the

comprises

to 13:45:00:01:095

shows

if the

recorded

order

side

price

print

contracts.

GMT,

of 12.7962

GMT

GMT

exchange

order

iceberg

a trade

on

13:45:00:01:060

is a trade

of 325

at 35.67,

What

best

recording

right-hand

prevailing

limit

the

average

buy

II data

is a 13:45:00:01:095

shares

is the

market

of

Level

exchange.

13:45:00:01:060

100

tick

GMT,

of 12.7962

I and

a well-known

aggregate

The

Level

in

and

a given
the

contracts

at

options

prevailing
2.85

market

is

ask

is 2.90.

best

passive

or

2.83.

aggressive?

Why?
8.

A quantitative

researcher

using

daily

closing

trader

use

to execute

prices.
the

(quant)
What
quant's

develops
order

types

buy-and-sell

his

investing

should
decisions?

an

models
execution

Is

CHAPTER

High-Frequency
Trade

Data

and

formats.
best

quote

best

Level

including

new

from

market

the

quotations

include

ask

II
limit

and

This

market

Every

time

limit

order

activity.

of X.
as

Market

When
all

the

other

the
posts

other
best

ask

asks

time

match
time

on

the
such

tick

the

order

book,

at

prices

away

details

contrasts

data,

the

are

of

data

data

with

records

a dealer,

s units

or

at time

into

tick

data

has

the

of

another

of a specific

is logged

quote
quotes

time

sell

live
entity

security

with

t b to buy

units

in a different

highest

in force,

s units

of

X at

units

of

X.

quotes

for

security

price

becomes

t b . Similarly,

sell

quotes

venue

the

from

where

directly

when

where

way,

when

price

If the

q , an

latest

X,

relative

to

known

as

trading
ask

entity

quote

is lower

becomes

is

than

known

as

all
the

ta .
to

the

as

to buy

bid
at

on

participants

situations
force

ask

happens

exchange

size,

the

and

a customer,

bid

to

t a to

posted

describes

price,

and

cancellations

chapter

known

arrived

order

on

to

ask

II

chapter.

available

at time

depends

best

price

changes

and

incorporated

arrived

available

What

at

newly

bid

at

all

quogte

are

in this

a limit

logged

q , a bid

previously

best

also

orders

discussed

price,

Level

Data?

data,

X at price

bid

I or

counterparts.

High-frequency

ticker

in Level

trade

methodologies

Is High-Frequency

a so-called

last

arrivals

sampling

posts

and

best

include

order
price.

distributed

the

size,

quotes

low-frequency

What

is often

I quotes

size,

available.

their

information

Level
bid

an

and
new

best

exchange,
quotes,

t b . Conversely,

the

should

moment

orders

are

exchange
other

exceeds
most

a trade
the

newly

they

arrive

posted.

will

parties

bid
,

executing

the

be

Best

exchanges
at the
arrived

bids

broadcast

tracking
the

largely

quote

best

ask
will

to

all

data.

In

already

in

immediately

preexisting
best

and

best
ask

fall

ask,
below

the

current

best

bid,

Most

best
at time

dark

may

bid,

not

match

broadcast
the

interdealer

Market

orders

an

to

buy

at

the

order

available
As

such,

bid
at the

bid,

the

and

trade

best

quotes,

if the
the

trading
the

When
size

of

best

market

the
the

ask

in

is

the

best

and

increments

of

exchange,

best

next

The

market

orders

certain
standard

each
sell

last

buy

best

venue.
the

market

best

buy
with

data

is
price

trading

matched

order

at

quote

matched

trade

number
trading

data

still

order
the

as

in

best

the

been

matched

market

buy

the

then

last

ask

on
are

the

lot

order

trading

that

the

market
than

consumes
be

the

matched

size

of

ask

the
quote

venue.
in

so-called

known
today

reflect

is bigger

to

of

most

order

until

that
on

with

lowest-priced

placed
units,

than

to

market

several
example,

force

reduced

proceeds

best

or

For

size
in

has

ask,

be

one

points.

remain

remaining

of

with

is smaller

available

available

other

to

the

the

will

and

ask

at

size

best

the

fulfilled.

be

incoming

entirety,

the

of

a market

on

the

may

quote

its

form

immediately

in

order

ask

of the

for

order

a security

market

quote

ask

mysterious

to

the

an

price,

with

last

buy

best

in

is

is posted

to

several

the

but

size.

need

ask

that

of

recorded

may

the
destined

data

executed

trade

spread,

disseminated

a certain

each

corresponding

order

limit

last

size

with

becomes

and
is

best

but

of the

preexisting

venue.

prices,

market

the

at

are

ask

(hence

be

order

order

ask

order

ask,

sequentially

Most

the

arrived

portion

the

best

venues,

order.
the

best

the

quantity
the

generating

newly

best

a security

orders

market

the

the

quotes

not

limit

a specified

and

at

crossing

quotes

may

moment

by

high-frequency

Unlike
of

asks

on

to

transaction

price

large

or

depending

or

executed

executed

Similarly,

may

market

available

is

arrived

pools).

information.
quantity

and

newly

contribute

specified

the

networks
participants,

trade

bids

dark

market

trade

ta .

pools

moniker,

, the

as
is

a
US$5

lot.

lot

sizes:

In

foreign

million,

considerable

reduction

high-profile
can

brokers

be

share

as

low

just

as

one

minimum

other

than

Small

an

and

broker-dealer

at the

delaying

but

dark

for

orders.

quotes

Recorded?

The

highest-frequency

data

are

information.

latest

Tick

data

the

an

to

100

amount

odd

lot.

through

by

obtain

process,

to an

lot

the

volume
the

brokers

executing

venue,

orders.

Data

the

for

packaged,

them

enforce

posted

In the

transmitting

still

order

by

exchanges,

is called

orders

venue.

Is High-Frequency

of

size,

in

entertained

order

or

How

arrivals

may

An

orders

of customers'

million
equity

pools

lot

execution

without

execution

On

aggregated,

larger-size

orders'

ago.

odd

be

of $25

of a lot

market

into

on

years

increment

may

sit

a few
share,

integer

broker-dealer

discounts

a minimum

requirement

limit

may

from

quote,

trade,

usually

has

collection

of

price,
the

sequential

order

following

size,

ticks,

and

volume

properties:

A timestamp
A financial
An

security

indicator

Bid

identification

of what

it carries:

price

Available

bid

size

Available

ask

size

Last

trade

price

Last

trade

size

Security-specific
The

market
price,

may

released
received
round-trip

data,
value

the
the

the

time

at

which

travel

At
time

volatility
as

the

for

actual

options
numerical

value

of

or size

date

or

implied
such

the

quote,
quote.

as

volume,

records

be

such

information,

available

A timestamp
It

information

price

Ask

the

code

the
the

of an

and

time

the

exchange

time
time
order

at which

when

this

or
the

article

quote

the

from

is

quote
the

broker-dealer

trading
written,
the

originated.

system
the

ordering

has

standard
customer

to the

exchange

of order
been

and

receipt

is 15

known

process

part

to

at

Another

of

the

The

the

customer's

last

executed
best

the

market.

security
ask,
the

Last

size

are

have

unable

to

quotation

microseconds

price

at

can

when

financial

as

or,

a ticker

followed

code

date,

and

which

the

differ

by

trade

size

the

can

last

by

the

consist

of

code.

bid

in

the

and

ask.

a favorable

without

shows

In

tickers

trade

the

posts

broker

for

exchange

from

a customer

matched

security.

ticker,

identification

price

Last

expiration

trade

is the
The

any

highest

best

ask

systems
quotes

than

the

best
to

Market

only.

each
depth

and

time.

In

the

may

bid

limit

broadcasting

actual

(Level

ask,

price

addition

to

as

ask

size

is

disseminated
the

best

I data)

on

the

well

as

bid,

is often

the

best

the

of

the

best

trading

bid

the

available

for

the
best

at prices

trading

premium
last

and

order

referred

ask,

buy

venue

the

in

information:

aggregate
on

security
to

depth

sometimes
as

of

entered

market

recorded

information

sale

lowest

posted

and

be

for

disseminate

and

In contrast,

information

the

entered
bid

available

is

may

ask

II data

price

particular

and

service

the

the

arise

be

the

exchanges,

futures,

quote.

corresponding

Level

even

of

can

multiple

shows

can

bid

books.

they

Sophisticated

and

identifier

futures

price

quotation

worse

Brokers

trade.

at

bid

is an

is immediately

The

acknowledgment

York.

if

speed.

code

security,

differences
that

in New

milliseconds

on

For

cleared.

order

the

customers

standard

quote

traded

trade

security

now

with

or less
their

include

symbol.

last

by

identification

underlying

The

this

the

simultaneously
exchange

customer

timestamps.

part

equities,

the

fired

therefore,

of their

to the

milliseconds

be

orders

systems,

back

sizes
venue's

to

as

the

subscription

trade

price,

a small

and

nominal

fee.
Panels

and

high-frequency
Arca

for

Standard

of

Figure

data

recorded
&

Poor's

4.1

illustrate
by

New

Depositary

a 30-second
York

Stock

Receipts

log
Exchange
(SPDR)

of

Level

(NYSE)
S&P

500

exchange-traded

fund

14:02:00:000
data:

GMT

best

bid,

corresponding

and

trade

last

Figure

4.1
from

Data

source:

and

are

of

data

have

2009.

Panel

trade

sizes

14:00:16:400
A

(best

bid

to

shows

information,

Data

while

size,

Recorded

to 14:02:00:000

securities
still

best

quote
panel

ask

B
size,

GMT

by

on

NYSE

November

Arca
9,

for
2009.

financial

advantageous
properties

Data
data

something

Unlike

much

last

position

of High-Frequency

they

from

Bloomberg

practitioners.

the

ask

SPY)
9,

I High-Frequency

High-frequency
Yet

November

14:00:16:400

Properties

ticker

size).

Level

SPY

on

best

displays

(ETF;

daily
research

distinct
and
of

have
of

or

been
novelty

monthly

and

properties,
intimidating
high-frequency

to

data

related

for

many

sets

to researchers.

many

commonly

Each

and
used

in

high-frequency

simultaneously
Table

years.

academics

applications,

which

data.

studied

can
4.1

property

be

summarizes
and

its

advantages

and

disadvantages

are

discussed

in

detail

later

in

the

article.
Table

4.1

Summary

High-Frequency
The

nearly

4.1

contained

quotes
and

Data

Are

two-minute

and

Table
by

2,000
best

4.2

NYSE

Arca

2009,

and

SPY,

the

day

Table

4.2

call

option

throughout

call
for

option.
SPY

omitted
SPY
The

alone

and

from

Japanese

number
would

the
from

euro

data

of

yen
quotes

comprise

over

for
of

and

SPY

9,

for

quotes

for

November

volume

tick

summarizes

on

I data

of

observations

ask

GMT

Level

of High-Frequency

Data

Voluminous

sample

over
sizes,

sizes.

provided

of Properties

SPY

shown

Level

sizes,

and

breakdown

data:
last

of

14:00:16:400

on

futures,
observed
160

yen

November

include

the
to

Japanese

on
years

best

bid

trade

prices

data

points

futures,
9,

greeks
November
of

Figure

14:02:00:000

cumulative
and

in

daily

and

2009.

Other

daily
for

the
9,

open,

trade
euro
2009,
high,

low,

close,

and

trading

days

Table

4.2

per

on

quality

reliable

data

ticker

does

tape

one

not

information

on

is

as

known

an

average

always

required

every

as

the

Level

match

its

data

on

bids,

to

archive

accurate

are

tick

minute

known

for

provide

of

within

assuming

of

252

Quotes

for

Selected

9, 2009

exchanges

records

quotes

Statistics

generally

equity

points,

year.

November

of

exchanges
U.S.

data

Summary

Securities

The

volume

of

the

limit

of

well

beyond

and

can

Centralized

asks,

and

and

as

to

Information

book

data

as

occurrence

Securities

II

law

data,

their

order

Level

by

quantity.

to

the

maintain

submit

U.S.

best

bid

be

obtained

best

centralized

Processor

the

size.

(SIP).

The

best

offer

and
via

special

subscription.
In

decentralized

interbank

money

given

time.

price

levels,

order

book.

tick

markets,

data

In
but

to

financial

quotes

from

There
quote

efficiency
are

at

different
of the
generally

discrepancies.

own

inventory.

$100

million

For
of

each

result,

given

and

dealers

and

USD/CAD

are

and

available

at

are

aware

of

own

prices

adjusted

dealer

may

quote
vary

Ridder,

any

current
for
his

on

its

own

from

among

disseminate

the

provides

specific

the

given

dealer

others,

to

collect

them

back,

improving

anomalies

in

interdealer

markets.

thought

example,

time

Knight

decentralized

First,

its

markets,

any

exchange

quotes

quotes

As

Telerate,

foreign

participants

institution

clients.

as

market-wide

decentralized

instrument
Reuters,

no

markets,

each

his

dealer.

the

market,
such

In

such

to
each

be

dealer's

a dealer
would

three

be

quotes
that

eager

has
to

reflect
just

diversify

that

sold
the

dealer's
customer

risk

of

his

position

and

however,

are

quotes.

To

dealer

to

dealer

incite

financial

bid

ask

quote

and

ask

prices

and

in an
as

sending
price

other

the

Dacorogna
market

to a minute.
dealers.

prices.

Some

The

advertise

their

particularly

true

telephone,

allowing

Fast-paced
quality

addition

to

and

The

difference

time

is known

ask

buy

the

tend

to

particular

whenever

a dark

the

same

orders,

in a

market

off

the

dealers

prices

the

they

pool,

dealers

information

by

previously

quoted

or supply.
that

is thought

quote

moving

provide

At

short

as
for

much

note

some

dealers'

to

vary

in
prices

were

delay

usually

data

discourage

lagged

so

This

negotiated
quotes

of other
do

feed.

between

may

milliseconds

of quotes

quotes

the

quotes

from

averages

delayed

order

markets

cover

Are

trade

their

spread,
to

are

and

fish

are

a considerable

Subject

price

and

to
was

over
and

quotes,

introduce

the

the

the

orders.

improving

buying
the
the

higher
spread

data
Bid

noise

in the

quote

spread.
and
the
along

selling
gain
with

and

the

The

bid-ask

the
other

available
bid

ask

and

data

in
ask
arrive

process.

and

the

long

comprise

sizes.

quote

Bounce

data

order

bid

bid-ask

Bid-Ask

volume

high-frequency

between
as

to the

and

associated

instantaneously

order

bid

presence

formats,

asynchronously

bid-ask

they

lag

most

Data

low-frequency
quotes

Thus,

tradable

of markets.

High-Frequency
In

USD/CAD.

may

(2001)

who

electronic

USD/CAD.
placing

demand

dealers

when

USD/CAD,

such

that

market

on

marketplace,

The

dealers

orders

instrument.

makers

al.

on

withhold

both

dealers,

clients

on

whenever

available
et

to

Most

their

in a financial

quotes

to assess

real

the

market

indicative

Third,
lag

long

sell

quote

on

USD/CAD.

with

bid

lower

anonymous

of

place

clients

the

disproportionately

well

to

the

his

more

transact

clients

raises

instrument

Second,

of

to

his

encourage

both

any

obligated

raises

raise

as

selling

temporarily

time,

are

avoid

ask

quote
spread

security.
security

The
must

transaction

at any

given

is the

cost

higher

the

produce
costs.

in
Most

low-frequency
spread

price

negligible

price

in

changes

Bid-ask

can

spreads

illustrates

the

institutional
2008.

changes

be

usually
average

Tokyo

The

then

reaches

New

York

and

active
of

buyers

October

and

for

the

shows,

trading

sellers.
2008,

subpoenas

issued

on

executives

in a case

reflects

relating

to

the

spread.

Figure

spread

of

October

is

of

has
the

concern

many

over

fraud

the

weekend

senior

securities

quiet.

overlap

over

the

increases

market

to

4.2
in

market
the

the

2009,

potential

bid-ask

spread

market

17,

incremental

weeks

during

the

October

two

when

bid-ask

observed

when

in the

the

day.

average

levels

spike

the

cycles

sessions,

The

1819,

than
the

hours,

lowest

make

however,

last

the

trading
its

data,

spread

market

during

tick

to

throughout

bid-ask

significantly

enough

or smaller

vary

4.2

London

In

comparable

Figure

spread

large

comparison.

EUR/USD
As

are

the

Lehman
at

Lehman

Spot

for

the

Size

of

Brothers.
Figure
Last
US$5

4.2
Two

Average
Weeks

Hourly
of

Bid-Ask

October

Spread

2008

on

on

EUR/USD

a Median

Transaction

Million

Bid-ask
uncertainty
bid-ask
JulyAugust

spreads
or

typically

instability.

spreads
2008

on

increase
Figure

EUR/USD

and

the

4.3 , for
in

crisis

during

the

periods

example,
stable

conditions

compares
market

of

of

market
average

conditions

SeptemberOctober

of

2008.
in

As

both

the

figure

crisis

and

significantly
at

all

increase
EUR/USD
pips)

not

and

New

the

Tokyo
three

at

hours

hours

trading

trading

by
12

hours,

times

the

4.3
of

GMT

the

and

average

is

the

during

shows,

day.

16

spread

increased

increase

during

spreads

are

conditions
the

spread

The

average

hourly

(0.48

basis

GMT,

From

persistent

normal

also

percent

overlap.

the

but

than

0.0048

sessions

pattern

conditions,

Figure

all

of

spread

months

As

increased

the

York

crisis

day.

uniform

spreads

between

over

the

intraday

market

during

of

is

the

normal

higher

hours

shows,

when

0 to
by
the

points

or

the

London

2 GMT,

during

0.0156

New

percent,

York/London

hours.
Figure

4.3

Hours

of

Comparison
the

of

Day

Average

during

Bid-Ask

Normal

Spreads

Market

for

Conditions

Different

and

Crisis

Conditions

As

result

uncertainty

of
and

increasing
crises,

decreases

during

EUR/USD

strategies

higher
normal
Asian

costs
market
hours

during

the

running

of

times.

For

over

Asian

September

resulted

spreads

profitability

those

conditions.
alone

bid-ask

and

A strategy
in

1.56

periods

high-frequency

hours

executed

percent

high-frequency

incurred

2008

of

strategies

example,

October
that

during

as

significantly
compared

100

evaporating

trades
from

with
during
daily

profits

due

to the

during

London

significant
even

and

daily

more

increased
New

profit
for

instruments.

For

example,

on

increased

comparison
While

with

tick

distorted
the

first

price

bounces

execution

of orders

high-frequency
(2001),

example,
bias

the

bounce

bid-ask

autocorrelation

bid-ask

In

addition

forecasting

spread

and
bid-ask

by

Roll

(1984),

to

equal

an

equal

to

when

the

a sell,

as

for

can
the

unobservable

next

and

(2001)
by

to

bid-ask

Dacorogna
a

calculate

percent
as

of

introduces

authors

bias

estimate

well

filtering

that

first-order
as

Voev

the

be

estimated

price

of an

bid-ask

market

order

in equation

of

time.

using
asset

and

data

from

s.

is a buy,

and

Future

bid-ask
realizations

model

at time

The

researchers

impending

the

value,

spread,

data,

the

ahead

fundamental

the

shown

to

it in models

where

of

the

al.

market

estimation

bounce

40

in
trade

during
into

also

valuable

sequential

Muller,

in

in

it is

so

quotes

The

adjustments

spread

half

et

shown)

to estimation.

techniques

adjust

the

Corsi

data
that

bid-ask

results

remedy

real-time

deploy

of

average

to

ask

estimates.

liquid

(not

dynamics,

distortions

the

be

2008.

the

Zumbach,

that

data.

prior

to

Corsi,

on
tick

and

can
less

NZD/USD

report

significant

volatility

propose
noise

bid

show

into

of

(2007)

the

the

for

still

SeptemberOctober

make

(2001)

but

situation

built

market

that
al.

introduce

considerable

Lunde

processes

parameters.

for

about

running

smaller

The

for

during

strategy

of JulyAugust

information

between

in

percent.

spreads

times

et

same

strategies

bid-ask

Dacorogna

the

resulted

of 0.48

conditions

same

place.

hours

three

carry

the

while

high-frequency

market

data
by

York

decrease

severe

average

spreads,

suggested

t , p t , is

m t , offset
price

offset

negative

when

assumed

by
is
the

value

positive
trade

(3):

(3)

If either
then

a buy

E [ I t ] = 0,

or
and

a sell
E [

order

can

p t ] = 0,

arrive

absent

next
changes

with

equal
in

the

probability,
fundamental

is

asset

value,

however,

m t.

The

is different

covariance

from

of

subsequent

price

changes,

0:

(4)
As

a result,

the

Numerous

expected

extensions

account
other

future

for

of

Roll's

contemporary

variables.

spread

model

market

Hasbrouck

can

be

as

been

developed

have

conditions

(2007)

estimated

along

provides

with

a good

follows:

to

numerous

summary

of

the

of

the

data

to

models.
To

use

bid-ask

standard

econometric

bounce,

many

midquote
quotes.
the
to

format:
The

market
meet

the

simple

other

midquote

can

of

willing
on

expressed

to trade
the
as

presence
the

the

to approximate

halfway
be

the

convert

average

is used

is theoretically
each

in

practitioners

the

midquote

techniques

tick

latest

bid

and

the

price

level

if buyers

and

sellers

price

spectrum.

ask

at which
agreed

Mathematically,

follows:

(5)
The

latter

condition

mid-quote
best

ask

Another

for

estimate:
quote,
way

weighing

tm

is updated
, arrives,

to

sample

the

accompanying

reflects

whenever

quotes

best

order

continuous
the

updating

latest

best

of

the

bid,

or

at t b or t a respectively.
tick

latest

the

bid

into
and

a cohesive
best

data

series

ask

quotes

best

bid

by

is

by
their

sizes:

(6)

where

and

recorded
the

at

best

bid

is the
time

best

bid

t b (when

quote

and

quote
became

the

best

bid

and
the
available

the
best

bid),

size

available
and

recorded

size

and

is

at time

ta

.
High-Frequency
Many
diffusion

classical
models

Data

Are

models
to

work

Not

assume

Normal

or Lognormal

lognormality

seamlessly,

and

of

prices,

resulting

allowing

in several

price
pricing

models,
of

such

market

for

sequential
in

of

do

trade

therefore,

not

ticks

the

prices

As
tick

section

data,

like

best

close

choice

fair

the

shows,

necessary

distribution

however,

midquotes

and

distribution,

to

of

that

normal

normal

for

approximations

instruments.
is

this

follow
is

considered

financial

of

the

trades,

sequential

are,

related

changes.

most

and

of

to be

lognormality

price

changes
quotes

Black-Scholes,

evolutions

condition

of

as

modelers

sequential
size-weighted

yet

normal.

of

distribution

Trade

tick

assuming

data

lognormal

prices.
Figure
from

4.4

compares

midquote

trade-price
as

neglect

the

the

the

quantiles

Figure

4.4

(Panel

A),

as

SPY

Midquote
Data

the

plots

Simple

the

A:

Midquote

Panel

B:

Size-weighted

Panel

C:

Last

trade

simple

price

simple

and

500

ETF

data

9,

2009.

The

data

quotes,

same

data

treating
Figure

4.5

sets

with

distribution.
Returns

Computed

(Panel

Recorded

B)
as

and
They

returns
midquote

B)

S&P

observation.
of

computed

(panel

adjacent

9, 2009

Panel

returns

November

independent

normal

of

Size-Weighted
for

an

distribution

Histograms

SPDR

between

of a standard

Processes

November

quote

simple
midquote

for

throughout

difference

quantile

of

size-weighted

processes

arrive

time

histograms

A),

C)

they

sequential

contrasts

C)

(panel
(panel

recorded

each

the

simple
returns

returns

from
Trade
Arrive

Midquote
Price

(Panel

throughout

As

Figures

4.4

constrained

by

midquote
tick

can

size

most

and
the

4.5

present
Figure

in all three

types

Quantile

November

notion

Midquote
for

SPY

the

A:

Midquote

Panel

B:

Size-weighted

Simple

Panel

C:

Trade

B)

returns

price

returns

changes

present,
midquote

three

Returns

Recorded

midquote

minimum
(at

and

9, 2009

Panel

distribution

distributions

also

illustrates

the

is
in the

minimum
forms

the

discussed.
the

fat

tails

distributions.

(Panel
Data

midquote

size-weighted

further

of data
of

the

increments

among

Plots

basic

size:

The

distribution
this

Size-Weighted
Processes

at half-tick

confirms

4.5

the

step

in equities).

continuous

Figure

show,

minimum

occur

is $0.01

4.5

returns

and
as

of

Midquote

Trade-Price
They

Arrive

(Panel

A),

(Panel

C)

throughout

Figure

4.6 Data-Sampling

A.

Last-tick

B.

Interpolated

sampling

sampling

Methodologies

As

clearly

shown

tick-by-tick
when

trade

heavy

values

past

normal

just
up

with

regularly

standard

deviations.

other
on

Relative

availability
since
ease

contrast,

time
to

sample
hour

from
the

tick
bid

last

on.

to

or ask

would
Figure

be

that

constant,

which

Figure
15-s

For

for

4.7
Intervals

to work

weekly,

daily,

reliance

have

of

published

daily

given

security

as

the

the

absence
not

have

Closing

short

are
the

and

time

returns,
intervals.

separated

irregularities

periods

of

if the

data

under

the

minute

the

current
this
of
to be
Quotes

the

be

converted

If

as

no

minute's

closing
This

approach,

determined

previous

idea.
new

to

prices,

the

the

quotes
closing
and

approach

quotes,

are

example,

traditional
be

varying
data

timefor

minute.

minute's

by
in the

are

would

particular

then

A illustrates

prices

8081)

that

minute,

data.

over

then

during

taken

Midquote

pp.

example,

any

certain

does

drove

bars,

arrived

in

follow

developed

traditional

spaced

predetermined

4.7 , panel

assumes

returns

monthly,

(newspapers

to overcome

minute

from

to:

observations

way

price

that

The

significantly

1997,

certain

during

prices

vary

minute.

data

quote

arrived

not

One

or

intervals.

whatever

O'Hara

it at

in

regularly

high-frequency

intervals.

every

that

did
and

trade

been

presented

data

deviate

1920s).

view

(Goodhart

daily

to

Midquote

in Time

have

is due

distribution,

ignored.

begin

Spaced

intervals

of processing

probably

By

time
of

the

outdated

it

fixed

normal

while

techniques

consistent

researchers

An

Irregularly

methodologies,

are

alike

to four

data,

the

deviations

deviations,

spaced

three

fit

midquotes

computational

or

Relative

standard

Are

the

closely

standard

Data

modern

quotes

four

of

two

High-Frequency
Most

4.5 ,

most

size-weighted

at

hourly,

Figure

returns

tails

and

normality
the

in

so

implicitly
prices

stay

case.

Sampled

at

200-ms

(top)

and

Dacorogna

et

sample

quotes:

quotes.

At

that

at

any

illustrates

linear

shown

methods

core

given
two

in

propose

of

the

time,

Figure
quite

unobserved

quotes

precise

between

technique

observed

lie

is

on

quotes.

an

to

adjacent
assumption

a straight

Figure

way

4.6 ,

line
panel

that
B

sampling.

4.6 , panels

different

results.

two

more

interpolation

interpolation

interpolation

the

a potentially

time-weighted

neighboring

produce

Mathematically,

(2001)
linear

the

connects

As

al.

sampling

and

B,

the

methods

two

can

quote-sampling

be

expressed

desired

sampling

as

follows:
Quote

sampling

using

closing

sampling

using

linear

is the

resulting

prices:

(7)
Quote

interpolation:

(8)
where

sampled

quote,

t is the

time

(start

of

last

a new

observed

of

the

last

of

the

first

the

minute,
quote

quote

value

prior

observed
of the

4.7

and

sampled

as

closing

and

15

s.

Figure

interpolated

from

prices,

as

Figure

4.8

Figure

to

the

4.8

Midquote

4.9

sampling
after

after

and

As

at lower

is

the

t , t next
time

sampling

time

histograms

of

interpolated,

plots

4.7

and

contain

frequencies.
are

4.8

q t,next

is

t.
midquote

data

of 200

ms

closing

prices

show,

oft-sampled

zero

At

same

more

value

t , and

more
the

is the

the

timestamp

the

of

of

is the

at frequencies

quantile

quotes

timestamp

t , q t,last

sampling

the

is,

the

time
time

Figures

that

interpolated
4.9

t last

sampling

compares

sparse,

Figure

15-s

the

compare

prices

sampled

computed

and

to

quote

4.9

are

distributions

(top)

prior

distributions.

distributions

example),

quote

first

Figures

for

continuous

returns

and

than

time,

returns

than

closing

at

200-ms

illustrates.
Time-Interpolated

Quotes

Sampled

Intervals

Quantile

Plots:

Closing

Prices

versus

Interpolated

Midquotes

Sampled

Instead

of

manipulating

regularly

spaced

the

distance

time

information.

are

ms

the

formats,

several

carry
the

yet-to-be-observed

intervals

most

the

good

news

quote

on

and

duration,
the

that

the

carries
intertrade

which
the

higher

whether

itself

agree
for

convenient

studied

arrivals

securities

intertrade

the

have

researchers

information
lower

into

researchers

subsequent

example,

indeed

disallowed;

interquote

between

For

intervals

at 200

short

more

sales

likely

impending

the
price

change.
Duration

models

between

any

processes
also

known

two

and

used

to

prespecified
trade

are

as

as

estimate
ticks.

the
well

increments.
processes;

time
as

the

The

the

Such

processes,

measure

price

to

sequential
trade

size,

volume

used

factors

models

are

respectively.
elapsed
time

the

between

known

price

between

working

models

the

Duration

interval

models

affecting

with

estimating

as

time
quote

models
changes

are
of

predetermined
fixed

price

variation

are
in

duration

of fixed

Durations
that

are

often

sequential

another.
+

volume

occur

hold

per

rate

of

constant,

constant,
times

or

the
t and

).

time.
The

it may

(t +

occur

between

of

any

two

time.

observing

processes.
that

assume

time

of one

points

t and

In a Poisson

words,
arrival

with

volume

independently

distribution.

In other

vary

as

processes

arrivals,

average

probability

known

Poisson

a Poisson

unit

(1/

quote

of arrivals

to have

are

using

like

number

) is assumed

average

modeled

events,

The

arrivals

increments

the

arrivals

rate

may

If the

average

exactly

process,

occur

be

(t

at

an

assumed
arrival

to

rate

arrivals

is

between

) is

(9)
Diamond

and

the

to

first

carries

that

news;

disallowed,
likelihood

of

markets

limited

duration

probability

of
however,

Easley

and

O'Hara
a

than

implications

of

conveys

whenever

possible,

and

news.

short

selling

the

is

higher

also

the

holds:

in

levels,

arrivals,

of

presence

liquidity

trade

further

point

have

in

the

complete

the

higher

the

absence

of

that

are

trades

different

information

proximity.
is

should

be

the

that

much

(1992)

and

strengthening

out

close

O'Hara

information

the

reverse

normal

ticks

of news.

occurring

Easley

sequence

and

a lack

interval

trades

The

were

presence

indicate

duration,

news.

bad

the

where

subsequent

(1992)
time

securities

selling

between

indicates

by

content

short

in

can

(1992)

subsequent

that

duration

good

O'Hara

between

intertrade

yet-unobserved

trades,

separated

the

and

posit

of

unobserved

with
the

models

markets

shorter

Easley

duration

intertrade

in

the

and

the

Their

constraints,

good

longer

(1987)

suggest

information.

short-sale
of

Verrecchia

One

that

the

used

argument

for

of

the

entire
in

its

price
entirety

high-frequency

trading.
Table

4.3

computed
4.3

shows
on

illustrates,

all

summary
trades

the

recorded
average

statistics
for

SPY

intertrade

for
on

a
May

duration

duration

measure

13,

As

was

2009.
the

Table
longest

outside

of

regular

preceding
Table

the
4.3

market

Hourly

May

13,

2009,

The

variation

several

liquidity,

trading

traders.

Foucault,
liquidity

strategy,

as

hit

by

Saar

(2002)

addition

price

hour

ET).

subsequent

trades

lack

that

may

Observed

orders

of

lower
and

on

spreads

in
time

intertrade

are

be

lack

levels

that

of
of

patiently
trading
for

form

of

the

durations

the

to

motivations

until

actually

contrasting

to low

a profitable

the

(2000)

due

to a

compensated

comes

Engle

due

be

due

consider

be

waiting

be

strategic

itself

should

the

be

(2005)
may

may

may

also

and

usually

durations,

of trading

Kandel

providers

Dufour

durations

have
prices

changes

the

order

their
bid-ask
limit

induce

and

Hasbrouck

higher

when

time-based

is

lower
and
traders

limit

order

hypothesis.
to

researchers

limit

takers;

short

p.m.

during

Duration

exchanges,
and

function

find

compensation

security

using

shortest

Intertrade

inactivity

on

Kadan,

However,

observe

In

trading

liquidity

spreads.

the

compensation
is

the

to 4:00
of

While

liquidity

The

(3:00

between

halts

providing

and

SPY

causes.

information,

and

close

in duration

of new

spread

hours,

Distributions

for

other

waiting.

market

been

and
of

between
modeling

volumes.

a specified

subsequent
durations

The

time

magnitude

trades

between
interval
is

known

and

fixed

between
as

quotes,

changes
subsequent

price

duration.

in

Price

duration

has

volatility.

Similarly,

changes

of

Volume

been
the

shown
time

has

decrease

interval

prespecified

duration

to

size
been

with

between

is

known

shown

to

increases

subsequent

as

the

volume

volume

decrease

with

in

duration.

increases

in

liquidity.
Using
de

a variant

Prado,

of

and

the

O'Hara

high-frequency

data.

define

unit

futures

clock

contracts.

is filled.

as

the

clock

twice

advances
information

biases

time

itself

estimates

of

Most

the

the

High-Frequency

Neither

Level

whether

a given

of

Level

recorded

a market

sell

order,

identifiers

as

inputs

yet

Do
II

some
the

models.

four

methodologies

have

been

proposed

was

a buy

from

Level

I data:

Tick

or a sell

too

high

in

Identifiers

of a market
for

overcome

to estimate

specifying
buy

order

buy-and-sell
this
whether

or

trade
challenge,
a

trade

rule

Lee-Ready

The

are

traditional

rule

Quote

Bulk

introducing

identifiers

call
To

the

series.

contain

applications

If

subsequent

Buy-and-Sell

a result

durations

a result,

prices

data

was

volume
however.

As

Contain

50
volume

variable,

price

bucket

whenever

between

independent

Not

trade

into

time

of the

tick

and

process,

transaction

the

50

is executed.

price,

estimation.

variance

Data

I nor

into

say

50-contract

trade

the
an

volume,

advances

of

researchers

whenever

The

trade,

be

bias

true

clock

estimation

to

variance

ticks

a 100-contract

the

trade

sequence.

of quote,

ceases

endogeneity

with

in

determines

substantial

comparison

then

volume

when

into

certain

Lopez

sampling

approach,

of

clock

arrive

information

trades,

bucket

Easley,

volume-based

volume-based

volume

content

methodology,

propose

50-contract

trades

available

The

Thus,

introduces

(2011)
In the

single-contract

The

volume-duration

rule

volume
tick

rule

classification
is one

of the

three

most

popular

methodologies

used

to

determine
in

the

whether
absence

popular
Lee

a given
of

such

methods
and

are

Ready

classification

trade

was

initiated

information

the

in

quote

(1991).

The

(BVC),

due

the

rule,

rule

the

and

newest

to

by
data

the

set.

or a

The

is

Lopez

de

seller,

other

Lee-Ready

method

Easley,

a buyer

rule,

two
after

the

bulk

volume

Prado,

and

O'Hara

(2012).
According
by

to

comparing

trade;

no

trade

the

bid

or

is then

Uptick,

tick

if

price

offer

of

trade

one

price

is

of

to

information

into

trade

classification

the

quote

classified
the

the

the
is

of the
higher

a trade

price

taken

four

of
into

is

performed

the

preceding

account.

Each

categories:

than

the

price

of

the

previous

of

the

previous

trade.
Downtick,

if the

trade

price

is lower

than

the

price

trade.
Zero-uptick,
was

an

if the

move

was

the

trade's

trade

if

has

the

trade

is classified

moved

direction

Michaely,

and

classified

77.66

tick

Figure
into
A.

moved,

but

the

last

recorded

move

has

not

moved,

but

the

last

recorded

of

from

that

an

uptick

up

or moved

as
the

O'Hara

or

percent

down.

nonzero

(2000),
of

in

all

the

previous

a downtick,

a zero-uptick

last

of

depending

If the
or

price

price

has

change.
the

trades.

Figure

last

whether

not

moved,

depending

According
tick

the

on

a zero-downtick,

19971998,

Nasdaq

trade,

rule
4.10

to

Ellis,

correctly
illustrates

rule.
4.10

Buys
Uptick:

order)

price

differs
as

price

the

the

price

is classified

the

not

a downtick.

the

on

has

uptick.

Zero-downtick,

If

price

An
and
The

Illustration

of

the

Tick

Rule,

Used

to

Classify

Trades

Sells
trade

is classified

as

a buy

(initiated

by

a market

buy

B.

Zero-uptick:

C.

Downtick:

D.

Zero-downtick:

The

low

The

are

trade

trade

to
and

due

stocks

be

rule ),

the

repealed

executed
rule

short

sales,

short

sales

on
the

2007.

et al.
may

be

as

a sell

that

uptick

the

nearly
(2008)
responsible

may

For

example,

observed

Exchange
30

percent

suggest
for

that
the

be

due

to

Asquith,

misclassifications

requiring

or zero-uptick
and

trades

equities.

regulations

Securities

Asquith
alone

to

Because

in

report

the

a sell

classified

issues

part

a buy

is classified

(2008)
in

as

as

correctly

regulatory

least

in

trade

of

Safaya

at

is classified

is classified

The

proportion

specifically
Oman,

The

that

short

(known

sales

as

the

uptick

Commission
of

equity

of

(SEC)
trades

are

regulation-constrained
observed

errors

in trade

classification.

In

preceding
On

trade

futures

indeed

find

trades

into

buyer

percent

Lee

and

al.

(2012)

is a buy

(sell)

bid

the

(1991)

happens

to

the

is not

trade

been

shown

(see

Ellis

subject

be

quotes

tick

rule.

and

trades

the

this

and

can

prevailing

of

the

situation

ahead

to classify

delay

in reporting

of

by

trades.

deteriorated

over

considerable

speed

Figure

4.11

shows

Figure

4.11

Example

using

trades.
the

past

since
an

five

seconds

However,

of the

two

(1991)

point

at

least

was

validity

of

out

they

show

are

often
to

seconds

Nasdaq's

median

the

rule

Ready's

study

was

of the

quote

classification.

Classification

that

five

and

Rule

result

propose

markets

rule

be

making

quotes

as

Quote

may

worse

They

has

Nasdaq

sequence,

them.

ask,

and

on

decades

Lee

example

the

the
price

and

quote

books,

quotes

of

often

Specifically,

the

a trade

trade
bid

deliver

triggered

In 1991,

the

is used

of

by

average

prevailing

difficult.

that

rule,

of all trades

out

electronic

this

If

Ready

reported

to

500.

prevailing

potentially
and

According

documented

the

rule

of

quote

trades

the

percent

Lee

often

introduction

ahead

mitigate

definition

example,
are

of the

with

recorded

For

the

also

time.

quote

in

classifies

S&P

(below)

of

76.4

the

(2012)

accuracy

correctly

Under

the

midpoint

classify

2000),

(2000).

at

the

on

the

precise.
al.

initiated.

quotes

is above

While

interpretation

determination
that

the

classified.

al.,

et al.

et

higher

rule

futures

prevailing
at

tick

of

more

Easley

seller

to classify

price

quotes

exactly

et

and

all

much

much

the

Ellis

trade

to correctly

to
the

way

be

rule,

delivers

of E-mini

constraints,
to

uptick

initiated

and

if the
ask

likely

calculations,

is another

Ready

short-sale

the
rule

of all trades

rule

and

from
tick

86.43

quote

free
the

et

of
are

that

Easley

The

absence

classifications

data,

classifying

than

the

may

have
gained

published.

The

so-called

Lee-Ready

rule.

The

trades

and

ask

quotes

occurring
are
classified

Ready

emphasize

at least

five

quotes.

Dufour

Ellis

et

may

al.

(2000)

differ

five-second
correctly

(2012)
of

for

every

shares
being

et

just

81.05

a small
increase

(10)
where:

traded),
as

al.

the

named
of

show

percent

of

time

BVC
follows:

generated
bulk
or
assigns

the

the

tick

by

probability

Ignoring

rule

either

Easley

or

buy

or

et

al.

the

estimate

a market

, works
bar,

of

the

Lee-Ready
as

buy

volume

delay

classification.

classification
(a

of
while

reporting

a probabilistic

a market

and

occurred

rule,

classification,

produces

volume
volume

trades

are

sequencing

system.

all

bid

Lee

that

trade

the

quote

and

again,

five-second

that

of trade
that

rule,

erroneous

the

the

prevailing

quotes

user's

(2000)

accuracy

with

that

end

the

Once

the

over

a methodology

rule,

follow

using

quote

rule.

to avoid

showing

the

the

trades

improvement

volume

unit

a buy

it,

Ellis

propose

The

to

delay,

a particular

order.

(2000)

on

classifies

further

Engle

depending

sell-initiated,
To

in order

object

tick

matching

first

between

under

the

prior

trades

midpoint

classified
using

seconds
and

classifies

at the

not

subsequently
(1991)

rule

say
observed

as

sell

follows:

every

100
volume

is the

total

is
time

volume

the

price

or volume
is the

price

observed

during

difference
bars,

standard

time

observed
and

or volume

between

interval

the

two

.
subsequent

deviation

of sequential

time

or volume-clock

based

changes.

Z is the
The

pdf

of a standard

buyer-initiated

normal

trade

distribution.

volume

can

then

be

estimated

as

(11)
According

to

the

BVC,

generated

by

a market

the

probability

sell

order

of

then

specific

volume's

being

becomes:

(12)
And

the

respective

size

of the

seller-initiated

volume

is then

(13)
Easley
and

et
show

trades
the

al.

(2012)

that

when

the

time

volume

apply
BVC

bars

clock

the
rule

are

is used

BVC

methodology

correctly

used,

classifies

and

instead

to

90.7

of the

86.6

percent

E-mini

futures

percent

of

of all trades

time-based

all

when

clock.

Summary
Tick

data

differ

tick

data

creates

frequencies.

are

2. What

types

series
4.
rule

and

the

of

possible

not

for

interpreting

properties

data
the

of the

Utilization

available

sampling

angles

and

key

are

properties

of data

data

most

What

opportunities

data.

and

at

resulting

Various

ticks
time

lower

interpolation

exploration.
discreet

of

of

data

series.

Questions

1. What

What

low-frequency

of

diverse

statistical

End-of-Chapter

3.

host

organizing

different

from

multitude

creates
of

deliver

techniques
methods

dramatically

the

quote

messages

sampling

closely

of high-frequency

fitting

are

technique
the

most

data?

frequent?

produces

normal

high-frequency

distribution?

key

differences

between

rule,

Lee-Ready

rule,

and

the
bulk

tick

trade

volume

classification
classification

time

rule?
5.

Consider

a trade

prevailing

at

time

executed

at

17.00.

initiated

or seller

completed
1 A version
of

Financial

Sons,

2012).

executed
t . The

initiated

of this
Models

t be

time

previous

Should

at time

at

the
under

trade,
trade

at

the

quote

classified

chapter
(3

under

appears
volume

t at

the

executed

at

time

the

in F.
set)

17.01,

t be

How

tick

rule?

(Hoboken,

time

classified

rule?

Fabozzi,

best

NJ:

quote

t 1,
as

should

ed.,

bid

was
buyer

the

trade

Encyclopedia
John

Wiley

&

CHAPTER

Trading

Costs

Trading

costs

trading

strategy.

long-term

can

make

or break

the

Transaction

strategies

are

profitability

costs

amplified

that

of a

may

dramatically

high-frequency

be
in

negligible

for

high-frequency

implicit

costs

setting.
This

chapter

impact

focuses

on

high-frequency

Overview

to

possess

finance,

following

are

uniform

Markets

are

consolidated;

immediately

Orders

of

infinitely

Traders
No

have

Market
In

real

securities

information
can

further
Costs

place,

Likewise,
to trading,
yet

distribution

and

implicit
inferred

financial

instrument

is traded.

information.

have

markets

the

limit

in

and
from

that

have

to

the

transparent

implicit

frictions:

differ

activity

from

in

are
order

book

size.

to trading

costs

instantaneouslymarkets

power.

to order

away

costs

they

exist.

markets

and

is,

orders.

borrowing

fragmented,

prior

that

instrument

queue

of limit

highly

known

implicit.

price

markets

time,

be

distort

explicit,

prior

over

the

executed

each

is invariant

life,

markets

be

constraints

price

a specific

wherever

number

unlimited

short-sale

of

all fundamental

can

with

infinite

seamless,

exist.

sizes

liquid,

containing

a price

reflect
costs

all

are

structure.

updated

transaction

markets

in their

is instantaneously

No

that

characteristics:

Markets

Prices

and

Costs

classical

the

transparent

trading.

of Execution

According

the

costs

can

be

the

data

be

their

are

not

estimated
of past

like

and

depth,

transaction

costs

textbook-perfect
referred

estimated
costs

incorporate

structure

issues

their
are

prices

to as

are
are
known
from

trades.

models.
transparent

known

known

with

before
the

as

latent

or

certainty

trading

costs'

or

takes

historical

Transparent

Execution

Costs

Transparent

execution

costs

they

comprise

section

broker

considers

Broker

are

generally

commissions,

each

of the

known

exchange

transparent

ahead

fees,

costs

of

and

trading;

taxes.

This

in detail.

Commissions

Brokers

charge

Providing

commissions

connectivity

for

to

exchanges,

dark

pools,

and

other

trading

venues.
Facilitating

best

according

to client

Serving

as

the

execution

opposite

Other
Some

to
as

or

counterparty,

to

clients'

orders

in

execution

to

settlement

and

reporting

may

as

with

move

away

from

execution

trade

are

can

be

value,

depend

on

the

total

various

trade

implicit,

transactions

business

bundling

the

options,
that

the

for

data
models,

or

volume.

broker

15

Broker

extent

variable

of
in

to
latest

to
build
best

book.
per

commissions

provides

brokers,
choose

The

from

of

presently

many

of this

receives
the

becoming

preferring

or

for

secrecy

often

instead.

month,

broker
and

retaining

revenues

in-house

order

is

execution

in Chapters

per

monthly

by

fees
information,

best

execution

discussed

or

research

of

engines

fixed

broker

high-frequency

best

execution

models
fees

of

premium

while

percentage

additional

other

compete

And

brokers'

proprietary

and

customers

understanding

customers

Paid-for

strategies.
a significant

their

data

research.

investment

clients

to clients.

charge

market

obsolete

for

margin

services.

proprietary

accounts

size,

their

streaming

increasingly

Broker

orders

arrangements.

and

broker-dealers

access

their

from

leverage

custom

their

client

activity.

Delivering

such

executing

of assets.

trades

trading

side,

(OTC)

custody

Clearing

orders:

specifications.

over-the-counter
Providing

of client

may
a given

soft-dollar,

addition

to

trade
also
firm,
or
direct

execution

services.

advance

of

execution.

executing

broker

understanding
Figure
Interactive

differences

another

in

can

be

favorable

the

are

broker

negotiated

cost

well

estimates

significant

in

from

and

one

are

worth

pricing.
costs

for

metals

trading

offered

by

Brokers.

Figure

5.1

Broker

Interactive

Commissions

on

Metal

Trading

Offered

by

Brokers

Source:

Interactive

Exchange

Brokers

web

site

Fees

Exchanges

match

communication
core

open

to

shows

commissions

The

to ensure

5.1

The

Broker

orders

from

networks
product

of

buy-and-sell

different

(ECNs)

every

broker-dealers

and

charge

exchange

interest,

that

is

traders

the
are

fees

for

liquidity,

or

electronic

their

services.

or

looking

presence

to transact

of
on

the

exchange.
Liquidity
prices

is

created

below

placed
hand,

at
are

the

by

the

current

prices

above

matched

exchange,

matched

immediately

below

order
the

available
To

attract

orders

buy

from

market
bid,

ask
the

as

liquidity,
consuming

provide

bid

liquidity.
at

with

the

or

a market

will
sell

best

the

available

immediately

orders

as

do

limit

sell
on

orders
can

market
limit

placed

limit

orders,

orders

Similarly,
be

sell,

a limit

orders

the

thus

matched

on

consume

price

sell

other

available

also
ask

at

will

be

removing

placed
with

at
the

or
best

would.

exchanges
liquidity

the

above

best

buy

Market

Limit

exchange.

price

limit

liquidity,
bid.

with

placed

the

orders;

current

consuming

a limit

sell

limit

immediately

liquidity;

the

open

or

may

charge

for

orders

fees

or

supplying

pay

rebates
liquidity.

for
As

discussed

in Chapter

orders

and

pay

exchanges.
charge
this
in

for

orders

book

was

either

or

order

fees
type.

iceberg

fees

equity

other

or algo-enabled

normal

liquidity
At

and

the

deployed

time

rebates

exchange

ECNs

in

vary

by

programs.
venues

such

as
like

fees

can

also

hidden-size

orders

volume-weighted

cost

exchange

as

inverted.

Selected

associated

commissions,

known

exchanges

orders,

carry

liquidity-removing

remove

called

trading

orders,

orders,

broker

are

that

rebate-based

of

for
are

orders

models.

offer

Complicated

(VWAP)

Like

also

and

orders,

price

U.S.

inverted

now

for

liquidity

most

charge
orders

pay

supply

written,

countries

Exchange

that

that

normal

that

liquidity-supplying

Exchanges

for

other

3, exchanges

like

average

premium.
are

negotiated

in

advance

of execution.
Taxes
According
be

to Benjamin

certain,

profits
the

except

death

trading

operation

operation

is

domiciled.

profits

investments
capital

of
gains

that
one

accountant

knowledge

pertaining

be

Proposals

to tax

time,

most

long

run.

on

every

as

much

year

in

one

contractions

Implicit

Execution

Costs

At

tick

of

level

be

of

to

which

fall

the

full

under

to

to

the

net

in which

rate,

unlike

reduced-tax

local

certified

or

wealth

of

Appropriate

tax

rates

activity.
surface

against

transaction

trading

tax

provide

said

generates

the

rates.

that
for

from

jurisdiction

transactions

stock,

be

trading

subject

taxation

can

charged

appropriate

of trading

of economic

data,

are

able

estimates
IBM

nothing

jurisdictions.

deciding

third

severe

the

most

trading

(2012b)
of

the

more,

proper

jurisdictions

transaction

Taxes

usually

individual

world

High-frequency

in advance

Aldridge

as

by

or

to

this

taxes.

should

determined

with

are

umbrella

chartered

in

and

of the

short-term

can

Franklin,

a 0.05

example,

volumes,

and

fade

taxes

percent

tax

is likely

to

over
in

imposed
wipe

ultimately

resulting

are

increasingly

growth.

transparent

costs

being

the

out
in

dominated

by

bid-ask

spread,

Additional
the

implicit

Bid-Ask

Spreads

Bid-ask

spreads

shape

or

best

bid

are

any

instantaneous
be

as

execution

via
the

liquidity

through

compensation

Bid-ask

the

cost,

market

price

difference

buy

The

at $56.50,

market
between

the

traded

is

as

by

the

ask

and

the

the

cost

of

the

patient

of

argument,

considerable
the

result,

limit

spread

the

providing

take

higher

risk

relation

in Chapter

that

traders'

uncertainty,

in

also

immediate

markets,
a

of

can

traders

traders

As
times

spread

desiring

side

the

to

made

until

trading
but

by

best

placing

the

the

to

fact

volatility.

4.

strategy

identifies

the

the

moves

up

to

the

desired

price

time

occurs

from

the

and

and

the

concept
a stock

market
In

this

and

price

buy
case,
obtained

is

by

the

the
on

in
an

executed.
to

the

the

trader

market

price.

latency

(e.g.,
order

time

refers

realized
of

change
the

trade

observed

$58.00.
the

adverse

orders

order
the

is the

time

quote

illustrates

price

that

market

the

example

slippage,

security

accompanies

prior

following

in

in detail

known

between

immediately

The

order

bid-ask

are

Costs

of

often

in

discussed

decision

Slippage

slippage.

are

rise
the

commonly

investment

The

of

or Latency

Latency

to

they

best

position.

the

limit

position

the

represents

flip

to

include

characterized

traders

volatility.

has

variability

spreads

Slippage

The

market
also

in

paid

losing

best

and

by
the

traders

values.

time,

On

impact.

Instead,

between

paid

orders.

are

a trading

orders.

limit

with

reflected

of

compensation

entering

increases

in

the

timing.

historical

point

market

and

advance.

difference

premium

market

is

of

given

the

spread

risk

the

liquidation

seen

in
that

of their
is

market

known

of investment,

related

researchers

and

not

cost

and

by

variables

spread
at

slippage,

appreciation

distribution

bid-ask

opportunity

tracked

random

of the

The

or

costs

of price

stochastic

of the

latency

implicit

costs

costs

cost

IBM)

to

be

is executed,
$1.50

or
a
the

differential

execution

is

the

cost

of latency.

In the

media,

divide

between

race.

In

trading

finite

major

liquidity

and

as

Figure

during
and

Figure

to

can
of market

5.2

Slippage

dependent

market

place

illustrates.

close

high
times,

b)

that

5.2
of

also

and

consume

periods

market

solely
is

the

market
As

as

as

forecasted

impact,

in

discussed

Illustration

for

times
The

at

result,

activity,

well

great
arms

the

function
of

the

of

a)

market
erode

the

in

the

speeds
is

example,

of

the

many

slippage

activity

of

orders

orders

high

speed

market

at

immediately

probabilistic
next.

on

When

orders,

of the

technological

number

price

trading

the

liquidity.

market

evidence

a direct

the

announcements.
be

the
of

not

move

as

have-nots

slippage

macroeconomic

participants
context

is

in the

arrived

portrayed

and

simultaneously

process,

open

haves

Instead,

available
have

is often

slippage

activity.

participants

larger

the

reality,

liquidity
that

slippage

typically
market
following

other

framework

market
in

the

Still,

latency

on

due

traders'

that

fast

of

have

Rolf

a substantial

(2012),

delivers

for

as

resulting

impact

example,

significant

Specifically,

latency

savings

(low-latency)

and

assumptions.

cost

dollar-denominated

can

infrastructure

forecasting

define

alone

Stoikov

trading

certain

(2012)

technology

performance.

ultra-fast

under

to

the

from

find

profitability

Stoikov

and

expected

executing

Rolf

value
with

of

slow

and

infrastructures:

(1)
where

S t + l is the

price

obtained

when

latency

of

milliseconds

10

algorithm
100

on

of

to

to

that

algorithm

cost

millisecond

latency.

The

shortcomings

may

not

Trader's

2.

of

and

can

Distances

and

private

orders.
the

cause

latency
the

increasing
centers
network

In

of

the

and

time

virtue
each

trading
world

connections

an
turn,

cost

as

latency.
incurs

costs
latency

network

an

just

may
can

or

occur

architecture

cause

geographic
and

the

quote

and

location

trading

can

of

has

Latencies
in Chapter
venue

delay
between

the

execution

communication

or trade

reported

network

differences

of physical

with

in

execution

latency.
are

In

with

interruptions

timely

addition,

by

latency

of

quotes.
and

servers

algorithm

cost

that

latency.

routing:

systems

trader's

increase

destinations,
financial

of order

disrupt

in

the

1 minute

Specifically,

interpreting

may

the

resulting

Congestion

of

than

that

latency

of

the

time

execution

a delay

greater

nodes

threefold

S t is

same

twice
of

1 millisecond
with

the

observe
is about

in

l and

at

1 millisecond

domain.

Slow

Networks.

venue

only

technology

following

of

delay

using

times

latency

authors

result

trader's

in processing

location

on

algo

four

in the

communications

the

run

t with

instrument

The

infrastructure

systems.

transmission

zero.

an

of

reside

of the

delays

financial

latency

of

using

at time

communication

of

associated

1.

same

l approaches

milliseconds

in any

obtained

the

configured

compared
An

price

delays.

to travel

to their

between
2.
servers

major

Co-location
may

help

alleviate
3.

these

Broker-dealers.

innovations
4.

issues.
Delays

of direct

Execution

delay

high-volatility

costs
of a trade,

trades

produce

can

strategy

an

may

prompted

experience

overloads

order-processing

Such

random

situations

distribution
the

activity

can

destinations

and

Appreciation

alternative

Both

price

incurred

of trading

and

of

backlog

and

most

often

occur

in

known

with

precision

during

Risk

and

timing

of

execution

child

orders

of
(see

costs

value

inferred

to

Fast

be

from

used

within

infrastructure,
human

past
the

backup

supervision

problems

and

route

backup

channels,

of

orders

to their

ensuring

information.

Timing

appreciation

be

continuous

network

transmission

Price

cost
process.

detect

along

cannot

of latency

expected

systems,

continuous

and

development

communication

series

in

execution.

are

in advance

trading

engines

environments.

latency

trading

venues

resulting
in

routing

access.

Execution

orders

subsequent

While

market

venues.

simultaneous

in broker-dealer

Costs
risk

large

costs

describe

position

Chapter

15

market

broken

for

risk

down

into

discussion

of

child

orders).
The

price

appreciation

investment
cost

value

during

to

random,

refers

execution
The

of a child

following

cost.

undervalued

at

placed

must

that

over

the

therefore
the

turns

the

trading
and

be

the

forecasted

of a large
price

following

two
per

illustrates

strategy
a buy

executed

to be

fundamental

execution

trade

1.3560,

bps

to

loss

position.

The

movement

of

timing

ahead

of

order.

out

50

refers

unforecasted

EUR/USD

appreciation

forecast

cost

and

minutes.

minute.

appreciation

concept

determines

that

EUR/USD

is

million

EUR/USD

is

order
over

correct,

the

of
the

$100
next

EUR/USD

three

price

The

price

appreciation

of

price,

cost
trading

The

to 1.3660

appreciation

the

price

minutes.

appreciates

The

not

of

cost
is

due

activity

is
to
in

EUR/USD.
The

cost

of timing

of the

traded

second,

10

limit

seconds,

risk

order.

However,

when

the

had

of

the

trading
percent
were

and

cost

is

order

cost

following

within

the

time

an

is executed.

The

activity,

cost

does

usually

not

apply

to

is fixed.

an

Cost
total

also

fail

been

to

not

the

the

(NBBO),

as

as

specified
for

sufficient

to

order
limit

limit

example,
fulfill

may

orders

discussed

the

complete

orderbased

execute,

market

have

to

limit

cross

liquidity

measured

the

fail

to

posted

at

in Chapter

profit

expected

3.

to

be

executed.

amount

dollar

higher
In futures,
market
market

In

(2010),

averaged
the

the

Market

cost.

dollar

comparable

yet

not

exchange,

Offer

inability

accompanies

the

does

with

does

can
have

order.

Review

costs
of

cost

measures

transaction

markets.

(2012c),

from

price

Costs

impact

markets;

lower,

not

Bid

the

Impact

dominant
Trading

risk

the

depreciate

order

associated

price

exchange

Best

generated

price

on

average,

timing

timing

price

cost

equity

the

opportunity

Market

so

market

The

orders

does

U.S.

when

Market

or

market

active

market

market

National

The

the

opportunity

the

market

On

is the

often,

when

price.

the

cost

Most

strategies,

execute

and

execution

on

Costs

opportunity

order.

much,

appreciate

until

orders.

the

how

minute,

to

market

where

Opportunity

an

applies

using

by

randomly

is made

cost

orders,

The

can

decision

executed

describes

security

investment
timing

risk

Per

percent
spent

U.S.,

transaction

on
EU,

market

impact

costs
observed

still

the

and

the

the

ITG

Global
percent

total

amount

with

only

figures

Japanese

reported
transaction
according
futures

in

of
0.089

These

and

Eurobund

market

0.387

commissions.

dominate:
in

to

turnover,

U.K.,

the

becoming

consumed

were

impact

in

rapidly

report,
of

costs

both

impact

impact

traded.

the

is

according

market

0.476

change

impact

equities,

volume
in

adverse

equity
emerging

costs
to

appear
Aldridge

(FGBL)

on

Eurex

is 0.020

costs

in

per

the

every

percent
futures

$1

executed

dollar

estimation

of market

trades

and

is

placing

his

particular
the
The

own

market

contained

in

in

an

with

welfare.

As
of

via

each
a

credible

At

$5

0.0010

The

fact

mere
the

to

$10

percent

of

cause

that

line

to

market

and

a trader

bet

on

participants

of

the

best

or

sell

at

market

trade

is

available.

orders

emit

The

idea

market

The

thought

market

time

that

limit

the
an

of

of whether

their

are

derive

implications

on

The

potent

larger

the

order

orders,
orders

usually

from

their

the

trader's

signals
order,

author
assumed

of

such,

and

the
name,

the

price,

about
the

and

more

both

and

limit

when

market

orders)

orders

or sell

orders
the

orders

orders,

limit

limit

to buy

price

market

the

are

opposite
and

limit

however,

limit

make

less

credible

and

even

the

direction

was

first

published

was

so

intensity

differ.

a credible
in

orders

commitments

market

a result,

may

signaling

are

Unlike

As

as

(market

a predefined

As

impact.

canceled.

of an

are

traders

as

more

Analysts

they

viewed

orders

at
buy

and

time,

profits

moves.

much

analyst.

irrespective

direct

orders

to

of market

news

same
of

passive

immediately

than

is

signal.
and

be

observation

often

are

signalsmarket

can

of

the

trades

commitments

under

the

the

other

trade

having

market

trading

a salary,

trade

result,

aggressive

signals

to

on

informs

percentage

impending

the

orders

in

considers

reputation

opinion

true.

via

trading,

security

section

or

of the

that

compensated

are

be

0.0005

information.

money

come

credible

or

to

belief.

prognoses

Both

convey

compensated

beliefs

This

tend

transaction

impact.

orders

than

typically

traded,

Institutional

Definitions

information

potent

volume.

markets

volume

direction

trader's

dollar

forex

volume.

and

All

the

and

million

Background

of

signal
markets

research

dared

Bagehot,

to
after

in

revolutionary

publish
a

the
famous

1971.
at

theory

the
only

English

nineteenth-century
The

mere

classical
of

journalist

existence

of market

finance.

market

most
assumed

In

the

efficiency,

classical

asset

interprets
the

and

the

upon

be

core

world,
the

the

principles

under

optimal

the

of

concept

steady

following

other

the

same

state

in

conditions

are

in identical

information.
way.

information,

or interest

rates,

such

affects

nonfundamental

relevant

information

prices

arrival,

as

security

information

already

contained

in

markets'

response

to

to periods
5.3

from

carry

earnings

prices.
has

Fundamental

impounded

no

an

into
a

no

Past

price

relevance

Figure
of
news

Finance
Information

prices

sharp

fundamental

information,

arrival
any

in

one

prices.

without

Classical

is

resulting

moving

themselves

Incorporate

to

receives

Trades

Figure

financial

information

information

as

against

to

prices.

security

well

runs

models,

fundamental

announcements

All

in economics.

to hold:

Everyone

security

idealized

pricing

simultaneously

trends

topics

impact

considered

Everyone

Only

covering

View

5.3
positive

step
level

as

all

immediately

the

function
to

the

of
next.

information

illustrates
fundamental

the

is
perfect

news,

arrivals.
on

How

Perfect

Markets

as

In

reality,

traders

investment

horizons,

subject

of

investors

the
are

and

other's

proverbial

market

analysis

Traders

poring
an

mutual

fund

professional
newbie.

statements
may

All

idealized

of these

conditions,

information

to other

In

most

addition,

from

have

factors
and
market
information

coexist

and

15

hours

a casual

only
a

point

of

trades

and

(see

allowed

both

eating

analysts.

are

divergent

also

likely

to

analyzing
a
from

markets'

each

technical

fundamental

Finally,

to the

long-term

deploying

briefly

the

investors

without

investor

view

On
of

has

a day

quarterly.

contribute
make

that

quants

divergent

experiences.

investors

information

of

short-term

fact
to

result

opinions

of

with

different

and

those

Similarly,

disagree

opinion

from

investors

lunch.

movement,

example),

long-term

the

effort

price

differ

for

over

differently,

education,

to

1991,

short-term

news

impending
likely

Hasbrouck,

have

interpret

his

seasoned
that

deviation

of

from

even

orders

able

to

into

prices

gradually,

carry

participants.
is

impounded

as

first

noted

by

classical

Kyle

finance

meeting

would

a sell

meeting

an

orders

(1985),

is

Often,

opposing

limit

theory,

each

short-term

reality,

traded

securities,

are

2007;

of

orders

price

a future
and

have

change

Hautsch

not

move

the

opposite

a buy

of
order

market

to form

supply

study

order

price-meeting

matched

the

of

of one

two

and

limit

a trade.

and,

in

demand,

markets

much.

In

the

holds:

followed

market

sell

(see

and

by

a rise

orders

on

in prices

average

of the

result

in

and,

When

for

(2012c)

of that

a comparable

positive

fundamental
decaying

news
level,

to its
5.4

from

Price

gradual
no

news

arrives,
with

the

the

overshoot

fundamental

level.

Reaction

to Trades

2009;

and

trade

is

result,

are

smaller

and

Huang

of a limit
order.

may

be

an

in

(2011)
order

to

The

direction

reversed

relative

on

how

far

is

the

price.
of
in

market

persistent

cancelable

market

adjustment
observed

result,

are

depending

market

Seppi,

orders

impact

order

order,

the

in

completed

market

a limit

As

by

Hautsch

of a similar

market

away

and

the

following

illustrates

news

orders.

estimate

percent

price

market

as

and

Kockelkoren,

interest,

of

positive

and

impact

2010).

followed

limit

change

5.4

Bouchaud,

be

signal,

price

Figure

to

credible

that

order

shown

Rosu,

most

than

limit

been

and

Parlour

magnitude

that

2008;

1997;

While

trading

of the

Harris,

market

2011).

of

25

(see

Huang,

indications

about

generate

George,

Eisler,

and

Aldridge

also

trade

also

irreversible

Figure

be

forces:

orders

Boulatov

limit

to

also

average

while

buy-and-sell

anticipation

be

is a product

the

prices.

Limit

and

on

as

is a result

although

may

does

orders

trade

order,

however,

buy

Every

opposing

trade

Market

lower

of

instantaneously,

a trade

direction

a product

not

prefer.

order.

of opposite

trade

and

actual

price
component

Observed

prices

in

response

trading

tends

to

conditions.
overshoot

gradually

in Real

to

Life

settling

its
or

Information

leakage

for

years,

to

tick-level

and

broker-dealers
data

infer

the

information

of

today's

fast,

ticker-tape
phenomenon

of

is

known

several

factors

including:

trading

world

and

as

market

with

information,

By

tape

continuously

to

process

resembling

trading.

As

the

ticker

their

would

markets
tape

impossible

for

manual
become

is

moving

a human

eye

to

time.

price

changes
(MI).

heterogeneous
is

ticker

according

impact

MI

access

quotes

literally

real-life

the

makers

adjust

in real

with

read
that.

new:

the

however,

information

is hardly

learn

the

it is now

orders

beliefs

market

high-frequency

and

to

from

and

computerized,

increasingly

In the

trades

trades

participants

able

forecasts
manual

and

market

being

predictionsin

increasingly

This

on

tape,
of

orders
other

price

ticker

short-term

parse

competed

content

variant

and

short-term

watching

their

accompanying

the

MI

following
can

be

news

and

attributed

traders

possessing

negotiation

or

different
ttonnement

to

process
into

via

the

Both

which

security

market

and

and

market

participants
market

on

ask

the

limit

orders

reputation,

Every

and
who

buy
(bid)

order

(sell)

side

of the

Estimation

of Market

Impact

The

of MI

the

the
MI

price

to

the

price

the

price

postorder

as

if he

time,

time,

the

before

say

revisions

or

estimate
Figure
Exchange

is

t* . In

10

tick

trades,

than

in the

driving

of

to

same

next

(lower)

other

direction.

liquidity

the

a higher

at

time

supply

market

best

time

the

when

the

buy

available

of

impact

trades

aggregate
least

trade-time

trading
MI

Market

Impact

is

order

measured

to

the

last

in
time
time

clock

time,
selected

t* . In

trade

the

order

z ticks,
volume

volume
units

quote

time,

the

following

(e.g.,

computationally

trade

selected

t is

following

in

trade

the

be

computed

Finally,

illustrates

5.5 . The

time,

trader

can

arrival

is

order.

of

measured

order

number

impact

Figure

In clock

the

5.5

t* relative

be

post

would

Figure

example,

in

or volume-time
of

t*

at

The

time

t can

the

past

clock-time

Estimation

t* , for

time.

much

trade?

illustrated

and

the

after

time,

How

order.

seconds,

reaches

at Time

single

t* , as

Tick-time

5.5

signals

depletes

book,

make

measured

interest

contracts).

to trade

question,

or volume

measured

of

order

to

trade

units,
MI

is

impounded

commitments

credible

temporarily

observed

time,

at

order

exchange

tick

observed

impact

choose

the

for

reference

x time

are

traders'

form

against

were

change

recorded

trade

answers

determination

incoming

beliefs

price.

metric

move

therefore

order

matched

as

to

order

and

represent

may

be

limit

information

prices.

money

(sell)

traders'

the

shares

or

easier

to

impact.
for

an

Order

Arriving

at

an

While
in

the

exact

advance

of

historical

precision,

on

some

can
on

trades

noted

earlier,

the

more

credible

the

impact
of

investment

can

and

temporary
prices

is

(see

broken
MI.

a losing

example,

may

worse

prices

profitable

into

O'Hara
two
MI

consistent

and

(1987)
distinct

with

the

in

an

the

the

that

by

with

trade,

grows

first
parts:

fundamental

MI,

the

trade,

the

higher

or
the

MI

the

classical

strategy

larger

evolution

functional

the

based

expected.

Waelbroeck,
were

impounds

with

measured

distorted

increase

when

Lillo,

forecast

formed

than

exact
can

to

for

pinpointed

be

be

be

The
trader

be

using
logic

used

one.

generates.

and

of

known

data.

by

Gerig,

Permanent

in a manner

into

The

can

conveyed

Farmer,

down

MI

be

estimated

can

information

remaining

Easley

price

size:

be

never

order

important:

while

size

be

trade

never

logic

historical

strategy

much

the

future

recent

for

can

can

expected

is a function

the

the
MI

The

to

price

of the

obtaining
MI

similar

future

trades,

MI

can

characteristics.

is

using

a profitable

As

(1987)

exact

trade

expected

MI

metrics.

repeating

sequential

trade

of

framework
turn

the

expectation

auxiliary

a future

trade-specific

the

an

event-study

form

trade,

and

while

full

relies

following

expectation

levels:

MI

the

data

estimating
price

impact

functional

size
slowly

2009).
to

of

note

with

Glosten
that

permanent
information

financial

his

theory

MI
and
into
and

the

efficient

the

noise

and

market

hypothesis

component

then

that

decays

(see

first

with

Fama,

1970).

overshoots

time

until

the

the

Temporary

fundamental

fundamental

MI
price

price

is

level

level

is

reached.
The

precise

been

of

of

competing

subjects

(2002)
is

shapes

and
linear

Mangegna

Kissell

and

(2003),

however,

Gatheral,

relationship

for

temporary

Figure
Temporary

5.6

impact,
Functional
Components

order

for

size

on

permanent
shown
Form

the

however,
finding
MI

and

in Figure
of

example,

a
topic

Market

MI

and

(see

MI
and

specification

Huberman
the

and
linear

order-size-based

time-based

that

Farmer,

law

reconciles

have

Korajczyk

suggested

V t* ).Lillo,
power

an

functions

Hodrick

( MI t

detected

by

as

temporary
Breen,

(2002),

2009),

specifications
the

the

research

power-law

for

Glantz

of

latest

and

theories.

function

). The
2004;

and

permanent

power-law

Stanzl,
and
linear
decay

5.6 .
Impact:

Permanent

and

For

market

size,

and

orders,
the

market

buy

versa.

function

the

signs

of

orders

appears

market

tend

model

as

MI

to

follows

orders

follow

unifying

the

(Eisler,

strongly

concave

are

other

market

buy

facts

and

the

autocorrelated

preceding

Bouchaud,

in

order

in

orders,

and

expresses

Kockelkoren,

time:
vice

the

MI

2009):

(2)
where:
is the

volume

is the

sign

is

the

of the
of the

by

at time

at time

noise

t*

t*

term

that

models

any

price

change

not

news

G ( t, t* ) is the

The

trade

independent

induced

prices

trade

propagator

respond

function,

to a single

MI

of an

individual

MI

propagator

trade

order

which
over

describes,

time;

recorded

on

average,

G ( t, t* ) decays

at time

with

how
time

t* is then

(3)
The

specific

way,

function
to

satisfy

subsequent

trades.

proportional

to

signs

in

Kockelkoren

the

If

the

autocorrelated

G ( t, t* ) has

did
of

time.

(2009),

high
not

the

decay,

trades

as

decay

with

time,

autocorrelation

of

price

and

According

G decays

to

to

a
of

would

would

Eisler,

in

direction

returns

returns

and

be

Bouchaud

be
highly
and

follows:

(4)
where

subsequent

and

is

trade

the

decay

parameter

in

the

correlation

of

signs:

(5)
G ( t, t* ) further

has

the

following

boundary

properties:

(6)
where

t =

standard
In

~ N ( 0, )

deviation

addition

shown

tv t

to

to depend

on:

distributed

with

mean

zero

and

trade

to depend

is normally

size
on

other

and

time,

variables.

however,

MI

In equities,

has
MI

been
has

empirically
been

shown

Intertrade

durations

Stock-specific
issuing

and

In futures,

MI

Liquidity

(see

Intertrade

on

best

for

on
on

Korajchyk,

of

2002;

Thum,

the
Lillo,

Hauptmann,

limit

experienced
orders

by

medium

experienced

small
times

5.7

summarizes

Figure

5.7

Market

Hautsch

and

with

a size
that

the

best

became

the

matching

the
best

best

would.

of a limit

became

Huang,

bid

to that
buy

order

best

depth
bids

Buy
quotes

comparable

MI

limit

the

bid,

at

the

were

than

depth
20

at

the

percent

bid)
of

the

were

on

followed

absolute

larger

than
about

depth
50

at

the

percent

bid)
of

the

were

MI

the

findings.
of Limit

Huang

(2011)

Orders

Observed

on

followed

absolute

orders.

Impact
and

The

order:

about

MI,

small

Figure

ask.

buy

with

orders

size

to

and

matched

of the

proximity
For

Hautsch

buy

to

orders.

positive
by

be

found

MI.

larger
MI,

MI

order's

became

of the

spread

negative

positive

(15

that

best

(see

to

been

market.

market

spread

size

the

times

by

Large

the

orders,

by

(seven

as

has

the

the

spread

spread

the

the

inside

small

with

inside

followed

Midsized

the

just

on

of

on

2012c)

order

as

side

went

MI,

Aldridge,

well

the

than

2006)

a limit

a negative

matched

buy

of
as

within

within

with

MI

order

orders

Lei,

(see

opposing

experienced

varied

average

of

a positive

placed

Source:

earnings

with:

and

volatility

Nasdaq

placed

Small-sized

the

posted

however,

and

Almgren,

to vary

Hanweck,

of the

details),

was

by

industry

2008)

found

direction

quote

orders

bid,

2000)

and

and

(Ferraris,

and

the
size

average

that

as

Hodrick,

2003;

been

the

orders

of the

Breen,

Burgardt,

experienced

limit
on

has

placed

2011,
ask

spread

and

depend

orders

Engle,

such

Mantegna,

durations

size

the

and

2005)

Volatility

The

(see

and

Li,

Dufour

characteristics,

company

Farmer,
and

(see

Nasdaq

MI

Empirical
Data

Estimation

Market

Impact

Preparation

MI

can

as

data

be

estimated

aggregate

an

arrival

liquidity

the

using

bid

of

Over

years,

buy

indicates

I or

trade
buyer-

and

discussed

Estimation
points

MI.

permanent

MI

decrease

an

arrival

tick,

on

the
in

bid

in

indicates

sell

several

in Chapter

increase

comes

Lee-Ready,

to

top-of-the-book
order.
across

orders

developed

well

one

size

of a market
one

as

allow
an

the

seller-initiated

quote,

data,

data

orders:

however,

and

in detail

and

the
trades.

methodologies
and

bulk

volume

4.

Model

separated

permanent

have

sells:

level

II tick

II

market

price

data,

Level

Level

and

order.

side

I and

only.

limit

researchers

buys

classification,

following

of

separating

separate

Level

a specific

limit

Level

challenge
the

at

both
stamps

arrivals

size
of

on

When

Data

trade

pinpoint

the

Basic

using

containing

precisely

to

of Permanent

Under
can

be

into

buys

and

the

assumption

estimated

sells

are

of

ready

for

estimation

of

form,

the

with

the

linear

functional

linear

regression

using

is the

number

specification:

(7)
where
time

t is the
units

measured,
the

as

of the

the

V t is

normalized

expressed
(8)

in

time

the
price

shown

trade,

event
size

window
of

the

change
in equation

at
trade

from
(8):

which

the

observed
time

of posttrade

at
to

time

trade
time

ticks
impact

t , and
and

can

or
is
is
be

Figures

5.8

each

and

trade

5.9

on

show

tick

data

Eurobund

futures

5.9

buyer-initiated

show,

5-tick

market

shown
on

on

on

the

trade

size

dependency

futures
size

10

is 99.999

on

trade

The

observed

next,

with

clear

statistical
possibly
Figure
Trades

5.8

by
results

of

in Eurobund

are
at
the

Intercept
Futures,

ticks

in

2009

and

As

exhibits

often

as

the

left

every

100

directly
0.001

the

actual

of

end

estimates

of

each
appears

the

the

trading

after

one

quarter
to

of the

Eurobund

ticks

from

MI's

is on

to

five

as

the

value

shows,

invariant

10,

dependency

contracts

percent

positive

While

attributable

largely
the

axis

and
The

reach

significant.
the

for

5.8

strong

that

persistent,

2010

properties.

very
of

following

Figure

following

5.9 , indicating

for

only

rollovers

the

statistically

price

trade

day

t -statistics

is highly

exceptions

to regular

Daily

the

five

(2012c).

trades

words,

futures

significance
due

buy

size

rose

trade.

display

, is small:

the

average

trades

percent

7 , in other

traded,

trading

of Figure

estimate,

of

every

Aldridge

size:

axis

estimated

per

of

trade

right

on

coefficient
order

(FGBL),

impact

dependency

of

and

day

each
to

the

when
dip

to

the
zero,

of futures.

of Model
20092010

(7)

Estimated

on

Buyer-Initiated

Figure

5.9

Daily

Buyer-Initiated

Figure

Trades

5.8

price

Size

plots

daily

observed

attributed
the

following
larger

trade

buy
the

contracts

impact.

The

persistent,
right

As

on

axis

accompanying

similar

scale

of

Figures

5.10

and

Figure

5.10

market
5.11

Daily

5.8
impact,

axis

5 , or
trade

5.8

about

FGBL

100
As

trades

Trades

impact
ranging

notes.

Seller-initiated
the

times

such,

the

impact

for

in

the

of 1 or

100

larger

than

100

size-dependent
is highly

t -ratios

increase

resulting

substantial

in

Figure

size-dependent

impact!

yet

be

price

size.

FGBL

cannot
in

of

contracts,

size-independent

Figure

10
to

generate

on

in

that

(left)

feature:

market

however,

of

of

dominates

specific

Estimated

increase

component

futures

comparable

with

value

attributable

Eurobund

observed

average
trades

order

impact

(7)

20092010

, the

the

the

Eurobund-futures

contracts,

Model

buyer-initiated

change

100

FGBL

the

is

price

incur

of

size-independent

market

under

following

Futures,

sizes.

trades

size-independent
trades

in Eurobund

estimates

average

than

of

following

to

shows,

Coefficient

from

statistically

about

5 to 20,

trades

opposite

incur

direction,

as
a
as

show.
Intercept

of

Model

(7)

Estimated

on

Seller-Initiated

Figure

Trades

5.11

Daily

in Eurobund

Size

Coefficient

Seller-Initiated

Trades

The

statistical

meaningfulness,

the

equation

(7)

however,
ranging

as
from

1 to

of

in Eurobund

applied

measured

Futures,

by

2 percent,

Futures,

or
to
the

the

20092010

Model

as

Figure

explanatory

power,
trades

R-squared
5.12

Estimated

on

20092010

buyer-initiated
adjusted

(7)

of

illustrates.

of model

in

is

quite

low,

the

model

and

To

put

the

to

2 percent

to

consider

adjusted
the

generalized
a

volatility

applications.
about
of

R-squared
Figure

(7)

Explanatory

of Equation

Additional

Explanatory

strong

invites
help

market

variables,

such

as

intertrade

durations.

where:

principle,

it

commercial

GARCH

usually

R-squared
is

of

the

comparable

to

models.
of

the

what

intercept
kind

variation

of

(9)

popular

of the

about

can

in

(GARCH),

of
the

of

Linear

Market

Impact

Variables

unexplained

one

power

many

while

needs

(7)

estimation

variables,

low

in

power

words,

Power

significance

questions
the

quite

of other

Estimator

The

is

predictive

used

predictive

In other

one

heteroskedasticity

model
of

though,

on

conditional

5 percent.

estimates

perspective,

R-squared

R-squared

equation

5.12

in

estimation

The

only

model

adjusted

autoregressive

popular

hits

R-squared

often

observed
To

extend

analyze
the

of

model

model

additional

captured

impact
the

in the

by

the

(7)

if any,

can

variables,
. Commercial

deploy
spread,

of equation

models

additional
short-term

potential

of equation

explanatory
volatility,

impact
(7)

as

for

of

follows:

and
these

As

before,
represents

the

-tick

change

in price,

(10)
V t denotes
is

the

the

several
low

of the

estimate
ways,

prices

before

size
of

as

short-term

standard

during

trade

tick

the

trade

at t

volatility,

and

deviation,

predefined
t,

recorded

as

or as

pretrade
first

can

a log

measured

change

period

suggested

be

of high

beginning

by

in

Garman

and
ticks

and

Klass

preceding

each

(1980):
(11)
is the
trade.

average
In

data

difference

spread

observed

containing

quotes,

between

the

best

during
the

offer

the

ticks

average

spread

and

the

bid:

the

average

best

is

the

mean

(12)
When
still
at

quote
be

intervals

the

hits

Aldridge

holds

is

assuming
to

true
buy

best

unavailable,

by

equal

market

the

spread

is

estimated

generally
as

data

the

that

sequential

effective
a buy

follows

is executed

at the

best

The

resulting

shown

in

equation

(19),

changes

occur

an

assumption

a sell

and

vice

for
an

can

Such

ask,

expression

spread

price

spread.

whenever

bid.

effective

and
the

the

versa,

market

average

approximation

sell

effective
noted

by

(2012c):

(13)
Finally,

is the

average

clock

time

between

every

two

subsequent

trades:
(14)
Aldridge

(2012c)

additional
impact,
intercept,
impact

estimates

explanatory
but

do

or change
of intertrade

not

equation
variables

eliminate
the

value

durations,

(9)

in FGBL

influence
the
of the

the

statistical
trade

, is consistent

data.

size

In

FGBL,

resulting
significance

coefficient,
with

that

the

market
of

the

V, .

The

described

by

Dufour

and

durations

Engle

are

observation

(2000):

shorter.

more

gradual

lower

price

changes

is

lower

According

is likely

allow

MI

due

to

to the

when

Dufour

following:

impounding

average

and

Engle

shorter

of information

in response

to each

coefficients

obtained

intertrade
(2000),

intertrade

into

the

durations

prices,

resulting

in

trade.

Causality
The

estimates

(7)

and

MI

is

of

(9)

can

trade

trading
the

several

and

For

trade

and

Engle

degree?

In

effects,

such

by

and

direction

addition,
as

and
the

any

it

VAR

model

persistent

impact

turn

would
on

the

sizes?
so-called

vector

value-at-risk,
later

answers

the

direction

us

to

irregularities

during

specification

of

VaR,

extended

by

following
of

influence

allows

with

truly
in

and

of trades

that

dependent

from

influence

since

relationship

the

(1991),

MI

equations

possible

which

deploy

model

it

a trade

trade

can

VAR

size

of

appear

(different

does

its

size

follow-up

Hasbrouck

is

trades,

make

one

The

question:

the

of

perspective:

trade,
and

following

model

(2000).

cause-and-effect
or does

the

questions,

proposed

each

could

models

following

impact

of the

(VAR)

first

trades,

of

instead

these

autoregressive

Dufour

example,

from

the

after
the

impact

size,

answer

model),

ticks

directions
market

original

from

determine

volume?

size

questioned

ticks

exaggerate

To

be

measured

other

MI

MI

subsequent
to a

test

key

for

greater
auxiliary

different

times

of

day.
The

Dufour

ascertains
be

and

Engle

causality

written

as

(2000)

within

five

ticks

from

the

the

time

VAR
of

model

each

that

trade

can

follows:

(15)
(16)
where
R i

is

calculated

the

instantaneous
as

one-tick

market
log

impact
return,

following
that

is,

a
as

trade
difference

tick

i ,
of

logarithms
tick

of

the

price

of

tick

i , and

the

price

trade

i , with

Q i =1

of

the

previous

trade

i- 1.

(17)
Q i is the

direction

of a buy

of

market

the

order,

D t,i is a dummy

and

variable

Q i =

1 for

indicating

when

a trade

seller-initiated

the

hour

is a

result

trades

of the

day:

(18)
b j and

d j , in turn,

are

functions

of trade

size:

(19)
(20)
Equation
the

(15)

measures

following

data:

trades,

time

lagged

direction

lagged

of day

(16)

predict

the

lagged

considers

how

considers

whether

the

directions

and
of the

Tables
(16),

5.2

and

5.3

shown

in

and

shown

significant

only

executed

between

due

executed
to

price

five

trade

the

one

and

the

product

trade

size

trades

of
QV

variables

words,
returns,

five

estimation

year

equation

(16)

time

of

day,

predict

the

can

other

11
hours
in

for

May

to

the

flash

is

Trade
on

effects

and

equations

indeed

of
to

future
the

11

time

had
the

response

of

a lower

to

This

The
day

2009,

lagged

hourly

MI

trade

trades

was

As
by

have
dummy

statistically

period.

The

market

impact

could,

and

crash.

of previous

time

day.

6,

and

MI.

GMT

GMT
of

sizes

(15)

determined

contemporaneous

size.

10

of

data
prior

MI

impact

the

trending

on

preceding

explanatory

other

futures

trades,

little

at

and
and

one-tick

of

Eurobund

10

In

previous

results

exactly

for

Q ,

same

trade.

five

R i

trade.

previous

measuring

trades

the

of

MI

contemporaneous

direction

5.2 , instantaneous

produce

impact

trades,

well

next

one-tick

occurred,

trade

the

show

concurrent

to

variable

day

Table

accompanying

of

on

trading

trade

past

impending

respectively,

regular

sign,

sizes

the

posttrade

market

preceding

and

of

direction

each

five

direction

of

one-tick

when

of

contemporaneous
Equation

dependency

potentially

macroeconomic

trades
than
be
news.

Interestingly,

the

Adj.

46.25

R 2 , was

Table
on

5.2

Table

by

significant

role

direction

by

the

preceding

presented

in

immediately

following

being

while

a buy;

percent

sequential

72.5

percent,

equation
a.m.

was

buy

happened
to

noon

hour

on

May

(15),

equation

indicating

6,

FGBL

Results
Trade

be

The

a
likely

the

2009.

showed

with

of
Tick

be

buy

market

impact

power

OLS

Estimation

Data

for

May

an

16

estimates

of

of
6, 2009

17

to

Q i

after

rose

trade
GMT

17

to
sign

(11:00

GMT

was

during

that

return

R 2 of nearly

VAR

a 62.5

and

trades
the

also

three

a buy

to

16

FGBL

given

trade

had

the

of
the

of

initiated,

buy

of the

to

following

in

sell,

each

FGBL

buys

dummy

from

adjusted

or

likelihood

trade

from

of
the

2009,

determined

trades,

buyer

hourly

play
Instead,

entirely

of observing

dummy

a high

explanatory

to

impact

trade)

consecutive
A

market

trades.

percent

buy.

R i

according

6,

46.2

significant

hourly

As

May

(15),

6, 2009

recent

upcoming

probability
The

only

the

or the

two

a preponderance

(16)

a high
5.3

to

following

69.3%

nor

Specifically,

had

number.

ET).

indicating

trades

being

trades,

a large

positive,

on

of

size

is almost

5.3 , on

buy

a trade

buys

four

Table

probability

consecutive

Table

May

preceding

generates.

by

Equation

for

of

trades

of those

trades

model,

most

direction

measured

Data

the

immediately

size

VAR

trade

(except

subsequent

of the

regression,
figure.

Tick

the

in determining

of

the

of

Trade

trades

direction

estimates

Estimation

neither

the

of

substantial

(FGBL)

past

of

independent
the

OLS

shows,

generated

the

of

futures

5.3

power

percenta

Results

Eurobund

As

explanatory

equation
40

percent,

variables.
Model,

Equation

(16),

question

that

conditions
of

change

May

6,

2010
market
As

shown

on

May

May
in

6,

flash

Tables

crash),

day

on

parameters

given

trade.

to

the

during

the

the

answers:

difference

of

those

MI

as

on

performs

also

percent

estimated

flash

crash
May

(15)

estimated

6,

regular

recorded

the

the

much

as

in

of

most
of
future

predictions

30

of

model

predictions
accurate

data

well

shows,

out-of-sample

99.9

market

from

equation

model

(2012c)

robust

whether

present

estimates

Aldridge

is

prior.

similar

delivering

later

little

year

that

makes

impact,

based

very

As

(15)

market

one

implying

conditions.

5.5

exhibit

(2012c),

are

year

and

values

Aldridge

2009,

equation

investigation
one

5.4

6, 2009,

crash,

market

further

dramatically

2010.

(flash

the

begs

of

minutes

ahead

for

high-frequency

MI
of

Summary
Transaction
trading
for

the

costs

systems.

Among

most
the

profitable

systems.

costs

2. What

are

all

costs,

cost

hurdle

however,

proportion

current

End-of-Chapter
1. What

a considerable

significant

measuring

3. Which

present

of

structure

market
costs.
is

impact

Understanding

imperative

to

accounts
and
designing

Questions
are

present

latent

type
kind

5. What

is market

6. Can

market

7. Can

expected

costs?

of cost

4. What

in the

of tax

How

is most

structure
impact?

impact
market

do

Why

impact

they

markets?
differ

dominant
do

following

financial

in today's

high-frequency
does

from

transparent

costs?

markets?
traders

face?

it exist?

a market
of a buy

buy
order

order
be

be

zero?

negative?

Explain.
Explain.

8.

What

temporary
9. What

is

permanent

market
market

market

impact?

How

limit

orders?

impact?
impact

accompanies

different

is

it

from

CHAPTER

Performance
Over

and

the

past

few

high-frequency
vary

Capacity

author

developed

time

to

and

required

to

estimation

of

studies

most

discusses

strategy

evaluate

strategy.

capacity

and

metrics

have

The

illustrates

results
been

performance.

approaches
capacity

measure

The

strategy's

popular

to

capacity.

different
a

Strategies

attempted
and

Many

illuminate

the

Trading

performance

author.

summarizes

measurement

several

(HFT)

to

over

chapter

years,

trading

from

of High-Frequency

for

and

capacity

performance

the

chapter

length

also

of

This

HFT

of time

discusses
with

specific

examples.
Principles
One

of Performance
can

manage

measurement

is

management

and

standard

several

evaluate

HFT

At

the

three

only

heart

something

therefore
of

investment

setting,

Measurement

HFT.

critical

While

metrics

of

metrics

apply
have

Performance

function

many

management

other

measurable.

well

been

in

investment

developed
a

for

high-frequency

developed

specifically

to

of

lie

activity.
of

successful

investment

management

HFT

P's:

Precision
Productivity
Performance
The

first

exactitude
from

the

strategy
solid

P,

precision

required
losing

Standardized
strategies,

mathematical

to quickly

ones.

apparent
predictor

of

and

Statistical

reliably
tools

profitability

is

metrics,

refers

separate

also

help

short-term

winning
ascertain

fluke

to

the

strategies
whether

or

statistically

all

investment

of performance.
metrics,
deliver

when
the

applied

second

process.

Statistical

tools

are

deployed

to optimize

a diverse

P,
highly
array

uniformly

across

productivity

of

scalable

and

of investing

ideas.

the

investment

can

be

rapidly

The

third

P,

systems.

performance,

In an

arrive

with

environment

and

and

sustainable

The

three

risk-management

P's

share

data,

consequently

on

back

longer-lasting
to
time

proper

from

high-frequency
multiple

measuring

sources

systems

equipped

with

capabilities

are

critical

for

data.

The

simplest,

tick

be

very

to

the

the

latest

are
portfolio

effective

and

without
the

and

improvement.

and

or

financial

tick

opportunity

has

data

can

to

data

instruments,

to limit

stronger,

allow

of

allow

strategy

forming

performance

and

identify

and

Finally,

least

Additionally,

accurately

development

the

informative

strategy,

events

evaluate

losing

trading

quickly

systems.

monitor

most

Trade-by-trade

strategy

market

strategy

is

valuable.

of

specific

Performance

data,

for

allows

into

along

requirement:

risks

trading

profitability
Basic

to

data

dependencies

in real

as

approaches

granular

managers

key

such

short-term

risk-mitigating

feeding

one

proven

the

points

to

investing.

processed,

most

systems

relevant

data

frequency,

Monitoring

tracking

particularly

where

nanosecond

necessary.

reveal

is

investment

trading

strategies

losses

and

enhance

way.
Measures

Return
Trading
one

strategies

may

characteristic

that

feasiblereturn.
prices

but

resulting

level

of

makes

Return

is most

The

come

often

can

financial

cross-strategy

and

shapes

comparison
be

as

under

cross-asset

sizes,

as

is

a dollar

and

of performance:

Figure

return
6.1

of equation

Illustration

(1)

is illustrated

of a Simple

Return

in Figure

6.1 .

share

strategies
difference

of

(1)
Simple

they

change

independent

consideration

comparison

but

different

a percentage

measure

instrument

and
across

expressed

considered

dimensionless

the

in all

in

in value.
the

price

allows

easy

An

equivalent

log

return

metric

is shown

in equation

(2):

(2)
It can
(1)

be

are

shown

that

nearly

choice

of the

mathematics

at

identical
return
of the

high
to

metric

frequencies,

log

returns

is often

estimation

simple
defined

dependent

returns
by

on

of

equation
the

equation
(2).

application

The
and

models.

Volatility
Volatility
down

of
around

known
and

as
6.3

its

6.2

measures
average

dispersion,
illustrate

comparatively
Figure

returns

is often
low-volatility

to each
Example

how
value.
taken
and

much

the

return

The

movement

as

a measure

high-volatility

moves
of

of a Low-Volatility

Market

returns,

of risk.
conditions,

other.
Condition

up

Figures

and
often
6.2
drawn

Figure

At

6.3

least

over
and

Example

of a High-Volatility

a dozen

a specific
so

on.

Simple

measures
period

The

most

standard

measure

of

of

volatility

time:

exist,

round-trip

common

deviation

Condition

trade,

measures
of

returns

each

metric

minute,

of volatility
is

by

far

assessed
day,

month,

include:
the

most

popular

of volatility.

(3)
where

is a simple

Weighted
used

average

average

of N observations

deviation,

emphasizing

preceding
later

time

t.

observations,

is also

often:

(4)

where

, and

to

individual

each

and

w i is the

to increase

data

returns
toward

importance

R i . All
the

most

weight

corresponding

w i are

often

chosen

current

observation:

to

add

up

to

(5)
(6)
Average

of

open,

high,

low,

and

close

prices

is

another

metric

of

of time

is useful

in

volatility
(7)
High

minus

analyses
low

metric

low
where

recorded
many

avoids

during
lagged

endogeneity

a specific
variables
problem

period
are

present.
inherent

The
with

high

minus

standard

deviation
from

and

lagged

lagged

standard

returns:

returns,

deviation

the

since

standard

analysis

is faulty

by

deviation

comparing

is

lagged

computed

returns

and

design.

(8)
Average
yet

square

another

returns

metric

particularly

of

well

conditional

recorded

over

volatility.

with

This

applications

heteroskedasticity

one
like

certain
has

period

been

shown

generalized

(GARCH)

(see

historical

loss.

of

time
to

is
work

autoregressive

Bollerslev,

2005):

(9)
Drawdown
Drawdown

is

maximum

loss

referred

to as

performance

measure

relative
the

to

water

fees

of

only

the
mark.

after

highest

previous

Investment
they

It

exceed

is

recorded

value,

the

managers
the

as

latter

typically

highest

water

often
receive

mark

on

record.
Maximum
history
Figures

drawdown
of the
6.4

identifies

investment,

and

6.5

and

illustrate

the
helps

biggest

drop

illustrate

computation

of

price

potential

of maximum

or

value

downside
drawdown.

Figure

6.4

Maximum

Drawdown

Computation,

Scenario

Figure

6.5

Maximum

Drawdown

Computation,

Scenario

in
risk.

Formally,

maximum

among

practitioners

observed
the

drawdown

in

lowest

minimum

that

historical

return

that

prior

occurred

on

data

at any

is then

the

past

drawdown
water

As

peak-to-trough

the

marks.

last

global

The

documents

data.

supersedes
the

is

the
such,

to

lowest

lowest

tail

maximum

the

the

last

global

return

in between

drawdown

is

popular
of

drawdown

records

high
two

known

to the

maximum

maximum
as

losses

maximum

global

is known

risk

severity

global

next
The

in time

of

maximum

from

maximum.

point

measure

that

measured

water

mark.

successive

as

the

A
high

maximum

drawdown:
(10)
In

risk

management,

related

to

described
Ratio

Win

ratio

months

concept

value-at-risk,
in detail

Win

the

explains
ended

measure

in Chapter

what

of

maximum
of

drawdown

worst

is

quantity

of

closely
losses,

14.

portion

of the

trades,

trading

days

or

trading

profitably:

(11)
Win
better

ratios

help

compare

forecasts

monitoring

run-time

result

accuracy
in

higher

performance.

of predictive
win
As

ratios.
a simplest

signals
Win

ratios
metric,

of

strategies:

also
win

help
ratio

in
can

be

used

to assess

with

prior

the

strategy

Figure

whether

performance.

Is Reaching

Similar,

but

run-time

performance

Average

Gain/Loss

Average

gain
to

cousin
average
a
the

high

may
win

observed
gain
win

in

as
ratio

loss

ratio
6.6

Shown

tests

for

is

may

consistent

indicate

that

illustrates.
May

Indicate

the

are

for

all

two

The

that

the

metrics,

loss

days

or

is

closely

also

in Chapter

profitability
the

of

statistically

discussed

Similarly,

trades,

consistency

14.

average

loss,

average

recorded.

evaluating

in Chapter

of expected

of the

14.

strategy

average

months

close

loss
when

The
when

computes

the

strategy

performance.

deliver

with

additional

may

tolerate

average

loss,

while

compared

with

the

and

win

Figure

described

conjunction

ratio

the

Ratio

drawdown.

is

negative

Win

average

measures

total

Considered
metrics

are

concept

result

average

delivered

in

maximum

positive

in
as

advanced,

and

gain

decline

record

Capacity

more

of the

performance

capacity,

Decline

Strategy

related

is reaching

6.6

present

average

gain/loss

the

win

ratio,

insights.

For

lower

average

low
average

win

ratio
loss.

combinations

average
example,
gain
requires

The
are

gain

and

a strategy
relative

outcomes
shown

with
to

a high

loss

the

average
of

various

in Figure

6.7

.
Figure

6.7

Outcomes

of

Win

Ratio

and

Average

Gain/Loss

Combinations

Formally,

for

following

the

strategy

inequality

must

to

deliver

positive

performance,

the

hold:

(12)
The

inequality

for

evaluating

manager

of

equation

(12)

performance

can

of

be

used

candidate

as

first

step

investing

litmus

test

strategies

and

credibility.

Correlation
Correlation

measures

another

strategy

co-movement

or financial

of strategy

returns

with

those

of

in a

portfolio,

the

resulting

instrument:

(13)
When

two

strategies

traditional

portfolio

portfolio
when
Simple
robust

is

diversified

performance

prices

of

correlations

with
financial
in rising

correlation

management
(i.e.,

however,

measurements
problem

low

of another

correlation,

main

with

of
simple

theory
allows

may

no

co-movement
correlation

in falling

combined

suggests

one

strategy

instruments
and

are

that

strategy

to

temporarily
longer

be

sufficient

financial

is

the

markets.

up

returns

declines).

of

display

pull

in delivering

instruments.

following
increasingly

Specifically,

The

observation:
divergent
when

broad

market

indices

500,

for

to

market

states

of
can

price

the
be

other

fall,

highly

entity

two

markets

of

financial

Such

by

one

of

from

of

as

& Poor's

any

most

the
the

instruments

the

in

or

sample

measured

broad

instruments

correlation

data

financial

When

asymmetric

[S&P]

two

financial

divergent

dividing

the

Standard

instruments.

returns

is known

computed

as

vary

however,

correlated.

returns

such

correlations

any

indices

become

(a market

example)

instruments

that

rise

tail

different
correlation

into

points

when

are

positive

and

negative:
(14)
(15)
In

both

equations

reference.
of

Equation

returns

same

of

for

the

correlated

losses

Alpha

Beta

and

performance
high-frequency
return

the

reference

S&P

500

that

is

alpha
avoided,
used
vice
Beta,

the

the

index.

Thus,
of

desirable.
unless

to trade

in

while

for

and

(15)

much

correlated

states

does

the

negatively

helps

is

as

up

returns

portfolio

states,

the

equation

delivering

existing

equally
At

buffer
more

returns

core
valuable

and

the

suited

most

strategy

or the

broader

alpha

reflects

the

prevailing

level,

abstracted

of

the

thereby

when

evaluation
alpha

any

performance

the

influences
by,
of the

alpha

signals
strategy

of

measures

conditions.

negative

strategy

investment

measured

market
a

of

for

basic

markets,

with

negative-alpha

reversebuy

measures

well

its

Strategies
the

1,

now-mainstay

strategies.

portfolio

correlation

positively

two

by

1 is taken

portfolio.

are

achieved

the

down

are

independent
is

of

instrument

A strategy

of wider

that

the

states.

in

beta

financial

computes

delivering

amplifying

Alpha

(15)

instrument

those

strategy

and

(14)

down

losses

the

and

financial

to

portfolio's
than

(14)

are

advises

the

strategy

can

say,

by

positive
generally

be

profitably
to sell

and

versa.
however,

is a multiplier

that

measures

exactly

how

the

strategy

responds
the

to the

strategy

reference

current

is

likely

portfolio

average,

the

portfolio

positive

Alpha

and

to

rises

strategy

declines.

either

market
deliver
in

is

trends.

beta

estimated

negative

the

may

beta

better

indicates

that

when
shows

when

investor's
be

using

beta

performance

perform

on

or negative
are

to

Depending

beta

positive

value.

likely

A positive

the

the

that,

on

reference

cumulative

portfolio,

preferable.

a linear

regression

(OLS):

(16)
where

R i,t is the

a unit

of time

over

the

t , R p,t

same

estimated,

return

of

i,t is the

observation

when

assumptions

of the

a high-frequency

is the

period

and

on

return
time,

on

the

equation

i are

is

of equation

portfolio

the

estimation

(16)

i observed

reference

i and

statistical

model

strategy

observed

parameters

error

applied

over

specific

to

(16),

the

errors

and

beta,

as

to

data.

be

to each
By

the

i,t average

to

zero.
Figure

6.8

graphically

scatterplot
plotted
on

of

the

HF

strategy

line

is

straight
Alpha
vertical
Figure

returns:

along

the

is

the

illustrates
the

horizontal

fitted
intercept,

alpha
returns
axis,

are

plotted

through
or

the
the

of
and

the

along
data,
point

the

computed

reference

portfolio

contemporaneous
the

the
where

vertical
slope

of

the

the

line

Graphical

Representation

of Alpha

and

Beta

a
are

returns

axis.

axis.
6.8

from

When

line

intersects

is

a
beta.
the

Skewness

and

Kurtosis

Skewness

and

kurtosis

the

distribution

tendency
Positive
more

of

Returns

the

returns

likely

to

post

possible
6.9

additional
of

strategy

skewness

illustrates
Figure

of

are

Skewness

to

of a return
positive
skewness
Values

the

parameters

strategy.

deliver

returns

Skewness

positive

distribution
than

used

implies
negative

or

to

describe

describes
negative

that

the

returns.

the
returns.

strategy
Figure

is
6.9

scenarios.
and

Their

Meaning

in

Distribution

of

Figure
Returns

6.10

Kurtosis

Values

and

Their

Meaning

in

Distribution

of

Kurtosis

measures

severely

positive

the

likelihood

and

severely

When

kurtosis

is

kurtosis

is

extreme

returns

the

of kurtosis.

idea

The

average

return,

prespecified
the

mainstays

trading

of

of time

In

kurtosis

of

distributions.

As

distribution

when
usual,
to

indicates

prevalence

of

indicates

that

indicates

whether

kurtosis

Comparative

the
more
been
risk

Table

the

measures

of

tails,

return,

the
a

a
are

different

volatility,

and

skewness

and

their

return

position

of

the

positive

skewness

negative

skewness

is

distribution

negative.
are

Kurtosis

normal;

higher-than-normal

high

probability

of

events.

not

in an

in a single

number
6.1

Performance

deviation,

performance

lend

to an

Several

developed

Table

standard

of the

do

strategies.

6.1

tails

while

returns

over

Ratios

a picture

strategies.

of

or negative

average

present

proportion

illustrates

for

of

the

average;

returns,

When

reporting

shape

illustrates

return

likely.

frequency

quote

the

of

normal

6.1 0

return,

sometimes

positive

fat

positive

average

is,

to

drawdown

and

describing

the

large

signifies

extreme

While

the

skewness

relative

Figure

at a predefined

comparison
to

are

maximum

measured

that

relative

returns

unlikely.

practitioners

returns

returns

and

addition

drawdown,

occurrences,

extreme
are

performance

strategies.

maximum

high,

volatility,

window

extreme

negative

returns.

low,

of

and

of a particular

easy

point

that

can

summarizes
Measure

be

used
the

mean,
to compare

most

Summary

popular

strategy,

among

performance

to summarize

drawdown

trading

comparison

comparative

attempt

maximum

two

metrics
variance,
different
point

or

have
and

tail

trading

measures.

The
in

first
the

generation

1960s

and

of
include

point
the

performance
Sharpe

measures
ratio,

Jensen's

were
alpha,

developed
and

the

Treynor

ratio.

measure

The

Sharpe

in comparative
metricsaverage

of capital

borrowed

Sharpe

winner

of

the

concept

textbook

definition

strategy

was

Nobel

enduring

of the
is the

most

evaluation;

widely

used

it incorporates

standard

in

Memorial

1966

Prize

in

the

deviation,

and

three
the

cost

leverage.

designed

used

, where

probably

return,

for

ratio

is

performance

desirable

The

ratio

the

by

Sharpe,

in Economics;

study

Sharpe

William

and

ratio

annualized

it is a

practice

later

remarkably

of

finance.

is

average

return

from

trading,

R is

the
annualized

standard

risk-free

rate

opportunity
with
HFT

no

zero.

Fed

as

trading

with

are

(e.g.,
cost

the

deviation

as

the

carried

Therefore,

that

the

is

position

It should

positions

trading

Funds)

well

activity.

of

be

returns,

and

R F

included

to

capture

carrying

noted

overnight,

that
the

high-frequency

costs

Sharpe

the

instruments

carrying

ratio

the

associated

in most

position

is

costs

is computed

as

follows:

What
in

makes

the

comparison

with,

Sharpe

ratio

efficient

securities.

Consider

is

line

point

to

It

has

Sharpe

ratio

trading

strategy,

Sharpe

ratio,

set

of all

the
of

all
or

the

the

portfolios

the

Sharpe
rate

portfolio

ratio

and
(M

The

and,
in

financial
security

the

the
is the

passing

for

bold

For

slope

of
a

portfolio),

line

tangent

is the

efficient

frontier

the

highest

any

instrument
is to the

classic

through

market

correspondingly,
set.

the

mean-variance

illustrates

combinations

slope

individual

closer

risk-free

portfolio

highest

the

performance,

Surprisingly,

selecting

which

security.

of

return?
for

figure,

given

individual

measure

absolute

example

the
a

appealing

metric

for

from

or

raw

In the

corresponding

mean-variance
itself.

frontier.

strategy,

an

effective

6.11

emanating

trading

ratio
say,

an

Figure

mean-variance
the

Sharpe

efficient

other

A , the

to

the

portfolio,
higher

frontier.

the

Figure

6.11

portfolio

Sharpe

has

the

Sharpe

ratio.

Sharpe

himself

Sharpe's

framework

for

the

percent

of

financial

security.

he

ranked

first

according

example
Jensen's
broad

the

for

mutual

each
in

picked
Sharpe
of

the

of a successful

market

is

the

then

influences,

develop

following
be

measure,

20

securities.
the

Sharpe

the

measure
capital

of

the

highest

portfolio

with
and

Sharpe

more

than

particular

portfolio

solution:

now
the

was

selection

to

is

best

invested

Equally

he

portfolio
no

what

known
performance

5 percent
weighted
ratios

as

of the
portfolio

is

just

one

application.

performance
asset

following

on

highest
ratio

market

which

allocated

securities

ratio

for

constraint:

universe
20

The

developing

fund

to

created

the

with

when

could

security

securities

alpha

was

Slope.

correspondingly,

metric

portfolio

Sharpe

then

and,

the

with

fund's

to the

into

fund

Mean-Variance

with

mandate

the

ratio,

allocation

up

slope

mechanism

consulting.

fund

as

highest

came

optimization

Sharpe

Ratio

pricing

that
model

abstracts
(CAPM)-style.

from

Jensen's

alpha

returns
The

third

from
While

the

the

chosen

CAPM

(2002),

tail

risk

in

ratio
an

deviations

are

performance.

New

the

unit

of

variability

average

risk

of

return

proxied

Sharpe

ratios

linked

to

they

adverse

primary
the

may

(2004),
on

use

in

by

beta

take

into

tail

the

Sharpe

example,

derivative

distribution

and

distributed
the

fat

of
that

tails.

and

have

the

Ignoring

overestimate

been

in

use

instruments

risk

inherent

Kat

present

surrounding

of

risk

and

nonnormally

measures

the

not

Brooks

for

underestimate

performance

do

returns.

concerns

return

capture

subsequently

returns

of

most

strategies.

A natural

extension

risk

standard

from

effort

to

ratio,

developed
measure

of

ratio,

for

replacing

average

and
a

adverse

that

strategies.

precisely
the

as

a proxy

Burke

the

the

tail

risk,

returns

the

ratios

average

metrics

are

ratio,

the

are

the
of

as

as

by

the

Kestner

volatility.

Finally,

the

is also

standard

The

meaningful

when

strategies.
Sharpe

adverse
lower

frequently

measure.

trading

extended

known

for

volatility

of

metrics

drawdown

ratio

returns

performance
of

Calmar

metric.

Sharpe
in

negative

The

uses

a volatility

in an
ratio,

described

(1994),

of

Calmar

maximum

drawdown
by

measure

The

that.

first

as

the

the

methodology

ratio,

class
with

change

Sterling

comparing

return

do

drawdowns

only

is to

the

uses

positive

Greek

volatility

of

developed

including

goes

risk
ratio

The

ratio

to a drawdown-based

(1991),

to ignoring

estimating
response,

Burke

of maximum

In addition

argument

tail

Young

the

Burke

criticized

the

volatility.

uses

deviation

deviation

and

by

(1996),

of

capture

Sterling

These

the

indices.

popular,

Sharma

normality

to

trading

per

and

asymmetric

from

developed

measures

extreme

researchers'

Sharpe

result

of

using

The

ratio,

remain

(2004),

against

returns.

market

benchmark

metrics

Mahdavi

cases

consideration

chosen

Treynor

three

the

into

estimation.

these

account

with

ratio,
of

the

the

takes

in co-movement

excess

the

implicitly

ratio

returns

partial

In
by
only.

moments

(LPMs)

and

mean,

are

standard

in computation
Thus,

and

deviation

LPM,

benchmark.
(1991),

uses

of the

benchmark,

in the

Sharpe

Kaplan

and

Sharpe

ratio

returns

below

ratio

standard

the

second

calculation.

with

the

LPM

benchmark.

Sortino,

van

return

above

average
per

unit

of

of

the

Finally,

der

Meer,

the

the

selected

and

van

that

der
short

volatility

developed

the

by

deviation

in

the

skewness

of

the

Upside

Potential

Platinga

benchmark

(the

first

of

returns

deviation

the

fell

of return

and

standard

the

below

standard

the

and

with

returns

returns,

only.

replaces

3 measure,

the

used

calculation

a measure

Kappa

data

Shadwick

Sortino

of the

replaces

the

by

as

The

third

by

fell

developed

LPM,

(2004),

that

(i.e.,

benchmark

ratio

deviation

the

(2004),

Sharpe

returns

ratio,

the

Knowles

by

moment)

Sortino

a specified

Knowles

in the
the

that

developed

and

of

of a distribution

except

below

Omega,

returns

average

moments

skewness),

Kaplan

of

The

produced

as

and

the

regular

returns

known

standard

the

deviation,

metric
(2002)

Meer

as

comprises

Keating

first

computed

(1999),

ratio,
measures

higher

partial

below

the

benchmark.
Value-at-risk

(VaR)

measures,

discussed

gained

considerable

popularity

risk

a convenient

point

in

VaR

measure

percent
often

Z-score
used

on

companion
expected
cutoff
of

MVaR

format

in

dollar

measures

Of

course,

the

the

original

whereas

the

issue,

a modified

this

normality.
in place

and

Gueyie
Gregoriou

of standard

returns

profit

and

Gueyie

of

to

(2003)

in Sharpe

ratio

is

also

VaR

known

as

return

within

the

distributions
fat-tailed.
was

account
also

99

also

be

measure

or

loss).

normal

into

tail
The

percent,

and

value

takes

the

metric

(CVaR),

(MVaR)
and

95
(the

known

also

framework.

returns

assumes

14,

summarize

percent,

average

(2003)

deviation

90

VaR

are

VaR

to

of daily

VaR

in Chapter

a statistical

of

conditional

(EL),

tail.

the

distributions

the

able

within

distribution

loss

Gregoriou

from

real

metrics

identifies

cutoff

measure,

returns,

address
by

essentially

as

in detail

To

proposed
deviations

suggest
calculations.

using

How

do

turns

these

out

performance

that

strategies.

all

Eling
performance

Sharpe

ratio

adequate

measures
for

each

rankings

(2007)

measure

attribution

benchmarking,

goes

and

has

and

hedge

fund

other?
of

compare

listed

nutshell,

analysis,
back

been

(1992)

strategy

and

i that

to

Fung

often

the

applied

performance

period

against

comparable

Schuhmacher
13

up

It

trading

hedge

fund

conclude

that

the

performance.

Attribution

Performance

Sharpe

stack

deliver

of the

is an

Performance

(1977)

metrics
and

ranking

metrics

arbitrage

to

and

attribution

invests

into

t , with

j = 1, ,

x jt

the

notes

an

as

of

Ross

performance

among

that

In

return

with

factor

by

others.

t- period

securities

underlying

to

theory

strategy

(1997),

individual

J , has

pricing

trading

Hsieh

referred

returns

on
r jt

in

structure:

(17)
where
portfolio

is

at time
. The

can

relative

weight

of

the

j th

financial

security

in

the

t,

j th

financial

security,

in

turn,

has

period-

t return

that

be

explained

by

K systematic

factors:

(18)
where
1,

F kt

is one

of

K , l is the

return

in period

to

broad

be

securities.
returns

K underlying

factor

loading,

t . Following
asset

as

and

as

factors

e jt is the

Sharpe

classes,

Combining

systematic

(1992),
well

equations

as

(17)

in period

security

factors

j
can

individual
and

(18),

idiosyncratic
be

stocks
we

t, k

can

assumed
or

other

express

follows:

(19)
reducing
underlying

the

large

strategy

Performance

attribution

the

returns

(20)

strategy's

number
i 's

returns
to

on

of
to

various

a basket

financial
a

small

factors
of factors:

securities
group
then

of

potentially
global

involves

factors.
regressing

where

bk

attributed

measures
to

generate

factor

strategy

the

basket

return

asset

classes

Three

Hsieh

IFC

Two

One-month
price

of

Performance

The

designer

Near-term

of the

and

lends

to

strategy's

value

the

eight

added

of

performance

global

attribution

equities,

of

groups

of

benchmarks:

MSCI

U.S.

nonU.S.

equities,

government

bonds

and

bonds
representing

cash
and

the

Federal

currencies

a useful

in

measure

Reserve's

in aggregate.

of

line

is

itself

to

strategy

returns

for

styles

of

by

the

investment

the

of

details

and

on

instead

reported

development

and

account.

forecasting

analysis

consideration
be

HF
costs
is,

the

strategies.

attribution

attribution

it may

That

of

of

(see,

for

2001).

under

transaction

value

other

allows

performance

performance

the

added

with

factors

Titman,

of

true

comparison

on

a benchmark,

benchmark

of

trending

based

in the

measure

to easy

strategy

into

capture

addition

persistence

high-frequency

taken

of

following

accurately

attribution

Jegadeesh

are

the

ability

strategy.

example,

dependency

idiosyncratic

measuring

is

strategies

performance

bottom

the

commodities

index

may

strategy

The

measures

be

reasons:

Performance
strategy

u it

can

equities

proxying

technique

black-box

U.S.

deposit

attribution

following

the

JPMorgan

dollar

that

persistent

in excess

government

gold

trade-weighted

the

market

Eurodollar

strategy's

performance

MSCI

nonU.S.

strategy

factors.
that

as

classes:

JPMorgan

The

well

emerging

return

find

classes:

bond

model,

strategy

(1997)

the

t.

strategy's

serve

equity

the
and

attribution

of weighted

and

of

i measures

in period

is the

Fung

performance

returns,

performance

the

and

k,

abnormal

idiosyncratic
In the

the

exhibits

cheaper

strategy,
as

well

when

the

is this:

to

invest

particularly
as

the

associated

performance

if

the
high

into

the
when
risks

attribution

beta

is

high,

into

the

alpha

into

benchmark

of aggregate
has

The

benchmark

is

HFT

strategy,

liquidation

any

effective

one

of

to

the

invest

following

can

be

lower

liquid

the

or

simply

costs

less

traded

in

in

costs.

drawdown

more

and

passive

risk.

than

the

instrument

benchmark

therefore

faces

the
lower

Evaluation

Over

the

called

past

few

years,

low-capacity

low-capacity
all

In
in

profitably

HFT

traders

such

fast

enter

an

trading,

liquidity

at the

best

or buy

order,

sell

reality,

as

strategies
orders

place
placed

times

later

in the

when

inventory

markets
has

orders

to be

as

Chapter

relying

solely

on

variables

in

performance
execution
Transparent

and

Market

take

an

then

offer

unable

to

capacity

of

amounts

at a specific

to
time

the
when

is placed.
book

describe,

orders,

orders

and

with

very

in HFT

are

wrong

turn

unexpected

most

HFT

few

market

generally
and

used

the

at

trader's

liquidatedfast.
much
5

limit

HFT

transparent
execution

this

limit

mix.

addition
of

of

complete

dominates

maximum

strategy

or best

to

traders

The

respectively,

execute

generate

orders,

impact

its

assumption

orders

high-frequency

bid

based

gaffean

market

trading

erroneously

example,

market

positions.

chapters

and

use

been

for

a major

solely

their

market-order

available

on

leaving

or liquidate

have

study,

analysis,

a high-frequency

market

strategies
One

for

high-frequency

multiplies

HFT

strategies.

conclusions

trades.

Limit

more

risk.

Capacity

In

be

when

transaction

benchmark

each

it might

particularly

The

their

low,

holds:

Investing

that

is

benchmark,

conditions

terms

and

of

smaller
this

orders

book
do

to

not

market

strategies.

impact

explains.
enjoy

These

also

costs.

than

Still,

infinite

impact

execution
costs,

market

even

capacity

also

variables

do

market

strategies
since

adversely
are

two
affect

probability

of

described

in

Chapter

of

this

book,
as

affect

they

tend

selected

to

to

1.

order

of

process
Be

limit
As

ahead

of

trade.

each

positive

execution

or

is simple
of

certainty

the

also

trades,

be

at

executed,

Become

round-trip

probability

with

to be

of

yet

are

predictable

Execution

negative,

costs

on

on

the

depending

model.

The

known

known
may

probability

not

2.

be

business

close

1.

profitability

venues

venue's
The

the

known

matched

execution

the

it needs

best

with

for

market
of

time

the

orders:

limit

order

it is

orders,

always

however,

is placed.

For

is

limit

to

available

pricebest

as

reaching

the

or

crossed

by

top

bid
of the

a market

or

best

askthe

book.

order

or

an

aggressive

order.
a

and

result,

the

depends

on

The

distance

Market
The

the

of

following

of the

limit

execution

for

limit

orders

is

variable

factors:

order

price

from

the

market

price.

at

same

volatility.
number

closer

probability

of

other

limit

orders

available

the

price

or

to market.

The

size

of the

limit

The

rate

of arrival

The

rate

of

order.
of same-side

arrival

of

opposing

placed

far

away

limit

orders.

market

orders

and

aggressive

limit

orders.
A limit

order

risk

nonexecution:

of

placed

far

crossing
higher

away
the

the

by
as

queue

and

venue;

as

schemes

limit

price

the

market.

The

in turn,
more

likely

limit

order.

The

each

limit

placed

at

the

according
in

may

price
never

on

is the

given

3,

the

(FIFO)

of the

the

market

market

price

congestion
a

faces
reach

probability

to a scheme

Chapter

first-in-first-out

market

depends

the

discussed
are

market

the

executed

the

the

order,

volatility,

specified
matters,

from

from

of
price

specific
most
and

tangible

limit

orders

market

price

volatilitythe
to hit

limit

will

be

orders
stored

to the
common

pro-rata

the

level
also
in

execution
matching
algorithms.

Under

the

risk

FIFO

arrangements,

of nonexecution:

before

the

queue.

market

With

executed
a long
The

on
time

the

a fixed

market

higher

of

close

of other
to market
the

less

the

higher

the

arrival

probability

Algebraically,
by

be

some

the

opposite

their

sizes;

face

only

partially

of the

limit

holds:

larger

executed
order

in the

all

orders

are

orders

may

take

at the

same

price

small

limit

orders

price,

the

likely

steeper

the

rate

arriving

limit

of

the

order

opposing

competition
to

be

market

for

matched.
orders,

the

of execution.

the

deployed

orders

in full.

and

the

limit

may

leaving

proportion

orders,

However,

orders

away,

number

trader's

larger

algorithm,

to execute
the

limit

moves

a pro-rata

higher

as

large

the

maximum

size

a high-frequency

of

strategy

limit

can

order

be

(capacity)

expressed

as

follows:

(21)
where
S

is

the

can

be

size
executed

is the
a buy,

of

direction

the

market

using

trades

market

the

impact

strategy,

and

impact

in

the

strategy,

of equal

order

when

is the
HFT

order

of the

and

the

each

placed

order

the

strategy

S.
t:

when

the

order

is

is a sell.
by

impacting
has

size
at time

generated

that

assuming

the

not

fully

or,

in

the

previous

market

order

conditions

decayed

by

placed

at time

time

by

t ; only

t needs

to

be

considered.
P t is

the

price

obtained

upon
is

executed;
Finally,

of

probability
probability

C t is the

cost

includes

this

book

cost

broker
that

order

case

of

a market

order,

the

price

execution.

the

this

the

affect

that
can

the

be

assumed

of execution
fees
the

and

order

to be

of the

other

costs

bottom-line

placed

order

at
1 for

placed

discussed

profitability

time

t will

market
at time

orders.
t;

in Chaper
of

high-frequency

be

probabilistically

strategies.
As

shown

in

Chapter

5,

market

impact

(MI)

can

be

this
5

of

estimated.
order,
the

In

the

addition

to

of execution

of

when

a market
from

various

points

away

signals

of

HFT

To

the

impact

further

orders
tree

an

of

market

Figure

system
and

be

then

the

MI

following

estimated

crosses

market.

well

by

noting

placed

that

positive
by

average

orders

Assuming
credibly

an

as

limit

determined

the

at
trade

results,

trade-off

the

between

of execution.

capacity

potentially

be

deliver

probability

evaluating

can

the

is

the

of

price

system

increase

may
for

magnitude

probability

statistics

capacity

the

of an

sequenced.
optimal

HFT

strategy,

Figure

capacity

6.12

Basic

Framework

for

al.

(2008)

conjecture

that

market

6.12

of an

HF

shows

trading

Evaluating

and

limit

decision

system.

Capacity

of

HFT

Strategies

Ding

et

deployed

is

performance
once

lower
may

capitalization
related

discusses

the

potentially

increasing

Most

of the

positively

to

trading

related

amount

research

managers
strategies

to

the

its

of

becomes

involved.
order

strategy
However,

performance

capital

capital

the

capitalization.

optimizing

capacity

amount

capacity,

capacity,
of

on

strategy

Evaluation

strategy

strategy

the

latest

portfolio

candidate

be

the

exceeds

negatively

Length

than

when

Chapter
execution

15
and

further.

Period
face
for

the

following

inclusion

question

in their

portfolios:

in

evaluating
how

long

does

one

the

need

strategy

produces

Some

portfolio

period:

six

record

at

Sharpe

If

ratio,

the

the

and

estimation

of the

Korkie
is

others

normally

long

that

is

evaluation

require

content

out

strategy

that,

with

track

just

one

statistically,

correct

if

it

to verify.

evaluation

any

is

The

properly

higher

period

of

the

needed

to

ratio.

strategy

(1981)

are

It turns

Sharpe

trading

arbitrarily

it is intended

the

confidence

investors

frames

ratio

shorter

the

Yet

time

an

Some

data.

Sharpe

validity

of

Jobson

years.

mentioned
the

returns

six

to gain

advertised?

adopted

years.

performance

with

ascertain

two

in order
ratio

have

to

least

previously

matched

Sharpe

managers

of daily

the

a strategy

the

months

of

month

to monitor

can

showed

be

that

distributed

assumed
the

with

to

error

mean

in

zero

be

normal,

Sharpe
and

ratio
standard

deviation

For

a 90

be

at

percent

least

Sharpe

1.645

ratio

evaluation

The

confidence

errors,

periods

Sharpe

correspond

annual

Sharpe

computed

data,

the

then

the

minimum

months

of

less

in

for

Sharpe

with

90

months

for

is
ratio

SR

daily

statistical
performance

data.

is

For

performance

the
of

however,

2/(12)

If

2,

data,

computed

observations
percent

T min ,
periods.

monthly

ratio

of

is

of

is

should

number

strategy

SR

monthly

of

minimum

estimation

ratio

ratio

deviation

verification

of

performance
month

the

trading

Sharpe

ratio

standard

of

Sharpe

Sharpe

calculation

basis

monthly

one

the

claimed

of

daily

the

ratio

frequency

the

claimed

result,

Sharpe

corresponding

for
than

the

on

number

nine

eight
6,

to

claimed

Sharpe

over

used

monthly
if

just

SR

the

than

As

for

based

However,

claimed

s.

ratio

corresponding

greater

used

ratio

should

The

times

level,

and
then

0.5

based
0.5

required

to

and

a claimed
data

daily

= 0.1054.
verify

the

is

then

just

Sharpe
is

is

0.5774.

confidence
data

it

the

on

is 2/(250)

the

required

over
ratio
to

verify

the

evaluation
values

claim.

Table

times

required

of Sharpe

Table

6.2

Required
Claimed

for

the

verification

Trading

Verification

minimum

performance

of performance

Strategy

Sharpe

Data)

Performance

of Reported

Annualized

Performance

summarizes

data

for

key

ratios.

Minimum
for

6.2

No.

Sharpe

Ratio

No.

of Months

Evaluation

Times

Ratios

of Months

Required

Required

(Daily

(Monthly

Performance

Data)
0.5

130.95

1.0

33.75

32.45

1.5

15.75

14.45

2.0

9.45

8.15

2.5

6.53

5.23

3.0

4.95

3.65

4.0

3.38

2.07

Alpha
In

Decay
addition

to

strategies
of

129.65

can

alpha

with

execution

in

latency

performance

be

evaluated

time.

The

case,
alpha

on
erosion

strategies

relying

infrastructure

either
The

the

used

alpha

decay

decay

can

observed

distribution

of

market

instrument

finite

basis

of

alpha

may

limit

orders,

on

of alpha

for

transmission

be

measured
to

costs
of

outlined

the

due

impact
period

metrics

time

due
or

of

latency

to

due

HFT

the

erosion

the

lack

to

market

the

of
poor

orders.

In

forecasted.

can

observed
after

decay,

be

and

earlier,

be

estimated

in a particular

each

trading

as

financial

decision

was

made.
Alpha
opportunity
of

the

assessed

decay

of
cost

associated

strategy.
based

strategy
with

Alpha
on

the

decay
signals

using

limit

failure

to execute

is

strategy

generated

orders

measures
on

specific
by

the

given

trading
and

the
signals

should

be

strategy.

Summary
Statistical

tools

for

strategy

evaluation

allow

managers

to assess

the

feasibility

and

portfolios.

As

HFT

appropriateness
traders'

deepens,

among

HFT

Sharpe

ratio

new

remains

As
the

high-frequency

investment

metrics

strategies.

End-of-Chapter
1. What

and

of

are
a

strategies

managers'

deployed

quick

test

to
of

to

their

understanding

capture

the

strategy

of

variability

feasibility,

the

favorite.

Questions

is the

Sharpe

ratio?

is the

Sortino

ratio?

What

are

its

drawbacks?

How

can

it

be

given

enhanced?
2. What
3.

Suppose

high-frequency
gains

on

trading
4.

Why

does

5. What
1 Strategy
and

trading

changes

strategy?

Why

asset

performance

does

attribution

strategy

in the

S&P

finds
500.

analysis

strong

What

on

dependence

are

the

of

implications

trading
for

the

Discuss.
market

impact

probability
is alpha
capacity
liquidity

enter

calculation

of

strategy

capacity?

of execution?

decay

and

has
by

why

been

Getmansky,

does

shown

it exist?
to be

Lo,

and

a function
Makarov

of trading
(2004).

costs

CHAPTER
The

Business

Contrary
(HFT)
pure

of High-Frequency

to

popular

is not

entirely

research.

As

the

the

considerably

the

most

of

economics

from
the

business

business

Instead,

such,

explains

belief,

Trading

of

those

of

economic

of

of

trading,

HFT

is the

building

nor

of

HFT

bolts

HFT

is enabled

floor.

of

of

technology.

operation

trading

and

trading

is it a business

business

an

traditional

nuts

high-frequency

differ

This

chapter

successful

HFT

organization.
Key

Processes

The

central

of HFT

processing
identifying

value

proposition

and

high

small

this

business

and

close

only

however,
7.1

fast,

illustrates

the

process

may

be

as

a 20-line

selection
Figure

systems.

becomes

MatLab,

data

The

to send

tick-by-tick

data

technology

stream

on

is

what

rapid-fire

reams

behind

differentiates
signals

of

to

open

and

like

with

shown

to

Algorithm

Design

and

fully

automated

HFT

and

any

model.
written

produce

and

by

is too

emotionless
simple

software

The

process

final

product:

concept,

hypothetical

systems,

used

to

activity,
a trading

or a prototype,

in a modeling

Reevaluation

complex

development

of the

making

decisions.

language,
profitability

of data.
7.1

separated

manner

a concept

of code

is

information,

deceivingly

a quantitative
piece

market

data

instruments

continuous

efficient,
sample

financial

Evaluating

Affordable

Just

begins

multiple

trading.

brain.

make

HFT

that

quote

in a consistent

develop
HFT

of

automated

human

can

idea

the

a trader

interpreting

decisions
for

tick

via

microseconds,

Figure

enables

every

possible

task

in

turnover.

by

positions.

Processing

trading

capital

changes

and

of

Process

such
on

The

next

tested

step

on

considered

The
a

concept

model

series

the

model

are

generally

tested

the
into

data.

of tick
are

on

Two

data

and

trades

unused

paper

trading.

is

generally

validity

of the

in Chapter

16

or

swath

of

proof

of

in

clean
the

initial

satisfactorily
Models

concept
is

to ascertain

discussed

performs

the
years

out-of-sample

model

successful
or

trading
the

programmed
sandbox

back-test,

the

paper-trading
activity

orders

order-placement

perfect

tick

whereby

of this

out-of-sample,

that

fail

the

back-test

discarded.
a

of

process

practices

quotes
If

preproduction

fully

is
of

of

amount

is moved

Following

volume

best

activity.

real-time

back-test:

large

Back-test

book.

track

a sufficient

concept.

data,

is

without

in

for

the

system.
testing

Paper

placing

actual

all

such,

other

moved
trading

the
aspects

but

Except

phase

is

paper-trading
of

the

into

the

emulates

orders,
log.

paper-trading

As

are

phase.

program-generated

functionality,
HFT

models

keeps
for

typically
stage

HFT

the
a
is a

system:

data

receipt

position

and

processing,

accounting,
the

correctness

algorithm

is

typically

collected

in

paper

back-test

and
paper

both

should

fully

reconciled

run

of

data,

and

risk

HFT

system,

of

the

paper

at

the

end

tradingany

paper

trading

trading

and

deliver

sufficient

generation

archiving

ascertain

When

run-time

with

to

move

transition

to

the

noted

and

results.

system

is
a

the

To
back-test

day

on

data

between

the

investigated.
properly

A month

into

signals,

management.

trading

are

results

trading

observed

back-test

back-test
the

each

deviations

are

identical

of

of

of

programmed,

clean

paper

usually

considered

low-amount

live

trading
to

be

trading

or

production.
The

challenges,

described

real-time
have

portfolio

to be

losses.
out

production
in

detail

Extended

can

perfectly

issues

loaded

Chapter
be

runs
and

with

16.

complex

to avoid

production

code

often
in

accounting

executed

various

is

little

own

set

of

Trade

execution

and

coding

exercises,

and

unexpected

with

ensure

its

malfunctions

capital

a smooth

at

and

stake

and
help

effective

iron

trading

functionality.
This

section

demonstrates

HFT

system,

human

Quant

HFT

back-tests:
Quant

The

HFT

identify

conditions,
causing

11

risk

as

successful

follows:

of

concept

of

this

examples

of

which

are

book,

the

road

maps

are

is a set

principles,

including

of persistent

described
for

indicators

in

trading.
built

to

model

validation

and

policy

percent
management,

harmless

large

split

of

opportunities.

10

or

core

models

management

seemingly

development

is functionally

the

of quant

trading

Competent

the

percent

models,

development:

in

development/proof

8 through

output

Risk

time

model
15

Chapters

that

the

glitches
like

losses.

from
The

discussed
in

in

the

code,

throwing

off

objective

of

Chapter

14,

market

data,

trading
risk

prevents
market

dynamics

management

and
is

to

assess
to

the

extent

mitigate

system

damaging

of

or

processes

the

be

of trading

Coding

trading

focus

of HFT

to

complex

conditions.

and
and

entities

the

HFT

This

Financial

flip

broker

of

be

quite

traders
of

work

however,
traders
Deutsche
prospective

and

some

for

traders
testing:

System

testing

20

react

today

are

to

human

be

a variety
be

tend
market

platform
to

multiple

(ECNs),
the

language,

data.
from

simultaneously.

use

of

special

With
one

and

FIX,

at

executing

Best

coding

16.

have

the

thus

discussed
by

the

effect

given
even

out

Chapter

complexities

it difficult

interface

making
audience.
can

monetary

as

To
FIX,

communication

of

platform

offer

of

(APIs).

the

offer

2.

interfaces

connectivity.

may

in

proprietary

capturing

the

of

through

changed

out

example,

to

trading

primary

systems

networks

induced

to

percent
the

interfaces

(FIX)

as

building

to defray

40

built

flexible

brokers

providers
in

require

comprises

programming

adapt

management

execution

financial

rolled

also

to

others

and

can

speed

platforms,

Brse,

System

for

application

interested

while

risk

infrastructure:

in Chapter

have

structures

Some

is accomplished

routing

in

providers

required

to

signals

cumbersome,

delay

to switch

goal

incorporate

or to several

FIX

proprietary

the

eXchange

the

and

with

communication

a switch

described

protocols

to

optimized

are

several

processes

operation.

systems

of codes

the

and

detect

Information

practices

counteract

code,

independence

to another

can

that

electronic

exchanges.

build

High-frequency

is,

broker-dealers,

to

risk-management

development.

infrastructure
and

risk-management

Most

sequence

the

to create

run-time,

damage.

a failure-safe

independentthat

the

into

and

during

breaks,

potential

built

to ensure

Coding

be

failure

limit

may

damages

conditions

warnings,

eliminate

presence

of potential

The
be

implementation

for
extent

extensive,

incentives

Some
much

The

as

for

divisions

of

$65,000

to

costs.

percent

is another

critical

component,

discussed

in detail

in

Chapter

16.

To

various

scenarios,

typically

deployed

performance

ensure

thorough

or

scripts.

to

with

discrepancies,

its

one

third

as

duties

also

ensures

gaffes

that

can

each

stated
as

coders.

cost

the

Run-time

monitoring:

5 percent

Run-time

monitoring

is

another

well-defined

task

of

monitoring

management

parameters

particularly

involved,

personnel

are

systems

and

acceptable

ensuring

and

Compliance

the

stage
start-up

programming
management,
steady-state
administrative
Figure

7.2

Development

tasked

with

the

run-time

whitewashing
money.

personnel.

The

with

risk

clear

duty

HFT

is

to detail.
of

performance

for

testing

procedures

attention

practices

administrativia:

other

not

Separate

watching

the

remains

monitoring

of

Keeping

and
of

are

within
discussed

HFT

spent

business

phase,

HFT

the

monitoring

production,

take
it

tasks

that

Allocation

of

each

task

take

is

outlined
as

the

Man-Hours

majority
during

varies

be

stage,
of

all

job.

in Figure
to

majority

of

guidelines,

here

shown

tend

monitoring,

function

is a full-time

ramp-up
the

regulatory

activities

businesses
In

are

of

development,

centered.
and

related
on

percent

tasks
track

all other

time

10

administrative

businesses.

proportion

the

other

and

documentation,
The

escalation
close

that

Best

system

requires

typically

and

16.

Compliance

trading

but

limits.

in Chapter

and

HFT

all

compensated

significant

requiring

its

document

coding

task

are

compare

of face-saving

business

HFT

to

often

Separating

through

testers

to

are

free

is run

code,

and

Testers

testing,

ultimately

of

mandate,

as

proper

system
software

block

bugs.

much

the

Professional

examine

known

about

testing,

with
7.2 .

data

and

testing,
effort.

compliance,

In

risk

Finally,
and

in
other

of time.
Different

Phases

of

HFT

In

the

traditional

often

broker-dealer

associated

only

teams

far

removed

policy

development

separating
an

are

that

occurred

testing,

using.

the

best

in Chapter

HFT

the

and

an

not

monitoring

of

of the
of

business.
software

as

the

by

as

have
HFT

and

expense.

By
creates

for

Knight

systems

systems
fiasco

that
are

functions

development,

the

Capital

shown

These

testing

management

the

is

technology

system

accountable

such
2012,

technology

unnecessary

feel

of

word

provided

such

business,

summer

practices

7.2

illustrates,
for

strategies

traditional

most

an

running

apprentices,

trading

the

thorough
key

to

follow

the

directly

discussed

through

the

time

HFT

trading

in

business

desks.

and

close

to

businesses

senior
trading

an

executing

costs

experienced

of

traditional

costly;

products
proprietary

moment

development

is

high-frequency

their

like

examples,

the

unusual

begins

tasks

the

detail

16.

Figure

process

gadgets

need

Real-life

profitability

shiny

seem

traders

in

where

business,

from

compliance,

long-term

new

may

that

they

As

from

technology

impression

from

with

world,

Designing

monitoring

nothing.
incurs

trader

with

desk

and

when

the

trained

curves

for

rolling

follows

proven

training

new
finished

By

fixed

contrast,

costs

from

track

promising
apprentices

the

record
young
replace

masters.

Figure
traditional

7.3

illustrates
trading

the
systems.

cost

The

cost

of

out

traditional

computerized
trading

and
remains

fairly

constant

through

necessitating

hiring

traditional

and

trading

trading

in

takes

on

trading

do

terms

of

labor

average

36
as

small

systems

engineer

and

systems

engineer

and

time.

One

staff,

trader

costs

up-front

investment

successful

moves

into

production,

of

simultaneously,

can

driving

be

system

ultimately
a

agent.

agent

is

computerized

includes

performance-monitoring
monitoring

that

trading

costs

typically

the

computerized

The

that

burnouts

of staffing

develop.

staff

systems

of

Developing

an

to

support

trader

change.

system

exception

requires

months

requiring

trading

new

and

the

the

not

however,

decline

several

With

training

desk

systems,

costly

time.

dedicated
Both

the

responsible

the

costs

for

closer

to

zero.
Figure

7.3

Trading

Markets

wide

range

trading
more
To

Economics

at

Suitable

of

high

met:
market

High-Frequency

versus

Traditional

than

ability
volatility

costs.

The

highly

interrelated

for

this

ensure

volatilities

of
and

market

Some

conditions

securities

fit

the

markets,

profile

for

however,

are

others.

to quickly
to

HFT

and

frequencies.

appropriate
the

for

securities

appropriate
be

of

Businesses

Financial
A

The

type

of

trading,

move

in and

that

changes

different

dependent

two
out

of positions

in prices

markets
on

requirements

the

have
volume

and

exceed
been
of

must

be

sufficient
transaction

shown

to

macroeconomic

be

news
as

reaching

well

as

liquidity

demand.
24

penny

trades,
less

a day,

5 days

the

prices

of illiquid

securities

High-frequency
security
find

a timely

Amihud

investors

optimally
the

investors
will

liquid

be

2006,
on

every

period

of

10

work

less

liquid

the

risk/return

is

higher

to

more
liquid

minutes

may

liquid

be

able

markets.

illiquid

that

longer-horizon

According

to

consideration;

to

While

or

these

longer-term

adverse

short-term

gains

by

on

where

the

average

assets.

securities;

liquid

show

assets.

the

Between
making

not

in illiquid
either

are
such

days.

most

(1986)

pairs

substantially,
with

with

and

securities,

few

change

the

supply

taking

market

more

risk

in

investments.
liquid
achieved

for

the

on

impervious

obtain

A perfectly
can

issue

liquid

compared

Mendelson

hold

(already

moves)
less

key

and

market

exchange

Less

may

focus

can

positions

factors:

available

counterparty

investors

into

two

foreign

once

as

reasonably-priced

securities,

major

only

risky

a holding

longer-horizon

authors,

more

by

readily

a week.

assets

strategies

requiring

as

trade

enter

execution.
by

such

may

to quickly

determined

characterized

stocks,

liquid

is in turn

securities

hours

ability

of electronic

are

Liquid

The

them

availability

assets

traded

markets.

to close

and

Liquid

as

the

market

of

treatment

of the

of

counterparties'

limit-order

to

expectations

trade
of

in

to
turn

impending

the

subject).
in

traded

the

on

movements,

as

risk

along

price

Bervas,

liquidity

market

their

or ask

(see

market,

The

depends

bid

Market

trade.

price

quoted

quantities

traders

willingness

willingness

one

irrespective

detailed

presence

is the

depends
well

as

the

participants'
aversions

with

and

other

market

information.
One

way

average

to
daily

compare
volume

terms

of daily

average

liquid

market,

followed

then

equities,

options,

the

liquidity

of each
trading
by

of

security

different

securities

as

measure

volume,

recently

commodities,

the

foreign

issued
and

exchange

is to

use

the

of liquidity.

In

is the

most

U.S.

Treasury

securities,

futures.

Swaps,

traditionally

traded
the

over

the

counter

Dodd-Frank

optimal

Act,

market

Economics
Costs

for

are

cost

HFT

way

the

electronic

to become

the

era

most

under

liquid

and

drivers

Business

in an

HFT

trading

costs,

important,

trading

losses.

compliance

business

to

business

limit

that

the

are

Costs

of Data

Data

tend

to

Effective

companies,

like

significant
tick
will

This

either

in

of

charge

of choice

at

and

and,

most

monitoring,
14

describes

or

Bloomberg,

data

obtain

costs,

Chapters

expensive

Broker-dealers

free

legal

equipment

and

the

and
16

costs

of

are
doing

to risk.

very
and

labor,

management,

section

premium.

to

risk

attributable

Reuters

need

instrument

latter.

data,
and

discussed

not

be

are

administrative

frameworks,

required

firms

their

entering

HFT.

software,

quality

on

but

of HFT

of Doing

The

(OTC),

offer
and

to

tick

two

trading

sale

venues

of

Multiple

for

traders.

years

initial

free.

data

trading

prospective

least

to generate

entirely

for

may

offer

Start-up

historical

HFT

data

in

the

models.

Hardware
As

discussed

in

component

of

effective,

Linux,

setup

is necessary

production,

to

computers

incidents

of

the

The

like

least

for

prominent

most

basic,

available

Windows

configuration
machines

of processing

are

commonly

system,

thorough

separate

costs

expenses.

operating

minimize

overloads

hardware

involves

for

to purchase

2,

operating

A professional

is wise

and

HFT

HFT

stores.

Chapter

and

retail

Red

testing,

code

Hat

access.

development,

unintentional

at

7 or

remote

yet

It
and

manipulation

power.

Connectivity
Connectivity
connection
operation
in

Chapter

is
with
receives
2,

important
a

for

sufficient
the

quotes

fullest
in

any

successful

bandwidth
set

of

particular

HFT
ensures

quotes
can

and
be

operation:

that

trading

trades:
lost

the
as

in

fast

described
cyberspace

congestion.

Connectivity

options

involve

Co-location
Proximity
Run-of-the-mill

connections

Co-location

or

venues

in

cage

their

for

wire

HFT

connecting

server

the

anyway

fast

like,
are

not

co-location,
wire

the

secure

Finally,

common

to

well

are

include

remote

co-location,
to facilities

the

same

facility

for

the

to

via

configure

they

the

are

close

trading

The

Internet,

run

by

to trading

venues.

a power

hookup).
general

often

accessibility.

servers,

fiber-optic
via

free

secure
network

most

located

as

actual

except

access

a cage

an

servers,

traders

trading

includes

and

exchange

other

Like

source,

and

wire,

however,

making

proximity

co-location.
HFT

until

network,

resulting

similar

and

typically

source,

the

may

exchange

than

as

and

offers

a shoestring

network

and

a power

and

(typically,

to

exchanges

with
in

proximity

connects
less

they

but

facilities,

HFT

companies

venues,

by

High-frequency

services

third-party

offered

server,

line.

Proximity

is

dedicated

the

dedicated

colo

its

start-up

systems

however,

in serial

can

quote

become
is

losses

survive

on

a premium

consistently

subject

to

and

information

following

software

profitable.

extreme
and

cable
The

congestion,

trading

delays.

Software
HFT

operations

built

in-house:

deploy

Computerized
of an

HFT

generates
and
in

loss

bound

The

portfolio

allocations

capacity,

The

significantly

and

trading

generator

Generators

business

to

of

system.

secrecy.

competitive
a finite

generation

(P&L).

utmost

the

and
are

a competitor

or may

core

accepts

and

processes

trade

signals,

and

often

every
armed

or

may

is the

built

requirement

considerations:

diminish

signals

most

secrecy

that

outright

investment
with
destroy

be

functionality
tick
records

in-house
stems

not

data,
profit

and
from

kept
purely

strategy

a generator
profitability

has

code

is

of

an

HFT

strategy.

Computer-aided
deployed
and

analysis

by

HFT

have

modeling

emerged

industry.

R,

efficient,

and,

and

news

moves.

that

Trading

software

deliver

the

best

software

enables

per-trade

licensing
of

sprung

of

up

to

a suite

extra

edge

increasing
pension

in

For

the

and

execution

altogether.

time

orders

into

foreign

traditional

hedge

funds,

algorithms

Chapter

15

New

lots.

discusses

on

is

now

the

bread

in the

most

MarketFactory
traders

investors,
their

bypass

latest

and

get

an

Furthermore,

developed

the

market

companies

orders

automated

to

This

Development

market.

bid

for

interval.

Yorkbased

buy-side

of

format.

decide

their

like

range

independent

to help

have

in

time

algorithms

exchange

of

newcomers

algorithms

a given

distribute

tools

capturing

of short-term

offer

Yet

the

facilitates

and

optimal

is

in R.

Promptly

trades,

execution

of software

number
funds

to

example,

that

modeling

execution

within

the

libraries.

forecasts

optimal

investors

in

proprietary

Jones,

broker-dealers.
help

known

in a machine-readable

price

best

many

manner.

provides

size

well

all of its

Dow

news

execution

and

butter

best

real-time

quantitative

software

AbleMarkets.com

traders

aggressiveness,

is

with

enhances

incorporates

achieving

efficient

but

of securities.

Reuters,
and

popular

MatLab

software

announcements

SemLab

software

models.

open-source

almost

pricing

Thomson

products

have

pricey,

extended

runs

fundamental

RavenPack,

and

be

trading

information-gathering

high-frequency

price

be

modeling

most

it is an

can

now

new

industry's

is free:

example,

financial

build

can

of all,

Internet-wide

rumors

the

MatLab

best

with

to

as

however,

for

done

operations

choices.

Citigroup,

is

an

like
own

large

suites

of

broker-dealers

developments

in

best

execution.
Run-time
stays
in

risk
within

detail

in

management

pre-specified
Chapter

14.

applications
behavioral
Such

and

applications

ensure
P&L

that

bounds,

may

also

the
as
be

system
discussed

known

as

system-monitoring

and

in-house,

but

advantage

of

modules.

are

customers

anything

built

of

other

an

The
of

the

review

more

by

sound

for

than

must

for

all

risk

limits,

monitoring
modern

power

related

generic

HFT
outages,

to

responsible

performance

be

monitoring

enough

to

warrant

software.
stream

advanced

forecasts

manner.

crash

can

can

For
be

be

information

incorporated

into

a forecast

that

example,

used

to

AbleMarkets.com

enhance
and

liquidity

HFTindex.com

of information.

Execution
may

rely

quickly

route

and

execute

Suisse

are

often

cited

execution.

UBS,

venues

for

Execution
advance.

The

in detail.
important

foreign

total
that

transactions
and

their
as

vary

eliminate

Clearing

and

income.

fixed
a

cost

to broker,
structure

settings
gains.

include

where

as
of

Chapter

execution
the

sheer

Credit

been

fee,
other

to

electronic

have

per-trade

ECNs
and

dominating
Brokers

however,

and

Sachs

Quantitative

broker

the

Goldman

charge

may,

from

brokers

broker-dealers

and

costs

executing

trades.

typically

high-frequency
can

their

exchange

Understanding
in

on

Barclays

providers

components

Custody

to

these

system.

practitioners

go-to

of

can

in a useful

kind

immediately

third-party

imminent

this

Electronic

tend

Sometimes,

HFT

provide

be

research

system

provision

shelf.

diligent
be

built

reliability

designed

systems,

systems

forecasts.

of

often

the

is the

may

breaches

should

third-party

warns

HFT

issues,

well-tested

a trading

result,

are

off

undergoes

are

risk-management

compliance

and

systems

problems

Like

Real-time

as

buy

software

applications

HFT

Any

purchasing

risk

software

mobile

organizations.

and

to

and

in-house.

performance

parties.

third-party

and,

and

or any

software

enough

third-party

multiple

Desktop

generic

buying

The

fault-tolerance

known

the

in

unobservable
5

describes

is

especially
number

of

In

addition

to

typically

providing

connectivity

offer

special

prime

trading

capital

(known

as

clearing

). Both

In

custody

a custody

for

the

insurance

Staffing

trading

development

pricey,

and

of

the

understand

include

trade

a certain

broker-dealer

clearing,

the

default

of

safekeeping

reconciliation

involve

venues

of

(known
degree

takes

the

as

of

risk.

responsibility

broker-dealer

may

trading

charge

act

as

counterparties.

custody

staff

high-frequency
required

PhD-level

econometrics.

In

enough

to

computer

and

addition,

any
make

and

and

clearing

fees.

issues

sure

all

in
should

and

needs

to

finance
be

system

and

experienced

interoperability,

efficiency.

financial

i's

are

sector,

dotted

departments.

is

of

risky

models

research
staff

both

Costs

in the

accounting

is

trading

programming

Legal

that

design

algorithmic

business

assistance

to

complex

security,

systems

quantitative

handle

Administrative

and

in

broker-dealers

Costs

Initial

to

clearing

the

and

that

) and

the

against

Broker-dealers

Like

and

whereas

exchanges,

services

custody

arrangement,

assets,

to

therefore

high-frequency

and

all

t's

Qualified

are

crossed

legal

indispensable

for

traders

and

need
in

legal

accounting

building

capable

operation.
Capitalization
As

with

with

of HFT

any

other

equity

and

contributions
investor

amount

of

be

the

Cost

the

initial

of the

structure

Sharpe
Maximum

ratio

trading,

founders

using

of the
of the

drawdown

or

equity

HFT

an

The

capital,

approximated

Profitability

of

leverage.

from

private

can

form

initial
of

the

capital
required

of

following

strategy

to be
operation.

strategy.
of the

strategy.

firm

equity
firm,
the

can

be

capitalized

normally
private
parent

depends

the

trading

HFT

comprises
equity

company.

on

many

with

live

factors,

variables:
traded

capital.

capital,
The
but

The

profitability

every

dollar

costs

and

During
be

of

be

market

impact

used:

or

Strategies

that

significant

costs.
small

annihilate
the
small.

traded

retail

In

brokerage

$40,000

every

trade

a retail

the

same

time,

for

every

$1

100

When

the

Leverage
traded
investment
lower
leverage
$100,000
a 1:9

helps

leverage,

until

the

subject

to

to

0.1

percent

than

refers

$1
costs

multiply
without
$0.1

of

faces

costs

one

is
to

small,

million

funds

with

trading

gains

of the

strategy.

percent

When
return

notional

one

little
of

as

the

retail
can

be

trade.

At

$3

$5

becomes

a $1

or

levered.
by
a

the

$100,000
obtaining

In

percent

strategy

close,

investors.

profitability.

the

to

possibly

A 0.1

And

notional,

leverage,

notional,
better

with

collateralized

a 1:9

retaining

common

notional.

and

as

when

round-trip

of

often

and

leverage.

the

to the

it

borrowed

Thus,

of

in

often

larger

trade

percent

offered

is

example,

every

0.0005

the

often

a position,
per

easily
true

notional

the

are

can

particularly

are

percent

open

gains

which

is

For

leg

0.05

are

per-trade

costs

every

than

capital

the

to unrealistic

exposed

traded

versa.

one

instruments.

is $100

per

institution

smaller

transaction

vice

tradedjust

becomes

and

profitability.

terms

the

5, trading

more

of

a large

financial

be

This

when

and

legs:

often

costs

due

costs,

profitability.

$20

or

available

not

transaction

operation,

trader

most

may

can
however,

basic

either

settings,

the

of

is small

two

caution,
for

performance

in Chapter

million

times

that

trading

profitability.

production

wrong,

on

performance

account

reducing

for

potential

Great

to

simply

gross

to

of

less,

has

emptying

cost

in

charges

and

fail

itself,
be

semblance

amount

of

return

account

back-test

high-frequency

described

not

gross

production.

well

phases

As

often

compared

any

start-up

the

do

does

picture

mistakes

into

the

performance.

can

coding

is transferred

return

investment,

strategy

back-tests

assumptions

typically

of

future

the

describes

high-level

back-tests
of

of all,

gross

stages

potential

should

of

earliest
as

strategy

The

presents

used

code

HFT

invested.

the

Worst

the

addition,
return

is traded
million

on
with

0.1

percent

= $1,000

or

1 percent

significant

improvement.

its

are

losses

unlevered

also

return

Of

course,

multiplied

$100,000

on

the

when

the

accordingly:

notional

$100,000
strategy

what

extends

to

investment,

is

loses

a $100

$1,000

loss

money,

loss

in

with

an

1:9

leverage.
Broker-dealers
observe

the

capital

by

when

the

known
but

are

in the

written
even
To

liquidating

said

account

in stone

and

be

each

help

traders

given

the

recorded

strategy.

capital,

and

bank

asset

large

hedge

his

an

optimize

much

traded

to

broker,

notional

are

not,

to

however,

account-by-account

the

strategy

leverage

and

leverage

be

later-stage
obtained

management
funds

the

value

and

and

that

Sharpe

the
the

works

ratio

trader

metrics

can

maximum

afford

downside

data.

leverage,

it can

their

collateral
broker

of

drawdown

how
of

in historical

can

Maximum

determine

to

of

happens

from

requirements
on

can

return

a safety

vary

negotiated

traders

volatility

addition

ensure

they

Liquidation

below

Margin

leverage:

basis.

liquidation,

for

and

percentage

times.

can

of

positions.

requirements

situation-by-situation
avoid

time

falls

a certain

at all

providers

in real

value

Margin

designate

natural

trader's

account

margin.

best

In

positions

typically

be

most

trader's

trader's

as

the

equity
from

be

established

divisions,

interested

can

a source
institutions,

fund-of-funds,

in boosting

of

their

trading
such

pension

as

funds,

performance,

and

the

like.
From

the

to invest
the

perspective
into

of

investments
the

term

inclusion

such
of
in

its

is used

alternative

larger

strategies
to

distinguish

long-only
status,

institutional

most

mutual
HFT
portfolios.

possibly

HFT

investments.

most

as

investor,

operation,

alternative

describes

investments,
a result

institutional

a high-frequency

category

umbrella;

of a large

The

under

funds

and

strategies
To

tends

to fall

term

alternative

a broader

strategies

hedge

from
bond
are

qualify

looking
into

fund

traditional
portfolios.

candidates
for

institutional

As
for

investment,

HFT

auditable

performance

with

as

The

compensation

little

paid

are

typically

The

performance

previous
Just
is

paid

increases
an

to
(or

incentive

Due

to

its

relative

institutional

other

as

far

percent

fund

hedge

the

three-year
per

annum,

fund

the

the

above

as

water

performance

when
fund

the

high

the

only
of

invested.

gain

known

HFT

Management

amount

manager,

value

model:

fees.

of

manager
the

fund

notional

shares,

hedge

point
but

may

high-frequency
assets

administrative

lower

gains

paid

Leverage

as

label

typically

the

manager

shares,

creating

(the

concerned,

in

turn,

classic

percent

a performance

emerging

results

in

higher

from

Where

into

an

arbitrary

fee

and

2-and-20
model,
as

of

the

translates

management

serving

30

an

is

a 1-and-30

management

as

fees.

percent

offer

and

This,

management
2

often

activity,

treated

view.

fee

under

often

longevity

of

command

traders

is

strategy

performance

20

model),
with

1 percent

compensation

for

above-the-highest-water

incentive.

of HFT

How

much

leverage

can

The

quick

answer

to

variables

characteristic

maximum

drawdown.

invariance

with

strategy

to

percent

performance

emerging

as

investor's

fund

HFT

12

the

of

HFT

The

performance,

mark

and

the

novelty,

risk

all

one-

to outperform.

perceived

of

to

percentages

increased)

business.

hedge

of

any

has

follows

high-frequency

manager

higher

then

are

value
with

of

percentages

fees

as

deliver

possible.

management
fixed

highest

mark.

as

to

record

of HFTs

are

fees

need

track

volatility

managers

fee

managers

strategy
levered

high-frequency

respect
is

exactly
100
Sharpe

an

HFT

this

business

question

to

each

The

beauty

to

leverage:
the

times.
ratio:

same
This
no

sustain

depends
HFT
of

and
on

strategy:
the

Sharpe

a Sharpe
as

the

is

due

levered

remain

two

performance

Sharpe

ratio

ratio

ratio

of

an

to

structure
positions

in

its

unlevered

ratio

HFT

and

lies

Sharpe
the

viable?

of

an
of
are

HFT
the
held

overnight,
in the

so

HFT

numerator,

increases,
the

the

as

the

ratio

pointed

expected

denominator

leverage,

Sharpe

out

ratio

not

include

in Chapter

return

of the

does

in

the

6. When

numerator

increase

the

risk-free

rate

Sharpe

ratio

the

and

proportionally

the

volatility

in

the

amount

of

must

take

into

an

HFT

by

L:

(1)
The

expected

account

return

the

business

costs

of
of

the

high-frequency

doing

is adjusted

business,

for

expenses:

and

assuming

operation

so

the

Sharpe

ratio

of

(2)
or,

using

daily

data,

250

trading

days

per

year:

(3)

Expressions

in

equations

profitable

business,

infinitely.

Annualized

variability.

The

severe.

To

required

for

actual

(3)

leverage

volatility

have

the

realizations

of

be

be

can

Sharpe

be

for

of

return

much

ratio

more

SR HFT

drawdown

increased

measure

losses

a maximum

positive

cannot

average

minimum

operation,

to

value

is only

the

a stable

that
of

is

the

and

comes

Ops
in as

metric.

Assuming

Ops

yet

determine

convenient

percent

(2)

the

worst-case

related

maximum

historical

to

maximum

scenarios,
the

drawdown

the

number
lies

away

drawdown
required

of

standard

from

the

can

occur

Sharpe

ratio

deviations

mean

in
SR

the

10
HFT

historical

of returns:

(4)
from

where

the

minimum

SR HFT

Ops

can

be

obtained

as

(5)
For

a levered

account
HFTOps
cumulative

strategy

operating
min

to remain
costs

solvent,

SR HFT

determined

by

equation

performance

of

sample

the

Ops

Sharpe

that

(5).

have

Figure

high-frequency

7.4

ratio
to

takes

into

exceed

SR

presents
strategy.

gross
The

strategy

delivers

drawdown
0.15

of 0.93

percent

minimum
when

be

in capital

maximum

daily

$3,432,729

for

return:

computed

Figure

(6)

if the

completely,
as

the

ratio

If

this

allowed

smaller

ignored

and

1.41.

250

sample

exceed

spending

considerably
average

to

in July

the

Sharpe

and

than
risk

the

then

annualized

HFT

Industry

history

of

Ops

this

the
average

Participants

min

cannot
per

number

of going

and

leverage,

operation
days

with

strategy

a nine-to-one

trading

2.78

2012,

SR HFT

particular

of the

of

SR HFTOps

leverage:
7.4

ratio

incurred
over

has

to

Sharpe

percent

operational
levered

million

gross

computed

computed

is

bankrupt
expected
daily

daily

the

volatility

strategy.

delivers

of

equation

(5),

is

allocated

$10

This
from

maximum

levered

of
The

strategy

exceed

computed

return

maximum

L = 10,

year.

via

then

the

$13,730,

or

number

is

the

daily

drawdown

performance
multiplied

by

is
capital

If computed

using

$6,959,034

drawdown

equation
misleading

is taken

If a stricter

into

can

nave

number

to

measure

drawdown
(5)

the

occurs

be

gain

works

spend

out

to be

when

the

nearly

risk

of

account.

performance

maximum
equation

(6),

is required,

in

rewritten

less

than

whereby

the

percent

of

observed
all

cases,

as

(7)
where

number

located

of

has

deployed

changed

in

2.41,

for

standard

deviations

from

strategy

example,

where

to

with

the

the

maximum

3.

maximum

For

minimum

the

drawdown

$10

operational
allowable

million

capital

Sharpe
annual

is

ratio

of

expense

is

$931,801.
An

HFT

operation

leverage

and

likelihood

of

risk

is

more

high

Sharpe

covering

costs,

of a catastrophic

Market
Like

likely

to

survive

ratios.
and

High

the

and

prosper

leverage

high

if

it

has

increases

Sharpe

ratio

the

reduces

the

forces

and

loss.

Participants
any

other

influences.

industry,

Figure

remainder

of the

7.4

HFT

is

subject

summarizes

section

the

discusses

to

outside

playing

market

field

of

participants

HFT.

The

in detail.

Competitors
HFT

firms

compete

management

firms,

competition
attracting
other
talented
direct
traditional
access

with

with
as

The

high-frequency

contest

as

for

market

traditional,

market-making

rivalry

and
with

firms

and

inefficiencies.

in the

involves
traditional

investment

broker-dealers.
hedge

funds

quantitative
also

strategists

broker-dealers

opportunities

more

mutual

trading

experienced

non-HFT
to profit

well

traditional

investment.

and

other,

hedge

includes

centers

on

funds

and

recruitment

technologists,
Likewise,
turf

The

as
the

wars

over

of
well

as

battle

with

first

dibs

market-making

arena.

Investors
Investors

in

HFT

include

funds-of-funds

aiming

to

diversify

their

portfolios,

hedge

mix,

private

and

create

eager

equity

wealth.

prime

Most

to

firms

add

new

seeing

investment

strategies

banks

to

sustainable
offer

their

existing

opportunity

leverage

to

through

their

services.

Services
Like

funds

and
any

specific

Technology

business,
support

to

the

providers

of

data,

Most

trading

common

high-frequency

staffing,

in more

detail

and,

operation
in many

business

hardware,

clearing,

described

high-frequency

services.

providers

custody,

Providers

earlier

critical
include

software,

administrative
in this

cases,

community

connectivity,
and

requires

and

execution,
legal

services,

chapter.

Government
Several
time

regulatory
this

initiatives

book

regulatory

was

were

written.

under

way

around

13

summarizes

Chapter

the

world

at

the

the

latest

thought.

Summary
Developing
include

high-frequency

issues

many

surrounding

systems.

The

decisions

requires

processes,

and

execution

costs

End-of-Chapter
1. What

are

2. How

much

3.

What

capital

fast

operating

box

The
with

on

generating

monitoring

deployment

time,

no

the

emotion,

and

-executed

short-term
solid

and

leaving

with

multiple

of

algorithm

and

Well-designed

capitalizing

nature

complex

consistently,

factors.

of

and

supervision.

considerably

capable

black

that

moves

profitability

in

of
all

markets.
Questions

the

key

time

kind
and

or

challenges

management

human

systems,

of electronic

of

risk

human

are

box

transparency

decrease

prices,

involves

gray

diligent

other

high-frequency

types

low

engines
or

security

the

constant

profit-generating
sickness,

business

of

a net

steps
is spent

operational
(after

in algorithm
on

monitoring

costs

transaction

development?

can
costs)

in a stable
an

HFT
Sharpe

with
ratio

HFT

operation?

$100
of 1.5

million
carry?

in

4.
with

What

is

the

a. Net

following
(after

b. Three
c.

the

costs)

officers

of $72,000

d. Co-location
are

HFT

Sharpe

earning
(office

per

of $36,000
the

needed

for

a breakeven

of

an

HFT

characteristics:

overhead

expenses)

5. Who

capital

transaction

full-time

Office

minimum

$150,000
space,

year
per

industry

ratio

year
participants?

of 2.8
per

year

networking

each
and

computer

CHAPTER

Statistical
Like

Arbitrage

human

broken
1.

trading,

down

high-frequency

into

Statistical

three

Stat-arb

relative

major

arbitrage

strategies.
to

or

traders

the

might

managers,

be

orders
capture

slowly

2006;

and

Angel,

1994).

traders

and

other

skill

in assessing

include

proprietary

available

to

Directional
be

after

which

set

its

typically
at

event-based
3.

the

premium

reducing

from

paid-for

other

an

forecasts

cease

potency.

As

execute

using

strategies

Market-making,

also

to

Kaniel

as

limit

and

Liu,

Directional

impending

market

money

managers

to information

Their
like

be

scheduled

may

close

an

access

sources,

well

trading.

forecasts

tend

event

sources

prices

of

deploy

and

information

can

Bloomberg,
based

not

on

yet

market

sources.

forecasts
to

as
(see

situations.

public;

money

may

discrepancies

superior

news

Stat-arb

arbitraging

traders

informed

on
models

instrument.

high-frequency

market

run

institutional

direction

with

cheap

based

managers,

as

often

general

and

related

are

immediate

traders'

may

the

become

traders

misvaluations

known

traders

the

microstructure;

and

traders

analyses

from

also

to

These

Stat-arb

be

value-motivated

security

financial

price

estimate

Directional

as

of

can

1998):

prices

high-frequency

evolving

strategies,

successfully

move.

each

fast-dissolving

to

Directional

security

discrepancies.

for

Harris,

low-frequency

as

orders

2.

of

strategies

known

indicators.

traditional,

well

(per

also
for

(HFT)

valuations

value

valuation

market

wait

statistical

fair

as

short-term

stat-arb,

proprietary

or purely

determine

trading

categories

traders

their

fundamental
to

Strategies

to
soon

result,

to
be

to

market

(see

are

discussed
as

distributed

specific

to

traders

Vega,

time,
obtained

the
are

or aggressive

in Chapter
liquidity

at

Forecasts

Information

directional
orders

sensitive.

occur

useful.

be

market

known

time

2007).

public,
impatient

limit

orders

Directional

9.

trading.

Market

makers

have

no

material

liquidity.

When

deploy
10

market
using

automated

and

11.

although
8.1

types

relative

Figure

8.1

Trader

Type

market

Graphical

Representation

(stat-arb)

traders

statistical

phenomena.

strategies

deployed

Stat-arb

phenomena,

statistically

persistent

price

level

of equity

company.

The

multiple
the

price

The

and

critical

suitable
to

level

with
of

variation

in the

stat-arb
at

in

the
from

and

90

its

and

scene

in

using

the

simple

common

stat-arb

fundamental
exist

reported
be

by

of

roots.

between

current

of the

between

price

and

traded

levels

fundamental

instrument

Such

the

earnings

some

statistically

of

variables,

the

volatility

of

values.
identification
the

or

process

of financial

relationship

among

percent

statistical

higher

is

observed

relationship

average

sophisticated

trading

detection

with

also

financial

confidence
tested

trader

book.

discusses

recently

linked

is that

least

well.
and

returns

may

may

one

as

orders,

Aggressiveness

the

function:

relationships

other

statistical

deviations
modern

many
point

for

hold

of

limit

Book

on

often,

instruments,

level

another,

most

relationships

financial

Order

Chapters

space.

primary

the

order

Order

chapter

use

orders

double-digit

HFT

its

and

limit

traders

in

aggressiveness

exploded

This

after

persistent

Limit

to

market

of

reaping

in the

is named

for

providing

liquidity

likely

in the

in the

from

discussed

of order

price

Distributions

with

call

distributions

arbitrage

1990s,

most

may

profit

algorithms,

are

situations

to

strategies,

makers

to the

aim

high-frequency

Market

illustrates

Statistical

and

market-making

selected

Figure

insights

cousin

of

the

within

Thus,

in

technical

variables

confidence.

when

remains

value.

the

instruments

way,

This

numerical

two
stat-arb

analysis

has

standard
is
strategy

utilizing

Bollinger

envelope

around

the

near-term

likely

Stat-arb

than
alone.

Price

price

as

ETF

and

While

future

values

point

of

statistical

the

significant

This

hardly

be

of the

stocks

used

such

of asset

chapter

as

discussed

later

make

under

shown

by

in this

prices
or

spurious.

predictions
a trading

different
who

relationship:

two

statistical

too

(2003),

of

of

From

not

can

chapter.

meaningful

Challe

occurrence

trader

significant

is

same

the

consideration.

spurious

the

of

between

random,

to

Law

the

between

highly

a relationship

the

from

illustrates
statistically

sunspots

and

the

returns.

explores

relationships,

exists,

purely

the

is

comprising

mismatch

mining

have

ETF

be

data

by

should

observed

produces

outrageous

predictability

traded

power

principles

instruments
a

economic

staying

economic

ETF

If

may

between

suggesting

on

arbitrage:

or

relationship

view,

link

index

as

longer

based

solid

financial

the

relationships

following

or

mismatch,

relationship

price,

of shown

and

on

index

stocks

it can

strategy
the

the

unrelated

dependency,

(ETF)

and

a statistical

the

based

accordingly.

arbitrage

completely

persistence

persistence

individual

basket

contrast,

about

of

of the

profitability

a traded

weighed

do-it-yourself

higher

stat-arb

finance,

two-standard-deviation

average

statistical

fund

bounds.

statistical
of

a basket

profitably

path

to have

example

of

showed

moving

measuring

exchange-traded

One

By

simple

detecting

An

that

price

tend

models

the

the

models

phenomena

bands

techniques

well

as

developed

presents

for

detecting

specific

cases

stat-arb
of

proven

in

stat-arb

dependencies.
Practical

Applications

General

Considerations

The

prices

models
but

they

individual

of

often

of Statistical

two
will

can
names.

or
be

more

In

financial

fundamentally

nevertheless
equities,

Arbitrage

instruments
related

span
the

variety

companies

traded

in some
of

fashion

asset

issuing

or

classes
securities

other,
and
may

belong

to

the

changes

in

actually

the

be

more

rights.

(e.g.,
same

and

infrequently

applied

constituents

asset

various

can

Table

8.1

Class

Presented

Asset

the

be

of
in This

Fundamental

as

deviations

from

two

options

on

8.1

arbitrage
itemizes

strategies

arbitrage.

The

fundamental

Strategies

Arbitrage

Strategy

trading

Different

Equities

Risk

Equities

Liquidity

Foreign

exchange

Triangular

arbitrage

Foreign

exchange

Uncovered

interest

equity

classes

of the

same

issuer

arbitrage

Volatility

such

statistical

selected

Arbitrage

Equities

Options

to

when

Section

Fundamental

ETFs

apply

the
are

list

is by

arbitrage

found.

Pairs

and

The

to expiration.

of

additional

derivative
rate.

of

trades

be

times

of fundamental
many

price

Table
The

well

profitable

may

examples

Equities

Indices

of

different

instruments.

ideas
and

part

pair

indices,

temporary

subsequently.

Summary

Class

with

numerous

exhaustive,

opportunities

be

the

exchange

Passive

same

profitable

and

may

of instruments

but

financial

to illustrate

means

pair

rate

by

the

have

exchange,

issue

differ

by

may

strategy

exhibit

the

discussed

intended

can

may

often

issued

foreign

securities.

discusses

to

strategies

no

its

same

to

securities

typically

class

exchange

trading

ETFs,

underlying

section

the

derivative

In options,

same

This

on

shares

foreign

a foreign

fixed-income

and

equilibrium.
the

contract)

the

exchanges
In

similarly

Companies

same

different

be

rebalanced

index

the

prices.

can

respond
the

company.
and

of

on

therefore

Alternatively,

shares,

in

underlying

will

same

shares

trading

futures

equities

the

of

chosen

the

class

deviations

instruments

and

market.

by

Even

but

intraday

broad

one

company

industry

issued

than

voting

same

rbitragea

Index
curve

composition
arbitrage

parity
arbitrage

(UIP)

arbitrage

by

Asset

Cross-asset

Futures

basis

trading

Cross-asset

Futures/ETF

arbitrage

successful

statistical

Equities
Examples

of

fundamental
following

equities
popular

equity

of

Trading

Pairs

trading

strategy.

and

is

the

trading

levels

or

the

variables

relationship

i and

based

on

j can

price

on

can

of

liquid

in

be

the

two

S i,t

development

financial

at

stat-arb
of

between

and

arrived

trading,

used

relationship

any

different

pairs

commonly

levels

be

pairs,

the

spillovers.

involved

characterizing

based

instruments

most

steps

are

reviews

strategies:

information

and

involving

section

market-neutral

large-to-small

signals

other

issuer,

simplest

strategies

This

trading

same

Mathematically,

stat-arb

abound.

stat-arb

the

arbitrage,

Pairs

equity

classes

liquidity

models

arbitrage

price

instruments.

S j,t

for

through

any

the

two

following

procedure:
1.

Identify

that

trade

example,
least
2.

the

universe

at least
for

hourly

once

every

Measure

the

and

once

within

trading

financial

the

desired

frequency

instruments:
trading

choose

instruments

frequency

securities

unit.
that

For

trade

at

hour.
difference

j, identified

in step

between
(1)

prices

across

time

of

every

two

securities,

t:

(1)
where

is

According

to

observations

sufficiently
the

at

minimum.

central
selected

The

hours

observations
500
3.

daily
For

is,
of

is strongly
observations

each

pair

trading

of

day.

data,

securities,

of

however,

Thus,

For

statistics,
the

may
can

larger

observations.

constitute

relationships

years)

daily

(CLT)

frequency

recommended.
(two

of

theorem

persistent

the

number

limit

intra-day

seasonalitythat
specific

large

of

robust

have

be
at

30
bare
high

observed
least

inferences,

30

at
daily
a T

is desirable.

select

the

ones

with

the

most

stable

of

relationshipsecurity
Goetzmann
of

the

pairs

and

that

Rouwenhorst

historical

move

together.

(1999)

differences

in

To

perform

returns

do

a simple

between

this,

Gatev,

minimization

every

two

liquid

securities:
(2)
The

stability

of

cointegration
Next,

and

for

each

of squares
4.

the

relationship

other

statistical

security

obtained

Estimate

can

be

assessed

using

techniques.

i , select

the

in equation

basic

also

security

j with

the

minimum

sum

(2).

distributional

properties

of

the

difference

as

follows.
Mean

or average

Standard

of the

difference:

deviation:

5. Monitor

and

act

upon

differences

time

, if

in security

At

a particular

sell

security

i and

buy

security

j . However,

buy

security

i and

sell

security

j.

6.

Once

gain,

the

close

out

direction,
Instead

gap
the

activate
of

arbitrage

in

security

prices

positions.

If the

stop

detecting
can

be

statistical
applied

two

securities

details

of

implementation

Pairs-trading
changing

factors

market

reverses
prices

are

anomalies

to

and

strategies

if

to
move

achieve
against

desirable

the

predicted

loss.

between

fundamental

prices:

other

conditions.

fundamental

in detail

be
The

trained
mean

levels,

statistical

as

correlation

such

statistical

discussed
can

price

variables,

traditional
of

in

relationships.

arbitrage
in the
to

based

following

dynamically
of

the

variable

The
on

text.
adjust

to
under

consideration,

to

assumed
with

to

the

tend,

latest

can

in

deviation

used
of

in

the

relationships

relationships

and,

to

the

issues

Different

classes

relatively

constant

common

equity

following

two

limited

the

latest

results

for

many

theory

solid

helps

and

profitability

and

arbitrary

reduces

risk

methodology.
in

the

estimation

strategies

Classes

of the

stocks

by

issued

standard

of

are

statistical

influenced

by

conditions.

expect

price

earliest

in economics

economic

the

arbitrage

issued

research

between

stat-arb

the
using

positive
of

average
the

reflecting

academic

embedded

Equity
to

than

computed

produce

use

market

reasonable

equity

that

adverse

Arbitraging

by

improves

statistical

numerous

is

in turn,

weight

observations,

distinguish

operations

relationships,

It

recent

are

weighted

Similarly,

be

understanding

analysts

addition

more

can

relationships

a moving

window.

consistently

quantitative

In

computation

Thorough

of trading

as
given

validated

may

statistical

computed
being

most

Statistical
finance

identified

computations

environment.

traders.

be

the

economic

and

the

observations

observations

number

which

same

range

from

the

characteristics

company
each

same
only:

Issuer

corresponding

the

by

Same

to

be

other.

company
voting

to

two

trading

Different
typically

rights

and

common
within

classes
diverge

number

in

of
the

of shares

outstanding.
Shares

with

superior

shares

with

inferior

shares

with

a degree

wider

of
and

(2003)

shows

exists

in

country
of investor

voting
voting

control

(1988)

voting

Smith

over

rights

or

the

the

countries.

to country

and

The

and

the
of

the

on

takeover

shares,

the

for
for

varies
legal

regulations,

than
given

Nenova

voting

privileges

substantially

among

that

Horner

example.

environment,

the

to exercise

companysee

premium

premium

more

shareholders

(1995),

price

worth

nonvoting

direction

stock

depends

usually

allow

Amoako-Adu

most

protection,

are

privileges

and

that

rights

from
the

other

degree
factors.

In

countries

voting

with

premium
premium

stock's

market

Stocks

with

a higher

class

than

The

Korea,

of

are

the

the

voting

usually

more

more

highly

them

Amihud,

2002;

Chordia,

and

same

time,

market
class,

follows:

it

Brennan

Subrahmanyam,
the

more

information

creating

few

if

be

liquid

significantly

opportunities

price

range

average

for

fair

daily

bet

that

widens
range

the

to

without

range

will

hours.
suffers

for

the

of the

may

strategy

work

of

in

open

the
of

funds

public

from

with

the

strategy.

two

main

shortcomings

substantial

assets

Chipotle;

& Sons;

daily

under

less

the

capacity

of

and

daily

volume

for

January

6,

closing

price

2009,

to sustain

for

does

2009.
not

a trading

For
reach

liquid
the

strategy.

names,
even

strategy

one
of any

Yahoo!

for

just

Exchange

share

dual-class
all

classes

Corp.;

(NYSE):

Moog,

class

is

Table

8.2

Class
million

eight

Greif,

and

shares

the

example,

Entertainment;

Lennar

classes

restricting

for

Stock

City

share

limited,

equity

York

Forest

dual

2009,

two

New

the

restricting

6,

for

K V Pharma;

volume

further

on

the

have

severely

January

data

on

that

is
In

historical

Inc.;

Wiley

companies

markets

trading

Blockbuster,

small

the

(AUM).

companies

too

percent

to value

the

deviations

share

carried

January

as

following

number

Finance

NYSE

At

share

reason,

the

not

applicability

The

Finland,

in South

50

1989;

Brennan,

work

good

may

trading

2.

1986,

liquid

standard

dual-class

John

as

outstanding

to incorporate

less

two

management
1.

to

investors

1997).

may

within

and

trading

is likely

sufficiently

The

whereas

of shares

1996;

trade

than

narrow

such

arbitrage.

typical

more

zero,
close

Mendelson,

the

information

number

Eleswarapu,

of shares

faster

to

worth

actively
and

and

close

be

Subrahmanyam,

1998;

transparency,

value.

Amihud

and

can

prompting

(see

greatest

is worth

voting

liquid,

the

Inc.;

Inc.

often

small,

shows

registered
B daily
in

reasonable

the
on

volume

shares
trading

the
on

and
size.

is

Table

8.2

Closing

NYSE

on

January

Risk

Arbitrage

Risk

arbitrage

models

and

Black

by

the

broad
a

returns

Shares

to a class
finance

built

Sharpe

are
CAPM

idea

each

on

on

of

trading

literature.

the

(1964),

returns

At

capital

asset

Lintner

security

may

(1965),

whereas

downward

sloping.

equation

is specified

the

returns

as

are

degree

and

can

have

broad

breweries

securities

experience

companies

whenever

all

The

security

luxury

positive

on

returns.

individual
of

well,

higher

that

market

returns
as

produce

the

stocks

positive

returns

refers

are
by

particular

for

example,

produce

to

Dual-Class

equilibrium

developed

on

that

For

tend

classical
models

based

is different

positive

on

is

co-movement

time.

of

(1972).

influenced

The

based

(CAPM)

CAPM

market

Volume

arbitrage

market-neutral

model

The

Daily

6, 2009

are

most

pricing

and

or market-neutral

that

core,

Price

been

the

with

the

through
shown

to

produces

movie

whenever

the

vary

market

and

of

companies

overall

market

follows:

(3)
where

r i,t is the

broad

market

interest

rate,

equation
regression.
abnormal
co-movement

on

security

index

achieved

such

as

can

be

The
return

return

in time
Funds

estimated

resulting
that

with

Fed

the

i at

rate,

parameter

market

period

using

is intrinsic

t , r M,t

is the

t , and

r f,t is the

valid

ordinary

in

).

security

time

least

estimates,

to the
(

time

t.

squares

) and

on

The

(OLS)

, measure
the

risk-free

period

and
(

return

security's

the

The

simplest

example

trading

pairs

market

conditions,

This

with

type

of

strategy,

from
Often,

the

the

although

securities

i and

the

estimates

produced,
the

that

going

pair

arbitrage

the

changes

to

different

intrinsic

referred
long

would

to

and

equities

in the
returns,

as

short,

neutralize

in

broader
or

alpha.

market-neutral

respectively,
the

is

in

resulting

two

portfolio

exposure.

j are

used

is not

for
with

using

belong

alphas

of

the

and

betas

difference

alpha

the

of

those

and

two

for

(3).

two
Once

securities

and

test,

similar

beta

point

alphas

means

or

equation

of

in

in the

same

CAPM

deviations

difference

the

The

from

standard

the

to

mandatory.

determined

significance

determined
betas

often

beta

this

along

statistical

is

securities

industry,

point

but

similar

two

response

beta,

idea

market

CAPM-based

same

or

with

broad

the

strategy

with

securities

of

are

estimates,

betas

is

described

then

here

for

in

the

only:

(4)
(5)

where

ni

and

estimation
The

nj

are

of equation

standard

t -ratio

difference

test

the
(3)

numbers
for

statistic

of

security
is then

observations

used

security

j , respectively.

i and
determined

as

follows:

(6)
The

outlined
As

for

with

statistically

betas

other

for

alphas

in equations
t -test

similar

follows
(4)

the

through

estimations,

if the

t statistic

same

as

the

one

to

be

(6).

betas
falls

procedure

can

within

one

be

deemed

standard

deviation

interval:
(7)
At

the

economically
to

exceed

statistical

same

time,
and

trading

the

difference

statistically
costs,

significance,

in

significant.
TC , and

with

95

the

alphas
The

t -ratio

percent

has

difference
has
typically

to

to

be

both

in alphas
indicate
considered

has
solid
the

minimum:
(8)
(9)
Once

pair

identified,
and

the

shorts

the

of

securities

trader

the

on

strategies

accounting

with

the

the

equity

pair

the

equations
the

for

with

lower

alpha.

in the

forecast.

other

(7)

security

market-neutral

For

following

in

used

basic

fundamentals.

that

long

horizon

Variations

equity

goes

security

predetermined

satisfying

The

pair

the

trading

three-factor

Fama

model

alpha

strategy

French

be

is

is held

for

include

factors,

and

can

(9)

higher

position

security-specific

example,

through

such

as

(1993)

show

successfully

used

in

trading:

(10)
where

r i,t is the

on

broad

difference
and
on

big

in

index,

returns

between

constructed

asset

returns

to

liquidity

is

more

higher

pricing
to

one

such
for

going

big)
or

t (high

long

t is the

minus

indices

and

timeis

the

portfolios

minus

low)

in stocks

going

with

short

return
time-

of

small

is the

return

comparatively

in

stocks

with

ratios.

literature,
the

nimble

investors

asset

investors

costly

several

studies

have

average

returns:

see

Subrahmanyam

for

(1996);

the

On

then

offers

should

offer

higher

the

trading

documented
and

that

flip
of

their
side,

limited

make

it

positions,
if liquidity

liquidity

is
may

opportunities.
less

Mendelson
Chordia,

Limited

levels

unwind

periods

profitable

Brennan,

that

liquidity

to

outcomes.

Amihud

security

inconvenience.

lower

investors

returns,
highly

financial
investors

inconvenience;

to

in

prospective

individual

leading
priced

In fact,

and

by

compensate

difficult

potentially

offer

(small

HML

book-to-market

inconvenience

indeed

and

t , MKT

Arbitrage

classical

some

market

ratios

low

i at time

SMB

stocks,

book-to-market

Liquidity

stock

market

comparatively

In

on

capitalization

a portfolio

high

return

liquid
(1986);

and

stocks

have

Brennan

Subrahmanyam

(1998);

and

stocks

based

however,

no

returns

Pastor

and

portion

of

attributed

well

risk-adjusted

Pastor

and

returns

have

indeed

deliver

Stambaugh

illiquid,

relatively

high
for

that

then

at

the

least

securities

market-wide

be

is priced

illiquidity

positive

may

liquidity

market

consistent

are

investors

financial

If the

higher

(2003)

higher

Pastor

find

exposure

To

and

than

following

is the

return

in

arbitrage

abnormal

measure

that

returns

stocks

on

whose

market-wide

are

sensitivity
(2003)

OLS

equities,

in the

stocks

Stambaugh

in the

that

to variability

returns

liquidity.

estimated

The

recognize

of

returns,
deliver

they

illiquid

securities.

however,

causes.

the

basis.

market-wide
liquidity,

returns.

liquid

illiquidity

asset

that

prospective

less

(2003),

to market-wide

Trading

information

abnormal

these

observed

may

(1998).

compensate

Stambaugh

strategies

the

positive

in holding

the

Radcliffe

on

simply

individual

the

and

exclusively

involved

into

Naik,

delivers

average
risks

Datar,

liquidity

insulated

of

stock

devise

from
i to

metric

the

market
that

is

specification:

(11)
where
for

r i,t

stock

of the

i at time

market

proxies
returns

are

the

orders

in

vice

versa.

The

time

shares

capitalization
smallness
2002,

the

market

prior

return

and

t , v i,t

is the

on

stock

i at time

. The

sign

order

reasonable

to

outweighs

the
return

flow

at

assume

be

t in

excess

excess

return

time

t ; when

stock

that

the

number

to

volume

of the

of

sell

r i,t is included

shown

dollar

number

of

orders,

and

to capture

persistent

the

in the

return

securities.
Spillovers

other

securities

considered

example,

the

effects

from

time

t:

financial

varies
for

is the

time-period

Information

are

i at

of

it is

of most

Large-to-Small
Equity

direction

autocorrelation

series

stock

at time

positive,

buy

first-order

t , and

return

for

on

to

exchange
small

with

be

relatively

small.
to

stocks

The

exchange.
were

limited

market

precise

cutoff

On

NYSE

those

the
with

for

market

in

capitalization

below

billion

to

stocks

were

Small

stocks

than

$10

billion
those

large

returns

Lo

of

this

and

of

attributes

with

cap

to

to

more

past

underreaction

to

slowly

traded

than

small
sluggish

large

inefficient

One

are

extremely

of

that

stocks.

stocks

an

returns

the

more

efficiently

documents

to

large

found

large

large

$1

and

example,

on

that

information

stocks

for

of

billion.

significantly

returns
is

further

medium,

(1990),

follow

capitalization

of $10

news

MacKinlay

(2006)

be

in excess

react

absorb

market

to

phenomenon

small

this

market

stocks

Hvidkjaer

reaction

stocks

considered

and

smaller

actively

stocks.

were

known

stocks.

interpretation

billion;

with

are

on

more

$1

stocks

behavior

and

of

small

investors.
A proposed
their

reason

relative

primary

The

because

invest

particular

will

in small

small

highly
and

stock

stocks.

features

(2002)

volume

and

found

bid-ask

spreads

enabling

moving

be

of

that
exhibit

whom

to make
small

further

stocks

of

raising

the

percent

or

momentum

for

the

stocks
daily

trading

decides

to

of

any

an
small

liquidity
more

risk
of

Securities

data

a
and

institutional

are

the
Llorente,

informational
firms

following

midcareer

traded
and

mostly
traditional

decisions.

make
trading.

market

share

complicating

use

smaller

and

small

stocks

profitable

of
manager

to

the

investors

market

further

a result,

into

the

reported

is

are

institutional

portfolio

of

stocks

who

well-diversified

ownership

studied

profitability

As

of

to

of

small

impounded

if a portfolio

(SEC),

analysis

Wang

up

must

technical

inefficient,

million;

addition,

many

market

size

of

investors

gets

typical

into

investors,

low-tech
The

In

that
unattractive

end

cutting

response

institutional

even

Commission

investing
by

portfolio

the

are

is $200
stocks,

U.S.

Exchange

small

position.

in

information

size.

significantly,

the

delay
to

stocks

manager

institutional

of

the

their

into

stock

of
small

of

institutional

the

unattractiveness

source

prices.

for

and

stocks

illiquid

Michaely,

and
Saar,

content

of

trade

stocks

with

large

high-volume

periods.

Stocks

of

large

exhibit

no

time

trading

of large

Foreign

Exchange

Foreign

exchange
to

work

arbitrage
parity

in the

applied
models.

the

flexible

and

the

profitable

to

portfolio

model
in the

Arbitrage

Triangular

arbitrage

three

foreign

triangular

arbitrage

example

described

synthetic

prices

be

between

on

EUR/CAD

as

that

have

to

been

statistical
interest

models,

price

the

periods.

uncovered

rate

such

monetary

generate

deviations

The

as

model,

consistently

from

following

following

Dacorogna

well

or

framework.

temporary

EUR/CAD,

correlation

summarizes

sticky

arbitrage

therefore,

preceding

exchange

used

crosses.

by

mispricings

the

statistical

of

arbitrages

and

following

as

models

arbitrage

can

exchange

classic
section

model,

exploits

during

foreign

of

stocks

This

fundamental

monetary

Triangular

term.

triangular

price

trades

of

results

large

records

however,

strategies,

the

of

a number
short

Other

on

stocks'

spread,
reversals

trading

returns

small

bid-ask
exhibit

based

lagged

has

small

Profitable

stocks

and

with

sometimes

periods.

with

volume

firms

and

small

cointegration

shown

and

momentum

high-volume
involve

firms

et

al.

the

market

that

are

fair

prices

in

example

illustrates

triangular

arbitrage

(2001).
prices

computed

The

strategy

on

EUR/CAD

as

follows:

and

(12)
(13)
If market

ask

the

strategy

and

wait

the

position

ask

and

for
is

for

buy

the

USD-rate

prices

is lower

market

market

capturing

two

price

to

the

overcome

sampled

EUR/CAD

and
the

synthetic

simultaneously.
measurement

profit.

to

compute
a

significantly

bid

be

to

align,

on

distort

small
the

reverse

the
to

market
at

least

USD/CAD.

synthetic
as

then

enough

and

delay

EUR/CAD,
EUR/CAD,

between

high

the

for

synthetic

difference

EUR/USD

Even
can

The

sell
prices

should

spreadson

synthetic

EUR/CAD,
synthetic

bid

used

than

rate
as

The

should

one

relationship

be

second

in
as

result
by

of
the

unobserved
time

adjusted

dealer

the

Wright

computation

UIP

parity

as

order,

in the

the

background;

prices

may

have

levels.

(UIP)

find

rates

is run

is specified

the

prices

Arbitrage

(2005)

exchange

the

equilibrium

interest

and

affect

receives

Parity

uncovered

foreign

that

no-arbitrage

Interest

Chaboud
in

the

to their

Uncovered
The

trades

at

that

high

between

is
the

just
UIP

one
best

frequencies

4:00

such

relation.

predicts

and

changes

daily

rates

p.m.

ET

and

9:00

p.m.

rate

on

the

domestic

when

ET.

The

follows:

(14)
where

i t is

deposits,
in

the

i t * is the

a foreign

one

one-period

unit

example,

one-period

currency,

of

foreign

the

UIP

equilibrium

interest

and

S t is

currency

if domestic

Swiss,

interest

CHF/USD

the

of

United

as

deposits

foreign

domestic

(14),

can

denominated

exchange

price

currency.

Statesbased

equation
rate

on

spot

in units

means

equation,

rate

currency

and

be

used

of

Thus,

foreign

for

means

to calculate

the

follows:

(15)
The

expression

regression

can

form

be

suitable

conveniently

transformed

for

linear

estimation:

of

this

relationship

to

the

following

(16)
A

statistical

statistical

arbitrage
deviations

trading

decisions

Indices

and

Index

underlying

price

should

underlying
the

is

sides

be

equal

index.

by

the

relative

Under
to

the

the

index,

Occasionally,

securities

following

driven

components.

composing
the

two

of

into

equation

(16)

and

mispricings

of

indices

the
make

ETFs

their

within

the

look

accordingly.

arbitrage

securities

of

would

deviate

arbitrage

price

Law

of

weighted
relative

from

the

the

opportunities.

of

One

portfolio

according
prices

Law

of

of One
If

the

Price,
of

to
the

their
index

Price

and

and

price

index
individual
weights
and

the

present
of

the

index-mimicking

portfolio

of

itself,

the

index

index-mimicking
index

then

index-mimicking

realize
is

portfolio,

and

Alexander

(1999)

shows

strategies

deliver

transaction

net

buy

portfolio

buy

of

also

portfolio,

pricing,

index,

net

of

costs

transaction

index,

hold

gain.

Similarly,

lower

close

the

position

of

the
the

when

price

sell

market

if

that

the

costs,

until

than

the

exceeds

the

corrects
price

of

index

itself,

gains

have

the

its
the
sell
been

realized.

portfolio

manager

manager

benchmark

index

becomes
2.

the

The

be

the

dependent

its

of the

next

assigned

indices.

independent

Far

forth

by

for

East

step:

benchmark.

For

example,

(EAFE)

a
the

Morgan

Outperforming

which

model
log

sets

the

EAFE

manager.

determines

and

or

arbitrage

step

equities,

Asian,

portfolio

and

technique

in international

error-correcting

as

is

constituent

objective

variable

currencies

or

a European,

and

manager

running

selects

index

returns

management

investing

can

Stanley

positive

portfolio

portfolio

cointegration-based

consistent

cointegration-based
1.

that

countries

(ECM)

prices

of

lead

with

log(EAFE)

constituent

(explanatory)

EAFE

by
as

indices

in

local

variables:

(17)
where

the

allocations

statistically

significant

pertaining

represents

the

the

residual

from

can

be

to their

expected
the

constrained

An

absolute

the

indices

1
respective

cointegrating

strategy

in proportions

country

outperformance

in estimation,

return

n coefficients

depending

can
identified

further

indices

of the

regression

in step

stationary.

obtained

and

benchmark

investor

2 and

optimal

x1 xn,

EAFE

is
on

be

indicate

if

preferences.

by

going

shorting

long

in

EAFE.

Options
In

options

structure,
instruments
different

and

other

statistical
written
characteristic.

derivative

instruments

arbitrage

usually

on

the
The

same

with
works

underlying

different

nonlinear

between
asset

characteristic

but
is

payoff
pair

having
most

of
one
often

either

the

strategy

expiration

date

development

or

the

proceeds

strike
along

price
the

of

steps

the

derivative.

noted

in the

The
previous

sections.
Cross-Asset
Statistical

arbitrage

statistical

arbitrage

instrument

and

fundamental
Basis

is

not

limited

can

its

to

apply

to

derivative,

or

single

pairs
two

asset

class.

consisting

financial

of

Instead,
a

financial

instruments

sharing

values.

Trading

Futures

are

stat-arb

models.

asset

and

financial

instruments

Futures

are

easy

of

prices

are

choice

linear

in

many

functions

cross-market

of the

underlying

to model:

(18)
where
the

F t is the

price

underlying

interest

r t is the

the

differential

The

asset

rate)

and

also

the

is

of

differences,

ongoing

Lyons
six

and

between

known

as

foreign

exchange

futures

foreign
foreign

interest

futures

of
of

the

the

or

futures,

r t is

rates.

As

and

with

the

equity

following

the
pairs
steps:

contemporaneous

price

of

expires,

contract

follows

price
rate,

contract

exchange

trading.

process

monitoring

t , S t is the

the

basis

distribution

documents
pairs:

FRF/USD,

and
the
The

whenever

USD/CAD.

spot

the

and

price

differences,

of a basis-trading

futures

USD/JPY,
The

futures
works

futures
level

the

results
DEM/USD,

strategy

predetermined
whenever

time

t. For

domestic

the

currency

values.

share,

at time

and

acting

differences.

(2001)

between

at time

basis-trading

estimation

those

equity

arbitrage

asset

contract

t , T is the

rate

between

underlying

upon

(e.g.,

at time

interest

statistical

trading,

of a futures

as

price
or
price

falls

bets

reverts

follows:

exceeds

more,

GBP/USD,

strategy

prices

strategy

sell
the

and

buy

short

of

to

its

foreign
spot
foreign

the

that

spot

involving
USD/CHF,

the
mean

difference
or

currency
price

by

futures
a

currency
price

by

median

certain
futures

at

least

prespecified

difference.

predetermined
values,

strategy

the

strategy

Futures/ETF
In

markets
Kawaller,

been

500

and

in

the

documented

for

react

to news

faster

the

median

basis

to 0.5.

Whaley

both

S&P

500

equities

markets.

clearinghouse

due

The
for

functional
equity

that

involved

As

for

human

faster
the

underdeveloped

written,

the

decreased

lead-lag
from

the

(1990)

opportunities

market,

effect
5-

to

exist

for

of
500

effect

measured
index

futures

hybrid

By

the

period

HFT

central
out

early

1990s;

clearing

mechanism

up

year

2005.

have

been

to the

equity
time

and

strategies

this

spot

book

by

However,

systems

with

was

markets

documented

advantage.

and

rolled

the

systematic

futures

10-minute

the

strategies

day.

equities

futures

America,

machines

between

to the

of the

during

while

powerful

similar

returns

relative

North

on
and

this

S&P

Exchange,

platform

to a 1- to 2-second
still

in

relied

to

of the

market

information-arbitraging

futures

remain

money

Mercantile

traders

prices

development

trading
still

that

for

markets

contracts

exchanges
both

futures

Chicago

futures

show

markets.

minutes.

historical

electronic

result,

perfected

to the

spot

prices

(1990):
and

5 to 10

of the

than

futures

than

specification.

and

by

quickly

example,

causality

Stoll

adjustment

is likely

Whaley

when

announcements,

more

(1993),

returns

markets

most

as

of 0.4

news

to adjust

Granger
by

market

quicker

fully

that

computed

ratio

Koch

intervals,

stock

The

reports

are

a Sharpe

shown

futures

itself,

in 5-minute
led

levels

macroeconomic

Koch,

S&P

was

trigger
obtains

to

have

index

(2001)

Arbitrage

response

the

Lyons

had

Stoll

and

profit-taking
low

transaction

costs.
Cointegration
Stat-arb

of Various
models

instruments,
Often,

such

cointegration.

can

potentially
multi-instrument
Cointegration

Financial
also

Instruments/Asset

be

from

built

on

drastically
multiasset

refers

to

Classes

two

or

different
models

a condition

are

more

financial

asset

classes.

developed

whereby

using
prices

of

two

or

more

simple

financial

instruments

specification,

two-instrument

move

in

parameterized

cointegration

model

tandem

on
can

be

according

historical

specified

to

data.

as

follows:

(19)
where

P 1,t

price

of

and

are

2 and

the

said

price

second

to be

terms

least

90

percent

considered

first

financial

instrument

in a simple

is,

of

the

financial

exist.

deviations

of

cointegrated

that

error

under

OLS

reverting.

Perhaps

the

the

of

error

observations,

away

from

the

tests

test

mean

of

for

lie

as

term

is
of

follows:
two

error

and

stationarity

within

, the

the

and

error

works

is

Instruments

generated

Several

simplest

P 2,t

consideration,

regression.

whenever

mean

instrument,

if

at

standard

series

can

be

stationary.

further

securities,
price

the

coefficients

stationary,

To

is

fine-tune

statistical

a stat-arb

researcher

changes

stat-arb

in

relationships

dependencies
may

between

include

lagged

vector-autoregressive
several

time

any

realizations

framework

periods

ahead

two
of

to

of trading

detect

time:

(20)
(21)
The

number

typically

of

th

in

based

on

and

. As

accompanying

lag

is

terminal
Yet

k , used

determined

coefficientsg
-ratios

lags,

in the

another

models

is

financial

and

considered

lag

regressions

(20)

statistical
of

drop

nonexistent,

and

to

(21),

significance

thumb,
off

and

if the
less
the

of

absolute

than
k th

value

two,
lag

the

of

becomes

is
the
of

the

t
k

the

regressions.

common
to

a rule

the

extend

way
the

to

enhance

regression

of

performance
equation

(19)

of
with

stat-arb
additional

instruments:

(22)
As

in equation

cointegration
equations

(19),
is

(20)

the
and

the

key

stat-arb

stationarity
(21),

equation

of

criterion
the
(22)

of

error
can

the

multi-instrument

terms,
be

extended

Similar
to

include

to

lagged

observations

of prices.

Summary
Statistical

arbitrage

provides

a simple

implement

in

Statistical
have

is
set

in

clearly

defined

of

systematic

arbitrage

longer

powerful

based

staying

in

solid

economic

than

settings

conditions

fashion
on

power

high-frequency

that

as

are

easy

high-frequency

strategies

to

settings.

theories

based

it

is

purely

on

likely

to

statistical

phenomena.
End-of-Chapter
1.

What

How

are

do

the

are

and

key

E-mini

4.

on

Suppose
are

through

(10)

are

by

this

past

minutes.

and

markets?

the

be

a long

by

and

arbitraged

(21)

in this

market-neutral

framework

relationship
How

has

reversed,

can

one

>

2).

prices

of

and

now

statistically

the

negative

can

SPY

and

on

two

scenario?
returns
of

In

S&P

mathematical

1. How

high-frequency

(18),

generates

range,

the

estimate

1 and

different

on

to equation

short-term
(20)

Discuss.

futures

linked

coefficients

the

E-mini

theoretically

500

arbitrage?

According

equations

S&P

(same

however,
30

of

over

linked

in financial

statistical

SPY

that

the

present

behind

model.

futures

that,

traders

trading

significant

futures

stocks

principles

models

statistically

of

coexist?

Suppose

cointegration

E-mini

and

in a stat-arb

relationship.

and

types

considering

contracts

of SPY

three

differ

are

You

500

the

they

2. What
3.

Questions

equations

the

short

(3)
term,

2 >

1 over

the

arbitrage

such

an

interested

in

occurrence?
5.

A particular

arbitraging
do

so,

returns
some
your

you

ETF

is updated

daily.

the

ETF

against

basket

run

cointegration

against
statistically
findings?

returns

of

significant

the

Suppose
of

you

securities

models

pitching

the

large

universe

of

dependencies.

How

are

it comprises.
high-frequency
stocks
do

and
you

To
ETF

pick
arbitrage

up

CHAPTER

Directional

Trading

Many

Around

traditional

idealized

low-frequency

market

particularly
public

as

prices

rational
hypotheses,

who

adjustment

to

of

price

the

never

volatility,

into

prices

band

is

and

error.

The

price

the

often

implicit

on

order

flow.

With

news

basis,

referred

to

of

fluctuations
strategies
closer

instantaneously
strategies
surrounding

to

updated
presented
market-wide

be

as

its

all

Fair

feasible

only
(see

efficient

markets

as

and
are

a common
idealized
with

the

chapter

events,

such

as

range.
a

on

The

degree

The

on

bring

the

news

for

trial

through

the

in

the

tick-by-tick
from

arbitraging

HFTs

they

of

price

to profit

By

news.

trade

volatile

is occurring

whereby

latest

news

French

placed

release,

state

the

conditions.

positioned

good:

fact,
follows:

the

that

markets.

news

price

happens

sellers

trades

on

the

because

price

ideally

that

a certain

from

optimal
and

prices

postannouncement

ttonnement,

new

financial

market

optimal

buyers

in this

is

In

described

accompanies

available.

noted

level

each

their

has

price

announcements

deliver

real-life

constant

traders

further

of

within

instantly

surrounding

the

settle

among

reported

and

eventually

toward

relevant

information

instantaneous.

finding

high-frequency

impact

ever

market

release

can

small,

negotiation

fundamental

evolution

hardly

that

ttonnement

all

all

is
goes,

assumed

details).

news
is

settles

of

theory

1961,

the

news

however

process

for

major

impoundment
swings

1970,
watched

the

several

is

incorporate

information

Muth,

assume

condition

the

reflect

of

has

surrounding

as

always

Fama,

Anyone

following

valuation,

expectations

models

instantaneously

soon

quantitative
the

The

markets

information

when

quantitative

conditions.

often:

long-term

Events

the
price

high-frequency
real-life
all

The
market

markets

prices

are

high-frequency
movements

announcements

and

other

occurrences.

Developing

Directional

Directional

event-based

strategies

that

events.

Event-Based
strategies

place

The

trades

events

of
increases

value

U.S.

Fed

Funds

and

profitably

The
make
The

rate

the

economic,
that

Fed

time

Funds

rates

The
are

of

reaction

industry,

or

For

even

the

the

example,

consistently

that

to

affect

raise

rate

for

announcements

events

trading

markets'

again.

raising

AUD/USD.

group

consistently

and

therefore,

event

arbitrage

profit

window

and

of

simultaneously
for

the

the

USD/CAD
of the

can

be

U.S.

consistently

arbitraged.

positive

event

dollar,

the

of

time

the

to

basis

time

in

decisions,

goal

the
be

interest

unexpected

lowering

refer

occurrences

instrument(s)

and

on

may

instrument-specific

of the

Strategies

as

scheduled

may

be

opened

the

is typically

ending

such

over

strategies

shortly

time

The

identify

For

events

the

portfolio

just

the

fully

event.

before

anticipated

or

positions

just

liquidated

the

ex-ante,

portfolio

announcement

is then

that

each

beginning

announcements,
of

portfolios

surrounding

period

afterwards.

ahead

to

window

a time

economic

announcement.

is

after

shortly

the

after

the

announcement.
Trading
to
with

positions

several

hours

low

can

be

and

can

volatilities.
the

trade

probability

that

the

momentum

wave

high-frequency
automated
Developing
on

equilibrium

tick-by-tick

event

an

from

of

will

the
be

to

to

are

the

most

profitably

often

higher

the

ride

the

profitably

are

second

event

equilibrium

strategies

and

outcomes

an

response,

able

of

profitable

response

faster

arbitrage

a fraction

consistently

post-announcement

applications
trading

the

strategy
the

in

speed

gain;

to

result,

anywhere

result

The

determines

As

held

price

well

level.

suited

executed

for
in

fully

environments.

event

arbitrage

pricing
trading

data

and
and

trading
leverages
events

strategy
statistical
the

instant

harnesses
tools
they

research
that

are

assess
released.

Further
the

along

impact

this

chapter,

of events

developing
Most

in

an

event

on

event

we

prices;

now

arbitrage

arbitrage

will

survey

we

academic

investigate

the

research

on

mechanics

of

strategy.

strategies

follow

three-stage

dates

and

development

process:
1.

For

each

historical

event

identify

times

of

past

events

in

data.

2. Compute
to

type,

historical

securities

of

price

interest

changes

and

at desired

surrounding

the

frequencies
events

pertaining

identified

in

step

1.
3.

Estimate

behavior

expected
surrounding

The

sources

the

past

can

collecting

be

the

as

and

greater

Dow

trading

Constitutes
events

news

about
key

are

repetitive.

estimate

events

of

that

the

day

and

U.S.
at

and

further

sites.
make

Eastern

Federal

Most

the

job

time.

of

times

the

Some

Open

irregular

SemLab,

news

in

announcements,

a.m.

collecting

RavenPack,

occurred

Internet

of

8:30

in

data.

Markets
during

the

Firms

such

HFTIndex.com

other

and

tradeable

simplifying

data

in

automation

of

Event?
arbitrage

activity,

requirement

for

The

historical

at

diligence

in event

economic

The

price

strategies.
an

used

historical

unemployment

occur

formats,

event-driven

time

changes,

distribute

machine-readable

on

various

U.S.

those

Jones,

AbleMarkets.com

The

as

specified

same

released

rate

based

from

the

always

interest
require

for

easier.

such

Reuters,

What

at

much

announcements,
Committee

times

collected

data
are

responses

events.

and

recur

example,

day

past

of dates

announcements

for

price

event

recurrence
impact

of the

strategies

market

disruptions,

suitability
of
events

can

is

events
and

that
allows
project

be

any

and
the

releases

other

events.

chosen

events

researchers
the

of

effect

to
into

the

future.
All

events

do

not

have

the

same

magnitude.

Some

events

may

have

positive
more

and

negative

severe
be

measured

the

expectations

frequently
example,

following

response

to

an
price

the

domestic

CPI

little

change

on

other

as

pure

of

key

component.

news,

then

the
the

cash

investor

the

foreign

should

change

in

the

country.

level

If,

market

of

however,

expectations,

figures

financial
and

and

on

political

the

well

before
are

figures

based

variables,

as

well

become

available,

of

forecasts,

basis

prices

well

economists

other

forecasts

into

in the

or

1982,

several

be
the

news,

of
from

the

before

for

slowly
months

released

on

the

formal

difference

between

expectation

formed

on

value
the

and

the

was

the

scheduled

component
the

studies

of
released

basis

the

separating

and

priced-in

(see

Frenkel,

assumed

time,

next

is

expected

example)

or quarters

unexpected

impact

event

over

the

news
the

macroeconomic

developed

past

estimation

or

earliest

Edwards,

to

The

and

steps

news

value

in

event

Many

market

expectations

The

the

future

with

pair

with

about

securities

change,

and

during

line

in

event-related

trade

unexpected

economic

stock,

occur.

the

1981,

be

is

earnings

example,

domestic

from

deviation

line

exchange

earnings,

observed

participants

announcements

figures

of its

Similarly,

of the

inflation,

When

their

in

move.

announced.

forecasting

impounding

are

expectations

are

continuously

market

the

form

news.

event

of a particular
value

changefor

to

an

lower-than-expected

of a foreign

out

have

occur.

statistics

with

the

price

may
of

economics,

present

not

(CPI)

turns

participants

tasked

the

index

should

formal

level

magnitude
event

The

net

events

realized

earnings

unexpected

consumer

the

or

should

the

the

Market

to the

some

The

In

higher-

price

market,

of

a surprise.

if

and

others.

event.

adjust
a

the

exchange

One

the

However,

expectations,

the

of

to as

announcement.

in

a deviation

should

prices,

than

as

referred

flows

on

consequences

can

for

impacts

that
trend

best

most

observed
predictor

of

news

release

day.

news

release,

was

in the

of autoregressive

announcement
analysis.

Later
and

researchers
Roubini

decisions

such

(1993)
of the

Once

bankers

the

not

at

market

and

figures
lead
The

the

empirical

best

seen

in

intended

announcements

the

the

USD/DEM

The

authors

particularly

significant

two

following

the

consumer

confidence

payroll

and

momentum
Surprises

in

relative

week

forecasts

U.S.

for

Reuters.

the
The

and
coming
forecasts

money

adopt
the

supply,

to

news

defined

in

the

indeed

can

be

impact

is

the

Goodhart,

and

Payne

of

that

news

exchange

trade

balance

rates,

but

an

are

Street

only

from

event

studies

non-farm
price

announcement.
be

measured

For

publish
as

developed

within

caused

forecasts.

announcements,

were

U.S.

can

Journal

at

announcements

announcements

of economists'

week's

sampled

However,

following

news

rate

announcements

Wall

(1998)

macroeconomic

and

news

or more

the

trigger

adjusting

Yet

announcement.

averages

Barron's

and

exchange

of

macroeconomic

may

securities.

employment

hours

economic

bankers

framework

found

predictors

12

to published

every

and

the

lasting

of

effect

to

central

to

central

of

the

significant

pertaining

hours

that

Almeida,

intervals.

the

changes

rates

Reserve.

that

gradually

impact

terms.

five-minute

action,

the

autoregressive

changes

the

Grilli

direction.

movements

on

Federal

is

major

interest

shows

future

as

of

shorter

documented

that

and

to predict

the

drastic

of

(1993)

U.S.

actions

Instead,

such

fashion

predict

given

course

evidence

autoregressive
to

control,

in the

the
behind

make

disruptions.

economy

used

to

Evans

a framework

including

bankers'

liberty

control,

such

rationale

longer-term

in their

and

using

central

their

large-scale
follow

been

main

of

under

Eichenbaum

bankers,

the

are

variables

have

central

again,

predictability

as

consensus

do

example,

Bloomberg

survey

of

field

economists.
Forecasting

Methodologies

Development
trading

data

of

forecasts

surrounding

involves
event

announcements

on
of

very

interest.

specific
Event

studies

measure

the

returns

surrounding

quantitative
the

impact

news

event

of

and

announcements

are

usually

on

the

conducted

as

follows:
1.

The

announcement

identified

and

events

and

should

be

and
can

be

As

very

the

of

returns

CPI

between

8:30:01

are

before

and

categorized

database

after

the

carefully

The

surprise

the

value

and

realized

the

realized

to the

times

if the

calculated

on

CPI

for

researcher

value

and

of
event

quotes

component

the

and

in

the

USD/CAD

historical

of

the

securities

prediction

the

analyst

on
are

in

calculated

past

is

the

consideration.
the

impact

USD/CAD,

from

CPI

always

announcements

under

change

is

data

surrounding

in evaluating

1-second

announcements

impact

the

of interest

is interested

on

change

U.S.

the

8:30:00

to

announcement

released
then

days.

at 8:30

estimated

a.m.

in

ET.)
simple

regression:

where

R t is the

security

is the
the

traded

the

frequencies.

between

corresponding

a.m.

The

the

simulations,

are

analysis.

announcements

linear

changes

consensus.

one-second

3.

useful

surprise

ways:

autoregressive

example,

(The

and

events

high

in two

announcements
For

create

with
at

difference

forecast
2. The

detailed,

captured

on

times,

of securities

difference

based

To

prices

measured

the

As

recorded.

the

trades

dates,

of

vector

order

vector
interest

that

captures

changes

announcement

surprises;

of the
in

announcement
equity

prices

in

and,
on
are

under
by

the

changes

to

regression
other

the

than
average

consideration.
in

t
in

pertaining

factors

measures

security

adjusted

to

of

arranged

error

intercept
due

for

announcements

idiosyncratic

finally,
the

of

announcement

announcements

estimated
returns

the

order

in the

t is the
is the

changes

surrounding
in the

of announcements;

announcements;

Changes

returns

arranged

of surprise

news

impact

of

the

overall

market
on

prices

equity

market

to account

values.

model

for

The

the

impact

adjustment

of Sharpe

of broader
is

often

market

influences

performed

using

the

(1964):

(1)
where

the

hat

expected

notation

equity

market

expresses

return

the

estimated

average

over

estimate

historical

and

data

is the

using

the

(1968),

and

model:

(2)
The

methodology

the

estimation

was
method

trading

opportunities.

During

a typical

made
a

around

includes

the

global
and

like.
country,

the

particular

to the

industry.

announcements
determined

With

is

from

country;

announcements
may

or,

like

and

usually

in

and

conditions
interest

economic

data,

and

regulation

contains

banks,

government-collected

news

economic

news

to

announcements,

industry

of tariffs,

related

mergers

launch

comprises

are

macroeconomic

releases,

product

central

be

Company

Macroeconomic
major

significant

rate

indicators

regional

gauges

of

performance.

the

alerts,

new

introduction

by

economic

earnings

news

Brown

statistically

consequences.

annual

Industry

and

announcements
or

announcements,

particular

economic

news

Ball

delivers

numerous

The

have

by

day

industry,

quarterly

acquisitions
and

world.

may

this

day,

company,

they

developed

to

trading
the

particular

news,

first

development

press

now

of information

wires,

feasible

and
to

news

well-developed

captures

news,

categorizes

based

on

historical

machine-readable
computer
companies

with

used

engines

events,

and

of
an

data
input

machine-readable

event
matches
Various

that
to

can

RSS
as

instant

be

readily

are

system
to securities
offer

parsed

strategies.
include

it

they

arbitrage
events

feeds,

Google,

companies

event-driven
offerings

as

such

the

automated

analysis.

as

such

announcements

streams

and

aggregation

capture

released.

technology

by
The

Thomson

Reuters,

Dow

A Practical

Jones,

latest

figures

a.m.

on

prespecified

USD

crosses

least

in

tracking

theory.

capture

changes

inflation

figures.

first

time

sample

windows

2008.

As

all

15-second

on

and

returns
the

between
and

changes,

the

USD-based

we

is announced,
expectations
The

sign

ignoring

the

then

tells

market's

select

the
data

inflation
through

at 8:30

a.m.

window

during

that

and

time.

We

30-second,

the

parity

and

impact

of

1-minute,

(PPP),

currencies

the

new
is

To

the

PPP

to

keep
news

a
is

When

adjust

inflation

to

of

2002

1-minute,

rates.

the

is

the

30-second,

spot.

foreign

versa.

consider

We

trading

5-minute,

latest

end

occur

measure

USD/CAD
vice

the

January

the

that

of the

U.S.

recorded

power

rates

of inflation
test

5-minute,

disturbs

rises,

will

into

inflation

and

as

trades

and

deviation

instantaneously,

recent

and

purchasing

foreign

rate

the

a.m.

of USD/CAD

exchange

inflation

window.

announcements

then

adjustments
strategies

to

from

at

opportunities

event

corresponding

domestic

domestic

example

the

trading

trading

data

9:00

We

announcement

to

to

further

blocks.

According

the

other

adjustment,
the

announcement

inflation

quotes

data

15-second

as

at 8:30

and

announcements

profitable

the

spot

quickly

profitable

tick-level

8:00

of the

the

of

monthly

one-time

following

from

the

released

how

construct

levels

all U.S.

define

download

rate

players.

USD/CAD

and

surrounding
in

we

can

known

announcements

release,

when

period

opportunity,

partition

we

are

instantaneous

in identification

trading

EST,

of smaller

inflation

On

an

in price

step
the

dates.

Identifying

in practice,

August

number

U.S.

undergo

happen

define

a large

Example

The

The

and

spot
the

U.S.

exchange
ratio

inflation

equilibrium
When

expected

to

in the

preannouncement

rate

the

the
U.S.

increase

simple,
same

the

and

levels.

matters

of

in

fashion
adjustment

this
as

it
to

figures.
us

during

which

time

intervals,

if

any,

the

market
our

properly

trading

when

and

consistently

window

inflation

from

rates

in Table

9.1 .

Table

Number

of

9.1

Following

the

Estimation

U.S.

seconds

4 1

15

seconds

5 6

Looking

at

five-minute

announcements,
in

the

instantaneous.

The

summary

of

Opportunities

Up

intervals

U.S.

Inflation

U.S.

to
with

potentially

8:30
95

inflation

adjustment
statistically

in

only

days

the

results

is

in

USD/CAD

Down

drops

in

significant

reacts
that

during

the

to

at

inflation

appears

response

only
is

interval

at

to

may
after

which

be

occur

lower

response

hypothesis;

a.m.,

to

of

The

to

indeed

five-minute

announcements

8:30

appears

inflation

reaction

adjustment

released

U.S.

persistently

confidence.

instantaneous
is

the

and

response

statistical

news

to

rate

decreases

percent
the

USD/CAD

inflation

a.m.

support

surrounding

that

USD/CAD

8:25

U.S.

includes

Announcements

Inflation

it appears

decreases

the

sample

Trading

Rate

U.S.

The

during

1 0

30

inflation

Inflation

to announcements

0 0

1 minute

from

announced.

Persistent

Frequency

5 minutes

8 to 9 a.m.

were

presented

responds

point

completed.
following

all,
the
No

rises

in

inflation.
Higher-frequency
occur

intervals

in short-term

adjustment
occur

to

from

therefore,

to

presents
the

number

of

announcements
30-second

showing

increases

8:34

Splitting

four

bursts.

tell

us

At

one-minute

in inflation

8:35

a.m.

a consistent

data

a different

into

tradable

now

profit-taking

30-second

inflation,

intervals,

in

inflation,

the

for
seen

example,
to

price

the

consistently
adjustment,

opportunity.
we

observe

increases

the

postannouncement

be

adjustments

postannouncement

opportunities

of rising

a decrease

intervals,

can

This

storythe

that

further.

adjustment

intervals.

For

the

price

adjustment

now

the
For

occurs

in

announcements
occurs

in

one

30-second

postannouncement

Examining

15-second

intervals,

time-persistent

trading

announcements,

there

USD/CAD

In

falling

a.m.

time

summary,

number
the

inflation

we

if

delays,

carrying

Tradable
This

at

implemented

in
and

periods

9:00

Six

15-second

number

rising

response

during
to

a.m.,

of

inflation
which

the

inflation

presenting

intervals
during

shorter

time

ready

consistently

the

intervals,

currency

same

costs,

and

we

8:30

nature

systematic

to

the

larger
to

opportunities

black-box)

reduces
errors

in response

of

(i.e.,

expensive

detect

movements

knowledgeably,

approach,

risk

of

in human

execution

judgment.

News
summarizes

various

financial

instruments.

various

academic

sources.

may

published
interest

in

persist,

have
due

however,

based

on

if

for

estimate

response

of

events

time

frames

for

since

the

strategies.

strong

periods

using

of

time.

in studies

is

the

impact
drawn

described

data;

is

of

the

tends

the
to be

the
first

general

impact

and

Some

of

were
and

on
from

impact

studies

factors,

low-frequency

used

their

news

The

fundamental

shorter

variables

and

of machine-readable

trading

impact

of the

more

of

types

impact

considerably

to proliferation
set

event

The
The

shrunk

this

even

studies

likely

is,
to

included

high-frequency
comparable

or

pronounced.

Corporate

News

Corporate

activity

and

annual,

the

announcements.

equity

15-second

short-term

to

specific

even

higher

For

8:30

The

conducive

section

news

five

significant

announcements.

which,

even

announcements

look

of statistically

them

an

frame.
as

makes

note

increased

opportunities.

accompany

we

are

between

tradable

interval.

opportunities.

consistently

announcement

9:00

time

prices,

such

significantly

and

as

earnings

impacts
Unexpectedly
unexpectedly

announcements,
equity

prices
positive

negative

of

the

earnings
earnings

both

quarterly

firms

releasing

typically
often

depress

lift

corporate

stock

valuation.

Earnings

announcements

announcement

that

consensus
to

its

new

equilibrium

value

unexpected

component

of an

event

Theoretically,
of

the

on

the

theory

are

the

or median
key

economists'

of the

security

price

component

of

the

difference

between

the

economists'

forecast.

The

variable

priced

as

discounted

capital

of

the

The

used

in estimation

of

the

prices.

company,
by

mean

forecasts.

from

unexpected

as

is the

equities

determined
pricing

the

analyst

adjustment

The

computed

and

by

different

in a rapid
level.

is

announced

preceded

materially

results

announcements

flows

is

forecast

impact

are

asset

Ross

present
at

the

pricing

(1977),

or

values
appropriate

model

the

of

(CAPM),

future

cash

interest

rate

the

investor-specific

arbitrage
opportunity

cost:
(3)
where

E [ Earnings

future

time

t,

discounting

adjust

Significant

market,

to

information
of

large-scale

of the

found

Unexpected

responses

whereby

forecasted

even

of

at a
for

changes
equity

to
prices

earnings.

from

can

company

appropriate

present.

earnings
and

flows
rate

about

impacts

earnings

cash

discount

price

movements

most

the

rapid

deviations

prevent

is

expected

dividends

to new

market

the

R t

generate

quickly

To

and

time

earnings

large

t ] are

result

can

in market

earnings

announcements

values

releases

are

made

cause

disruptions.
on

the

after

the

overall
markets

close.
Other

firm-level

splits,

for

and

Roll

following
Event

news

example,
(1969),
a split

arbitrage

announcements
documented

has
who
relative
models
affect

firm-level

also

affects

been
show

equity

documented
that

to their

the

factors

by

share

equilibrium

incorporate
each

prices.

company
for

Fama,

prices
price

the

The

Fisher,

of

stock

Jensen,

typically

increase

levels.

observation

differently.

evaluation

effect

include

that
The
the

earnings

most
size

widely
of

the

firm

market

1987;

capitalization

and

Fan-fah,

Industry

News

Industry

news

with

Mohd,

consists

in

that

macroeconomic
systematic

and

Nasir,

of

entire

market

in

that

fashion,

Atiase,

is

industry

1985;

Freeman,

2008).

and

regulatory

products.

the

news

see

of legal

new

throughout

securities

details,

mostly

announcements

reverberate

(for

These

sector

same

collected

and

usually

along

announcements

and

the

news

decisions

tend

to

move

direction.

all
Unlike

disseminated

emerges

in

in

an

erratic

fashion.
Empirical

evidence

relaxing

rules

higher

equity

findings

of

equity

price
the

of

of

Administration

was

Macroeconomic

agencies

for

are

example,
Federal

by

accompanied

by

set

as

and

on
by

Reserve

such

reported

rules

decisions

government

variables

and

prices.

that

the

rising

constricting
includes

of

equity

controls

introduction

(2005)

U.S.

the

ascertained

price

the

Boscaljon
by

rules

who

elimination

in

of

found

Food

that

and

Drug

values.

News

Macroeconomic

U.S.

or

result

evidence
(1999),

decisions

industry
of

The

values

advertising

that

particular

Emanuel

equity
equity

suggests

introduction

relaxation

depressed

relaxation

and

in

down.

Bowman,

upswing

controls

the

values

announcements
an

of

whereas

Navissi,

in

decisions

activity

values,

pushes

resulted

regulatory

governing

activity

that

on

or

observations

predetermined

economists

CPIs

statistics

some

the
are

of

typically

agencies

schedule.

at the

Bank

are

central

England.
not

affiliated

set
with

Interest
banks,

On
but

by
rates,

such

the

are

the

made

other

as

hand,

observed

countries'

the

and
central

banks.
Other

macroeconomic

departments
ICSC

Goldman

International

of

indices

both

for-profit

store

sales

Council

of

and
index,

Shopping

are

developed

nonprofit
for

example,

Centers

private
is
(ICSC)

by

research

companies.
calculated
and

The
by

is

the

actively

supported
tracks

and
weekly

consumer

promoted
sales

by

the

at sample

confidence:

the

Goldman

retailers

more

and

their

future

spending

and

the

higher

is

the

aspects

of

economic

different

prices

of McDonald's

and

earnings

hamburgers

in different
levels.

Table

an

schedule

shows

announcements

for

European

news

European

trading

Most

most

session

are
that

Australia
hours

in Asia.

Table

9.2

March

3, 2009

New

Ex-Ante

North

distributed

in
with

the

the

Zealand,

are

Schedule

of

the

of

released

the

trading

Pacific

in

region,
during

Macroeconomic

retail
indices

from

relative

to oil

supplies

news

trading

morning

U.S.

are

day.
of

the

closed.

and

Canadian

North

American

Europe.
which

the

the

Other

markets

the

morning

is their

a typical
in

of

about

macroeconomic

2009,

of

afternoon
Asia

ranging

American

announcements

from

and

3,

higher

index

indicator

are

index.

of

The

an

countries

released

while

coincides

announcements

March
often

session

macroeconomic

governments

ex-ante

Tuesday,

is

the

activity

employment

9.2

the

of

to industry-specific

as

consumers

potential,
value

Group.

serves

confident

economy

measure

Sachs

morning

Announcements

Most
includes
trading

for

Many

announcements

are

which

are

of

aggregates

financial

institutions.

major

media

various

forecasts

The

and

data

accompanied
made

consensus

every

week

consensus

by

figures

companies,

economists

by

forecasts,

economists
are

such

as

and

calculate

of

usually

various

produced

Bloomberg

LP,

by

that

average

poll

industry

expectations.
Macroeconomic
impact

news

on

arrives

from

currencies,

commodities,

derivative

instruments

is

technique

that

of securities

of Event

Event

trading

each

event

may

of

and

estimated

persistent

world.

The

fixed-income

using

impact

the

event

of news

and

studies,
on

the

prices

Arbitrage

be

to

many

different

asset

for

documented

classes,

every

yet

financial

persistent

the

impact

instrument.

impact

of events

of
This

on

various

in

foreign

instruments.

Foreign

Exchange

Market

responses

exchange

Markets
to

were

Edison
Love

the

is applicable

considers

financial

equities,

usually

measures

corner

of interest.

Application

section

every

studied

(1996);
and

macroeconomic
by

Andersen,

Payne

(2008),

foreign

exchange

that

news

about

real

economic

surprise

increases

by

documents

little

impact

of inflation

on

et al.

(2003)

conducted

their

quotes
intervals.
quickly
releases.

percent

interpolated
The
and

authors
efficiently

Volatility,

based
show
to
however,

on

such

At

new

takes

(1997);

(2003);

and

shows

that

time,

exchange
on

for

every
USD

the

author

rates.

foreign

to create

exact

exchange

rate

to taper

to
payroll

employment,

according

longer

significantly

nonfarm

same

analysis

average
levels

the

foreign

timestamps
that

as

payroll

average.

on

Vega

most

(1996)

nonfarm

appreciates

Andersen

0.2

reacts

Edison

in

and

Payne

others.

activity,

In particular,

and

Diebold,

many

finds

100,000

Goodhart,

among

(1996)

figures.

Almeida,

Bollerslev,

Edison

employment

announcements

to
off

exchange
five-minute
levels

the
after

adjust

information
the

spike

surrounding
that

most

bad

news

news

usually

et al.

(2003)

announcements.
has

The

more

authors

pronounced

also
effect

document
than

good

news.
Andersen
International
for

Money

estimation

authors
rate

of

then

R t as

use

the

Market

Services

surprise

model

the

consensus
(MMS)

component
five-minute

forecasts

of

as

news

changes

compiled
the

by

expected

the
value

announcements.

in spot

foreign

The
exchange

follows:

(4)
where

R t-i is i -period

is

surprise

the

periods,

component

and

is the

seasonalities.
following

value
of

the

time-varying

Andersen

et

of the
k th

fundamental

volatility

al.

(2003)

five-minute

that
estimate

variables:

GDP

(advance,

Nonfarm
Retail

preliminary,

and

payroll
sales

Industrial

production

Capacity

utilization

Personal

income

Consumer

credit

Personal
New

lagged

consumption

home

Durable

sales

goods

Construction
Factory

orders
spending

orders

Business

inventories

Government
Trade

expenditures

budget

deficit

balance

Producer

price

index

Consumer

price

index

Consumer

confidence

index

final

figures)

spot

rate,

variable
incorporates
the

impact

S k,t-j

lagged

intraday
of

the

Institute

for

Supply

Association

Management

of Purchasing

Housing

(ISM)

Managers

index

(formerly

[NAPM]

the

National

index)

starts

Index

of leading

Target

Fed

Initial

indicators

Funds

rate

unemployment

Money

supply

claims

(M1,

M2,

M3)

Employment
Manufacturing

orders

Manufacturing

output

Trade

balance

Current

account

CPI
Producer

prices

Wholesale

price

Import

prices

Money

stock

index

M3

Andersen,

Bollerslev,

following

currency

CHF/USD,

and

30,

1998.

positively,
following

EUR/USD
authors

with

99

index,

Managers

(NAPM)

responded

and

different
(2008)
the

percent

nonfarm

and

the

index.

(2008)

countries
studied

United

the

States,

3,
all

1992,

surprise

money
document

affects

different

impact

of

the

Eurozone,

the

that

the

industrial

balance,

currency

consumer
of

Purchasing

pairs
in

considered
the

initial

M3.
macroeconomic

currency

pairs.

macroeconomic
and

in

employment,

increases

stock

responded

increases

Association

the

December

pairs

surprise

the

DEM/USD,

through

currency

trade

National

considered

USD/JPY,

to

orders,

All

to
and

(2003)

payroll

goods

claims
Payne

that

significance,

negatively

unemployment

January

document

durable

Vega

GBP/USD,
from

variables:

confidence

and

pairs:

The

production,

Love

Diebold,

the

news
Love

news
United

from

and

Payne

originating

Kingdom

on

in
the

EUR/USD,

GBP/USD,

authors

find

EUR/USD,
in

the

respective

trading

domestic

equity

and

Payne

news

by

are

the

and

also

the

originating

document
regions

in Table

the

on

their

9.3 .

Announcements

Payne

The
on

news

three

shown

News

effect

the

(2008)

from

Region-Specific
Love

largest

affected

findings

per

day
and

to

to

firms,

two

is

filled

foreign.

Expected
For

classical

with

How

on

the

(2008)

macroeconomic

do

the

announcements,

macroeconomic

changes

The

likely

news

determined

by

riskiness

of

U.S.

equities

U.S.

Treasury;

the

is

discount

as
central

rate

of equities.

changes

sales,

impact

of

risk-free
share.

risk-free

by
rate

short-term

bank.

of equity

The
earnings

is,

target

and

those

consumer

earnings

prospects
signal

and

the

risk-free
higher

tough
a result.

bare

minimum,

the

idiosyncratic

rate

pertinent

bill

interest

the

as

its

to

firms.

and

risk-free

significant

traded

conditions.

at

rate

are

in market

may

three-month

lower

publicly

expectations

The

the

prices

with

however,

finance

proxied

of

uplifting

earnings

the

equity

confidence

costs,

equity

the

by

classical

level

often
the

retail

decrease

a particular

is taken

country's

in

associated

labor

and

rate

rates

affected

boost

in

earnings

consumer

Rising

conditions

discount

country

to

outfits.

business

be

changes

expected

discount

increasing

are

retail

in

in the
may

example,

theory,

changes

earnings

spending

financial

factors:

and

prices

the

has

pairs.

markets?

According

that

news

of

their

of

exchange-rate

Markets

both

for

type

Currency,

typical

due

Love

the

Effect

Respective

the

of

EUR/GBP

is most

Kingdom.

currencies;
9.3

Equity

U.S.

GBP/USD

impact

Table

the

while

United

specific

that

and

issued

equities
rate

in

the
the

the

another

announced

rate,
are

by

to

lower

theoretical

by
is

How

does

macroeconomic

empirical

evidence

interest

rate

other

shows

returns

with

90

percent

percent

announcement

prices

reaction

to

mixed.

Positive

inflation

independent

of

1983,

for

1985,

business

cycle.

are

news

during

periods

Roley

and
stock

rates

Cutler,

and

Poterba,

and

suggests

seconds

to

or

that

minutes

Several

for
of

the
high

stock

lower

(see

other

on

news

induce

conditions

conditional

of

the

activity,

returns

and

Roley,
variables

contemporary

during

usually

stock

macroeconomic

the

market

is

Pearce

state

industrial

economic

changes

of

production

recessions
according

rises,

response

but

only

but
to

the

figures
bad

news

McQueen

equities

Jagannathan
is

not
for

industrial

The

(2005).

The

limited

to

production

have

and

following

U.S.

observe

the

figures

when

news

of

the

rising

asymmetric
when

actually
by

Boyd,

response

to

macroeconomic

Finnish

Lflund
response
equity

the

presents

confirmed

asymmetric
the

For

During

hypothesis:

markets.

in

found

economy.

expansions.

in unemployment

asymmetric

were

increase

attributes

been

the

the

returns

overheating

increase

findings

instance,

drop

the

of

economic

(1992)

to

statistics

state
that

during

returns

is overheated,
news.

the

finds

Orphanides
of

economy

on

(1992)

contractions,

unemployment.

in unemployment

dependent

Orphanides

economic

in

tend

Higher-than-expected

unemployment

(1997),

to

monthly

evidence

within

manner,
long-term

macroeconomic

market

details).

reactions

example,

news

occur

to

long-term

(See

Anecdotal

shocks

other

unexpected

cause

good

for

strongly

(1993).

Similarly,
to

both
affect

rates.

nonmonetary

reactions

good

in

Ample

time.

Stock

produce

pronounced

confidence

example.)

of

a less

practice?

respond

positively

short-term

in

prices

Decreases

indeed

for

for

adjustments

in

statistical

confidence

Summers,1989,
most

news.
rates

equities

equity

and,

macroeconomic
interest

affect

that

announcements

short-term

99

news

and

Hu,

and

Nummelin
to

surprises

market;

they

found

that

sluggish

higher-than-expected
states

Whether
the

or

of the

not

macroeconomic
are

volatility.

volatility

While

is

not

macroeconomic
been
Kuttner
the

to

interest

Stivers
the

for

(2005)
Dow

higher

tells

announcements

are

announcements

to
price

figures,

or

interest

shows

that

during

periods

volatility.
that

Different
frequencies.

The
with

Karceski,

and
for

of

U.S.

if

higher

real

largest

information

increase

Lakonishok
and

and

Japanese

European
to

impact
in

(1998),

equities

risk,

and
by
equity

on

no

major

consider

news
of

employment
(1999)

example,

news

drives

asset

price

Hogan

(1998)

found

has

considerable

markets.

equities

equity
of

in an

news

Veronesi

stock

affect

frequency
for

U.S.

that

(PPI),

news

appear

to

macroeconomic

which

Errunza

macroeconomic

the

to

macroeconomic

of the

and

announcements

FOMC.

uncertainty,
markets,

are

the

sensitive

show

(2008)

index

more

are

Market

accompanied

when

price
of

in

comprising

higher

days

they

decisions

Open

response

be

Wilson

producer

and

macroeconomic

on

and

major

(CPI),

volatility

sources

than

Bernanke

Connolly

(2008)

U.S.

Savor

European
and

the

increase

be

rate

of

In the

on

made.

investors

monetary

impact

returns

index

higher

should

Wilson

major

have

component

implies

risk

and

news

of equities
in

market
other

volatility.

volatility

in

of

Federal

(DJIA)

higher

with

volatility

U.S.

prices,

stock

unexpected

volatility

Savor

are

announcements

consumer

that

that

volatility.

return

Average

us

days

the

in the

Higher

to

an

equity

spikes

Indeed,

on

in

increases

macroeconomic

that
of

increase

news.

returns.

stocks

stock

by

out

market

show

Industrial

theory

returns

in

increase

document

macroeconomic
financial

surprises

announcements

Jones

pointed

related

example,

(FOMC)

move

surrounded

(1989)

necessarily

rate

Committee

usually

Schwert

significantly

(2005),

bolsters

announcements

always

factors,

shown

growth

economy.

announcements

market

production

at

data

equity

appears
data.

analyzed

arbitrage

different

pricing

to
Chan,

monthly
theory

setting

and

equities

are

using
past

found
most

equities

of

the

Wasserfallen

equities

Flannery

and

and

announcements
to

be

in

rate,

the

and

decreased

and

the

al.
a

with

(1998,

poor

p.

job

in

frequencies.
news

sensitive

current

real

news

companies

from

PPI,

monetary

on

equity

returns

account

deficit,

domestic

the

model
following
on

aggregate,

both

balance

figures.
stock

a result,

for

markets

in

the

news

foreign
rates

and

example,

notes

that

in

response

AUD/USD
(GDP);

domestic

equity

exchange

increased

product

rising

return

macroeconomic

As

(1992),

of

generalized

influence

USD-based

Sadeghi

of

the

the

types

foreign

to

on

several

starts

that

the

returns

that

significant

States.

to

gross

following

by

have

housing

United

daily

(GARCH)

overreact

markets,
the

that

found

document

balances.

Australian

increases

et

estimate

and

typically

from

found

authors

CPI,

(1995)

account

the

individual

macroeconomic

heteroskedasticity

report,

countries

domestic

The

volatility:

Mehdian

tend

size,

most

return

impacted

announcements

employment

markets

of

of

do

monthly

impact

variables

macroeconomic

developed

Chan

By

that

(covary)

factors

(2002)

conditional

and

show
yield

tandem

individual

frequencies.

(1998)

However,

significantly

independent

Ajayi

low

dividend

in

at

no

news.

autoregressive

in

finds

are

returns

al.

macroeconomic

Protopapadakis

macroeconomic

of trade,

move

at

of

data.

equities

equity

et

and

covariation

(1989)

quarterly

with

that

the

return

Chan

equities.

that

returns

ratio,

corresponding

explaining

characteristics

future

portfolios,

factors

document

U.S.

of

book-to-market

are

returns

idiosyncratic

predictive

factor-mimicking
return,

182)

that

to

exchange
equity

inflation

returns

or

interest

rates.
Stocks
react
(1987),
exhibited

of

differently
for

example,
higher

to

different

industries

macroeconomic
pointed
sensitivity

have

been

announcements.

out

that
to

stocks

of

announcements

shown
Hardouvelis

financial

institutions
of

monetary

to

adjustments.
as

The

well.

Li

extent

and

Hu

capitalization

are

are

stocks.

small-cap

The

size

of

impacts

the

surprise

average,

markets

than

did

Fixed-Income
Jones,
employment
while

days

of

and
the
the

PPI

volatility

CPI

decline

in

bonds

issued

move

the

prices

news

for

example,

standard

deviation

and

foreign

studied

the

of

exchange

and

incorporated

and

in

the

U.S.

Treasury
the

goods

impact

(2001).

orders,

nonfarm

on

the
the

information

employment

is

figures,

documents

higher

yields

market
those

in

that
bills

a
and

that

prices

just

two

of

curve.
10

GDP,
payrolls,

distinct

information

and

retail

sales,

and

is

fully

following

its

estimate

classes:

jobless
and

the
on

Remolona

announcement
starts,

and

(1993)

announcements

housing
PPI,

Lee

(1999)

Fleming

and

1999);

minutes

Remolona

macroeconomic

yield

and
new

to

Ederington

1998,

Ederington

show

responses
by

(1997,

(1998)

and

of

find

beyond

(1996)

bond

Remolona

Green

impact

authors

persist

that

include

and

Fleming

high-frequency

not

of

Treasury.

bond

announcement.

note

causes

of

Remolona

effect

markedly

announcement

Krueger

announcements

Elton,

The

increased
did

the

(1996)

prices.

U.S.

Fleming

bonds.

volatility

that

Edison

studies

(1993);

indicators,

bond

unemployment

the

macroeconomic

durable

than

prices.

bond

U.S.
by

Treasury

the

into
and

High-frequency

measure

one

(1998)

U.S.

indicating

(1987)

Balduzzi,

on

of the

promptly

and

entire

market

surprises

(1992),

in equities

Lumsdaine
data

day,

Hardouvelis

Fleming

large

macroeconomic

Schirm
within

changes

announcements,

incorporated

Lee

and

the

matter

surprises.

and

announcement

PPI,

of

to

Markets

Lamont,

that

with

stocks

macroeconomic

those

larger

large

that

component

surprises,

caused

appears

to

Aggarwal

small

capitalization

show

sensitive

prices.

that

market

(1998)

more

the

equity

document

of

rate,
trade

the

(1999)
CPI,
leading
balance.

Fleming
be

the

and
actual

forecast
All

of

Remolona
number

for

Fleming

and

authors

then

same

10

macroeconomic

document

the

to

variable.

The
a

percent

average

to a 0.9

percent

statistical

the

2-year

average

percent

drop

released

from

the

consensus

in the

impact

8:30

to

the

Table

yield

the

the

As

in

The

bond

was

and

yields

in

drop

confidence.
30-year

in

in

macro

Table

9.4

rate

led

on

bill

with

95

yield

of

3-month

percent

news

a.m.,

jobless

of the

The

the

8:35

9.4 .

by

a.m.
of

change

in

a 1.3

8:30

changes

in

percent
of

at

surprise

and

99
yield

to

studied

the

average

increase

drop

with

of

curve

surprise

confidence

in

were

reproduced

percent

note

announcements

positive

are

surprise

Reuters

news

significant

results

Thomson

significance

yield

statistically

the

macroeconomic

release.

the

entire

the

less

(1999)

measure

response

shows,

news

Remolona

on

define

released

the

the

releases

(1999)

the

corresponding
not

statistically

significant.
Table

9.4

Documented

Futures
The

Effects
by

of

Fleming

Macroeconomic
and

News

Remolona

Announcements

(1999)

Markets
impact

market
Ederington

has

of
been
and

Becker,

Finnerty

(2004)

document

the

macroeconomic
studied

Lee

by

(1993);

and

Kopecky

that

news

announcements

Becker,
and

Finnerty,

Simpson

(1996)

and

announcements

and

on
and

the

futures

Kopecky

(1996);

Ramchander

Simpson

and
regarding

(2004).
Ramchander
the

PPI,

merchandise

trade,

bond

Ederington

futures.

adjustment

of

within

first

the

volatility,

minute

have

emerging

Tamirisa

(2007)

whereas

effect.

However,

conducted
mature

Istambul

Stock

of
price

futures

happens

News-related
15

minutes.

significantly
as

influenced

by

the

however,

to

USD-based

Malaysia,

and

(GNPs),
exchange

Philippines,

stock

markets.

between

CPIs,

Singapore,
At

the

rates

the

U.S.

same
money

had

that

emerging

are

not

on

the

market
variables

little

effect.
markets

depending

consideration

and

the

on
degree

economy.
(ASEAN)
by

(2002)

find

supplies,
ASEAN

and

Thailand)

time,

Bailey

supply

U.S.

lira/USD

domestic

predominantly

of

while

world

variables,

money

that

Turkish

supply

found

Sharma

(2006)

to

the

while

Nations

influenced

Wongbangpo
products

Asian

ij

impact

and

money

world

much

markets

the

major

generate

found

equity

macroeconomic

the

instantaneously

rate,

under

affect

had

and

that

(2000)

with

not

and

equities,

market

be

did

found

and

Southeast

appear

variables.

global

news

macroeconomic

Turkish

integration
of

U.S.

markets

interest

Bigan

emerging

market's

Association

Turkish

and

the

almost

and

and

announcements

emerging

production

Taskin,

of

equity

news

Bannister,

Sahlstrm,

estimated

affect

industrial

Muradoglu,

size

on

Exchange,
the

Andritzky,

Omran,

respond

(2008)

of macroeconomic

announcements

analysis

rate,

returns

relation

prices

news-induced

following

impact

that

announcements,

Kandir

local

the

example,

found

markets

affected.

the

move

released.

macroeconomic

authors

similar

exchange

how

Nikkinen,

macroeconomic

national

that

is

for

the

For

domestic

equity

of the

CPI

exchange

news

persist

the

find

foreign

considered

study
The

impact,

the

(1993)

the

often

economies.

spreads.

were

and

after

may

authors

such

rate

Lee

and

Economies

Several

bond

payrolls,

and

interest

however,

Emerging

on

nonfarm

and

countries,
their

that
interest

countries

stock

local

gross

rates,

and

(Indonesia,

significantly
(1990)

domestic

influence

found

no

causal

returns

of

Asian

Pacific

markets.

Commodity

Markets

Empirical

evidence

Gorton

and

activity

and

inflation

and

affect

inflation

prices,

except

on

Chambers

(1985),

and

relationship

between

Real

Investment

Equity

real

traded

securities,

market

capitalization

1991
an

estate

and

at

invested

in

percent

of

payout

ratios,

news
The
by

Simpson,

the

returns

well

established

by

when

Stevenson

it is

75
estate,

of
the

may
than
on

Ramchander,
REITs

as

REIT
and

examine

on

the

inflation

response

of REITs

is comparable

the

Federal

increases

rises.
response
policy.
to that
the

themselves.

about

the

publicly

1960.
$9

million

like

peculiar

assets

should

out

to

in

is traded

following

pay

The

at
of

be

least
their

90
high

macroeconomic

equities.

when

monetary

in

A REIT

differently

the

novel

Because

(2007).

the

prices

must

Webb

in U.S.

Summary

have
(1984),

details

fairly

REIT's

ordinary

changes

rates

have

dividends.

unexpectedly

(2007)

REIT

which

rates.

was

to

performance

increase

inflation

the

commodity

Bond

Congress

2006.

the

respond
would

are

by

REIT

unanticipated

depressing

on

more

REITs

billion

earnings

inflation

Funds

news
real

rates,

interest

U.S.

required

and

REITs

on

real

of the

See

(REITs)
the

percent

taxable

of

effect

for
and

U.S.-based

to $300

but

announcements
impact

both

effect

interest

(2006)

trusts

least

its

rising

prices

of

higher-than-expected

valuations.

Frankel

that

The

a positive
by

findings

Trusts

grew

real

have

the

document

mixed;

investment

equity,

structure:

be

commodity

includes

prices.

commodity

of all

steadily

ordinary

can

accompanied

impact

Estate

who

commodity

generally

when

markets

(2006),

however,

activity

cooling

commodity

Rouwenhorst

announcements,

in the

volatility

has

been

documented

The

authors

found

unexpectedly
Bredin,
of
The

falls

O'Reilly,
REIT

find

of equitiesincrease
of

REIT

as
and

returns

authors

that

prices

to
that
in
while

Directional

trading

around

windows

immediately

of

market

other

enable

Which

news

and

Estimation
profitable

of

in

preceding
the

trading

the

impact

of

decisions

narrow
reaction
historical

surrounding

the

following

is/is

not

tradable

gain

on

event

in

the

HFT

Why?

The

S&P

previous
b.

the

profitability

Questions
of

sense?
a.

generates

announcements.

End-of-Chapter
1.

following

participants.

announcements
market

events

500

registers

a positive

market

open

relative

to

close.

Announcement

of

the

QE3

(quantitative

easing

led

by

the

U.S.

in

HFT

Fed).
c. Regular
2.

announcement

What

financial

of employment

instruments

can

figures.
be

traded

on

events

setting?
3.

Suppose

a particular

announcement
latest

of

positive

announcement

change

is

stock

usually

changes
figures

negative.

How

rises
to

have

the
just

can

your

within
U.S.

15

minutes

nonfarm

been

payroll.

released,

system

of

trade

an
The

and

the

on

the

announcement?
4. Intuitively,
prices
5.

in the

Does

more

why
short

does

like

a change

in CPI

affect

futures

term?

high-frequency

or less

something

efficient?

directional

trading

on

events

make

markets

CHAPTER

10

Automated

Market

Most

MakingNave

high-frequency

automated
was

trading

(HFT)

market-making

entirely

mode.

human,

This

successful

Inventory
systems

services.

but

chapter

are

Some

is now

moving

considers

the

market-making

Models

30

deployed

years

ago,

to a nearly
basic

fully

to

provide

this

activity

computerized

principles

behind

the

models.

Introduction
Every

veteran

market

trader

maker

instrument.

who

With

profits

rare

lived

human

traders

market

turned

the

next

USD/CAD

An

foreign

markets
bonus

against

for
by

trader's

taking

bet

setting

fell

Automated

properly
into

sour

of

such

futures

and

periodically
bets

off

its

at his
mark,

trading

as

well
stay

and

tested

actions.

As

of

market

incidence

market

makers'

bottom

making

is cost

efficient:

then

bank,
his

trader

trader

who

in
made
and

Once,

the

instantaneous
was

prohibited

fired

profitability

in nearly

The

the

was

discretion.

resulting

day,

and

was

certain

was

one

loss,

personal

provides

makers

the

once-potent

immediately

from

ever

again

floor.

making

institution

and

enhanced

and

reduce

heavy
extensive

these

burnout

at

incident

discretionary

financial

generated

a major

the

programmed

human

particular

about

suffered

employer.

market

dominant

commanded

note:

to his

any

markets

stories

losses

market

automated

he

radically

on

market-making

him,
example

following
a foot

same

directional

multimillion-dollar
dismissed

the

time

in

employers,

all

exchange

a long

of

lifestyles.

nearly
on

story

financial

their

luxurious

end

day.

of
for

exceptions,

transactions

kingpins

transaction-based
and

recount

owned

These

bonuses

can

line.
once

as
on

several
other

market

script.

computer
a

result,
crashes

Second,
the

advantages

to

participants.

the
First,

Unlike

human

traders,

systems

do

deviate

automated
and

execution
human-intensive

not

market

negative

surprises

of automated
programming

makers
for
market
and

testing

stages

are

development

and

shareholders
reduced

the

best

on

can

instrument
venue

limit

or

market

current

portfolio,

the

or only

limit

market

making

orders

placed

When
with

the
the

quote.

Similarly,

incoming

makers'

profitability

and

deploy

as

posts

and

maker

the

market

and

whether

sell

orders.

market

maker's

only

limit

however,
activity

describing

order

portable.

to post

of

Most

limit

limit

choose

new

methodology

maker,

in the

sides

the

description).

a market

to a continuous

term

book

participants,

both

that

extremely

positions

limit
financial

centralized

buy

may

on

centralized
another

and

follows:
limit

conditions

order

their

market-making

provided

technology

works

general

with

definitions

world

market

any

is

with

the

consider
open

market.

market

buy

limit

Liquidity

making

as

well

limit

buy

as

trading.

and

Market

book

strictly

price

highest
is

ask,

makers'
market

reaches

price,

this

distributed

to

when

limit

or best

market

exchange,

limit

simultaneously

market,

offer

little
are

strategies

equipped

centralized

to refer

market

lowest-priced

HFT

reductions

Almost

class,

Some

order

of

require

enhanced

asset

orders.

is a more

limit

market-making

market-making

on

provision

of

another

the

process

sell

of

order

computerized,

Depending

makers

automated

form

on

in

making

A market-making
human

the

the

another

exchanges

model,

to

exchange

deploys

3 for

today's

details

head-count

markets.

run

even
also

Chapter

an

be

and

trading

book

feature
various

running
book

for

costs.

across

strategy

from

in

16

market

directly

clients

transaction

scalability

most

savings

passed

and

Perhaps

of

are

Chapter

automated

The

significant

order

(see

processes),

compensation.

(see

completed

sell
and

the

order

orders

are

market

maker's

price

becomes

the

other

market

market
in the

is quoted

orders

the

executed
of

market,

the

by

bid

participants

maker's

to other

best

limit
his

market
virtue

opposite

order

as
sell

order

on

that

a Level

order

becomes

is
a

I
the

best

participants.
of being
direction.

matched
The

with
market

maker's

bids

market

maker's

incoming

are

ask

market

market

maker

inventory
track

over

or picked

little

and

Market

order

just

below

price

order

short

Similarly,

if

participant,

or

market

compensation
market

to

the

as

market

profitability.

avoid

being

a profit;

too

quickly

run

much

liquidate

the

market.

providing
market

the

the

market

his

market

limit

order

sell

maker's

maker

short

or liquidity,

traders

account).

buy

market
order,

If the
position

collects

the

spread

to traders

are

the

executed

another

limit

buy

with
is

account.

the

the

a market

order

yet

buy

above

maker's

from

on

a limit

it is matched

of

orders,

orders

just

When

maker's

size

order

market

market

limit

placing

sell

limit

arrives

the
the

the
in

order

neutralizes

The

a limit

recorded

of

maker

and

and

to the

placement

participant,

order,

with

portfolio,

orders.

to

to generate

market

market

sell

is added

for

price

another

is

maker's

known

therefore:

in order

inability

makes

position

position

traded

profitability.

describes

market

sell

position

the

the

are

enhance

are

by

order,

inventory,

and

maker

risk

price.

limit

of

lifted

loss.

market

it is matched

long

market

from

quantities

the

Principles

the

maker's
a

a long

trader

creates

arrives

market

the

the

be

and

markets.

market-making
of

risk

insufficient

a certain

to

limit

the

information

the

be

the

Key

sides

market

by

may

Making:

both

(i.e.,

over

face

a nutshell,

to

said

quantities

acquiring

sufficient

orders,

executed

These

it, to reduce

respond

makes

position

In

and

divests

of a market

sell

are

every

account.

to ensure

inventory

inventory

With

upon

market

orders

or

his

functions

Keep

Too

in

by

limit

orders.

to manage

broad

Manage

hit

offer

Immediately

two

be

accumulates

begins

The

or

instrument

inventory.

to

buy

maker

financial

said

known

as

and

size
in

of
the

as
placing
liquidity

takers.
Of

course,

maker

places

market
his

making
limit

orders,

is subject
he

to

risk.

immediately

As

soon

faces

two

as

the

types

market
of risk:

Inventory
Risk

risk

of adverse

Inventory
incur

risk

when

selection
describes

the

value

market

movements.

position

(buying)

experience

addition,
his

other

parties

of

buy

order

imminent

maker

and

Galai

produce

the

significant

sell

commit

wishes

to
from

resulting

includes

the

misses

while

inventory
the

to

losing

entering

limit

about

Per
an

the
the

that

the

Copeland

informational

by

better-informed

adverse

makers,

the

that

and

trade.
from

than

maker's

down,

off

market

be

to

market
set

maker

case,

and

it is possible

market

picked

due

and

a market

suffer

are

loss

information

likely

this

to

orders

for

also

maker

order,

the

in

they

losses

is
when

than

of

In

time,

maker

better

maker

a market

risks

same

to

term.

competition

risk

market

example,

limit

long

likely

short

potential

possesses

market

whereby

selection
the

risks

can
can

be

controlled.

Simulating

a Market-Making

Trading

with

orders

execute
difficult

limit-order

all

the

market

to

maker

face

market

measures

taker

the

natural

orders.

between

about

While

profitably

are

of

(1983),

investors.

the

to

is

the

at the

the

For

in

may

Inventory

gains

due

might

accumulating

market

inventory.

limit

with

the

positions

information

is

disadvantage,

their

selection

trade.

direction

in price

least

maker

market

maker

market

at

the

the

better

market

maker

when

of his

the

market

is incurred

market

the

declines

position,

his

is matched

has

to sell

maker,
of

inventory

reflecting

the

end

market

as

loss

downward-trending

differences

market

seller

risk

adverse

When

losing

his

execution

informational

the

on

loss

costs

risk

taker.

liquidating

for

The

in

looking

opportunity
waiting

of his

positions,

in difficulties

potential

Thus,

inventory

close

the

limit

orders
and

to

generates

sometimes

model.

modeling.

Strategy

This

nonlinear
they

section

do

not.
delves

payoffs;
As

sometimes

a result,
into

the

limit
logistics

limit
orders
of

Strategies

that

assumption

that

observed

in

contrast,
solid

when

the

market

each

data

order

at

simulating

Most

deploy

the

process

in

when

the

the

of the

simulated

processed
the

Figure

10.1

price

Market-Making

This

section

strategies

forecasts
Fixed
The
orders

order.

limit

buy

A limit

sell

price

or

limit

price

placed.

additional

By
work.

executed
the

ask

an

trade

is

limit

order

best

with

latest

order

crosses

the

trade

simulated

the

order

limit

or

simulated

requires

or

. A

price

last

is

the

order,

in

considered

quote

order

only

falls

can

be

best

bid

below
marked
quote

order.

Limit

Order

Execution

Strategies

concern

the

discussed

details

themselves

Enhancements
are

given

of Simulating

explains

that

inventory.

reaches

the

near

orders

quote

10.1

be

simulated

limit

of the

the
of limit

trade

Mechanics

Nave

of

last

whenever

exceeds

time

Figure

executed

can

executed

consider

opposing

shown

price

is

execution

simulations
best

orders

of
only

based
in Chapter

on

practical
with

market-making

effective

short-term

management

directional

market

11.

Offset
most
at

nave
a

market-making

predetermined

strategy
number

of

is continuously
ticks

away

placing
from

the

limit
market

price,

on

orders
the

both

being

quotes

likely

passive

allowed

to
price,

market

from

nave

strategy,

more

frequent

capital.

the

can

bears

can

found
they

Sandas

are

proportional

are

liquidity.

The

thought

went,
who

the

that

time

profit

order

number

away

current
from

the

the

be

risk

the

limit
The

market

the

market
the
maker

market

making

Exchange,

for

increases

study,

written

profitability

providing

to the

to

orders

makers'

spend

maker's

market

of

a
the

spread

Stock
on

in

and

key

times

A study

profit

price

cumulative

Stockholm

tied

are

the

market

of

the

market

is directly

of

execution,

to

frequently.

they

(1997).

market-making

content

to

longer

the

the

models

be

is

liquidity;

frequency

was

not

change

the

compensated

orders.
and

theories

waiting

higher

did

submitted

equilibrium
Seppi

more

is

makers

market

shown

orders.

the

the

the

lower

expected

shows

early

were

providers

the

his

price,

by
not

instead,

with

which

their

executed.

contrast,

makers

on

matched

to

makers'

orders

stocks
that

(2001),

market

of

the

market

market

far

of

higher

and

execution

traded

example,

the

to

of execution

percent

from
of

the

limit

volatility.

been

higher

capital,

the

quotes

order

has

current

market

reallocation

capture,

for

actively

when

his

stub

of

proximity

probability

10

probability

trading

price
at

from

within

limit

the

The

the

of extreme

resulting
of

flip

waiting

higher

the

market-maker

they

of

profitability.

maker

By

at times

order

instruments,

only

probability

placed

away

so-called

offset

Frequency

makers'

on

prevent

the

orders

far

the

limit

whereas

orders

executing

smaller

Limit

the

financial

limit

to

on

those

most

place

market

The

In

Naturally,

executed,

orders,

zero.

market.

price.

to be

limit

to

the

depends

market

are

close

of

executed

current

for

sides

included

the

for

time

until

expected
their

Such
Rock

presumed

limit

time

that
spent

order

providing

execution,

compensation
order

ideas
(1996),

were

market

to

specifications
known
Glosten

the
liquidity
once

as
(1994),

static
and

The

assumptions

exchange

of

conditions.

equilibrium

Changing

the

canceling

orders

expected

a tidy

compensation

adverse

position

once

cancellations
most

was

static

and

markets

at

high-frequency
human
The

to

the

be

Slower

of

orders

limited

improve

orders

orders

be

fixed-offset

shown

speed

hit.

As

up

in

Limit

free

of

an

order

charge

a result,

profitability

bid
of

and

the

on

today's

than

limit

execution

their

deliver

away

from

orders

prices,

market

maker's

makers

to ensure
and

to

their

high

profitable
the

ask

market

rates,

still

best

HFT

allows

strategy,

it a function
away

spot

profitability.

market-making

going

may

A sample

in equation

the

market

vary

price.

way

the

are

makers.
to

be

determination

likely
In

offset

to change

in high-volatility

market

need

may

intuitive

volatility:

market

orders

one

One

of

from
for

limit

executed.

best

their

conditions.

premium

however,

indeed

of ending

written.

and

Offset

farther

higher

the

high

offsets

market

make

order

makers

were

better

the

resubmit

can

larger

the

with

is to

at
by

book,

Volatility-Dependent
To

enjoy

technology

and

the

via

risk

performed

was

limit

market

orders

book

makers

technology

strategies

the

early

counterparts.

cancel
top

bearing

be

reflected

of

and

limit

this

Super-fast

continually

details

costly,

could

time

place

may

technology.

the

their

market

ability

for

revisions

market-making

however,

on

prohibitively

models

the

limit
offset

conditions,
to

be

hit,

closer

limit

generating

low-volatility

placed

of

conditions,
to

the

market

of volatility-dependent

to

offset

is

(1):

(1)
Offset
Still

Is a Function
another

offset
Parlour
limit

of Order-Arrival

way

to improve

dependent
and
orders,

Furthermore,

on

Seppi

(2008)

both
all

the

limit

existing
orders

the
arrival
point

Rate
nave

market

frequency
out,

and
execute

limit
those

making
of

orders

market

orders.

compete

submitted

against

is to make

future

in

with
the

market

the
As
other

future.
orders.

The

market

order

arrival

of market-making
The

market

another,

and
of

order

arrivals

with

a
trains

lift

orders

can

be

at an

average

for
the

matched

can

a given

interval

or

important

then

arrive

of

street

determinant

using

ask

with
to

orders

certain

top

like

the
of

statistical

Probability

of a top-of-the-book

be

expressed

as

a function

to

rate,

of

of time,

one

randomness

arrive

average
the

of

exponentially-distributed

market

repopulate

the

well-specified
of

the

Much

though,

assumptions

example,

independently

randomness.

cars,

modeled

assumed
rate

to

degree

the

times,
to

is an

assumed

be

Under

and

be

great

can

distributions.

bid

can

subway

inter-arrival

therefore,

profitability.
orders

arrival

rate,

the

of

hit

the

. The

limit

limit

order

book

limit

order

being

= 1/,

= 1/

and

t:

(2)
Parameters
data

and

can

be

1.

When

t =

increases,
the

market
an

ask
buy

explicitly

due

following

the

key

other
traders

passive

buy

or

an

moves
order

or

for

Foucault,

trader

at

rates.

determining

trader

how

place

his

the

Level

intervals,

the

best

arrived.

ask

bid
When

decreases,

Stoikov

(2008)

present

optimal

offset

of

Kandel

should

at

have

best

arrival

data.
time

size
to

the

and

tick

minimal

the

and

and

or

Avellaneda

Kadan,

whether

for

up

size

deploying

order,

recent

is assumed

the

is recorded.

model

on

(2005).

This

place

from

many

ticks

away

limit

order.

The

limit
model

passive

or

the

an

market

model

makes

the

from

passive

to

assumptions:

in

determining

calibrated
parameters

bid

up

execution

aggressive
The

to

limit

should

All

limit

suggests

aggressive
price

best

of strategy

is

be

calibrate

moves

analytical

orders

can

to

the

order

example

Another

All

used

a new

best

parameters
the

market

execution,

and

factor
aggressive

have
are
vice

in whether
order,

been

free

to

selected.
switch

versa.
or
is the

not

a trader

so-called

decides
reservation

to

place
spread,

defined

as

follows:

(3)
where
is the

market-maker's

incorporate

expectations

is the

minimal

is the

arrival

1/
arrivals.

rate

buy

computed
time.

In Foucault

al.

of the

et

short-run

equilibrium,

2.

with

market

transaction
3.

the

market

current
the

sequential
of

of

orders

orders,

order

aggressiveness

market

sell

recorded

both

reservation

all traders

different

different

per

is

market

spread
achieve

three

orders,

per
and

unit
limit,

of
are

j g R to
their

different

establish

optimality.

types

To

do

of markets:

traders.

types

types

can

of

of

spread,

the

of

arrival

the
and

trader

further

model

cost

of

is

avoid

placing

rate

spread

the

the

spread

trader

of nonexecution,

crossing

two

size.
the

behind

spread

the

intuition

reservation

market

with

unit

traders,

facing

diverging

traders,

with

heterogeneous

costs.

main

costs

arrival

sell

all

of time;

per

levels

(homogenous)

two

with

transaction

spread,

of

consider

identical

orders
between

market

may

costs.

market

The

use

(2005)

with

rate

(2005),

where

et al.

1. A market

the

that

equities.

market
time

same

(2005)

for

to determine

of

et al.

to be

Foucault

is

of execution

impact.

$0.01

is used

a number

Foucault

so,

say

average

orders,

dollar-cost
market

of matching

model

as

assumed

size,

the

If the

limit

about

tick

representing

of

expected

additional
order.

opposing

places

follows:

the

given

than

the

trader's

costs

associated

When
orders

an

improving

as

executing

smaller

a market
of

works

are

aggressive
his

the

order

trader's
slice.

If

reservation
with

the

risk

trader's

execution

smaller

than

limit

cost

The

structure

order,
by

the

shrinking
avoiding

spread.

In a market

with

transaction

costs,

identical
and

traders,
face

all traders
common

[1 . . . q ] have

arrival

rates

for

the

same

opposing

orders,
=

resulting

j R . In

ticks

such

away

greater

from

the

market

from

and

market

Similarly,

spread

potentially

the

market

a market

with

[1

dynamics

the

i th

trader

orders.

All

market

spread

s,

reservation

following

similar

with

j 1 R and

from

the

spread.

hand,

2 places

ticks

away

market

limit

than

order

the

spread.

j1 R < j2 R < . . . < jq

market

j i R , all

when

In this

places

traders,
the

j1 R

market

the

j 2 R . Yet,

order

ever

j1 R .

from

S < j 2 R , trader

the

away

than

away

than

limit

reservation

other

ticks

ticks

spread
traders

spread
place

s is
[ i , ,

q]

smaller

limit

smaller
place

than

orders

the

at

their

in

transaction

costs

of

traders

may

be

due

to

the

factors:
Large

smaller

traders.

Market

impact:

institutional

Large

market

impact

Taxes:

Individual

The

j2 R

s is larger

spread

the

costs,

larger

trader

size.

s is

shrinking

is

traders

spread

with

market

places

j R

spread.

differences

Other

on

outcome,

transaction

order

order

Such

j1 R

whenever

traders

otherwise.

= jq R

spread

order

away

reservation

order

tick

q ] heterogeneous

hold:

market

rates

market

a limit

no

or farther

limit

1 places

orders,

inside

different

s exceeds

price,
sized

minimum

spread

market

identically

Fees:

market

submit

competitive

limit

trader

with

than

places

while

the

R , similar

Trader

the

market

facing

order,

from

behind

traders

inside

market

to the

j 1 R = j 2 R = ...

would
the

a highly

the

whenever

market

whenever

trader

market

trader

submit

place.

spreads:

whenever

by

two

take

the

The

market

spreads

with

dynamics

In

each

jR ,

In a market

market,

is characterized

shrinking

the

reservation

than

market

in equal

on

their

than

Foucault

child

orders
order

investors

may

al.

likely

are

to

face

lower

expected

to

fees

bear

than

higher

slices.
be

subject

to

high

marginal

tax

gains.

transaction
et

are

smaller

trading

quantifiable

traders

(2005)

costs.
model

takes

into

account

only

the

aggregate

transaction

among

various

Key

types

implications

include

the

Higher
as

following

overall

trading.

Lower

trading

rebates

arrival
lower

are

et

rates

al.

for

also

facilitate

lead

market

good

costs

help

and

does

not

distinguish

(2005)

model,

observations:

result,

orders

a trader,

Foucault

order

market

facing

of costs.

from

market

and

cost

to lower

spreads.
the

lead

As

markets,

to

execution,

and

market

competing
the

spreads,
not

spreads,

result,

lowering

lower

cheaper

reservation

costs

of

suggesting

to make

that

trading

more

expensive.
High

execution

result

in

wide

creating

long-dated

from

adopted

in

traders

from

While

market

and

foundation
the

Foucault
predicting

model

deviate
For

are

always

opposing

market

with

time.

As

and

no

limit

price,

eliminating

markets,

and

al.

the

orders
consequence

and
all

limit

behind

the
orders

the

in many

execution
in the

market
today's

large

stuffing

ban
prohibits

away
to

all

markets

are

in

assumptions
thus

limit
the

limit

and
arrival

costs

price

trade

groundbreaking

and

model

from

agricultural

several

that

in

process.

presents

that

to

at

percent

in the

and

due
traders

farmers,

reality

size,

in

orders,

ability

behavior,

same

orders

10

hurting

assumes

observed

20102011

the

model

by

wide,

in

than

market

model

of

markets

quotes:

quote

traders

(2005)

from
the

of

a result,

farther

micro-trader

one,

number

States

consumption

et

for

applicability.

impractical

other

be

of stub

so-called

United

however,

destroy

quoted

The

orders

rates,

can
ban

traditionally

the

limit

wide-quoted

the

orders

and

placing

processors,

low

market.

Europe

prevailing

naturally

of

the

may

the

resulting

have

arrival

conditions

following

the

futures

deviations

the

and

and

Such

markets

demand

order

spreads,

voids.

futures

low

low

reservation

no-trade

long-dated
the

costs

are

shrink
ever
where

the

market
of

the

invariant
spread,

placed,
the

its

an

spread

has

already

contracted

Trend-Dependent
An

improvement

to

reducing

market

conditions

however,

side

may

frameworks

for

market

maker

needs

the
To

demand

as

of

Intuitively,

orders

may

Many

measures

of

the

liquidity
are

The

tightness

Market

depth

Shape

of the

like

better

a
such

markets,
orders

on

exit

the

or

is trending,
the

the

event-based

intelligent

are

forecasting

is

high,

have

metrics

bid

By

to block

Price

sensitivity

to order-flow

Price

change

and

over
supply

1968).

50
and
This

particularly

reverse

aggressive

been
are:

best

ask.

book.

unit

in

Demsetz,

profitability

spread.

and

sensitivity

per

(see

low.

defined

useful,

and

particularly
logic,

whenever

the
high,
the

liquidity-consuming

profitability.

bid-ask

at best

formally
immediacy

strategy

liquidity

of the

order

was

liquidity.

particularly

popular

support

market

predictable

the

Price

Technical

the

within

of limit

whether

liquidity

levels

of

most

size

exposure,

and

In

trending

the

markets.

making

hedge

making

adding
and

deliver

Among

the

analysis,

comprises

service

levels

In

involve

down

and

making.

exchanges

premium
the

and

selection

Chapter

market

security

immediacy

whenever

up

may

a Service

predictable

liquidity

determine

in

models

to reduce

markets,

provision

on

associated

allowed.

11.

objective

ago

tick

mean-reverting

directional

in Chapter

key

minimum

bounces

perfect

of

as

price

adverse

deploy

Making

and

of

presented

discussed

years

the
risk

altogether.

systems

of the

trending

the

a market

wrong

The

market-making

market,

range,

Market

basic

between

mean-reverting

markets

size

Offset

differentiation

the

to the

transactions.
imbalance.

volume.
resistance

levels.

developed

over

the

years.

Market
The

resilience.

Tightness

The

of the

tightness

competition

of

competing

to post

post

limit

resulting
for

In

most
Depth

many

value
hours.

the
at

the

relative
the

best

can

be

be

at the

available,

of as

book,

Order

Book
Level

Using

as

cumulative

of

sizes

limit

the

order

top-of-the-book
Such

makers

seeking

selected

price

indicate

presence
when

book

to
points.

large

of

investor

the

traders

will

be

test

the

suitable

wide

spreads,

its

minimum

the

be

and
orders

most

equities,

was

the

its

using
at

the

spread
In

such

potency,

and

market

best

bid

depth
and

of traders

that

size

smaller.

used

to

were

calculate

available
demand
posted

market

written.

loses

tick

demand

helps

on

the

be

the

ask

desire

a given

can
on

separate

to

exact
trading

determined
sell

and

immediately

deeper-in-the-book

may

be

fine-tune

large

will

buy

respectively

from

High

orders

supply

also

separation

of limit

can

book,

liquidity

liquidity.

that,

order

are

reaches

U.S.

sizes

limit

traders

with

through

The

data

of

more

tighter

approximated

ask.

and

of

liquidity

spread

minimum

II data,

easy

book

if the

supply

Level

be

sizes

II

this
bid-ask

can
best

of the

venue.

Shape

of the

the

of aggregate

of the

traded

the

When

sides

commonly
time

the

often

in

tightness

thought

spread
value

the

degree

Ask

minimum

at

and

Best

the

closer

markets

the

bid

of the

the

at

the

is an
in

bid-ask

levels

top

and

the

liquidity

Shape

amount

options.

The

price,

metric

equity

$0.01

conditions,

This

Bid

orders,

market

particularly

example,

often

the

reflects

traders.

limit

and

at Best

stays

For

execute
to

securities,

and

was

U.S.

spread

orderplacing

spread.

markets,

as

Market

limit
and

Spread

bid-ask

orders

bid-ask

most

such

the

among

their

Bid-Ask

their
liquidity

interesting

algorithms

investors.
is

behind-the-market

particularly

observed

present,

to
deeper

Academic
all

available

human

post
in the

research
and

to

market

orders
book

at
may

suggests
automated

market

makers

choose

to

limit

orders

on

Schwartz,

and

aggressive
Handa,
order

book

is

makers

on

market

power

and

Limit

trading

order

skipped

others.

may

for
the

investor

of

the

example,

presence

side

better

also

of

placing

order

find

that

large

book

from

by

limit

of

the

prices

contain

when

documented
The

to

opposing

obtain

specific
side

(2003),
due

the

that

opposite

Tiwari

investors

empirically

to

book.

when

the

investor,

exert

greater

investors

on

the

that

are

side.

books
by

the

imbalanced

market

populous

cater

of

ranges

submitting

by

impact

holes,

Biais,

holes

on

limit

of prices

orders.

Hillion,

and

Spatt

trading

has

not

Holes

were

(1995),

among

been

thoroughly

of the

order

studied.
Price

Sensitivity

When

the

can

Level

still

analysis,
Due

estimated

like

support

many
limit

parsing

pointed

out

by

linear

in

minute
the

Kavajecz

trend

one

of

case

of
the

minute
minute

as

well

prior

trading

algos

and

resistance

Odders-White

for

choose
levels.

liquidity
This

to

peaks

fact

was

first

(2004).

resistance

levels

minima

in

To

determine

would

compute

minute,

averages.

as

pinpoint

book

technical

moving

II information.

price

algorithm

and

from

methodology,

helps

and

hail

resistance

resistance

resistance.

the

and

support

recent

and

that

market

and

shape

levels

Level

support

ahead,

past

traders

and

projections

maxima

support

the

the

techniques

of the

cumbersome

Mathematically,

available,

resistance

at

support

without

not

and

human

orders

Computing

are

using

popularity

example,

Transactions

II data

be

to the

place

to Block

and

are

determined

as

support,

and

case

of

the

support

minimum
then

project

level
prices
the

one
within

minima

ahead:

(4)
Similarly,

resistance

level

can

be

computed

as:

(5)
In

addition

to

support

and

resistance

levels,

indicators

based

on

moving

averages

a short-run
the

moving

buy-side

market

help

identify

the

average

liquidity

skewness

rises

pool

above

in the

limit

of the

order

a long-run

order

book

book.

moving
moves

When

average,

closer

to

the

price.

Kavajecz

and

remains

Odders-White

popular

order

books,

help

traders

deploy

in

like

(2004)

dark

markets

foreign

liquidity

to Order-Flow

Price

sensitivity

to block

large

market

market

induced

net

by

Kyle's

eat

the

lambda

after

idea

through

the

Price

sensitivity

trading

size.

Kyle

centralized

technical

techniques

limit

measures

main

directional
trade

without

order

books

and

how

the

market

price

low

liquidity

Imbalance

The

substantially.

the

the

analysis

strategies.

transactions

order.

order

price

between

an

technical

markets

because

provision

Sensitivity

following

that

unobservable

Price

moves

and

exchange,

reverse-engineer

profitable

speculate

The

is that

limit

order

book,

measures

volume

and

sensitivity

moving

the
the

makes

relationship

price

parameter

the

is

change
known

as

(1985):

(6)
where

P t is the

to the

market

sizes

recorded

Parameters

impact

specification
individual
is

the

(7)

and

of

price

the

Change

modified

less

bid

t are

then

(6):

to
the

liquidity

available

per

Volume

Kyle's

lambda

time

t is the

best

error

net
price
on

proposed

term

the

trading

by

time

with

regression,

The

each
and

measuring

directional

t .

regression

associated

imbalance.
large

at

linear

interest,

t due

between

quotes

of the

of

to

difference

of the

intercept

volume

t 1 to time

ask

components

parameter

changes

Unit

and

is the

the

from

NVOL

t is the

P t,

sensitive

price

and

best

lambda,

the

higher

follows:

the

of equation

Kyle's

lambda,

Price

of orders

at
,

in market

observation

sensitivity

the

change

lower

trades,

the
the
and

venue.

Aldridge

(2012d)

works

as

where

P t is a change

predefined

time

executed

at

the

best

during

best

during

by

market

a result

Technical

t:

Absolute

price

is

the

was

in force

at

trade

time.

as

the

sell

net

volume

while

buy

orders.

per

volume
each

volume

of

trades

The

difference
flow

was

matched

between

bid

trades

trades

trade

imbalance
at

the

during
of

time

order

orders,

observed

at the

executed

Resistance

change

trades

aggregate

S t a is the

the

and

executed

t , and

of market

Support

on

S tb

that

interpreted
period

generated
often

bid

prevailing

be

time

t,

period

ask

S t a can

are

period

at the

processed

in prices

is

S t b

observed
likely

to

processed

be

at

ask

Levels

unit

volume

traded

is yet

another

measure

of liquidity:
(8)
This

metric

ratio.

is

the

deeper

is

requires

average
the

(8)

and

accounts

volume

traded.

Market

Resilience

Market

resilience

how

quickly

order.
this

lambda,

available

equation

times,

Kyle's

price

of

of

(2002)

in price

attribution

metric

Amihud

to
change

smaller

the

the

order

per

the

another
book

market

that

unit

recovers

the

illiquidity

change

shape

is discussed

The

volume,
Kyle's

and

liquidity

estimates

instrument.

trading

of liquidity
its

resilience

of

bid

the

ratio

modified

absolute

metric

as

financial

prevailing

assumes

known

illiquidity

While

to

for

is

traded

change

trades

is yet

of

of

and

liquidity.

only

Estimation

ask
is

the
lambda

quotes,

the

balanced
in

and

at

price

per

unit

measures

following

all

just
market

in Chapter

15

of

book.

Profitable

Market

ensure

that

conditions

must

relate

arrival

and

to

Similarly

size-induced

To

due

the
prices

Making
the

market-making

hold.

of traded

The

rates

of

operation

conditions,
limit

instruments.

orders,

dating

is profitable,
back

probability

several

to Garman
of

gain

key

(1976),
and

loss,

Temporary

imbalances

differences

between

optimizes

his

budgets,

risk

appetite,

are

these

less

is to

latter

cash.

Both

which

may

access

to

than

buy

and

and

orders

due

way

host

order

other

imbalances

market

maker's

bankruptcy

maker

has

as

independent

arrive

in

problems

The

avoiding

dealer

of

optimization

create.

to

differences
a

aggregated

market

are

the

markets,

while

the

sell

and

underlying

help

profits

orders

reflect

the

differences

whenever

sell

trader

important

maximize

arise

and
trader

individual

optimization

objective

buy

individual

flow,

The

themselves

The

the

order

idiosyncrasies.

that

between

or

no

failure.

inventory

or

stochastic

processes.
The

model

solution

estimation

of

clip

million

of 10

when

and

a is

denoted

sell

securities

as

is

the
to

the

market

cash

when

the
the

at

(pays

money

a customer
the
the

or
the

bid

solution

to

a classical

maker
to

the
in

of

an

ask

price

departure

of

a customer

at

or

probability

of

the

comes

market

maker

in

dollar

market

probability

customer,

to the

Suppose

a security

lies

(e.g.,

securities

maker

buy

to

of

prices

arrives

customer).

maker

based

Gambler's

Gambler's
certain

Ruin
initial

all

known

his

as
the

certain
Under
will

buy

market

ask

price

P b,

order

to

can

be

b.

the

that

market
to

solution

loses

to

and

unit

exchange)
in

order

a . Correspondingly,

the

denoted
The

pays

bid

which

the

a customer

from

at

comes

dealer

clip

optimal

in foreign

departs

(the

arrival,

rates

customer

dealer)
to sell

the

for

lose

the

of wealth,
Gambler's

all his

money

In

the

a gambler,

position
This

unbounded
bets

the

Problem.

Problem,

money.

gambler
level

Ruin

wealth

an

on

and

wagers

version

of

problem.
until

he

at which
Ruin
is

or

either
point

Problem,

the
The
loses
he
the

dealer's

a dealer,
(stays

version
starts

out

in business)

Gambler's
bounded
all

problem

his

Ruin

of

the

with
until

Problem

problem
money

known

that

he
is

assumes
or reaches

exits.
probability

the

gambler

(9)

where

Initial

probability
Gain

is

losing

an

of

) is the

From

Wealth

the

probability

Gambler's

failure

is

certain

whenever

gaining.

In

the

other

( Loss

of gaining

an

It

can

) is

the

) of

initial

wealth,

and

Pr(

the

( Gain

can

see

losing

in the

exceeds

term

Ruin

following

two

if he

runs

out

of cash.

2. The

market

maker

fails

if he

runs

out

of inventory

satisfy

client

the

Loss

variables

other

words,

Gain

= 1

Loss

= 1

maker's
is

the

of

are

equity,

perspective,
probability
a of
one
The

Ruin

of

we

units

this

of

unit
unit

the

probability

selling

of

Gambler's

arriving.
inventory
Ruin

be

may

for

the

assume

the

may

the

a buyer
unit

Problem

inventory,

single

exchange,

gaining

arriving.

out

of

probability
of

Gambler's

running

foreign

for

of

positive

)> Pr( Loss

).

to

the

and

is unable

to

demand.

In modeling

of

is

ways:

fails

case

failure

Problem

maker

the

of

probability

is Pr( Gain

market

In

that

condition

long

probability

the

1. The

through

the

shown

Gambler's

in the

).

that

further

minimum

the

business

Pr( Loss

be
of

the

applies

cash,

we

probability

failure

start-up

amount

Problem,

words,

(1976)

market-making

amount

positive.

of avoiding

Garman

gambler's

Ruin

always

probability

the

be

unit

that

a share
a clip.

inventory,

is

Problem

the
b,

equation

and
logic,
probability
(9)

now

of
it

and

asset.

the

the

unit

ruin

Gain

In

from

one

same
the

stock.

Then,

of

the

financial

of

losing

maker's

both

underlying

of

By

market

In

case
market

inventory

equals

the

the

probability
of

seller

becomes

(10)

where

E 0 ( P a , P b ) is the

initial

average

price

of an

underlying

unit

of

inventory

and
is

the

initial

number

of

units

of

the

financial

instrument

applied

to the

in

possession
of the
The

market

maker.

Gambler's

Ruin

probability

of

maker's

failure

due

perspective,

when

a buyer

willing
the

Problem

to

market

market

at

maker's

seller

of

the

Gambler's

of

arrives.

probability

security
Problem

now

the

the

market

dollarhappens

the

arrival

of a buyer

a . As

a dollar

is P a . Similarly,

giving

security

takes

From

maker's

probability

or

the

market

before,

with

losing

selling

cash.

cashsay

of gaining
of

for

of

As

P a happens

probability

Ruin

out

a unit

security

price

maker's

running

gaining

of the

buy

to

is further

away

at

price

following

result,

dollar

P b is

the
to

b.

The

shape:

(11)

For

market

equations

(10)

words,

the

maker
and

to
(11)

remain
need

following

in

business,

to be

the

satisfied

two

first

conditions

simultaneously.

inequalities

In

have

of
other

to

hold

contemporaneously:
b >

a and

a pa >
For

both

true

at

spread

apb
inequalities
all

times:

allows

sufficient
true:

the

market

to hold

same

p a > p b , defining
the

market

inventory
existence
maker

at the

of the

satisfies

the

maker

positions.

The

bid-ask

profitability

time,

to

the

bid-ask
earn

reverse,
spread

following

spread.

cash

conditions

not

The

while

however,
does

must

be

bid-ask

maintaining
does

not

guarantee

of equation

that

hold
the

(11).

Summary
Market
and

making
deserves

feasible

with

and

any

compensation.
very

simple

liquidity

provision
Profitable

models

that

is a service
automated

only

take

to the
market

into

account

markets
making
inventory

is

on

the

market

maker's

therefore,

desirable,

extracted

out

of

books.
the

Level

While

information
I data

Level

II data

delivered

using

simple

by

is informative,
Level

techniques

II data
like

and,
can
technical

analysis.
End-of-Chapter
1.

What

Questions
is

market

making?

How

does

it

differ

from

liquidity

the

common

provision?
2. Why
3.

What

is market
is the

making
core

inherently

market-making

profitable?
strategy?

What

are

extensions?
4. What
5.

What

strategies?

are
are

the
the

common
minimum

measures
profitability

of liquidity?
conditions

of

market-making

be

CHAPTER

11

Automated

Market

Making

Inventory

market-making

and

manage

help

imbalances
with

immediate

spot

to

models

do

contrast,

this

chapter

demand

in

in

Chapter

prices

or

portfolio.

The

an

inventory

level

and

consider

proceed

and

actions

10,

resulting

adjust

start

to

the

satisfy

inventory

conditions.

information

the

usually

sufficient

market's

explain
from

models

to

the
and

information-based
carefully

Such

available

in

expected

future

as

What's

the

well

This

to

other

market

and

time,

and

between

the

show

timing

cases

The

some

may

Case

1: Market

optimize

supply

and

present

and

order

book,

order

outcomes.

that

manage
cases

for

sample

book

apply

to

for

horizontal

its

market
to

all

matching

axis

is time.

The
observed

dashed

line

shows

the

midquote:

trades:

occur

in the
Does

some

best
trades

vicinity
Not

Move

ask.

The

occur

of either

I data

figures

observable
from

venues

solid

offer,

the

The

that

given

deploy

operation.

best

and

detect

accordingly.

situations

Level

trading

to

positions

illustration.

a.k.a.

bid

is sufficient

one's

ask,

best
of

of limit

trading

subscribing

cases

order

best

the

shape

for

in

market

and

available

account

and

and
four

the

the

of various
them,

to

information

presents

limit

the

to

the

carries

participants

figures,

available

of orders

movements

venue.

all

patterns

models

including

data

price

centralized

trading

addressed

Data?

accompanying

trading

models

responding
The

histories

section

all

news
by

trading,

available

short-term

the

markets.

as

in the

Publicly

the

process

demand

flow,

market-making

evaluate

deduce

market-making

and

book,

demand,

not

participants,

bid

variations

inventory,

client

discussed

participants.

By

In

transitory

client

according

by

models,

in a dealer's

a sweet

market

II

at
bid

lines
at the
the

are
given

simple

stars

indicate

ask,

some

or ask.

the
point

in

average
the

at

best

bid,

price
and

In

Figure

the

first

finally

at

11.1

, the

trade

moved.

the

best

11.1

bid

is the

The

recorded

in

the

best

then

at

best

bid

again.

dynamic

11.1
that

arrived

trade

was

recorded

market

sell

time

move.

to

question

The

bid,

Neither

the

at

in the

work

the

following
the

sequence:

best

nor

market

ask,

and

best

ask

the

shown

at

order

about

the

first

are

executed

market

sell

order.

bid

side

of

and,

sell

in size

carry

best

any
the

in

Figure

their

risk.

buy

therefore,

orders

than

the

size

the

trade,

bid

order
must

the

market

is

be

likely

shows

that

The
a

trader

11.1

orders

sell.

would

next

trade

Once

bid

buy
such
be

two

not

from

the

quotes
been
or

the

revised

additional

order

and

because

it

matched

with

is recorded
again,

at

did

ask

the
have

as

not

The

participants

would

market

bid

best

they

identified

buy
book.

of

be.

best

best
least

first

sellthe

at the

at

in

The

would

the

and

shown

a market

sell

afterward:
order

market

happening.

was

other

Figure

askmarket
limit

is

knowledge

market,

to Trades

the

information,

Had

the

order,

In

trade

best

levels.

shortly
The

of the

bid,

specialized

to minimize

orders

ask

market

not

movement

at the

unusual

arrivedfollowing

their

the

or

as

original

Following

none.

the

at

quotes

is

the

smaller

did

in Response

in

perspective:

imminent

Move

resulted

market

concerned

Not

interesting

was

order

remained

recorded

this

and

order

the

broader

their

Does

, nothing

order

the

Market

answer

Figure

best

at

are

short

the

trades

occurred

What

Figure

11.1

Figure

is

at the
11.1

shows

no

The

size

sizes

changes
of

both

at the

The

in prices
orders

best

orders

following

bid

were

was

and

not

either

small

the

best

perceived

order,

implying

in

comparison

have

any

that

with

aggregate

ask.
to

special

informational

advantages.
Case

2: Market

Figure
bid.

11.2

Moves

presents

Following

over

and

Rebounds

a different

the

bid,

the

scenario:

quotes

drop

a trade
lower,

is recorded

yet

at

gradually

the

recovers

time.

Figure

11.2

The

Market

Moves

and

Shortly

Rebounds

Following

Trade

The

trade

in

Figure

triggered

by

that

followed

1.

was
The

bid;

a market

trade

the

11.2
sell

by

size

occurred
order.

gradual

was

at
The

the

relative

market

sell

order

had

Despite

its

large

size,

the

drive

the

and

subsequent

recovery

large

bid,

drop

suggests
to

the

to sweep

was,

therefore,

in the

quotes

that:

size

available

through

the

at

book

the

best

to be

fully

matched.
2.
that
levels;

the

In Figure
may

due

inventory
quote,

trade
11.2

be

just
a

pricethe
carried

, the
to

best

the

market

quotes
no

maker

via

did

not

slowly

move

the

recovered

fundamentals

to

their

original

information.

ask

market

acquired

trade

quote
makers'
the

can

sell

also

drops

activity
trade.

promptly

following
seeking

By
sell

the
to

lowering
the

newly

sell.

This

disgorge

the

the

ask

best

purchased

inventory,
Case
The

realizing

3: A Trade
trade

above,

in

the

Figure

sell.

containing
Case

levels,

but

price

level

the

in

placed

came

in at the

bid

the

sell,

warrant

ask

quotes

their

Figure
and

not
new

recover
lows.

as

is

hence
a

trade

price

decline.

their

original

to

The

of fundamental

11.3
was

however,

a permanent

did

result

ahead

of
may

to

protect

counterparty.
to

as

market

participants
traders

determine

ways.

sell

referred

informed

again

at

may

superior

broker-dealer

losing

Figure

is observed.

in

quotes

the

several
the

possesses

executed

and

activity

remained

makers

information

orders

in

new

persistent

information

carried

trade.

market

who

bid

market

to

the

likely

both

Case

trade,

of

information

instead

Unlike

the

of quotes

interpreted

above,

bid.

Markets

once

markets

is most

sell

Moves

the

following

recovery

trade

at

shifted

explanation

enough

Unlike

occurred

recovered

immediate

The
The

also

substantially

likely

information.

Markets

quotes

A Trade

most

profit.

11.3

the

No

11.3

The

The

Figure

trade

downward.

by

Moves

where

11.3

a quick

Often,

market

be

the
himself
Such

prehedging.
may
through

the

order

research
other

that

first

to

against
activity

the

the
the

information-extracting

the

trader

the

trader

places

quotes

perfectly

Alternatively,
determine

that

situations,

accepting
is

possessed
of

routinely

such

adjust

trade

observe

and
In

sell

broker-dealer

may

skills

traders.

the

profitable
the

after

further

trader's

the

trade

trades

legal,
broker-dealer

probabilistic
models

and

is

with

is

often

and

other

presence
described

of

later

in this

chapter.

Case

4: The

Quotes

Figure

11.4

illustrates

widen

following

Figure

11.4

The

can

The

be

to

the

time

the

news,

closest

shown

in Figure

Cases

following
Market

described

where

spreads

Information

following

have

been

the
take

autocorrelation

Order

flow

aggressiveness
order

the

markets

natural

known

of

uncertainty

traders

response
and

as

to

quoting

pull
wide,

book

specific

market
trading

market

are

maker

behavior

also

possible.

events,
systems

observable
to

extract

makers.

Flow
the

chapter

developed

behavior
into

scenarios

to other

of

flow

of the

makers'

result

announcement,

move

market

automated

in Order

sections

models

Other

solely

on

to

a practice

predictable

helps

available

based

news

likely

of such

better-informed

illustrate

data,

that

example

Market

price,

often

behavior

quote

most

scheduled
is

by

is

An

other.

over

Order

Shape

the

situations.

remaining

moves

markets.

market

11.4

distinct

techniques

a case

a Trade

11.4

released,

or

run

through

information
Modeling

way

Following

a major

once

to the

makers'

public

scenario:

Figure

in the

being

orders

The

in

preceding

one

avoid

another

Widen

depicted

considerably

in

Quotes

uncertainty

where

is

yet

a trade.

scenario

increased

Widen

account:

of

describe
to

other

identify
market

four

classes

impending
participants.

of

market
The

Sequential

evolution

Autocorrelation
Order

of Order

flow

is

information.
extract

of quotes

Flow

result

of

Information
and

then

as

a Predictor

end

models

trade

upon

of Market

customers

receiving

are

to observe

trained

information

available

Movement
and

acting

order

to

on

flow

and

various

market

trades

initiated

participants.
Order

flow

with

is the

market

noted

buy

within

orders

market

buy

volume

caused

Equation

difference
and

trades

predetermined

order

are

by

(1)

in trade

market

as

sell

the

between

triggered

period

known

illustrates

volume

of

by
time.

buyer

orders

market
Trades

initiated.

of order

orders,

begun

to as

all

with

Similarly,

is referred

definition

sell

trading

seller

initiated.

flow:

(1)
where

is the

matched

with

volume

the

volume

ask

triggered

order

by

side

50

Around

the

from

order

sell

orders

research,

percent

news

of

releases,

order

Love

and

Payne

Eurozone

and

the

U.S.

surrounding

the

directionality

of

expected

to

Euro

securities

lift

market

buy

book,

and

hitting

the

the
the

orders
is

the

bid

According

Other
Lyons

(2002a),
to

(2008)

being
trading

side

of

the

Lyons

that

studies

Thus,

with

Jones,

Kaul,

(2001),

order

and

Lipson

flow

is

prices.

directional.
following

For
a

major
flow

closely

follows

the

good

U.S.

buy

U.S.

findings
and

for

order

by

similar

Perraudin

market

the

rate

is dominated

(1995),

and

highly

announcement,

exchange

responsible

into

estimate

news

dollar,

is directly

becomes

announcement.

U.S.

orders.

flow

impounded

flow

EUR/USD

include

Lyons

order

information

example,

and

of

market

to academic

least

sell

resulting

book.

According
at

trading

news,
dollar

about

Vitale

or

various

(1996),

Evans

(1994).
informative

for

three

exposing

their

reasons:
1. Order
equity

flow
to

can

their

be
own

thought

of as

forecasts.

market

participants

Market

orders

are

irrevocable

commitments

to

information.
a

buy

Limit

result,

carry

participants'

or

sell,

orders

can

and
also

information.

honest

be

Order

beliefs

therefore

carry

executed

and

flow

about

most
be

therefore

the

upcoming

with

limited

powerful

costly

and,

reflects

as

market

direction

of

the

market.
2.

Order

can

flow

directly

networks.
can

data

partially

investors

may,

the

this

from

data

the

infer
in

receive

decentralized

observe
End

described

is

and

process

internalization

viewed

available

it before

the

3. Order

flow

the

possesses
again,

information

any
the

to

order

and

regardless

who

superior

information,

observing

capitalize

on

or
the

contain

the
due

modeling
market

as

in

the

full

is

costs

order

by

flow

order

position

is
flow

to exploit

prices.
that

will

originator
to

they

orders,

broker
the

positions

whether

as

internalization

in a unique

nontrivial

data,

internally,

The

market

but

exchange

investor

possess

into

all,

data

The

Because

it are

of

flow

orders

to

at

market

order

flow.

possible.
those

large

traders

positioned

the

interdealer

flow

from

match

is impounded

shows

market

to

model

order

of

Brokers

and

participant.

function

everyone,

or successfully

direct

numbers

seek
of

clients

possess

significant

whenever

to

any

market

a necessary

exchanges

information

move

as

their

information

other

miss

increasingly

not

see

flow

of

Exchanges

broker-dealers

avoiding

seldom

section.

however,

presently

flow

order

brokers

called

order

distribution.

of

market

the

the

trades

impact.

order

movements

temporarily

flow

Once
are

best

surrounding

the

transaction.
Lyons

(2001)

further

order

flows,

with

information,
subjective
(2001),

and

transparent
order

to

extract

flow

between
order

opaque

analysis
order

distinguishes

transparency

flows
market

Pretrade

versus

posttrade

flows

failing

information

and

providing

to produce

beliefs.

encompasses

dimensions:

transparent

immediate
useful

According
the

opaque

following

data
to

or

Lyons
three

Price

versus

Public

versus

Brokers

to

the

the

of

information.

information

the

End

trade,

by

the

customers.

Undoubtedly,

the

of

superior

efficiently.

Order

Is Directly

As

Flow

noted

and

by

the

among

financial
difference
Order

flow

When

the

is

computed
buyer-initiated
trades.
the

case

in

difference

flow

measures

are
or

attention
Jones

have

et

given

price

al.

(1994)

time

posttrade

available

flow

to
to

to

to

and

all
use

obtain

use

to

the

quantities

the

exchange),
the

as

and

in

standard

actually

that

selling

observable
can

particular

be

pressure.
can

be

size

of

seller-initiated
(as

is

measured

buyer-initiated

the

interest.

flow

of

by

as

cumulative
size

often

as

trades

the
and

interval.

the

empirical

most

orders

size,
would

found

order

the

time

since

run-ups

or

number-of-trades-based

used

of

specific

for

trading

the

flow

centrally

measured

buying

directly

order

in each

of

not

number

flow

Evans

previously

viewed

formally

cumulative

are

be

and

was

seller-initiated

between

and

(1996),
flow

can

is

difference

comparable

and

and
dealer

positioned

Vitale

Order

observable,

been

parcels

venue.

referred

trade-size-based

order

clips,

trades

price

the

order

order

are

between

seller-initiated
Both

the

foreign

have

and

resources

but

buyer-initiated

sizes

trade

or

better

and

others,

trading

any

trades

When

Perraudin

sometimes

as

public

appropriate

participants

at

trade

only

the

the

public

see

much
in

firsthand

both

both

becomes
are

among

or

between

it

the

market

security

see

embedded

(1995),

broker-dealer

observe

generally

dealers

(2002b),

dispersed

time

flow

Observable

Lyons

Lyons

can

can

given

information

interdealer
can

and

information

returns,

and

pretrade,

customers

information

wealth

customer

information

quantity

price

information

dealer

observing

access
the

quantity

that

primarily

measures
literature.
are

These

transmitted
to

accompany
order

of

avoid
larger

flow

measured

in
undue
trades.
in

number

of

trades

measured
The

predicts

in aggregate

importance

announcement

(2008),

for

into

and

the

the

Payne
pairs:

order

(2008)

is

flow
shown

causes

Table

11.1

Following

Table

Trades

at

Average

flow

is

necessarily

only

the

of

and

USD/GBP.

Payne

exchange
find

the

that

information

flow
The

by

Love

measure

by

number
Payne

Eurozone,

percent.

Currency

Number

as

seller-initiated

or 0.626

One-Minute
the

of

flow

and

from

of

Payne

the

Love

release

0.00626

impact

order

interval.

three

and

and

excess

on

of

Returns

Buyer-Initiated

Trades

Observable
transparent

from

customers,

their

order

found

news

in

brokers

and

of

in

in

executing

bid

all

impact

trade

of Seller-Initiated

Order

see

time

Increase

Directly

coming

and

foreign

of buyer-initiated

Changes

Is Not

orders

flow

following

and

of

authors

to increase

Flow

example,

Love

in

half

one-minute

the

Trade

not

least

. The

in each

USD/EUR

in Excess

flow

rates

number

Order

For

at

buyer-initiated

a Single

order

level

announcements

the

11.1

the

that

additional

order

respective

trades

document

trades

for

price

empirically.

GBP/EUR,

between

of seller-initiated

each

the

studied

the

in

than

prices.

(2008)

on

better

at a new

verified

news

USD/EUR,

difference

(2008)

been

accounts

market

volatility

in arriving

examine

directly

impounded

currency

flow

macroeconomic

flow

and

of trades.

has

example,

surrounding

Love

size

of order

a news

order

prices

offer

can

prices,

to

all

directly
but

market

participants.

observe

generally

the

buy-and-sell
customers

and,

possibly,

the

depth

sprung

up

extract

can
of

the

market.
As

result,

information

various
from

the

models
observable

have
data.

The

most

to

basic

order

algorithm

flow
tests

autocorrelation

of

trades

over

the

The

identification

recorded

buys

and

the

Lee-Ready

book.
of

sells.

or

Trades

+1,

past

volume

each
the

direction

signs.

identified

and

computes

trade

as

sell

First,
period

clock

rule

buys

are
is

variable,

can

30

be

described

minutes,

into
using

Chapter

trade

as

all

performed

in

assigned

function

separates

T , say,

trades

noted

autocorrelation

algorithm

of time

of

trades

the

4 of

direction

this

value

1.

Next,

the

algorithm

(ACF)

for

lagged

tick

in the

trade

xt :

(2)
where

t is the

evaluation
time
with

sequential

interval

interval.
a lag

Figure
financial

T , and

An

ACF

, reveals

11.5

plot,

trade

shows

of

is

a given

the

linking

total
the

trade

number

of

computed

ticks

chosen

within

the

autocorrelation

dependencies.

comparative

autocorrelation

figures

for

two

instruments.

Figure

11.5

Autocorrelation

(MSFT)

and

Source:

Sotiropoulos,

Hasbrouck
returns
time

number

Sunbeam

(BEAM)

in

by

Order
on

Flow

October

Observed
31,

for

Microsoft

2011

2012.

(1991),
caused

of

estimating

previously

placed

order

autocorrelation,

orders

as

well

adjusts
as

effects

for
of

the

of day:

(3)
where
given

x t is
trade

otherwise;

the
was
r t is

order

flow

estimated
a

one-trade

observed
originate
return;

at
from
and

time

t , set

a market
D t is

the

to
buy

+1

when

order

dummy

the

and
indicator

controlling
Ellul,

for

the

time

Holden,

Jain,

autocorrelation
order
depletion

firm

(2004);

Foucault,

Kadan,

and

(1995).

Spatt

Order

and

flow

price

of

large

buy

large

number

order

placed

positive

traded

of

the

(2007)

market

higher;

sell

will

depress

orders
the

sell

order

will

a buy

find

studies

of

Rosenqvist

and

(2004);

Biais,

Hillion,

profitably.
inevitably

order

at

the

the

time

Similarly,

and

observed

push

gains.

prices,
is

but

Sandas

in positive

flow

strong

trade

placing
result

to

market

believe

to the
as

the

The

results

that

market

Viswanathan
situation

different

timely

will

sell

generate

announcement

the

the

opposed

available
price

of orders

to

the

limit

more
price

of

the

publicly

available

aggressiveness

percentage

orders,

best

the

Movement

the

security

is

are

The

aggressive
and

of

that

prices.

data,

is the
more

likely

about

those

trades.
submitted

higher
trader

the
in his

the

trader

to

move

away

of

Foster

price.

of

Vega

(1996),
where
degree.

from

monitoring

refers

the

of Market

information

of market

from

a Predictor

proposes

capture

to

as

prices,

percentage

orders

confirm

and

(2005);

of buy

liquidity

Other

and

to

will

when

Aggressiveness

is

Miller,

is observed

Aggressiveness

Vega

bid

security

volume

extract

at

Niemeyer,

easy

within

frequencies,

Hedvall,

Rosu

competing

results.

Order
To

the

high

frequencies.

number

of

(2007)

at lower

is

large

at

short-term

waves

events

al.

flow

(2005);

information

disproportionately

et

Hollifield,

Kandel

as

market

order

include

Ranaldo

interpret

flows

current

correlation

t falls.

(2007)

Ellul
in

autocorrelation

time

order

to

correlation

order

(1997);

which

Jennings

replenishment.

serial

order

and

responding
and

negative

into

in high-frequency

flows

positive

of day

(2007)
who

evaluate

different
For

is made,

are

market

example,
it is common

based
the

on

average

response

participants
before
to

are

an
see

of prices
informed

expected
a

and

consensus

in
to

economic
forecast

that

is developed

The

consensus

forecasts
under

2009

2.2

while

all

percent

12,

Foster

and

degree

of

speed

with

profits

of

ability

the

scheduled

news
prices

through

however,

comes

nonexecution.

better

a result,

about

the

executes

participants

trader's

faster

the

is

subject

to adverse

impending

trade

the

also

from

each

of

the

instead

of

waiting

shows

more

aggressively.

the

the

the

In

other

faster

the

following

information
result
price

in

risk

of

executed
orders

rapidly

preserve

more

advantage,

are

Market

more

the

impacts

associated

moving

the

in
affects

hand,

and

are
quick

trading

gains.

aggressive

his

prices.
convey

security

about

(2007)

the

themselves

direction

Vega

a.m.

other.

the

and

enters

information

at 8:30

determines

orders

are

and

may

from

securities

pricing.

capture

0.8

ranged

prices,

range,

other

prices

information

orders

convey

going.

when

information

aggressive

may

on

to

and

and

orders,

be

correlation

into

orders;

wait

the

United

to

participants

Limit

market

costthe

of

announcement

trading.

achieved

immediately

trader

market

be

the

the

be

the

of

actual

than

trading,

must

trading,

can

in aggressive

execution

the

at

but

The

analysts

the

number

of

percent).

that

prices

range

all

in

revealed

forecast

market

prices

the

+1.0

learn

by

sales

impounded

to

consensus

market

is

active

Market

immediately

various

analysts.

announcement

number

analysts'

fair

retail

information,

release.

execution

The

the

in

show

participants

at

favorable

used

of

possessing

arrives

enters

(1996)

narrower

the

for

to be

information

market

to

analysts'

actual

happened

informativeness

forecasts

estimates
(the

Viswanathan

the

market

analysts'

2009,

which

the

by

of

prior
change

percent

traders

of

words,

the

to 0.3

February

example,

reported

market

accompanied

dispersion

month-to-month
LP

of several

typically

the

For

Bloomberg

percent,

forecasts

is

measures

consideration.

States,

As

averaging

number

that

January

on

by

price

move.

for

a more

his

beliefs

that

If a

trader

favorable
about

better-informed

Mimicking

information

aggressive

price,

where

the

market
trades,

therefore,

may

Measures

of

informed

result

in

consistently

aggressiveness

traders'

of

information

profitable

the

and

order

flow

facilitate

trading
may

strategy.

further

generation

capture

of

short-term

profits.
Anand,

Chakravarty,

Stock

Exchange

limit

orders

price

perform

orders,

Anand

details

of

larger

November

et al.

times

al.

of
as

t and

limit

(2005)

use

various

January

1991.

Diff t , the

characteristics

difference

between

the

bid-ask

To

evaluate

the

NYSE,

the

the

execution
spanning

equation

the

than

inside

the

on

better

market

and

regression

York

than

orders
on

New

better

orders.

trail

following

the

perform

placed

all

audit

on

or

smaller

trading

order

at

orders

sampled

the

that

orders

orders

outperform

through

(2005)

impact

measured

than

find

limit

individuals,

three-month

1990

Anand

by

orders

et

(2005)

institutional

better

and

Martell

(NYSE),

placed

spread,

the

and

to estimate

price

bid-ask

changes

midpoints

at

t + n:

(4)
where

t is the

time

after

order

minutes
particular
the

order

Institutional
institutional
stock-specific

is
or

submission,

by

the

dummy
the

and

dummies

is the
daily

the

orders.

with

the

if

in

the

traded

in

orders,

the

zero

order

is

otherwise.

the

value

D 1

to

particular

60

buy

and

takes

individual

associated

For

value

then

shares

of shares

quote
that

5 and
of

period.

standing

for

number
volume

takes

variable
0

n equals

sample

that

dummy

orders

mean
the

than

Size

over

better
is

order

submission.

stock

Aggressiveness
at

the

divided

particular

placed

of

for

D n -1

are

stock

that

the

results

was

traded.
Table

11.2

robustness
institutional
for

stock

dependent

, from

Anand

et

regressions
and

testing

individual

selection
variable

(2005),
for

orders.

by
in

al.

the

institutional
regression

summarizes

a difference
The

in the

regression
and
is

individual
the

change

of

performance

equation

controls

traders.
in

of

the

The
bid-ask

midpoint

5 and

Table

11.2

then

minutes

Summary

Difference

in the

According

to

of

after

several

can

of Institutional

be

to

to

orders

that

Testing
Individual

future

from

Orders

exhibits

of

either

realizations

autocorrelation

originate

for

aggressiveness

forecast

The

thought

and

market

used

aggressiveness.
is

submission.

Regressions

researchers,

that

aggressiveness

order

Robustness

Performance

autocorrelation
market

60

of

of
market

the

following

sources:
Large
an

institutional
extended

period

of

are

time

transmitted
at

in smaller

comparable

slices

levels

of

over
market

aggressiveness
Simple

price

Research
was
on

momentum

into

detecting

performed
the

least

Paris

the

Biais

Bourse

aggressive

aggressive
that

by

by

orders

distribution

state

orders

other

orders

empirical

of

of

certain
the

See,

(2000)

for

Swiss

Stock

the

for

of
that

the

that

detected

same

of

level

of
the

example,

Griffiths,

Stock

Exchange;

Cao,

separated

orders

prices

book.

the

to

The

the

diagonal

effect
are

aggressiveness.
for

Smith,

and

Wang

on
are

whereby
followed

by

Subsequent
different

Turnbull,

Ranaldo

found

depends

submissions

findings

most

authors

order

Exchange;
Hansch,

observed

aggressivenessfrom

aggressiveness

confirmed

Toronto

aggressiveness

of aggressiveness

and

level

market

move

current

in terms

authors

research

exchanges.

orders

market

The

who

degree

outside

the

of

(1995),

the

of orders

of

autocorrelated.
initial

et al.

market

limit

the

autocorrelation

stock

and

White

(2004)

for

the

(2004)

for

the

Australian
Stock

Stock

Exchange;

Exchange

CAC40

of

stocks

Shape
Several

Order

studies

used

to

predict

example,

tends

However,

a liquidity

market

Whitcomb

that

11.6

to Push

Rosu

(2005)

the

the

of
Foucault,

depth

of

limit

around
limit

the

lower

Caglio

far

et

(2003)

for

the

al.

to

order
et

the

market

book
al.

phenomenon

for
peak.

to

pull

Schwartz,

and

gravitational

evolution

for

the

tends

Maier,

be

price,

from

price

can

(2004),

market

away

Cohen,

sample

Direction

Cao

price
the

peak.

Liquidity
Price

the

such
away

pull

of

of the

market

depth

the

Market

Price

the

market

and

book

depth,

can
the

find

that

as

earnings
the

the

shape

Peak

the

lead

to

Theissen

the

expected
carry

announcements:
current

market

order

market

book
orders.

hump-shaped

(2005)
future

orders

limit

arriving

forecast

lower

limit

of
for

orders

Moinas,

Near

from

distribution

large

from

Peak

Away

that

order

(2004)

events
orders

from

probability

books.

and

away

Market

probabilities

the

peak

Asymmetric

order

prices:

market

the

the

limit

close

the

determines

on

the

moves.

peak

this

for

of Market

how

the

(2001)

liquidity.

An

Tends

a Predictor

price

shows

Handa

Chan

Bourse.

push

call

11.6

Paris

liquidity

(1981),

associated

High

and

and

considered

toward

and

depends

as

price

Figure

Figure

Book

to

quotes.
the

the

short-term

find

Bae,

Kong;

on

have

example,

the

Hong

traded

of the

Ahn,

find

volatility

limit
that
of

volatility.
private

the
asset

Berber
information

a concentration
price

is likely

to

of
reflect

someone's
Cont,

postannouncement

Kukanov,

can

be

price

and

used

to

a new

Stoikov

(2011)

generate

movements.

define

valuation

To

of the

traded

suggest

that

instrument.

even

Level

successful

predictions

of

predict

the

movement,

Cont

order

flow

variable,

price

imbalance

the

data

impending
et al.

(2011)

(OFI):

(5)
where

e n represents

liquidity,

and

an

is defined

instantaneous
as

change

in the

top-of-the-book

follows:

(6)
where

is

condition
the

best

the

indicator

is

true,

bid

and

Equations

(5)

Imbalance

and
the

and

change
increased,

best
the

new

best

next,

the

size

To

ascertain

be

on

the

OFI

Cont,

bid

at the

predictive
figures

Order-Flow
Kukanov,

to

as

by

the

at

the

new

best

the

present

shown

tick,

best

the

sizes

at

the

the
by
from
bid

ask.

the

the

tick-to-tick

best

bid

price

the

size

at

the

one

tick

to

the

size

recorded
the

If

OFI

Flow
in

decreases,

the

OFI,

in Figure

Stoikov

Order

the

at

OFI

ask

is

price

is increased

by

ask.

short-term

Imbalance

If

best

price

of

on

decreases

ask

-vis

and

bracketed

change

prices.

the

power

follows:

increases

price

previous

vis-

as

depends

Imbalance

if

tick

turn

offer

Similarly,

relationship,
11.7

in

best

size

the

q A are

instantaneous

is reduced

the

when

q B and

interpreted

which

best

respectively.

can

Flow

to

and

OFI

last

recorded

the

ask,

and

the

tick.

from

the

Source:

If

by

increases

Figure

bid

associated

decremented

a linear

(6)

Order

bid.

previous

map

best

liquidity,

in

equal

0 otherwise,

depends

top-of-the-book

the

function,

Cont
price

11.7

versus
(2011)

et

al.

(2011)

changes,

and

next
obtain

Short-Term

Price

Changes

Evolution
The

of Tick

advanced

intent

and

include
trade

class

future

as

flows

to

actions

informed

trading

their

and

core,

also

use

observed

models

informational

markets.

addresses

participants.

the

Such

models

reverse-engineer

quote

a market

possesses.

maker

or inferred

markets,
exchange

describe

order

Differences

flow

to

and

make

and

is

such

slower

in

on

information

flow

arising

during

the

in information

flow

markets,

and

trading

asymmetries

Information

centralized

electronic

information

of information.

transparent

foreign

the

specifically

market
to

Movement

decisions.

possible

different

models

approaches

information

dissemination

of Market

of various

discover

models

a Predictor

of information

game-theoretic

Information

At

Data

as

the

persist

comparably
most

(OTC)

in

faster

equity

opaque

over-the-counter

flow

markets

markets,
markets

in
and

such

in bonds

as
and

derivatives.
Asymmetric

information

selection,
market

or

ability

participants.

Odders-White
asymmetric

the

and
information

present
of

in

informed

According
Ready
have

to
(2006),
been

the
traders
Dennis
the

proposed

markets
to

leads
pick

and

over

off

Weston

following

to

uninformed
(2001)

measures
the

adverse

years:

and
of

Quoted

bid-ask

Effective

spread

bid-ask

spread

Information-based

impact

Adverse-selection

components

Probability
Quoted

of informed

Bid-Ask

The

bid-ask

(1971)

spread
of

and

spread

reflects

market

maker

using

dealer

receives

to the
in

market

order

price

spread,

the

asymmetry

between

dealer

the

dealer's
of

the

asymmetric

given

point

Effective
The

bid
and

and
ask

the

dealer

dealer

higher

is

clients

access

dealer's

and

the

to

quoted

public

bid-ask

dealer

may
the

relative
he

do

serve

market

as
at

quoted

information

Given

as

quotes

the

of

himself.

an

adverse

wider

information

in

quoting

potentially
the

the

from

spread

estimate

spread

available

the
come

the

result,

the

by

a disadvantage

against
a

the

movements

increases

As

by

researchers,

may

at

readily

suggested

When

suspects

himself

information

that

the

most

of

the

measure

large

at

any

in time.
Bid-Ask

effective

between

leave
the

his

numerous
of market

he

most

First

information.

movements.

same

clients,

by

that

compensate

in

has

flow

yet

information.

asymmetric

may

crudest,

expectations

movements,
to

uncertainty
bid-ask

and

the

developed

the

order

trader

is

asymmetric

later

bid-ask

informed

spread

trading

measure

Bagehot

bid-ask

Spread

quoted

observable

of the

bid-ask

the
ask

Spread

latest

spread
trade

prices,

is

price

divided

by

computed

and
the

the

as

midpoint

midpoint

twice

the

between

between

difference
the

the

quoted

quoted

bid

terms,

the

prices:

(7)

The

effective

latest
markets
midquote

spread

realized
are
is

price
balanced
the

natural

measures

how

fell

away

from

no

information

and

trading

price.

far,

in

the

percentage

simple
streams

When

the

midquote.
through,
limit

order

When
the
book

true
is

skewed

or

closer

to

of the

book.

imbalanced

the

side

with

Information-Based
The

in

limit

way,

orders

the

traded

located

price

at

or

near

moves
the

top

Impact

attributable

to

impact

measure

Hasbrouck

specify

estimation

other

excess

information-based

(1996)

some

the

of the

(1991).

following

of

asymmetric

Brennan

vector

and

impact

is

Subrahmanyam

autoregressive

information-based

information

(VAR)

measure,

model

for

(8)
(9)
where

i,t is the

, V i,t = sign ( P
security

time

propose

Adverse
The

adverse
to

into

selection

Order-processing
Inventory
Models
and

time

volume

Bid-Ask

components

the

Harris

following

of
(1988).

three

t 1 to time
in trading

and

of equation

of the

time

recorded

t . Brennan

in estimation

and

i from

the

Subrahmanyam

(8):

K = M = 5.

Spread
the

The

bid-ask
model

spread
separates

is
the

components:

risk
costs

risk
in a similar

George,

Glosten
(1997)

to

of security

is the

Components

Glosten

spread

Adverse

lags

selection

attributable

v i,t

t 1

five

Selection

bid-ask

in price

i,t ). v i,t , and

i from

(1996)

change

Kaul,

and

Harris

aggregates

specified

as

spirit

were

and

proposed

Nimalendran

(1988)

model

inventory

risk

by

Roll

(1991).
popularized

and

(1984);
The
by

Stoll

(1989);

version
Huang

order-processing

of
and

costs

the
Stoll

and

is

follows:

(10)
where

i,t is the

t , V i,t = sign ( P
security
as

defined

to adverse

i from

i,t ).
time

previously,
selection.

change

in price

v i,t , v i,t

is

t 1 to time
and

the

of security

i from

volume

recorded

t , S i,t is the

i is the

fraction

time

effective
of the

traded

in

t 1 to time
trading

bid-ask
spread

the
spread
due

Probability

of Informed

Easley,

Kiefer,

distill

the

The

model

dealer

Trading

O'Hara,

likelihood

and

of

Paperman

informed

trading

reverse-engineers

to

obtain

(1996)

the

probabilistic

propose

from

quote

idea

of

sequential

data.

provided

order

to

quote

sequence
the

model

flow

by

seen

by

a
the

dealer.
The

model

occurs
a

is

that

select
of

analyst.

The

on

impact

of

and

then

place

Thus,

all

same

side

investors

the

the
is

of

the
trades

the

the

place

an

).

informed

market

at

is then

event

of the

the

event

a rate

as

their

having

or

will

keep

brilliant

to

sells.

At

on
at

prices;

orders
the

rate

on

the

same

placing

orders

of

informed

probability

positive
informed

have

place

effect

occurs,

knowledge
will

suppose

having

likely

to

controlled

a negative

event

event

buys

. The

determined

is

a
by

event
only

. Furthermore,

event's

the

event

be

finding

is

the

an

is observable
may

of its

to

marketeither

but

event

When
the

Suppose

research

of

according

of

probability

impact

investors
of

an

or

probability

(1

uninformed
of

the

levels

Such

of such

occurs,

concept.

price

investors.

probability

know

following

information

prices

investors

taking

to

is

on

sides

is bound

event

prices

they

the

selected

if the

effect

on

group

release

that

built

time,
on

both
trading

follows:

(11)
The

parameters

likelihood

function

and

over

are

T periods

then

estimated

from

the

following

of time:

(12)
where
within
(13)

is
a specific

period

the

likelihood
of time:

of

observing

buys

and

sells

Summary
Understanding
can

the

unlock

whether

profitable

moves
End
If

the

original
the

levels,

2. What
from
3.

data?

Does

the

following

can

be

flow?

said

How

information

immediate

is

uninformed

trade,

and

about

the

is it measured?

order-flow

and

increases?

about
profitability

or

following

his

then

revert

informational

to

their

content

of

Stoikov

imbalance
(2011)

How

metric

increase

when

can

it be

developed
the

size

estimated

by
at

the

Cont,

best

bid

Explain.

Suppose

you

Who
the

What

falls,

what

tick

utilize

if he

in

understanding

information.

widen

is order

MSFT.

example,

Explain.

Kukanov,

4.

result

participant

Questions

quotes

trade?

market

possesses

may

trader

each

For

participant

the

superior

of Chapter

1.

5.

has

of

strategies.

movement

engaging

if he

motivation

market

market

either

and

trading

a particular

impending
from

type

what

observe

is most

information
is

adverse
does

this

likely

high

trading

to generate
selection?
mean

for

autocorrelation
and

why?

positive
When

the

a market-making

of

How

can

order
your

flow
algorithm

gains?
risk

of

adverse

algorithm?

selection

in

CHAPTER

12

Additional

HFT

As

Chapters

and

Strategies,

Market

8 through

large

11

automates

however,

perceive

potential

to

discusses

these

illustrate,

human
a range

detection

market

crashes.

of

HFT

continue

spectrum

we

find

employers

open

the

Street
for

Jobs

Journal,
HFT

invest

the

The
Wall

profitable
the

is

on

belief

hackers,

that

can

Mr.

prey

providers

the

strategies
markets.

of

(often

the

that

the

deem

the

future

end

of

industry.

careful

they

hiring

one

postings
are

that

and

section

These

find

individuals

the
Just

in the

Wall

seeking

talent

whitest

shoe

firms

generally

worthwhile

only

and

hiring

advertised

is

enormously

industry

some

who

traders.

the

and

markets

world?

are

Purportedly

are

traders
supposedly
malicious

of
identify

evidence
HFT

are

unfair

advantage

Cuban

high-frequency
a

range

of

how

than
of

divergent
has

financial

strategies

based

more

such

a
that

nothing

Mr.

uncompetitive
out

proclaimed

have

one,

Cuban,

fears

take

Stanley
For

Mark

Cuban's

and

Morgan

scapegoat

high-frequency

Mr.

as

recently

traders

unscrupulous
a

that

such

businessman,

Cuban

making

Opponents
HFT

we

high-frequency
to

job

Stanley.

) implies

to game

So

fallen

find

high-frequency

investors.

the

Investment

chapter

methods

services

employers

high-frequency

and

of

On

the

to stay.

systems

views

with

HFT,

have

manipulation

gamut.

by

of

This

along

financial

mostly

their

Journal

seeking

how

of

Dallas-based
of

the

something

extreme

afraid

his

into

here

other

will

the

and

Morgan

extent

and

successful
he

like

Street

in

advertising

resources

legitimate.

At

you

The

banks

run

in Money

and

roles.

investment

in

page

to

(HFT)

that

dynamics.

market

Crashes

opponents

outcomes

in detail,

high-frequency

on

HFT

market

threats

Opinions

Market
trading

The

adverse

impact

perceived

and

high-frequency
trading.

of

negatively

enable

Manipulation,

likely

services
traders.

of

purported

HFT

destroys

compiled

by

one

of

the

workgroups

(CFTC)

of

subcommittee

spread

scalping,

As

chapter

this

associated

market

with

HFT

Strategies

not

Strategies
activity,

financial

markets.

HFT

CFTC

into

be
a

lit

direct

the

be

to name

as

a few.
and

often

categories:

discovery.
exchanges;

of

technique

tasked

following

the

pools.

consequence

HFT

names

malicious

following

in dark

manipulation

identified

ominous

just
to

Commission's

marketsregulated

feasible

on

sniping,

of the

price

Trading

such

thought

one

in

subcommittee

Latency

included
and

serving

are

market

strategies

that

with

pump-and-dump

is

banned

identifying

in

most

undesirable

instances:

arbitrage

Spread

scalping

Rebate

capture

Quote

fall

can

that

Futures

strategies

feasible

strategies

HFT

ignition,

shows,

strategies

same

Commodity
on

Legitimate

The

the

matching

Layering
Ignition
Pinging/Sniping/Sniffing
Quote

stuffing

Spoofing
Pump-and-dump
Machine
Each

of

learning
the

sections.

orders

excess

for

is discussed

market

typically

unprofitable

Arbitrage

activities

via

profitability

methodologies
Latency

some

markets

advanced

orders

proposed

While

manipulate

any

manipulation

participants

special
carry

of such
to

detection

HFT.

in detail
claim

that

front-running
a heavier
orders
This

of pump-and-dump

HFTs

order

price

in the

in the

tag

long

chapter

can

types,
that

term,
also

activity.

following
also
such

annihilates
rendering
discusses

the

Latency

arbitrage

most

direct

obvious
HFT

is often

example

the

however,

Law

states

and

other

deployment

the

belief

well-defined

the

without

unlike

11,

to

as

one

arbitrage,

Contrary
a

of HFT

race,

8 through

has

Price

their

trading

will

market

market

has

then

be

the

foreign

When

of

of

some,

market

benefit,

of

for
in

latency

the

at

price

as

the

markets,

the

market

price

IBM

of IBM

Price.

given

financial
of

words,

in New

The

investors

price,

no

in

that

matter

in ideal

stock

the

trades.

low-frequency

fair

One

instrument

In other

of

of

regardless

financial

the

to trade.

Law

price,

to assure

be

the

theoretical

London

York,

in

should

after

adjusting

exchange.

prices

diverge

serves

the

same

is

same

where

decide

conditions,

always

the

always

they

theory

well-functioning

of markets

of One

the

In

the

high-frequency
previously

the

IBM

in the

his

frequencies

can

instruments

market

are

to

markets.

be
consistent

it where
and

the

the
that

across

prices

the

globe,

is

trades

the

prices
then

and

investors

on

traded

upholding

too

by

market
trader

HF

temporarily

produced

the

gain,

example,

stock

supply

high-frequency

capture

assured

stock

markets
arbitrageurs

For
the

equilibrate

The

to

different

latency

where

demand

serves

position

in

discrepancies.

buying

the

traders
divergent

reverses

price

simultaneously
process,

instrument

high-frequency

away

sell

while

financial

reason,

trade

traders

cheaply.

same

whatever

and

overpriced,

in

quickly
of

all

financial
the

Law

of

Price.

Latency
on

arms

Chapters

financial

in

always

Law

One

opponents

in latency

pivotal.

of

that

characteristics

jump

in

arbitrage

concept

instrument

for

the

next.

important

which

succeed

is

latency

described

The

To

technology

by

technological

discussed

fastest

An

of the

consequences.
strategies

pinpointed

taking

market

arbitrage
advantage
participants

is

an

example

of high
and

of

speeds.

regulators

a trading

strategy

A question
alike

is

that

commonly
how

much

is

based

asked

by

speed

is

enough?
much
of

When
speed

view,

the

race

for

when

an

extra

return.

dollar

Until

will

Spread

Scalping

that

continue

High-frequency
market-making

activity

two-sided

continuous

in

the

spread

Chapter

adverse

10,

subject
positions
of

rise

liquidate
presented

profitability

of seemingly

discussed

fraught
more
superior

the

forecasting
bound
all

to leave

other

automated

think

is

generates

of

HFT.

As

the

hardly

be

profitable

in

stagnant

whereby

the

orders,

the

market

10

and

even

in their

asymmetries,
maker.

such

market

spread-scalping

In

may

maker

normal

state,
some

profits

be

the

to

market

to

ensure

markets

or

losing

know

may
just

have
superior

traders

end

market

are

traders

better-informed
the

able

of

traders

fundamentals

on

absence

strategies.

whereby

situations,

not

is necessary

Better-informed

industry

the

analysis

spread-capturing

is

accumulated

markets.

11,

simplest
making

maker's

maker

and

its

market

Significant

in Chapters

11,

market
in the

positions.

nave

market,

or

discussed

can

nearly

simple:

that

and

about

the

an

liquidity

account

variations

In

to

inventory

market

skills.

of trading.

participants

of

generates

high-frequency

area

refers

market

with

information

longer

extensive

informational

than

profitability:

among
in the

point

equilibrium

to

in Chapter

with

is

trading
no

how

economic

there

and

often

the

as

technology

innovation

some

the

on

subject

his

conditions,

As

competition

a limit

is

fall

market

profitably

time,

for

risk,

and

opposing

on

be

activity

risks

inventory

soon

capacity

provision

in

there
From

as

spent

that

such

Even
to

end

scalping

value

selection

incarnation.

will

to foster

spread

Should

markets?

technological

additional

traders

end?

in the

speed

increasing

scalps

race

is acceptable

the

between

does

of trades,

maker

are
erasing

may

have

accumulated.
For

a specific

alleggedly
market,

example,

consider

spread-scalping
ahead

of

HFT
the

a news
has

impending

announcement.

positions

on

announcement

both

Suppose
sides
on

the

of

an
the
jobs

figuresinformation
preceding

on

month.

high-frequency
accuracy
the

that

variety,

may

jobs

number

the

large

market

matter

to

portfolio.

In

summary,

some

buy

may

end

may

seem

yet

a legitimate
Without

market

orders.

on

of

sides

Rebate

work

of the

be

of a spread

market

on

be

the

market

able

daily

the

that

is

In

a considerable

a predatory

a
our

loss

in

strategy

in its

market

most

to
nave

inventory

and

participants

call

integral

part

either

side

capable

of

is a tiny

profit

to

of

traders,

with

on

not

to

presumed

announcement.

is the

or

forecast,

side

the

profitable

sitting

required

his

opposite

with

most

low-

Suppose

The

in

like

that

would

Compensation

amount

both

provide

of market

of the

spread,

executing

their

comparable

the

limit

orders

the

dangers

basis.

Capture

Another
HFT

strategy
is

traders

the
are

costs
The

immediacy

up

what

orders

on

market.

the

reasonable

of lower-frequency

activity

limit

the

increased.
bet

enhanced
is

during

with

news

it is hardly

scalping

of

the

the

maker

considerations

the

on
to activity

informational

desiring

the

due

participants,

traders

to

quantities

Spread

to

order

scalping

whether

to

large

spread

functionality.

decides

considerably

or lost

have

selling

and

making,

to

order,

incarnation.

market

next

added

forecasted

takes

market

market

have

then

of seconds,

his

trader,

maker

rise

high-frequency

were

market

informed-trader's
about

jobs

is likely

trader

spread-scalping

just

many

better-informed

better-informed

sending

how

and

fragmentation.
improve
profitability

the

forth

capture.

presumed

to

benefits

and

put

rebate

strategy

value,

often

is
a

of

frequent
In

reality,

profitability
without

other

an

Under

and
to

be

example
as
of

this
trading

forecasting

of

simply

orders

empty

of

strategy,

profit

market
an

example

this

generate

limit

thought

as

high-frequency
by

on

section

is

arbitraging

various

exercise
what

with
wrong

illustrates,

strategies,
methodologies,

of

but

the

exchanges.
no

economic

with

market

rebates
cannot
such

help
deliver
as

the

ones

presented

To

be

and

in Chapters

profitable,

hold

Chapter

a high-frequency

position

equities,

where

from

rest

equity

by

example,
orders.

the

directional
the

value

so.

The

order
trader's

limit

a
when

expected

trade

them

and

U.S.
the

providing

(NYSE),

to traders
of

1 tick

or

for
posting

each

trade

$0.01

is p up

high-frequency

marginal

in

markets,

orders,

rational

only

in most

ahead

up

the

exist

in

differentiate

shares

going

order

outlined

Exchange

100

an

As

rebate

estimates

a stock

only

to
per

Stock

who

gain.

normal

every

environment,

buy
the

for

of

NYSE

exceeds

York

execute

presently

gain

posting

New

seek

In

traders

a trader

probability

a limit

$0.01.

to

orders

minimum

to $0.30

Consider

realize

market

The

to

needs

exchanges

currently

$0.13

current

post

of

rebates

to

and

pack.

is

pays

limit

11.

trader

enough

myriad

the

doing

limit

will

of

pay

liquidity

for
a

markets

exchanges

In

long

3, rebates

the

8 through

trader

cumulative

rebate

return:

(1)
where

$ rebate

is the

costs

) represents

the

execute

his

trades.

clearing

fee,

NYSE

(1)

of the

costs

the

rebate

per

trader

needs

Transaction

transaction

pass-through

Equation

value

costs

fee,

fee,

and

to

$( transaction

pay

equities

FINRA

in addition

is equivalent

in

share,

per
may

pass-through

to broker-dealer

share

to

include

fee,

and

commissions.

to

(2)
As

the

inequality

orderposting

(2)

trader

has

with

probability

greater

that

transaction

costs

NYSE

rebates

executed
NYSE
or

as
fee,

another

structure,

run
a market

a negative
$0.0023
a rational

shows,
to

than
for

in
predict
50

selling

about

30

the

rebate,
share.

high-frequency

for

absence
direction

percent.

$50,000,

order,

per

the

million
or

30

million

removing
To

of
of

rebates,

the

price

correctly

broker-dealer

shares

shares

profitable

trader

needs

considering

per

share.

incur

the

liquidity

be

advertises

without

$0.0016

of about
under
to

limit

have

When
additional
$70,000

such

cost

forecasting

that

reliably

least

70

predicts

probability

of

market

movement,

shares

as

p up , to

be

at

percent:

(3)
When

posting

receives

the

same

$60,000

nonrebate

in

30

rebates

transaction

$0.0003

in profit

directional

or

costs

per

high-frequency

million

trade.

trader
forecasting,

$0.0020

and

This

to

lower

but

just

limit

per

orders,

share,

generating

the

trader

offsetting

about

the

$10,000

or

rebate-driven

profitability

allows

his

probability

of

required

to 48

the

correct

percent:

(4)
In

other

words,

while

high-frequency

forecasts,

correctness,

the

profitable.

As

capture
as

minor

respective

allow

of

trading

strategy,

market-making

accuracy

probability

random

spread-scalping

improvement

so-called

thought

of

of

forecast

strategies

to be

a successful

operation,

with

performance,

quote-matching

to mimic

trader

is

then

generate.
of

worsening

limit

thought

to

and

The

capable

of

certain

predicated

on
generate

key

in

take
profit.
the

the

rebate

rebates

not

as

serving
a

by

the

primary

limit

orders

short

of the

Specifically,

positive

the

strategy

market
is

impact

the

the

and
child

of

trader

is

markets

in

high-frequency

reversing

ability

block

subsequent

move

success

trader's

or negative
of

move,

orders

impact

high-frequency

allowing

the

original

market
on

always

term,

high-frequency

assumption

obtains

is

high-frequency

the

the

the

trader

negatively

amplifying

that

impact

could

investor

advantage

high-frequency
trader.

market

a strategy

which

direction

of another

assumes

identifying

the

ride

the

strategy

capturing

orders

investor

pricing

to quickly

strategy,

such

large

the

trades.

trader

the

If feasible,

trades

The

the

not

required

Matching

orders

do

the

the

of profitability.

Quote

the

with

decrease

and

rebates

is a complex

source

In

rebates

positions
the

to foresee

and

strategy
which

is
limit

impact.
the

primary

reason

for

its

infeasibility.

Most

the

of

identity

follow

5,

movement

of

in

following
even

a
for

the
the

limit

term,

case

of

buy

order

orders

is likely

are

the

to

ability

limit

followed

orders,

positive

by

behind

be

the

relying

by

the

and
in
price

no

means

market
it

mostly

is

impact

very

small

negative

market
on

tag

a positive

average,

can

protect

discussed

is

while
on

orders
solely

as

movement

and

to be

They

Furthermore,

orders,

strategy

anonymous:

HFTs

the

is

for

matching

the

orders

short

persistent

quote

are

entity.

buy

top-of-the-book

statistically

limit

specific

many

In

exchanges

disallowing

while

guaranteed.

traders,

orders

Chapter

of today's

price.

copying

or

As

not

a result,

other

trader's

at

different

a disappointment.

Layering
In

layering,

price

levels

the

high-frequency

away

near

from

future,

objectives

of

suspect

market

then
often

enters
price,

to

limit

often

resubmit

confound

orders

to cancel

the

casual

the

orders

orders

in

again.

observers,

The

who

in

turn

wrongdoing.

Some

layering

layering

may

of

the

orders

the

order
with

available

of

many

executing

participant
limit

and

demand,

The

layers

orders

of changing

and

other
The

such

layering

however,

is

legitimate

manipulative

either

then

buy

or

promptly

traders'

and

in highly

sell

cancels

inferences

Securities

penalized

brokers

price

price

queue

a broker's
broker

as

price-time

a broker

at different

reaches

intent

layering,

with

layering,

with

manipulative.

about
Exchange

publicized

cases

2012.

Much

given

the

(SEC)

in September

be

a market
book

supply

Commission

books

indeed

is one-sided:

side

The

the

and

layering

trader

may

well

as

priority,

or a market
points
of
order,
use

his

described

the

limit
the

market

maker

with

strategy,

intent

priority

book.
may
to

limit
3.

placeholder
a time

When
pursue

execute

in

Chapter

of securing

order
broker

makers
in

leaves

practiced

the
one

slice

In

by
order
most

limit

orders

priority

in

market
of two
of

price
paths:

an

order,

securing
In

a preferential

the

the

absence

of

placeholder

Similarly,

cancel

it,

price

for

customer

his

customer.

orders,

the

broker

may

simply

cancel

order.

market

maker

depending

may

on

his

decide

to

estimates

execute

of

on

inventory

the

and

order

or

information

risks.
As

such,

most

manipulative

but

clogging

the

Mercantile

schedule.
orders

when

the

order

in

the

posted
price

the

size

secure
removes

from

the

order.

priority
need

in the

for

in

the

is

are

bid

matched
strategy

given

queue,

as

the

execution

to

at

of

pro-rata

the

ask.

all
same

Each

limit

proportional

to

the

to

eliminates

means

of

schedule,

best

as

Chicago

executed

and,

order

the

partially,

Such

by

with

pro-rata

or the

not

markets

by

priority

level
best

layering

the

changes

3, under

are

engine

price-time

the

queue

noise

example,

price

becomes

limit

time
the

a given

price

each

the

for

strategy

implemented

in Chapter

level

layering

matching

solution

(CME),

at

given

of

the

successful

described

the

unwanted

and

engine

As

of

create

Exchange

matching

limit

do

networks

cancellations.

the

incarnations

need

a result,

securing

entirely
priority

of

execution.
Ignition
In

an

the

ignition
location

against

In

the

today's

may

or ignite

out

his

exchanges

work

from

them.

Next,

have

positions,

only

the

investors'
and

other

as

To

match

market

result

price,

the

to

all

is

stop-loss

the

thought

orders

strategy

substantial

trader

of long-term

pump-and-dump.
away

will

trader

investors'

high-frequency

closing

expense

a high-frequency

long-term

positions

allowing

the

of

them,

stop-loss

swiftly

strategy,

the

while

match

that

on

market

detect

and

assumes

impact

ride

to

the

large
market,

impact

capturing

wave,

small

gains

at

losses.
lit

trading

of

market

against
one

venues,

manipulation,

someone's
needs

such

to

orders
move

the

strategy
such

placed
market

as
well
price

substantially

in

manipulation

has

methodology

described

While

the

direction

always

been
later

a manipulation-free

venues,

such

pools.

The

as

institutional
under

the

and

have

can

orders.

be

Market

screened

is not

feasible

it can

be

been

without

beware

loss

using

chapter.

exchanges,

explicitly

buyer

stop

strategy

however,

investors,

operate

in the

regulated

pools,

said

illegal

ignition

all

dark

of

created

in lit trading

deployed
for

regulatory

in

dark

sophisticated

protection,

and

principle.

Pinging/Sniping/Sniffing/Phishing
Pinging,

sniping,

same

general

type

ignition

strategy,

against

those

small

in

Automated
and

lit

for

as

traders

away

from

price

away

market

the

in dark

as

the

selectively

present

market

the

with

first.

of

the

matches

market

impact

a strategy

is

for
often

Citibank's

screen

for

pingers

render

generally

pinging

possible

accompanied
ignition

execute

at

unless

they

price,

market

it is

the

mechanisms.

it is not

case

and

charges

market

to
like

example,
to

Such

unless
in

for

ways

much

orders

such

pools,

pools,

exchanges,

cannot

from

by

illegal

strategies,

random

in

price

expressly

in

levels

move

movements

lit

the

Such

price

construe

some

market

conditions

are

than

others.

Avoiding

conditions

manipulation.

discussed

conducive

later
to

favorable

to

associated

with

Stuffing

Quote

stuffing

in this

market
market

Quote

trader

pingers

Just

limit

strategy,

behavior.

discourage

such

temporary

designed

refer

strategy,

riding

dark

have

pinging

pinging
of

and

Some

typically

pools

ignition

Desk,

manipulation.

markets

pools.

is feasible

markets,

market

As

dark

pinging

hidden

the

monikers

The

creating
like

and

phishing

strategy.

Much

them

unprofitable

and

identifies

Trading

charge

While

of

orders,

gains.

observed

sniffing,

manipulation
manipulation

ignition

refers

intentionally

chapter,

and

to

other

a purported

clogs

the

may

help

variants

traders
of pinging

high-frequency
networks

and

to

eliminate

risks

strategies.

strategy
the

more

matching

whereby
engine

with

large

number

of

layering,

where

the

execution

priority

in the

thought

to send
of

markets.

Quote-stuffing

slowing

traders,

engine

order

and

orders

the

quote

the

quote-stuffing

are

stream,

further

thought

priority
then

the

thus

manipulating

to

effectively

are

expressed

to do

access

ensure

traders

with
and

and

to

quote-stuffing

traders,

stuffers'

Unlike

seeking

cancellations

other

quote

cancellations.
is

queues,

and

traders

order

trader

book

down

ensure

and

orders

high-frequency

in rapid

purpose

other

limit

so

the

to

delay

matching

front-run

other

traders.
The

idea

any

network

cannot

of

engineer

selectively

participants,
venue.

When

cancellations,
irrespective

matching

it

is

of

who

for

down

trail

leads

for

whom

to

the

up

the

such

If

lines

low-frequency

traders

voids

culprits

manipulation

indeed

can

be

with

with

indeed

benefits

of

may

be

desire

to

the

fast

result

fast

suspected

manipulation),

unequipped
may

the

who

anyone

and

such

maneuver

those

(a market

trading

participants,

equipped

only

traders,

the

Obviously,

network-clogging

capabilities.

clogging

for

some

orders

market

problem.

clogging

low-frequency

matching

intentionally

little

network

However,

advantageous

the

all

with

as

account
for

to

clogged
for

flaw:

network

access

gets

clogged

do

critical

communication

high-speed

caused

if anything,

technology.

slow

equally

one

individual

network

engine

exercises

technology;

an

receiving

contains

confirm,
down

still

network-clogging

fast

will
slow

while

strategy

of
natural

technology,
in

increased

profitability.
Spoofing
The

spoofing

radically
the

different

order

other

strategy
intent.

book

traders'

is

without
inferences

resulting

prices.

In

detected

in

lit

most

similar

In

to

layering

spoofing,

the

execution;

in the

about

available

principle,
markets:

spoofing
while

but

trader

the

supply

layering

with

intentionally

process,

can

executed

be
calls

and

trader

for

distorts
changing

demand,

screened
fairly

for

and
and

balanced

entry

of

limit

one-sided

has

of

been

Dodd-Frank
the

spoofing

at the

In summary,
reasoned

to

or

market
and

be

are

and

discussed

for

which

consequence

conditions

in the

Global

show

accounts.

United

States

prosecuted.

In

Markets

to be
HFTs

of

Identification

market

trading

in the

considered

cause

a direct

Bunge

would

under

2011,

for

$550,000

for

activity

and

open.

manipulation.
the

actively

been

spoofing

selected

has

strategies

the

in
illegal

fined

market's

ranges,

expressly

CFTC

most

price

orders

made

U.S.

all

limit

Act,

example,

exist

across

peaks

Spoofing
the

orders

to

HFT
be

illegal

market

conducive

following

should

already
of

illicit

banned

activity,

do

not

such

as

manipulation

market

activity

manipulation

are

section.

Pump-and-Dump
Pump-and-dump
high

is a truly

or low

frequencies.

portrayed

in

pumped

or

to dump
other

the

it at the

investors.

whereby

first

instrument,

only

opportunity,

all while

In

the

thought

only

to

the

at

and
absence

posttrade
size

leaving

for

expense

Gatheral

(2010)

seller-initiated

of

other

and
impact

trades,

for

at

or down

the
and

traders.

function

seller-initiated

example,

raid,

financial

first

available

traders

prices

of

securities,

capitalize

Huberman
necessary

on
and

false
Stanzl

conditions

for

opportunities:
should

trades.

buyer-initiated

of

dust.

pump-and-dump

for

computer-assisted

developed

impact

bear

of

the

just

expense

is the

in the

positions

have

at the

price

a profit

investors

their

market

buyer-initiated
market

at

up

high-frequency

permanent

permanent

the

well

brokers

instrument

pump-and-dump

position

drive

the

financial

at

was

unscrupulous

a profit

pump-and-dump,

reverse

of

where

depresses

other

implemented

pump-and-dump

realizing

of the

his

to momentarily

momentum
(2004)

side

close

whether

a particular

artificially

to

promptly

of

opportunity,

high-frequency

are

Room,

value

flip

trader

activity,

low-frequency

Boiler

the

The

the

The

film

raised

adverse

be

When

trades
high-frequency

the

symmetric
the

in

posttrade

exceed
trader

that

for
could

pump
level

the

security

and

profit.

then

The

impact:
would

dump

gain

the
differ

12.1

value

from

following

that

not
of

following
A

of

pump-and-dump

12.1

Aldridge

formally

processed

permanent

at time

t, where

and

size

V .

of

then

his

a new

change

the

negative
V , as

opportunity

market
buy

other

price

size

at

asymmetric

the

high

positions

following

On

Rules

trades

hand,

when

following
of

the

shown
exists

dump

Conversely,

Out

in

price
Figure

when

the

High-Frequency

pump-and-dump
impact

trade.
by

f(V)

a trade

a buyer-initiated

If f(V)

first
if

strategies

f(V t ) of

V t > 0 indicates

artificially

trade

to

that

market

a seller-initiated

same

equal

describes

V t < 0 describes
pump

the

to

is violated.

Impact

Manipulation

(2012e)

trades.

trade

condition
Market

of

is

the

impact

arbitrage

Pump-and-Dump

a measure

sell

out

from

market

feasible,

size

purchases

closing

solely

a buy-initiated

no-pump-no-dump
Figure

security,

absolute

trade

repeated

originate

is

sell-initiated

through
the

would

pump-and-dump

change

price

buying
<

using

of

trade

> - f( - V) , a trader
and
- f( - V) ,

size

V t
and

could

then

selling

at the

the

trader

could

manipulate

the

markets

by

first

selling

and

then

buying

the

securities

back.
To

examine

market

the

impact

window

evolution

within

order

Figure

Denoting

by

event,

12.2

as

Sequence

market

market

different

is determined

market

of

impact

event

windows,

a number

shown

impact

ticks

12.2

Used

time,

where

of trade

in Figure

of Events

over

f,

the

consider

length

before

and

of the
after

the

in Market

function

we

we

Impact

Computation

obtain

the

following

specification:

To

evaluate

the

specification

for

, consistent

with

Glantz

(2002);

Huberman

feasibility
the

market

Breen,
and

and

of

pump

and

as

a function

and

Korajczyk

impact
Hodrick,

Lillo,

Stanzl

the

Farmer,

(2004)

and

and

dump,

we

of trading

linear

volume,

(2002);

Mantegna

Gatheral

use

Kissell

(2003),

V
and

following

(2010):

(5)
where

V t is

the

size

sizedependent
impact

of

market
each

from

hypothesis,
financial

is

that

trade

executed

impact,

trade

pump-and-dump
different

of

and

recorded

feasible,

then

time

for
for

pump-and-dump
can

t.

t,

trade
If

is

the

specified

as

trade

high-frequency
trades

seller-initiated
in

the

sizeindependent

buyer-initiated

exists
be

time

is the
at

estimated

instrument,

at

will

trades.
trading

be

The

activity

null
of

follows:

(6)
And

the

specified

alternative

hypothesis

ruling

out

pump

and

dump

can

be

as:

(7)
The

framework

manipulative

above
activity

allows
in various

for

straightforward

financial

instruments.

screening

for

market

What

does

pump-and-dump

illustrates
(symbol

the

analysis

FGBL)

trades,

millisecond

The

the

trade

Eurex

contain

the

trades

whether

the

tick

discussed
12.3

Trades,

In

trade

was

computing

observations,

function

of data
reports

month

of

20092010
of

12.4

the

shows

observed
Table
and

for
12.1

Small

buy

on

estimates

results
and

is the

the

data

official

copy

distributed

buyer

to

or

or

traders.

identification

or seller.

buy

with

To

identify

a market

sell,

3 is used.
of

Market

(MI),

Impact

that
that

sell
of

the

day

of

Buy-and-Sell

the

returns
MI

on

(5)

for

are

treated

specific

as

day

is

only.

Figure

coefficients
of

overnight

of equation

period.

Estimation
Trades

impact

volume

timestamp,
data

a market

futures

time-stamped

information
the

12.3

20092010

recorded

12.1

The

by

by

ensuring

Table

relationship

initiated

Coefficients

market

missing

size.

the

bid/offer

initiated

Futures,

to

Figure

Eurobund

and

is commercially

best

in Chapter

Volume

FGBL

trade

in

Eurex

sequentially

and

were

example

of

addition

and

not

of whether

rule

price

The

sequence

In

tape,

does

recorded

trading

data

Figure

on

granularity.

includes
of the

detect?

12.3
for

difference

trades

graphically

buy

and

tests

of

of all

sizes

illustrates

sell

trades.

volume

by
the

Figure

coefficients

trades.
Size-Dependent

in Eurobund

Futures,

Market
by

Month

Impact

for

Large

Figure

12.4

Difference

Buyer-Initiated

The

Trades

observed
change

Based

on

The

results

Less

differences

impact

possibility

in

the

from

the

of high-frequency
presented

That

of Seller

in buyer-initiated

month

results,

Volume-Attributable

to month
FGBL

and

futures

Market
Initiated

and
lack
data

Impact,

Trades

seller-initiated
statistical
does

market

significance.
not

support

pump-and-dump.
in

Table

12.1

indicate

of

another

interesting

the

phenomenon:
the

trade-size-related

trade

size

variation,
order

rises

intercept
of

single

of

up

trade

of 1 contract.

fund

managers

tradesin

no

To

in

that

posttrade

MI

the

market

Hodrick,

and

and

permanent

that
bid-ask

inputs

to

that

function

volatility,

impact

in

futures

change

the

created

by

variables
trade

on

studies

do

not

MI

of

their

of

size
leaves

market

is

and

well.

and

the

the

the

and
the

vice

on

equity

that

MI

use

trade

current

study,

duration

help

explain

or

the

statistical
the

as

variables,

trades.

intercept,

the

stock-specific
longer

Ferraris

volume-dependent

component,

(2003);

the

auxiliary

Breen,

that

find

versa.

the

such

showed

(2000)

seller-initiated

show

equities.

In

the

value

(2006)

of

trade.

of

additional

time

intertrade
None

(2005)

for

spread,

Mantegna

dependent

models

between

alter

Li

Lei

on

and

Engle

MI,

at

of

is

such

focused

and

auxiliary

characteristics,

Farmer,

equities

and

sizeindependent

impact

volatility,

and

liquidity

have

Lillo,

to lower

buyer-initiated

also

dependent

prevailing

symmetry

large

several

analysis:

Hanweck,

commercial

as

to the

Burghardt,

Dufour

spread,

other

FGBL

robust,

impact.

in

leads

are

market

Hauptmann,

of

as

average

Eurex

temporary

Thum,

forecasts

the

and

on

found

Other

spread

size

the

10 7 ,

impact

trade

studies

is also

several

single

(on

of

much

the

Other

futures,

duration

reports
and

for

characteristics.

intertrade

as

trading

added

(2002);

MI

liquidity

be

Korajchyk

Almgren,

order

institutions

unexplained

is large

the

about

results

can

depth.

on

until

capacity.

the

variables

for

register

The

incur

markets,

duration.

example,

the

larger

is

market,

equities

variables

For

news

median

to

equation

MI

concerned

the

begin

contracts.

may

is great

than

not

impact

contracts

futures

the

intertrade

100

market

are

whether

explanatory

as

100

FGBL

to a much

explanatory

about

does

trade-related

This

Unlike

check

and

to

larger

trace.

resilient

the

who

the

considerably

to
, in the

10 5 ), and

trade

MI

(2008)
volatility
predictive
we

find

market
however,

MI

coefficient

The

auxiliary

significance

of

leaving

the

dominant

size-independent

Machine

Learning

Machine

learning

accompanying
in

the

is often
HFT.

about

machines

machines

before

are

similar

to

to

the

then

as

evidence

of nested

as

simple

functional
the

and

of

As

the

iterative

models

may

in

clustering

of data

supervised

particular

financial

with

little

learning

theory
be

and

is

of

as

set

the

same:

is less

is often

parametric,
a

type
the

unsupervised

basic

free-form

analysis

used,

uncover

threatening

data

to

minimize

from

regression

below.

to

patterns
intelligence,

of

major

data

to

neural

learning

whereby
deviations

in

from

of

an

important

to

seeks
any

boosting
to

identify

relationships.

unsupervised
signals

the

learning

networks

devoid

each

supervised

learning

under

categories:

Supervised

the

fit

data,

information

identification

two

relationships,

Unsupervised

unstructured
to distill

into
learning.

of

used

by

learning
observing

points.

boosting
variable

the

down

Models

so-called

include

character

Machine

can

remains

seeks

range

used

of

broken

estimation

umbrella

dependent

and

discussed

Techniques

be

iteration

algorithms
patterns

can

analysis.

that

intelligence

film

nonparametric,

learning

belief

mining.

learning

subsequent
previous

data

learning

supervised
is

machine

shut

intelligence.

analysis

learning

The

participants

in control

or

example,

should

and

technology.

Independent

machine

basic

investors.

event

for

investors

market

machine

regression,

a result,

more

Machine

linear

Kass,

Terminator

of

Each

worrisome

reasoning

originates

analyses.

most

that

attack

traditional

learning

the

Doug

independent

of such

estimators.

nature

in data.

of

of

proposed

fundamentals

machine

series

one

commentator,

HFT

some

exposure

In reality,

as

unexplained.

Schwarzenegger's

among

cited

impact

machines

capable

Arnold

resonates

cited

A CNBC

conversation

down

market

algorithm,
Y , for

instrument,

for

instance,
is fitted

example,
a

time
with

works
series

a function

as
of

follows:

returns
G ,

a
on

expressing

dependence

of

parameters

Y on

returns

of

another

financial

instrument,

X ,

and

(8)
Next,

the

boosting

error

term

is computed

as

follows:

(9)
where
,

I st<>0

is the

) matches

series

indicator

Y t precisely,

w t represents

time,

with

later

function

all

and

1 when

boosting

weights

in

iterations

recomputed

machine

learning

taking

weights

the

first

on

value

of 0 when

t is not

equal

to

assigned

iteration

according

to

set

to the

to

0.

each
1,

The

time

observation

and

following

G(X t

weights

in

calculation:

(10)
where
(11)

This
G(X t ,

) that

learning

and

Machine

testing,

devoid
may

term.

and

dependent

trading

the

variables.
was

application

and

function

researchers

running

the

machine

such

suffer

from

underpinnings
in the

not

the

but

capable
was

of

not

are

this

for

stable

in the
as

beyond

long

spurious

behavior

written,

threatening

data

relationships

future

was

intelligence
not

flaw:

known
for

book

certainly

sizes

critical

produce

are

predictors

time

may

past,

reliable

window

etc.

relationships

At

algorithm

produces

predictors,

stable

are

ultimately

parameters

may

accidental

inferences,

learning

select

economic

been

Such

Y t . Human

algorithms

of

have

fits

additional

learning

mining
that

closely

simulation

training

methodology

no
its

of

machine
immediate

to humans.

Summary
This

chapter

adverse
fears

discusses

outcomes
surrounding

however,

HFT

are

manipulation,
reducing

from

a
yet

the

risks

direct
even

algorithms
HFT.

As

strategies

often
the

chapter

are

unwarranted.

consequence
those

to all market

presented

strategies
participants.

of
can

as

illustrates
Some

evidence
most

screened

of

the

strategies,

high-frequency
be

of

market
for,

greatly

End-of-Chapter
1. What

Questions

is latency

2. Suppose
NYSE
or

at 125.07

negative

impact

of IBM

is simultaneously

in Tokyo.
on

the

Does

price

latency

of

IBM

trading

at 125.03

arbitrage

deliver

from

the

market

on

positive
efficiency

is spread

What

kind

of

scalping?

What

layering

is

is rebate

manipulative?

capture?
What

kind

of

layering

benign?
5.

the

of view?

3. What
4.

a stock

and

point

arbitrage?

How

can

one

detect

manipulation?
6. What

is machine

learning?

pump-and-dump

high-frequency

is

CHAPTER

13

Regulation
At

the

time

already

this

Indeed,

other
as

traditional

of HFT

machines,
May

6,

ATS

BATS

and

the

Group

imminent

directions.

call

public

Trading

This

traditional

for

stricter

of Regulators
issues

the

of

discusses

such.

flash
of

March

loss

current

as

(IPO)
on

of

monitoring

as

offering

same

automation

regulated

System)

chapter
and

than

such

the

the

instrument.

be

failures

was

at

much

financial

more

$10-million-per-minute

HFT,

main

little

(HFT)

tracked

follows

given

should

initial

1, 2012.

to

is

of market

Alternative

August

however,

spectacular

Currently,

HFT

botched

activity,

in

regulation,

the

Initiatives

trading

and

2010;

trading

of HFT

processes

examples

relevant

Key

market

shows,

citing

(Best

on

high-frequency

regulation

of

book

trading

Proponents

of

Current

forms

this

written,

regulated

level.

as

was

heavily

broker-dealer
rules

book

crash

electronic
23,

Knight

modern

of

2012;
Capital

legislation

approaches,

and

likely

Worldwide

on

the

regulatory

table

include:

Jurisdiction
Stability

of systems

Investor

protection

Efficient

trade

Market

structure

This

section

matching

considers

each

of the

issues

in detail.

Jurisdiction
The

long

shadow

cast

their

mandate

and

objectives

due

to philosophies

Most

of

regulators
certain

to

by

provide

of market

U.S.

regulation

prescribe
behaviors

and

market
legal

regulators
oversight

regulators
histories

have

observed

to

evolved

in the

markets.

as

as

at
on

part

large.

of

Roles

a regional

basis

jurisdictions.

a rule-based

remedies

HFT

markets

of individual

represents
specific

covers

well
By

approach,
as

in

punishments

contrast,

regulators

which
for
of

the

European

system,

Union

whereby

conformance
The

with

to

systems

in

the

been

markets

for

alike.

Behaviors

grain

of

the

tenet

action

that

of

of

integrity.

The

approach

to

to

dealt

are

the

be

objective
equal

widows

European
to

to a contingent

orphans
against

actively

Union,

of

access

and
to go

are

be

regulatory

the

considered

distilled

can

between

allowing

for

its

systems.

models

States,

behaviors

In

can

developed

U.K.

the

identified,
however,

fairness

of

of traders

regulation,

the

gainsan

may

international

run

and

Exchange

Commission

afoul

Swaps

and

under

the

itself

is not

within

jurisdiction
regulated

of

to

be

future

market

developments

regulators

financial

key

forward-looking

seek

regulation,

International

by

country,

rules

several

(SEC)

like

futures

of

Canadian

the

each

regulatory

Commodity
contracts

more

The

to

and,

Organization

of

(IOSCO).

adopted

products,

by

styles.

tends

anticipate

standards

arise

developed

example,

impact.

promulgated

regulatory

taken

to

resulting

example,

futures

has

working

the

own

for

government

issues
for

their

regulators,

Commissions

Jurisdictional

The

access

with.

Australian

those

Securities

on.

and

Such

unfair

have

the

specifically

related

investors
fair

exist

field,

in

market

regulatory

United

playing

evaluated

desired

that

the

the

markets.

assessing

States,

EU

regulatory

regulators.

objective

conform

level

regulation

countries

then

and

In

is

of

differences

large

is deemed

of European

and

to

U.S.
and

U.S.

regions.

principle-based

case

principles

the

blocking

documented,

Other

two

has

main

general

philosophical
the

regulation

the

regulatory

between

the

established

each

differences

traced

to

have

as

for

various

agencies.

The

oversees

trading

equity

options,

exchange-traded

Futures

Trading

Commission

and

related

on

fixed
of

the

outside

markets,

such

income

and

CFTC,

even

of

basic

well.

as

United

and

equities

and

in

funds,

exchange
foreign

investor

are

Securities

options

though

the

markets

(CFTC)

foreign

common

In

and

deals
on

so
with

futures.
also
exchange

frameworks.

fall

In

addition

to

the

includes

industry

Industry

Regulatory

equities.

In

sent

cases
to

Association
best

is

to

subject

the

Order

all

all

no

sorts

of

members

of the

such

and

quotes

electronically,

levels.

Regulation
further

execution.
daily

open,

newspaper,
Reg

NMS,

prices

the

as
all

within

from

may

too

seem
States

be

large
can

assured
prints

now.

guarantee

After
the

all,

execution

been

years.

the

first

by

validity

recorded

on
dearly
ATS),

SIP

of

and
prices

NMS),

store
at

tick

enacted

in

a broker-dealer's
order
in the
the

had

of

their

in

the

only

next

broker.

Processor

day's
Under

execution
centralized
(SIP).

under

a market

The

Reg

broker-dealers
of

make

(Reg

of

printed

most

time

only

process

a market

into

orders

electronization

to track

given

The

to

Systems

Information
quotes

have

paying

(Reg

of

asset

limit

for

transparency

price

one

bestowed

receive,

quotes,

submitting
by

to

executing

trade

15

display

mandated

ability

the

Securities
for

promotes

broker-dealers

Systems

close

for

a minute

window

were

in great

and

can

and

individuals,

Trading

ability

a client

investors

one-minute

including

investor's

a reference

tape,

the

Market

low,

Industry

past

rule

The

further

bringing

high,

ticker-data

United

to

1980s,

in the

small.

selected

1998,

an

next

Futures

example,

to

Automated

National

are

from

for

privileges
and

led

abuses.

FINRA

FINRA

exchanges

entities,

such

enhanced
In

equities,

how

trading

has

to

of

licensing

market

by

dramatically

required

in

of

The

regulations

matter

force

industry

detected

vary
U.S.

Regulation

into

exchanges

the

The

exchange

ability.

went

2005,

can

Previously,

world

traders.

groundbreaking

customers,

Financial

the

supervision

equivalent

system

(SROs).

examination.

futures

Rule

regulatory

administers

further

futures

Display

markets.

that

the

next.

to three

allowed

for

(FIA)

for

U.S.

navigates

manipulation

approaches

regulator

of

SEC

(FINRA)

first-level

market

the

the

organizations

FINRA

maintains

among

Regulatory

1997

role,

of

practices

CFTC,

Authority

and

over

and

self-regulatory

this

examinations
Potential

SEC

NMS
in

order

in

the
as

little

as

one-hundredth

selected

U.S.

send

the

Reg

of

window

execute
the

market

may

choose

departing

most

some

exchanges,

executions.
to

DirectEdge
execute,

and

improvements

observers

should

they

While
with

the

trade
venue

flash

slow

voluntary

be

much

on

When

the

forwarded

at

DirectEdge

can
on
as

the

DirectEdge,

rendering

hand

to
must

to

orders

desiring

DirectEdge
from

the

when
intent

potential

OMX,

their

set

up

as

it

flash
orders,

DirectEdge.
for
is

investors
based

or be

special
pack.

orders

where

on

to
price

Most

to flash
a

traders

their

flashing

fair

order

for

direct

is not

yet

flash

table.

At

orders,

exchange:
flash

the

instantly
data.

tactic

immediately

is

exchange

in

to Boston

system

the

flash

with

of

exchange

exchange

persist

traders

line

and

banned

a whole

exchange

the

new

U.S.

incoming

orders

the

traders

NBBO,

of

NBBO

the

or better.

on

DirectEdge

limit

system

equity

exchange

created

negotiation

the

NBBO,

to take

in SIP

the

the

all

orders

NBBO,

the

typically

the

and

equities:

NBBO,

limit

DirectEdge,

are

separating

traders

to

distributed

the

flash

really

only

and

end

Under

NMS

voluntarily

participation:
at

Still,

peculiar

satisfying

as

communication,

and

orders.

contemporary

order

Reg

matching

inclined

with

have

have

invite

even

subscribers

orders

not

the

their

over

DirectEdge

order

to

is

does

of

at the

the

flashing

place

quarter

enough

flash

such

The

choosing

of

or

exchanges

as

little

it,

exchange.

orders

orders

as

price

have
order

broadcast

market

present,

not

exchanges:

to

match

(NBBO).

to another

limit

order,

structure

at the

rule

possess

while

a market

investors

rule

market

for

not

Offer

order

faster,

prices.

does

order-forwarding

does

delivers

execute

incoming

differentiation

from

only

or

microseconds).

another

and

exchange
an

route

Bid

may

If a given

(250

NMS

introduced

Best

exchanges

in

benchmark

has

National

The

Reg

milliseconds)

receive

a second

execution
NMS

can

(10

acknowledgment

one-thousandth

needed

a second

exchanges

back

1-minute

of

on

flashed

negotiation
A

study

by

Brodsky

(2011),

price

The

reducing

U.S.

known

SEC

as

DMA,

exchanges
sense

the

traded

also

without

to

the

and

under

broker-dealers
of market

are

(LEI),

issue

by

of

end

implemented,
privileges,
The

this

U.S.

As
of

servers.

in the
To

CFTC

facilities
disclose

proximity

the

flash

the
without

or

ability

Chapter

2,

in the
however,
the

that

shortest,

and

to

all

entity

to

resolve

and

measure
Once

fair

and

as

not

DMA

proximity

implies

that

the

exchange's

the

machines

in the

access,
and

their

of

broker-dealers.

traders'

average

legal

LEIs.

by

facility

co-location

fees

type

reinstatement

but
and

which

is likely

co-location

exchange,

all

The

co-location

leaves

pricing

ID-lending

to track

to

same

traders
checks

the

individual

lead

use

many

caused

section,

with

may

exchanges

surveillance.

the

paying

broker-dealer's

trader

following

execution

avoid

When

on

of

and

exchanges
which

access
complete

own

most

trades.

Under

to

their

however,

also

identification

makes

book)

taken

uniform

clients.

this

also

transparent

longest,

of

involvement

recommended

their

develop

smaller

machine

implement

can

DMA

even

in

in

flashing

(DMA).

information

checks.

exchanges

hosting,

ensure

prospective

Following

discussed

ban

the

existing

initiative

with

a trader's

risk

numbered

has

Proximity

building
facility.

time

CFTC

services.
location

the

LEI

use

the

in the

traders
the

their

stamp

complicating

granting

traders

to discern

discussed

access,

access

15

DMA,

reaction,

naked

market

ID

unable

on

direct

present,
to

result

provision

to

who

broker-dealer's

records

risks

At

identification
same

identifier

any

Chapter

broker-dealers.

broker-dealer
use

in

orders

steps

to

traders,

(discussed

flash

liquidity

lend

traders

to sophisticated

fees

and

would
allow

that

volume.

taken

access

directly,

systems

and

disincentivize

broker-dealers
and

confirms

yet

has

sponsored

information

the

example,

improvements,

exchanges,

this

for

same

exact

in June

2010

proximity

hosting

customers,
latencies

in

to

and
all

to

current

crash,

the

SEC

and

CFTC

introduced

clear

rules

under

which

rules

erroneous

were

trades

successfully

can

applied

be

corrected

in the

or busted.

Knight

Capital

followed

the

The

fiasco

new

of August

2012.
European
have

regulators

voted

against

discussed

in

controls

treaded

HFT

has

transaction

been

Given

trading

in

FIX

jurisdiction
the

world

would

such

free

of transaction

likely

from

to

fail

trading

to

U.K.

creating
government
potentially

a minute

the

temptation

agreement

defect

the

that

can

Europeans

to the

if all

would

the
be

tax.
such

a tax

language

the

in

a few

can

just

around

the

measure
benefits

all
great

in

in

highthe

collect

generating

lines

countries

too

one

execute

tax,

is

using

instantaneous

would

quickly,

considers

trading

transaction
and

finds

mark-up

trader

all.

Greek

communicate

changing

identical

for

Parliament,

and

little

Even

woes,

money

however,

the

equal

proposal

from

agreement

move

the

markets

systems

change,

an

and

The

as

error

life,

EU

Changing
as

taxes.

the

activity

settlement

takes

and

by

a computer

6.

stuffing,

risk

free

broader

capital,

they

financial

in response

(FIX),

impose

as

breaking

jurisdiction

The

after

the

trading

Chapter

in another

of

activity.

little

eXchange

quote

financial

shrink

for

introduced

trading

of today's

in

called

model;
quote

of European

by

European

and

source

of international

detail

agree

the

ratified
all

will

Most

to trading

code;

In light
as

U.S.

however,

proposal,

the

mobility

also

Overall,

some

on

activity

orders,

minimum

HFT.

recently

in

Information

country

by

tax

to backfire.

discussed

have

taxation

riches

Financial

the

viewed

today's

likely

on

small

substantial
the

They
systems,

and

flash

services.

lightly

parliamentarians

that

12.

co-location

have

for

access,

algorithmic

to

calls

generally

naked

Chapter

for

access

have

first

the

global

trading

and

premia.
regulators
rules

for
has

caused

have

taken

envisioned
identified
by

a most
future

the

computer

following
trading:

of

proactive
computer
four

approach,
trading.

dimensions

that
The

of

of
U.K.

instability

1.

Nonlinear

code

sensitivities

of trading

to

change,

systems

or matching

information,

where

whereby

small

engines

perturbations

have

large

in

system-wide

impact.
2.

Incomplete

to assemble

a more

3. Normalization
be

seen

accurate

4. Internal

risks

as

by

systems,

news,

in obtaining

13.1

participants

markets

unexpected

are

than

and

able

others.

risky

events

can

normal.

amplified

a delay

market

of the

where

risk-management
and

picture

of variance,

increasingly

some

system-wide
changes

feedback
in

reference

loops

market
data,

as

that

include

volatility,
illustrated

market
in Figure

Figure

13.1

Computerized
Source:

Zigrand,

Key

Findings

Trading:
Cliff,

of

the

U.K.

Problem-Amplifying
Hendershott

(2011)

Foresight
Feedback

Initiative
Loops

on

Stability
The

of Systems:

stability

markets:

of

a poorly

markets.

system

was

at that

moment.)

In

case
that

imagine

of
bore

how

similar

To

algo

issues,

regulators

with

the

polices

goal
roads

themselves
The
As

surveillance
the

before

million.
a task.

collect

In

volumes

are

to be

could
the

of

Knight

affected

stability,

much
who

to

like

the

may

fact

Capital

is

easy

other
of

ways

to the

it

occurrences
mulling

minute.

oblivious

However,

naturally

Capital's

every

portfolio

have

in the

employees

Knight

million

incident.

drivers

for

erroneous

of

Knight

algos

to

market

such

rampant

police

markets

highway

present

Modern
the

tick
day,

perfectly

suitable

for

requiring
next-day

shows,

problem

patrol

dangers

States,
data

the

identification

multiday
data

even

to

both

FINRA

every
T+1

basis.

of

market

detect

as

time.

45-minute

delay

shareholders

are

ill

and

CFTC

executed

recurrence,
cannot

in real

Knight's

however,

from

or on

performed

cost

regulators,

United

of

is best

Capital

Knight's

business

subsection,

US$10
appeared

havoc

of trading,
that

Group,

others.

following

activity

the

eliminate

errant

pinpointing

US$440
such

example

or

ensuring

for

and

of

in

Capital

wreaked

start

the

Nothing
erroneous

Knight

noticed

it was

scenario

limit

of

still

of

surrounding
of

the

scenario

errors.

issues

after

Capital,

operational

algorithm

average

brunt

participants.

of

Exchange

Capital

the

ideal

incident

minutes

on

Knight

host

ill-operated

Stock

at Knight

the

Algorithms

algorithmic

recent

and

losing

(Employees

the

45

York

to

inadvertent

the

tested

New

itself

than

A whopping

the

the

of

illustrates

better

Error-Prone

refers

free

memory

algorithms

at

systems

trading

recent

when

Detecting

equipped

for

currently

transaction

2 While

the

T+1

manipulation
discussed

algorithmic

the

in

data

is

or

other

the

next

problems

in

real

time.
Instead
real-time

of

implementing

surveillance

complex
at

their

offices,

data

screening

however,

systems

regulators

may

for
be

much

more

productive

example,

regulators

trades.

Yet,

such

most

Surveillance

clearing
would

at

feasible,

services
in

can

surveillance

also

observe

to be

the

real

time.

was

the

minutes

to

modern

unresolved

persist:

for

track

cross-asset

timestamps

cite

to

participants'

technology

of

inherent

choice.

in GPS-based

exchange,

addition,

issue

Currently

Measures

present,

real-time
to

most

instrument
several

circuit
for

percentage

Intercontinental

All

several

Commodity

that

minutes

points.

System

about

on

the

exchange,

time

45

frames.

surveillance
exchanges

to

synchronizing
positioning
however,

freedom
point

to

on

to

deploy

the

latency

the

differ

location
by

of

several

Stability
enabled
for

trading

following

Exchange

alarm

it

markets.

today,
halt

Futures

debacle,

NYSE

may

already

exchanges

breakers

in

the

depending

fast-paced

have

accounts

in global

opponents

clocks

for

exchanges

surveillance.

implement

the

in today's

to

took

among

forfeit

level

positioned

synchronization,

to

GPS-wound

microsecondsan

At

time

synchronization:

even

Deployed

of

reluctance
In

an

shorter

used

brokers

Group

proposed

equipment

is

matching

exchange-level

Some

opponents

the

sound

data

and

utilize

across

multiple

in much

strategies.

day,

markets.

best

it

their

traders

Capital
to

aggregate

using

at

patterns

patterns

trading

many

While

trading

and

trades

first

for

broker-dealers,

exchanges

surrounding

how

is,

as

trading.

detect

The

that

Knight

was

firms,

fast-moving

across

of the

that

trading

(GPSs).

market

the

case

of exchanges

systems

level,

irregularities

issues

example,

many

with

clearing

of the

aggregating

erroneous
can

end

Surveillance

Knight's

establish

technology

Still,

and

of

counterparties

complicated,

NYSE,
of

level

the

for

solution,

in the

exchange,
behavior

tardy
firm

market

all,

unusual

at the

a challenge.

detect

After

happens

be

best

the

all

brokers,
be

At

observe

be

can

can

and

field.

executing

multiple

time

prove

the
yet

of

real

may

in the

an

exchanges
(ICE)

some

example,

forms
are

in a selected
intraday
like
have

the
also

of

required
financial

price

drop

CME

and

introduced

of
the
the

following

measures

Interval

price

described

range

Protection

points
orders

Message

on

system

throttle

disconnect

limits

Maximum

quantity

limits

Real-time

position

validation

Price

reasonability

Interval

Price

Limits

Interval

price

limits

extreme
each

The

traded

level

define
in Figure
is halted

Source:

the

upper
13.2
until

13.2
ICE

and

, when
one

The

plus

lower

IPLs,

the

of the

Illustration

the

price
two

and

of Interval

determined

minus

Price

used
the

illustrated

following

as

the

Limits

by

follows.

For

lower
occur:

a
on

in

variation

in Figure

situations

triggered

computes

window

below

are

work

exchange

data

falls

that

halts

variation

moving

price

breakers

IPL-based

a normal

in the

average

circuit

instrument,
and

evidenced

moving

are

moves.

financial

price

lows

Figure

(IPLs)

short-term

average
and

next:

limits

No-cancel

Cancel

in detail

the

moving
highs

computation.
parameter

13.2

. As

shown

IPL,

the

trading

The

price

The

IPL

the

reverts

computational

previous

The

width

limit,
of

determined
bands

more

often

No-Cancel

Range

No-cancel

range

where

the

period

of

and

quote

limit

the

life

(MQL).

example,
be

limit

best

ask

with

an

orders

due

be

to natural

Introduction
markets.

canceled

of

MQL

A study

markets

were

algorithm

by

much

than

the

popularity

than

falling

soon

ICE

of

orders

EBS

they

fall

be

of

little

for

example,

minimum

best

for

cannot

order

are

was

bid

and

matched

instituted
range

and

out

best

EBS,

ask

also

no-cancel

as

the

the
the

has

both

the

best

time

limit

book,

only

broker

the

the

order

predetermined

as

a second,

in the

case

known

the

the

recomputed

limit

interdealer

of

The

be

comprises

and

is

activity.

for

bid

bands

instrument:

in the

from

order.

may

ICE,

of

the

the

not
limit

no-cancel

movements.
to

Chaboud

other

is lower

price

will
little

also

best

when

(2012),

impacted

introduction

with

is

quarter

appears

more

IPL

financial

cancelled

(ms)

the

as

price

IPL.

the

with

only

orders
In

the

exchange

the

market
limit

new

range

at the

change

altogether.

can

range

gain

may

whereby

canceled

foreign

within

incoming

be

milliseconds

a result,

the

of quotes

measure

placed

250

quotes

measure
be

for

As

the

lower

instrument

no-cancel

the

orders

canceled

placed.

the

At

of

depth

cannot

ask,

the

computing

instrument

to the

When

best

for

traded
the

orders

time.

and

frequency

for

refers

shifts

used

frequently

those

above

trading.

trading

than

period

window

the

for

level

allowing

the

by

price

bid

to a higher

by
of

showed

overhaul

in the

The

no-cancel

MQL.

exchanges

consequence

or

be

EBS

to

the

that

the

matching

range

mandated

in

the

may
near

future.
Protection

Points

Protection

points

ticks

a large

order

book.

stipulate

incoming
When

the

market

a market

maximum
order

buy

order

number

can

sweep

sweeps

of

price

through
through

levels
in

the

the

or
limit

maximum

number

of price

remainder

of

the

order.

Protection

points

were

in

the

levels

created

as

100

protection

points

Cancel

Orders

on

Exchanges

and

messages

from

client

to

fails

further
been

are

is

they

Message

Throttle

For
every

10

canceled
on

sweeps
received

markets
the

continuously

regular
client

were

sweeps.

left

Today,

as

to

in

Chapter

heartbeat,

CME

is

take

that

monitor

cancel

then

misses
to

limit

have

orders

of

cancellation

clients

their

When

assumed

The

the

heartbeat
16.

and

measure.

ensure

unable

monitor

connection

protective
to

do

positions,

not

of

execute

reducing

the

executions.

limits,

also

of order

may

orders.

be
At

problems

known

the

the

1 order

for

limits

are

throttle
with

that

the

properly

ratios,

executions.

at least

message

ensure

impacting

fill

to order

in consultation

limits

without

minimum

to execute

ICE,

basis

throttle

as

cancellations

required

a case-by-case

message

algorithmic

The

as

community.

described

such

ratio

a trader

Appropriate

the

trade

maximum

determined

as
the

throttle

example,

sweep

Limits

message
the

trading

occurrence

traders

and

buy

Protection

could

once.

of

limit

CME.

alike:

result

the

alike

with

as

are

of unwanted

dictate

by

unfilled

Disconnect

performed

incidence

The

as

pings,

clients

when

liquidity,

Exchanges

disconnected

orders

orders,

in

scheduled

orders

large

clients,

check

terminated.

limit

their

broker-dealers
their

the

at

markets

the

into

that

book

the

welcomed

entirety,

common

orders

order

and

System

on

to a previously

the

on
no

in use

market

of

in its
converted

currently

traders

or

filled

automatically

are

prices
little

not

in response

the

with

is

ticks

average

bare,

is still

marketslarge

disadvantaged
poor

order
points

futures

many

and

each
limits

operating

trader.
detect
trading

strategies.
Maximum

Quantity

Limits

Maximum

quantity

limits

finger

errors

by

enforcing

help
the

prevent
maximum

human
order

and

algorithmic

sizes

and

fat
trading

positions.
with

Maximum

algorithm

operation

At

designers,

of the

Real-Time
present,

The

real-time

check

trading

secret

prohibited
some

of

the

Price

end

in

into

consultation

account

optimal

of each

markets

value

in

the

client's

be

In

by

Knight

continuous

credit

worthiness

given

account

algorithm

extreme

liquidated
similar

withstand

the

trading

positions.

markets,

trading

is

market

may

to

incurred

-vis

new

holdings

one

vis-

limit,

equity

the

values

the

into

positioned

like

futures

market

margin

most

to

situations,

satisfy

checks

is

are

margin
performed

day.

Reasonability

Under

price

within

a predetermined

reasonability,

U.S.

exchanges,

than

10

Overall,

the

additional

Both

regulators

market

and

surveillance

expected

to be

rolled

switches

Legal

entity

Switches

identifiers

place

market

price.

On

most

limit

orders

higher

or

lower

market

orders

prices,

known

as

measures
of

described

errant

levels

the

to prevent

proliferation

deter

at price

price.

exchange

Surveillance

from

only

market

measures,

and

orders

prevailing

low

Wider

identify

not

the

crash

existing

near-term

Near-Term

Kill

flash

algorithms.

markets

from

allow

away

may

at unreasonably
many

errant

range

away

after

executing

exchanges

traders

percent

instituted

Kill

stable

entering

In

at the

take

uniquely

systems,

When

account

requirements.
only

to

critical

from

to

are

of position

the

determined

system.

trading

account.

exceeds

order

markets

clients'

Capital.

often

Validation

futures
of

are
in

trading

Position

blow-ups

of

quantities

stub

are

in crisis

robust

will

before

rule

was
from

quotes.

measures,
next,

algorithms

The

in
as

ensure
problems

catching
well

as

that

the

occur.

Measures
exchanges
and
out

are

stability
in the

near

seeking

with
future:

the

to

further

following

improve
measures,

Kill

switches

entry

are

at the

designed

following

Execution

to

automatically

block

and

unblock

order

levels:

firm

Account
Asset

class

Side

of market

Product
Exchange
At

the

flow

execution

from

corrupt.

Knight

have

An

turns

off

trading

kill

switch

Kill

switches

risk

takes

guidelines.

ability
are

be
In

directly

whenever
exceeded.

Chapter
Legal

entity

Entity

to

be

upon

for

operated

by

exchanges

a client's

while

algorithm,

algorithm's

particular

risk-tolerance

the

specific
may

down

tolerance
are

switch

venue.

switches

metrics

kill

exchange

some

shutting
risk

in

options,

kill

execution

kill

trading

product

following

API-based

by

side-of-market

The

a given

of

trading

instance,

The

switch

detection

disallows

instrument,
on

kill

be

trading
conditions

described

in

Identifiers
regulatory

be

aided

identifiers

market
Financial
Banks

to

switch

capability.

to trade

Various

determined

14.

Real-time
likely

the

are

disables

kill

financial

addition,

all

markets.

ability

into

switch

other

selling

particular

of

trading

instrument,

in

termination

execution-firm-level

asset-class

or

away

an

kill

financial

buying

may

programmed

the

allows

algorithms

Knight's

trading

in one

switch

case,

all

of

allowing

switch
bars

Capital

while

type

potentially

kill

account-level

account,

specific

whose

stopped

irregularities.
specific

level,

broker-dealer

In the

would

firm's

surveillance
by

(LEIs).

participants:
intermediaries

one

of
An

the
LEI

at
key
is

the

exchange

level

new

regulatory

initiatives:

a unique

identifier

is

assigned

further
legal
to

all

Finance
All

companies

listed

companies

Hedge

funds

Proprietary

trading

Pension

organizations

funds

Mutual

funds

Private

equity

Other

funds

entities

There

are

required

expected
to

to

obtain

LEI-based

be

an

no

LEI.

surveillance

to

thresholds
Such

all

(capitalization

configuration

trading

or

will

entities,

other)

extend

except

the

for

natural

persons.
Presently

proposed

characters,

with

using

a check

official

name

LEIs

special

the

system.

legal

investments,

the

incorporation,

the

update

information,

of LEI

The

LEI

system

Organization

for

validation
LEI

registry,

system
the

of

U.S.

have

date

will

date

of expiry,

be

administered

review

authorities

in Canada,

expressed

intention

expected

over-the-counter
extending
of LEIs
Protection

to

(OTC)
to all
for

phase

in

by

derivatives,

asset

all dealers

classes.

LEIs,

of

LEI

apply

like
The

executing

asset
credit

will

of

last

take

on

maintenance
records.
level;

the

date

of

International

ISO

international

to

pooled

country

the

of

Hong

the

if any.

The

assignment

with
for

the

by

(ISO).

regulators,

are

Investor

the

on

The

use

the

and

validated

associated

headquarters,
assignment,

an

be

manager

LEI

annual

can

be

first

the

alphanumeric

fund

to operate

jurisdictions.

the

as

will

the

20

that

the

applications,

respective

later

with

Standardization

well

LEI

the

of

digits

is expected

already

LEIs

or

of

of

of
Each

entity

address

LEI
as

composed

sequencing

character
of

are

Kong,
LEI

CFTC

OTC

derivative

LEI

in addition

in

beginning

their

with

swaps,

already

to

Australia

system

default

U.S.

The

and

class,

of

and

requires

transactions.

Investor

protection

like

U.S.

the

safeguard
in the

is one

SEC.

traders

of

The
and

the

explicit

SEC

goals

and

most

by

minimizing

investors

of

other

several

regulators,

regulators
the

seek

following

to

activities

markets:

Market

manipulation

Front-running
Market
This

crashes

section

considers

Detecting

Intentional

Whole

classes

distortions
12.

principle

is the

the

2.

The

to

through,

market
and

two
be

was

high-frequency
as

detection

market

discussed

manipulation

perspective,

should

activity

thought

screening

requires

activity

attributed

HFT

regulatory

manipulation
1. The

that,

ill

issues.

Manipulation

strategies

prove

said

as

of these

Market

of

often

Having

From

each

in

is

of such

as

Chapter

feasible

in

manipulation.

establishing

credible

market

principles:
recurrent.

performed

with

the

intent

of

manipulating

markets.
In

other

words,

regulators

need

manipulation.
harmful

Chapter

size.
for

As

Investors

markets

in the

following
regulators
manipulation

next

impact.
be

examined

to

same

market

of

consider

potentially

previous

entity

the

and

blueprints

section

can

be

buy-and-sell
can

on

and

detect

high-frequency

symmetry
of

monitor

markets
trading

the

orders

successfully

in

of systems,

by

market

screening

discussed

stability

measured

Account-level
for

intentional

trajectory.

previous

by

activity,

instance

same

following

alike

of
an

the

manipulation

market
can

the

shown

high-frequency

asymmetric

observing

by

detected

and

pattern

need

be

impact

market-manipulating

after

along

of market

market

establish

activity

can

extent

detect

regulators

of actions

12.

credibly

so,

market

Manipulation

do

activity,

sequence

of

to

To

subsequent

the

to

equal

markets

in real-time
in

asymmetric

manipulation.

for

Front-Running
Naturally,

unscrupulous

choose

to

front-run

impending

dispose

of

ensure

strategies

moniker

of

The

proposing

now

executed

exempt

interests

any

rule
brokers

first,

broker-dealer's

clients

can

effectively

desk.

the

information

The

Australian

Small

traders

capital

other

designed

to

Investment

One
the

in

liquidity

traders

in

obtained

intent

of

were

the

execution

area

where

the

same

HFT

money

on

the

the
to

prevent

has

about

disadvantage,

for

The
has

various

initiatives

to

execution

but

client's
as

few

brokers.

of

market

on

the

integrity

table

of

Securities

concerned
orders,
after

the

transparency,

Australian

response

the

pretrade

specifically

in

brokers,

however,

goal

by

front-running,

broker

key

banks'

bonanza

each

the

shut

execute

diversify

the

not

a front-running

with

the

problem

and

placed

to

nature

To

requirement

(ASIC)

the

hands

front-running.

Commission

before

of

to

own

this

forced

obligation

positions

have

is

stem

shifts

at

to establish

issues.

regulators

their

are

regulators

Australian

prices

execution

limiting

enough

detecting

creating

into

have

all

essentially

matters

flow.

their

with

however,

complicates

from

take

order

above

further

large

information,

into

clients'

banks,

function,

with

may

were

with

some

directly

prehedging

deal

banks

high-frequency

moved

detect

order-flow
In

data

to

operations,
client

of

tried

example,

sector.

was

Dodd-Frank
to

customer
at

HFT

for

using

operations

they

has

trading

banking

the

are

behalf.

the

trading

new

rule,

for

of

Instead,
a

Volcker

order-flow

whenever

regulation

proprietary

stability

possessing
clients

The

incentives

proprietary

with

the

their

minimizing

own

move.

Under

down.

their

price

problem.

brokers

and

itself

moving
they

with
market

placed

their

orders.
Predicting

Market

Following

the

research

focused

forward.

Two

flash

main

Crashes
crash
on

of

May

advanced

streams

of crash

6,

2010,

prediction
predictability

of

considerable
such
have

body

events
emerged:

of

going

1. Based

on

asymmetry

2. Based

on

abnormal

Crash

Prediction

of liquidity
trading

Based

in the

limit

order

books.

patterns.

on

Asymmetry

of Liquidity

in the

Limit

Order

Books
The

first

Prado,
6,

stream

and

2010,

(2011)
was

the

limit

eaten

the

market

To

estimate

research

O'Hara
crash

markets:
were

of

by

activity
the

volume-based

takes

developed

root

direct

orders

up

first

the

market

precipitated

into

incidence

of

bid

incoming

of

probability

in the

result

on

by

side

of

that

orders

crash,

de

the

May

liquidity

in

for

example,

E-minis,
a few

a full-blown

of informed

Lopez

observation

asymmetric

sell

Easley,

hours

the

before

crash.
the

trading,

authors

or VPIN

develop

metric:

(1)
where

and

are

respectively,

computed

Lopez

de

each

time

Prado,

time.
and

of
Tudor

may

Crash

and

in

predicting

Normal

Mandelbrot-like

evolve

coefficient
value
in most

from
market

growth

exponent

returns

on

returns

in

indicates

points
normal

to

a trending
real-life

applied

considers

of

known

market.
markets,

the

Hurst

that

VPIN
of

a patent

is
crash

in VPIN

the

in

as

Hurst

the

market

while
et

to

the

manner.

al.

exponent

trading

predictors

calibrated

market,
Florescu

by

patterns
as

describes

mean-reverting

show

Patterns

are

random

each

methodology.

normalities

0.5

where

within

ahead

for

normal

movements

completely
a

hours

Trading

parameter,

of

a few

Easley,

clocks

authors

measured

of the

Abnormal

those

The

orders,

unit.

contracts:

as

has

utilization

of literature

deviations

crashes.

Hurst

for

Based

stream

crashes

market

clock

E-mini

volume

Corporation

fees

Prediction

uses

extreme

buy

volume

contracts.
trading

Investment

and

consider

50

, then,

require

A separate

to

sell

volume-based

(2011)

corresponds

unit

by

each

O'Hara

asymmetry

capable

initiated

within

and

unit

volume-clock
extreme

volumes

fit

of
the

exponent.
where

A
the

lower

Hurst

higher

Hurst

(2012)
value

shows

that

has

been

shown

to

likely

be

to precede

that

ahead

exponent
2012f

of

days

in

Efficient

Trade

their

the

own

example

match

one

been

Market

where

March

shows

0.85.

Hurst
Aldridge

that

hours

is

2008,

reached

crashes

and

can

be

sometimes

trade

may
at

a market

sell

internalization,

broker-dealers

incidents

customer

and

are

the

bid

for

another

for

as

trades
wash

money

of

the

against
An

placing

exchange,

customer.

laundering

to

trades.

of a broker-dealer's

best

order

feasible

execute

known

consist

seeking

an

only

to

Wash

trades

are

closely

and

discouraged.

Structure

Regulatory

dimensions

two

main

lit

versus

the

table.

Swap

areas:

Following

new

dark

Execution
the

traded

surrounding
markets,

pools.

Dodd-Frank
in

and

in specialized

traders

new

market

due

section

regulation,

electronically

trade

as

structure

swap

presently

execution

span

facilities,

briefly

considers

the

swaps

are

asset

and

issues

on

Facilities

swap

will

market

such

This

computerized

have

(2012)

methodology

encouraging

Such

deemed

monitored

stocks

market

et al.
of

crash-predicting

actively

situations

it with

have

mini-crash

exponent

Matching

orders.

for

Hurst

Florescu

services

of

of

events.

of a wash

order

value

example,

services

onset

are

minimize

For

a separate

of the

Regulators

heightened

financial

identify

ahead

financial

develops
to

0.6.

a crash.

values

shown

be

about

to their

the

trading

United

falls

under

States

The

the

jurisdiction

of

facilities

(SEFs)

swap-execution

structure,
electronic

are

a new
newly

bound

to

attract

and

will

require

nature,

class

to

established
the

CFTC,
that

will

high-frequency
new

regulatory

rules.
Lit

and

Regulation
SEC

in

Dark

Pools

Alternative
1998

streamlined

Trading

Systems

definitions

(Reg
and

ATS)

applicability

introduced
of

lit

by
and

the
dark

pools.

The

trading

term

capital

exchange.
to

pool
for

The

is

of
it may

large

impact
order

and

price

in secret.

Dark

pools

dark,

dark

Large

associated

leakage.

create

their

executedtrade

As

items

are
orders.

prints

are

The

disseminated

in

venues

11,
of

lit
the

prefer

that

to

do

order

books

investors

and

not

in

the

market

information

signaling

revealed

to all

to

market

participants

are

some

desiring

directions

limited

the

in Chapter

their

orders

While

order-related

large

by

order

those

market

keeping

enticed

limit

confidence

impending

to

refer

participants.

trading

By

pooling

usually
the

discussed

many

advantages

investors

where

avoid

about

books.

lit market

induce

to

or

regulator-supervised

particularly

unregulated

order

with

may

information

limit

market

book

flow,

attracting

venue,

seeking

largely

pools

often,

others,

and

order

are

their

makers.

order

information

and

retain

disclose

lit

contain

more

engaged

disadvantage

other

books

market

all

volumes

than

trading

by
the

venue

formal

and,

exchange-like

transparency

trade

less

lit pool

observable

investors,

to a trading

matching

terms

traditional

book

refers

only

participants

when

of a given

market.
Large

investors

Market

makers

markets.
instance,

to

their

the

tend
to

quotes,

Trading

Desk

dark
than

makers
the

do

become

venues

earn

lit

less
higher

of
able

dark

pools

in
market

rents.

When

dark

dark

pools

and

trading

pools.

than

in

George

lit

(2011),

informational

revisions

longer

on

Citigroup's

higher

their

Automated

profitability

equilibrium

pools

makers'

competitive

11,
include

offer

an

dark

hide

from

offerings.

to

venues,

in

Boulatov
to

Chapter

tend

engaged
of

in

benefiting

more

Liquidnet's

pools

majority

venues

and

earn

are

discussed
of

group

research

makers

(ATD)

though

only
to

the

Examples

participants

Once

also

market

information.

market

not

According

for

Even

are

and
the

controls
lit

trading

moves

to

the

value

the

of

operations.
dark

makers
of

market

proportion

versus

market

to

in

market

pools,

lit

these

lit

makers'

rents

in the

lit markets

dark

markets,

more

until

an

by

the

is reached.

the

informational

market-makers

equilibrium

which

exceeds

equilibrium

moves
Figure

between

advantages

from

dark

13.3

illustrates

the

dark

to

in

lit

the

the

venues,

mechanism

and

lit

markets

Dark

and

is

achieved.
Figure

13.3

Market-Making

Equilibrium

between

Lit

Trading

Venues

Canadian

regulators

suitable

only

Dark
the

to large

Order

Size

opposite

investors,

have

investors

In

and

success

out

dark

trading

Threshold.

direction
the

singled

of which

amounts

the

created

pools

United

as
greater

States,

a segregated

is yet

trading

venues

than
the

dark

a cut-off

NYSE

pool

to be

determined.

proactively

tackling

issues

relating

to monitor

markets

in real

time

went

for

small

Summary
Regulators

worldwide

New

have

ideas

for

issues

emerged

such

Expanding

substantial

entity

identifier

are

market
activity

manipulation
at

improvements
initiative

End-of-Chapter
What

as

monitoring

most

1.

are

the
of

is bound

and

exchange
current

to be

helpful

level

to

and

market
will

screen
crashes.

likely

enforcement;
in the

HFT.

the

deliver
legal

process.

Questions
the

latest

key

regulatory

developments

in

the

United

States?

In the

2. What

kind

selected

U.S.

markets?

3. What

is the

interval

4. What

is the

message

5. What

is a legal

entity

1 At

time

this

book

was

weighing

the

Banking
the

United
of HFT

idea

that

to focus

only

positives.

As

traffic

to

one

a computer

bandit

venues,

his

The

and

lacking

does

funds

reluctance

to

agreement
of

The

on

FINRA.

may

the

is sent

and

obtain
so.
the

data

FINRA.

At

market
average

systems
tick

Part

manipulation
prosecutes

of

data
the

capability
present,
and
every

of

text;

trading
has

Most

of

sending

it

communications.
on

financial

over

markets:
redirecting

process.

tremendous

the

Without

annihilating

the

an

machine

communication,

in

create

provides

trader's

outright

traders

most

security

attack

accounts

of financial

do

outright
trading

ignore

computer.

of

intercept

tend

co-location

in plain

to

hearings
and

participants:

with

advantage

Senate

the

on

altogether,

of co-location

exchange's

an

Committee

data

The

secrecy

would

Senate

2, however,

and

not

U.S.

operations,

imagine

build

with

instances
SEC

to

in

it work?

unfair

aspects

to

personal

SEC

deployed

is it determined?

participants.

ensures

stability

already

co-location

offers

markets

therefore,

security
2

line

the

disallowing

market

broker-dealers,

co-location,

on

co-located

can

How

finalized,

all

in today's

co-location,

does

limit?

in Chapter

networks

to

was

negative

In

How

are

identifier?

in

private

co-located

throttle

detailed

In Australia?

mechanisms

limit?

the

on

dedicated

funds

price

market

benefit

trading

protection

savvy

communications.
a

In Canada?

co-location

technologically

enormous

Kingdom?

trading
A

gap

ban

in

on

market

worldwide.
and

is

reason
is
FINRA
forwards
eighth

commonly

cited

behind
the

the

SEC's

agency's
detects
them
case

as

implicit
potential

to

the

forwarded

SEC.
by

Chapter
Risk

14

Management

Media

coverage

tends

to focus

detailed
real
or

in

risks

of HFT
of risks
on

clearing

the

said

Chapters

and

existing

risks

facing

mitigation

of
by

strategy

risks

HFTs

investors.

development

of HFT

embedded

in

for

HFT

As

problems

recent

(BATS)

systems

alone

on

risk

exposure

practices

Nasdaq

for

measuring

All

risk

can

the

risks

in

covers

discusses
that

can

be

such

as

practices

placed

on

and

in

minimizing

Chapter

Best

16

suggests

also
best

from

Alternative

poorly

multimillion-dollar

therefore

is

ongoing
risk

detail

different

the

types

oversight
depends

broken

Regulatory

and

Credit

counterparty

and

such

executed
losses,

and
critical

Trading

incurred

management
to

ensuring

HFT

of

risks

operational

enterprises.

for

be

and

the

result

sections

for

of

15

best

risks,

and

14

(MI)

implementation.

Understanding

HFT

of HFT

following

Chapter

participants,

consideration

operational

illustrate,
may

in

The

nature

Chapter

covers

by

venues

the

impact

mitigate

incurred

trading

them.

market

to

the

of HFT.

instantaneously.

success

describe

market

specific

ways

to

Risk

Systems

embedded

16

as

paid

those

their

16

is

the

and

with

other

technology

execution

Measuring

with

of

and

(HFT)

manipulation,

attention

include

dealing

Chapter
with

no

traders.

and

minimization

practices

almost

for

associated

both

discusses

through

high-frequency

institutional

risks

14

or

risks

themselves

trading

of market

strategies

These

traders

risks

little

HFT

risks.

high-frequency

the

However,

in many

parties.

used

overstate

12.

inherent

high-frequency

risks

and

Chapter

minimize

accompanying

legal

on

down
risk
risk

into

quantification

and

of

risk.

Chapter

of

risk

exposure.

the
the

type

of

following

management

16

risk

documents
The

under

categories:

of
best

methodology
consideration.

Market

risk

Liquidity

risk

Operational

risk

Regulatory
and

and

liquidity

legal

risks

describes

are

of

risk.

Regulatory

and

Legal

Regulatory

and

legal

may

affect

Chapter

13,

surrounding
and

indicate,
such

as

(As
for

the

to

borrow

the

as

much

three-to-one

of long-term

the

typically

broker's
along

with

HFT

latest

discussed

to

the

markets
hearings

regulatory

reform,

co-location,

still

co-location

stability

controls

Senate

HFT

is

of market

in

risk
both

U.S.

13,

issues

those

generally
the

of

amount

exist.

imperative

systems.)

the

overnight
are

margin
by

the

and

trade

at

the

In

most

or

do

is considerably
broker-dealers'

to

HFTs

responsible

more

in

account,

on

of

the

not

need

cheaper

market
and

is

for
times

perspective,

blow-ups
of

leverage

discretion
HFTs

to

equities,

in their

the

ability

have

three

available

trader's

trader's

traders.

changes

subject

intraday

positions

generally

Because

From

that

HFTs

of cash

leverage

high-frequency
to

margin,

their

refers

other

borrow

greater

The

in

needs.

unsupervised

leverage.

As

to

of banning

therefore

investors.

investors

the

Leverage

overnight,

that

on

legislations

strengthened

adverse

broker-dealer.

positions

long-term

of

trading

or

margin-advancing

systems.

1 in Chapter

to

capital

as

hold

his

can

focuses

of new

beneficial

idea

leverage.

HFTs

15

Risk

comparable

example,

risk,

Chapter

16

demands

HFT

therefore

potential

for

market

below.

Chapter

reforms

As

and

specifies
secure

risk,

sections

the
of

risks

Counterparty

capital

abilities

are

in footnote

risk

ability

and

security

and

Credit

operation

ill-thought-through

computer

Credit

comprises

themselves.

discussed

the

impact.

regulatory

however,

counterparty

in

market

risk

recent

HFTs

and

Risk

the

HFT,

the

credit

discussed

mitigation

operational

that

risk,

to
than
it is

value
defaults

tightly

monitored

HFT

oversight

of
on

employee,

at

markets,
by

the

least

exchanges

in real

time.

reflects

the

risk

An

situation

broker-dealer,
Lehman

and

counterparty
$300

was

insolvency.
conditions

memory.

many

the

their

exposure

diversifying

futures
enforced

should

not

the

live

up

a counterparty
is

bankrupt.

with

The

the

collapse

most

of

spectacular

to Reuters,

close

proceedings

as

a result

of

hedge

funds

to the

brink

prominent
traders

may

creditworthiness

of

among

prevent

their

different

to

failure

custodied

According

high-frequency

tracking

to

was

in bankruptcy

pushing

by

goes

loss

equation

money

2008

in recent

The

due

fund's

and

financial

losses

October

frozen

collapse,

of
trading

of

In

monitored

in the

broker-dealer

in

failure

billion

bank's

as

the

Brothers

automatically

partners

which

configurations.

probability

example

in

practices

are

trader's

obligations.
a

best

positions

high-frequency

is

the

margin

Counterparty

their

in

the
of

similar

brokers,

brokers

to

as

and

well

trading

venues.
Market

Risk

Market

risk

is

movement
E-mini

of

the

risk

the

traded

futures

of

following

market

risk

as

market

movement

position

will

cost

position,

the

trader

loss

as

order
order

at
is

to

an
A

Even

system

will

before

to pay

any
of

close

need

in

incurring

liquidation

immediately

price

position

begins

executed.

to

adverse

long

1446.02

instantaneous

money:
trading

due

instrument.

place,

trader

or the

buy

the

takes
the

capital

financial
a

soon

of

down
the

the
the

bid-ask

spread.
The

proliferation

market

risk

strategies.
read

by

market

However,

on

the

every

The

than
bulk

tick

of

market

high-frequency
do
of

trading

carried

nanoseconds,
risks

of automated

their

human

high-frequency

has

makers

not
and

per-trade

basis

data

and

traders

face

changed
other

the

nature

intraday

and

due

react

in

considerably

trading

to their
the

of

ability

to

matter
lower

of

market

counterparts.
market

risk

management

focuses

on

the

following

four

1. First

order:

2. Second

Stop

and

4. Higher

fourth

of

to

with

risk

management

methodology

functions

deviations

takes

into

account

third-

and

fourth-order

related

to

other

instruments

is

linear

to
and

any

functional

price

of

and

computed

of

the

shape

of

the

therefore
squared

metric.

Finally,
price,

traded

from

trading

parameters.

noted

are

a second-order

kurtosis

distributional

the

is

as

previously

in price,

Volatility

referred

skewness

with

are

price.

(VaR)

methodologies

losses

of

and

value-at-risk

relationship

Stop

price

be

Short-term

Hedging

instrument.

first-order

cutouts

order:

the

the

financial

losses
Volatility

order:

order

refers

aspects:

order:

3. Third

The

key

and

VaR

returns,

the

hedging

may

is

therefore

higher-order.
Each

order

of risk

First-Order

Risk

Stop

losses

fixed

or

management
Management:

denote

variable,

absolute

maximum

for

each

trade

can

be

limits

absolute

or

trailing.

a given
for

Absolute

stop

can

to lose

relative

amount
has
14.1

The

trailing

a strategy

achieved

the

illustrates

Figure
Stop-Loss

14.1

can

lose

highest

the

difference

Difference

Thresholds

loss,

gain

within

level

however,

after
between

between

hard

the

price

Simple

the
the

was

and

trailing

(Fixed)

same
losses

can

other

be

strategy

position

the

opened.
stop
and

related

hard

at which

position
fixed

and

amount

at which

be
the

are
stop

and

stipulates

to the

and

strategy

can

outline

Variable

the

and

losses
lose

volatility

price

relative

stop

strategy.

specify

to the

next.

position

potentially

market

losses

each

Fixed

can

trade
of

stop

for

trading

each

function

variables.

opened.

position

in detail

Losses

loss

within

strategy-specific

Stop

hard

each

determined

afford

is discussed

was

price

an

strategy
Figure
losses.
Trailing

Determining
The

Stop-Loss

optimal

Parameters

stop-loss

parameter

should

satisfy

the

following

three

requirements:
1.

The

stop

winning

loss

should

limit

losing

trades

without

affecting

the

trades.

2. A stop

loss

should

not

be

triggered

losses

should

be

executed

due

to natural

market

volatility

alone.
3. Stop
The

preceding

conditions

where

requirements

for

stop

E [ Profit

=E ( Gain

translate

) + E ( Loss

| Loss

> StopLoss

+ E ( Loss

| Loss

StopLoss

* Pr( StopLoss

can
( Gain

gain,

Pr( Loss

| Loss

| Loss

During
price
of

estimated

), and

Loss

may

volatility
the

Pr( Gain
>

average

following

mathematical

of

in-sample

the

well
) +

above

stop
The

following
back-test,

the

cumulative

Pr( StopLoss
as

and

E ( Loss

volatility,

false

as

simulation,

) and
high

strategies.
is via

StopLoss
the

> StopLoss

), as

losses

> StopLoss

trigger

| Loss

StopLoss

from

periods

trading

In

| Loss
of

be

the

* Pr( Loss

loss,

into

losses:

) * Pr( Gain

Probability

immediately.

can

below

| Loss
be

the

average

stop

loss

StopLoss

natural

oscillations

simplest

adversely

values,

of

the

for

volatility

E (

market

account

the

).

performance

to

),

gain,

affecting

way

of

StopLoss

the

| Loss

losses,

probability

variable

analysis:
estimate

parameter

over

rolling

window.

can

be

Within

estimated

weighted

match

the

average

used

window,
standard

function.

time

multiplier

result

in

(better)

triangular

of

the

or

window

can

previous

step

strategy.

obtained
for

or

of the

larger

parameter

deviation,

duration

parameters

create

volatility

using

The

holding

volatility

should

the

observations

position

to

volatility

time
simple

later

of the

be

higher

weighting

Distribution
can

as

toward

exponential

each

in the

the

stop-loss

absolute

parameter:

value

of

the

stop

loss.
An

out-of-sample

back-test

stop-loss-enabled

which

Risk

cutouts
the

Management:

refer

HFT

to

systems

rules
are

in high-volatility

conditions,

optimize

capital

performance,

of some

strategies

orders

of

realized.
by

computing

index

over

estimates

volatility

conditions

cluster

or

plausible

if not

to

volatility
on

the

considered
Determining

past

will
for

or

security.

in the

profitability

of

conditions

during

value

can

be

estimate,
the

the

as

data.

they

are

into

the

periods

the

Cutouts

implied

of

traded
risk

state

of

allow

world

is

empirically

as

the

standard

or

market

backward-looking
that

the

(volatility
so

the

cutouts

derived
are
volatility

tends

can
such

from
tied

past
to

assumption

volatility,

security,

orders

the

asset

volatility

To

and

determined

time,

cutouts

metric.

place

future

volatility

volatility

takes

assuming

forward-looking
an

through

Such

better

volatility.

pass

such

of

work

in low

underlying

Alternatively,

a second-order
Volatility

different

world

persist
long

best

cutouts

window

Since
of

work

strategies

of conditions

of

risky,

HFT

others

returns

bulletproof).

(VIX)

traded

changes

volatility

are

Some

when

measuring

index

market

set

rolling

persist

variable

one

of the

certain

volatility

surrounding

volatility

states

short-term

Cutouts

halted.

strategies

Volatility

of

higher

Volatility

while

when

other

deviation

confirm

strategy.

Second-Order
Volatility

should

is
be

as

tied
the

the

options

to the

squared

cutouts

can

be

Many

trading

strategies

independent

of

of a strategy,

for

1. In the
rolling
be

2.

execution,

one

Within
as

The
time

strategy.

strategy

gains

at

time

previous

step.

sampled

at regular
the

used

successful

operation
should
week,

be

the

technique:

parameter

or

over

parameter
(better)

match

the

can

weighted

exponential

obtained

of

and

tolerated
dollar

moving

can

weighting

average

position

volatility

estimates

using

completed

round-trip

trade

is

as

and

year

personnel
may

be

volatility

amountfor
Volatility

is
the

be

obtained

returns,

the

mark-to-market

in

R t on

strategy

(negative)
better

estimates

obtained

process

should

to
side

take
of

worst-case

scenario

and

include

should

percent

Examples

the
gain

the

Ex-Ante

$15

(low)

the

the

equation.

the

risk

that

the

The

tolerable

equity
fiscal

per

risks

per

costs,

worst-case

per

above

strategy.

event

operational

million

volatility

establish

losses

of the

statistically

in step
for

in

of organizational

example,
Cutouts

high

switch

willing

and

in

volatility

losing

costs.
10

estimate

strategy

positive

performs

operation
on

volatility

intervals.

management

up

last

of realized

of

quantified

month,

Determining

can

the

a turn-on/turn-off

ends

overhead

a hard

time

risk

that

is the

strategy

as

of

regression

A median

budget

be

of the

the

conditions.
be

the

gain

Instead

estimate

significant,

can

the

t , and

side

If

deviation,

window

conditions

following

volatility

or

strategies

equation:

left-hand

3.

the

performance

volatility

volatility

triangular

on

the

the

conditions

of such

the

use

window,

using

of the

execution

may

standard

of the

R t represents

realized
the

simple

duration

following

where

time

volatility

enhance

determine

estimate

each

observations

Regress

the

back-test,

To

to limit

strategy

later

function.

desirable
To

in-sample

in certain

parameters.

conditions.

estimated

holding

be

better

volatility

window.

toward

stop-loss

it may

in adverse
optimal

the

perform

day,

such

as

losses

to

month

or

year.

Forecasting

volatility

addition

to

some

and

shown

into

tools

deviation

of

for

recent

can

is

shown

that

the

next

or

when
low,

developed

up

into

clusters

of

be

ones.
has

been

peaks

and

high-volatility
to

surrounded

for

may

Volatility

are

square
or

the
another
by

be

more

or

predict:
less

high

by

similarly

standard

be

technique,
(2011),

mean.

then

in
a

weights
metric

returns;

it has

of
been

computations.

as

if

to

the

the

last

latest

extrapolate

the

forecasting

tool

trend
is

assume
period's
volatility
into

the

called

heteroskedasticity

software

question

The

popular

same

calculate

Since

computing

it is reasonable

the

volatility

research

Aldridge

may

many

the

the

observations
by

Another

conditional

into

than

observations
from

from

calculations.

intraperiod

will

and

trending

deviations

deviationbased

volatility

trend,

square

to cluster,

one

simple:

relevant

later

squared

of volatility

quite

of volatility

more

to standard

is built
key

metric

exponential.

autoregressive
and

of

deviations

of

period's

popular

average

weigh

average

form

generalized
estimator

call

in high-volatility

by

are

basic

Alternatively,

losses

followed

volatility

can

linear

superior

observations

Yet,

of

tendency

volatility.

future.

most

researchers

the

to be
the

stop

volatility

also

straightforward

cases

simple

observations

either

volatility

Given

(a

average

be

is

usually

measuring

the

some

weighted

in some

principle.

in

volatility,

models

builds

volatility

low-volatility

returns

presents

past,

in

resulting

are

better

looser

volatility

result,

arbitrage

In

figures.

Popular

the

work

and

applications.

directly

strategies:

simple

time:

trading

management

the

gradually,

while

volatility

most

be

observations

observations,

mean)

in

As

high-volatility

low

can

valleys

observations.

of

conditions

cluster

reverses

risk

treatment

volatility
to

may

Many

in low-volatility

Forecasting

many
that

strategies

in others.

volatility-dependent

in

strategies

futures

than

tighter

important

option-based

spot

conditions

is

the

(GARCH)

packages.
is whether

known

as

may

work

the

Markov
best.

volatility

state
The

Markov

is

high

dependency,
technique

divides

historical

then

assesses

and

vice

1. Run
2.

observations
probabilities

versa.

Examine
low

cutoff

of

Specifically,

a linear

groups:

into

distribution

and

high

of

errors,

low

from

technique

of price

the

and

transition

the

regression

high

can

changes

error

on

terms;

based

on

volatility

high

to

be

used

past

states,
low

the

probability

as

price

separate

follows:
changes.

them

arbitrary

and

yet

into

two

appropriate

point.

3. Estimate

historical

changes

from

a. For
was

each

low

low

b.

Count

to high

sequential

a change

the

transition

from

state,

totals

and

in the

and

based

vice

determine

a change

from

high-volatility

express

on

the

sequential

versa:

observation,

to high,

or a stay

the

states

error
low

probabilities

whether

high

the

to low,

error

a stay

in

state.

them

in

percentage

probability

form.
4.

During

or

low.

Given

assess
the

run-time,

the

many

for

current

transition
change

volatility

level

determined

in

in the

next

is

high

step

period.

3,

Adjust

loss.

typically

used

market

exposure

If the

strategy
paper

including
and

VaR

hits
trading

the

fourth

has

effective

in

HFT

loss

that

Value-at-Risk

in

by

now

potential
of

the

the

intraday

floor

review

historical
and
of

returns

applications

and

for

moments
risk

of

HFT

entire

skewness

a fourth-order
of

Management:

properties

considers

third

and

distributional

VaR

security,

fast

metric

into

traded

very

probabilistic

is moved

notice.

the

is

intraday

further

concept

be

models.

Risk
is

VaR

the

strategy

represents

can

consideration

mark-to-market

The

of

other

(VaR)

Intraday

returns,

the

a volatility

Fourth-Order

into

ceiling

the

of

models

and

Value-at-risk

HFT.

probabilities

switching

and

takes

whether

accordingly.

applications
Third-

the

likelihood

trading

Markov

assess

for

the

VaR
of its

of
As

the

set

the

intraday

threshold,
stability

distribution

kurtosis
returns.

to

of

until
of

the

the

security

a result,

VaR

measure.
emerged

as

the

dominant

metric

in

market

risk

principal
is

management

measuresVaR

the

value

of

should

determined

as

can

scenarios
are
worst

at

returns

best,

then

the

all

percent

the

to

of the

fifth

lowest

observation

or

the

average

below

is the

the

to

If

the

100

14.2

from

sample
the

illustrates

sample

then

VaR

scenario

threshold.

among

all

For

returns

observations

ES

is

the

the

the
cutoff

100

highest,

the

returns

worst-case

return

highest,

5. Figure

if

prespecified

average

percentile.

the

the

to the

words,

lowest

lowest

1 through

other

if

order

the

ES

lower

In

scenarios

all

corresponds

is

at
are

average

concepts

VaR

14.2
(

99

= 5 Percent)

compute

steps:

The

VaR,

Percent

VaR

Computed

the

trader

on

or

risk

the

Percent)

Sample

manager

and

Return

may

use

95

Percent

Population

the

of

of VaR

ES.

Figure

To

or

from

observations

at

percent

VaR

scenario

and

ascending

VaR

specified

example,

from

determines

scenarios
a 95

arranged

and

value

measure

percent

percentile.

arranged

to the

example,
the

fifth

strategy
in

the

historical

For

two

(ES).

the

of

returns.

value

95

with
of

distribution

spans

shortfall

scenario

a particular

realized

corresponds

among

from

lowest

the

framework

expected

probability

portfolio

their

are

ES

of
or

the

VaR

negative

The

by

the

The

and
a

a percentile

are

observations

The

case

occur.

strategy

to the

return

in

be

arranged

itself

loss

probability

that

estimation.

following

1. Compute
strategy
2.

daily
either

live

Determine

strategy
3. Set

net

(after

transaction

or simulated

the

cut-off

costs)

historical

(back-tested)

returns

of the

returns.

corresponding

to

the

worst

percent

of

5 percentile

of

returns.
the

shutdown

strategy

returns,

until

cause

the

threshold
place

of

equivalent

the

the

low

strategy
return

to the

on
is

lowest

probation

in

ascertained

and

paper

trading

the

strategy

be

found

is

adjusted.
An

analytical

approximation

parameterizing
that

the

the

the

fall

represent

the

respectively.
-1.65,
average

of

all

can

be

computed

the

99

percent

values
of

the

how

true
close

the

distribution,

values

an

VaR;

integral

the

illustrates

Figure

14.3

The

the

Distribution,

95

The

applicability
distribution
this
Percent

2.33

of the

as

is

to

-1.65.

average

parametric
of

ES

as

the

VaR

the

parametric

resembles

left

the

is then

distribution

is computed

is computed

sample

to

VaR

-2.33.
the

from

assumes

where
of

the

by

Specifically,

in the

VaR

parametric

VaR

fashion.

deviation

parametric

percent

ES
to

standard

percent

parametric

14.3

Corresponds

of

distribution

Figure

of

-1.65

95

as

in a normal
5 percent

and

parametric
from

distributed

parametric

the

95

the

The

can

that

mean

while

VaR

assumes
at

The

true

distribution.

are

VaR

observations

percent

sample

observations

parametric

to

tail

of

and

observations,
computed

computed

is an
VaR

Similarly,
while

of all

the

average

function.
-2.33,

as
as

The

normal

the

the

99

distribution

approximation
depends

on

distribution.

idea.
Parametric
While
of the

the

VaR
99

Distribution

Corresponds

Percent

to

Parametric

1.65
VaR

While

the

average
the

VaR
of

many

absolute

have
normal

tails,

limitations

New

York

Einhorn,
that

VaR

that

works

article

author

The

measure

of

events.

of corporate

pictures

extreme

available

data,

analytically

from
of

VaR

of

the

Once

worst-case

distributional

comparable

recently,

VaR,

many

tail

distributions

the

tail

Jorion
a

on

faulty

extreme
mainstays

daily

VaR

has

trading,

quickly

quantitative

outfits

to

develop

is parameterized
events

functions,
were

air

floors.

extreme

severity

fraud.

been

active

trading
the

extreme
losses.

of

an

best-selling

bet

have

of

accident.

presents
to

Most

like

the

both

ES

is

and

sense

a car

metrics

and

years

many

shortcomings

have

strategists

management
on

This

Taleb,

VaR

pushes

stated
tool

false

David

Capital,

you

approach

for

risk

creates

In

2009,

risk-management

Nicholas

criticism,

parameterize
of

events

the

of choice

the

to

actively

beyond

a secret.

Greenlight

when

VaR

management

a tool

alleviate

the

lie

2,

watchdogs.

calling

distributions

been

January

and

, as

hardly

use

Nassim

that

into

as

trading

catastrophic.

fund

except

Swan

and

risk

becoming

the

quoted

all

forays

began

the

truly

on

its

time,

Despite

be

useless

all

risk

events

hedge

managers

out

extreme

the

indicates

entire

very

have

when

Black

points

made

the

senior

also

of

(2000)

of

destroy
return

can

and

measure

financial

published

relatively

among

The

article

catastrophic

security

can

methodology

founder

was

potentially

bag

VaR

location

Most
the

and

the
neither

that

that

bounds

Times
the

scenarios,

markets.

meaning

of

summarize

scenario
and

distribution

The

metrics

worst-case

banks,

fat

ES

worst

operations,

To

and

even
ever

based

can

be

though

observed

fuller

determined
no

in

on

extreme

the

sample

data.
The

parameterization

value

theory

modeling
distributions

(EVT).

functions.
belong

of
EVT

the
is an

tails

umbrella

Dacorogna
to

the

is

et
family

al.
of

performed
term
(2001)
Pareto

using
spanning
note

the

extreme

a range
that

distributions.

all

of tail

fat-tailed
A

Pareto

distribution

family

is described

as

follows:

(1)
where

the

tail

from

the

from

financial

the

return

same

tail

For

security

to

the

of

events

process

Figure

14.4

parameter

raw

the

tail

index

for

really

a is determined,
the
did

occur

of using

tail

parameterization:

Using

Tail

Index

in

the

the

Even
can

be

tail
for

vary

can

events

that

to

raw

from

may

sample.

returns
one

the

to

of

quoting

magnitude

occur,

Figure

Parameterization

varies

estimations.

estimate

that

estimated

index

high-frequency

we

extreme

needs

returns,

security.

especially

all

that

security

financial

security,

index

probability

extreme

the

data.

to another,
the

and

is

financial

institution
When

index

given

the

14.4

illustrates

Predict

Extreme

Events

1.

Sample

live

results

2. The

tail

sample
3.

return
are

arranged

index

value

return

Using

the

probabilities
According
would

observations

to
occur

either

back-test

or

order.

is estimated

distribution

on

function

of

observing

the

tail

with

the

bottom

obtained

with

with

The

index

approach
the

extreme

distribution

a probability

register

from

the

index

would

distributions

in ascending

from

5 percentile

of

the

distribution.

percent
tail

obtained

of

a probability

allows
sample

us

events

function,

0.5

to

the

while

of 0.001
deduce

distributions

are
a 7

percent

tail

index,

the

estimated.

percent
a return

return
of

11

percent.

the
of

unobserved
observed

return
returns.

Although
one,

the

the

tail

tail

index

index

approach

returns

with

theoretical

sparse

(and

it usually

be

representative

procedure

sample

based

on

process
The

is),

as

sample

manager

the

index

data

The
sample

fills

returns.

The

distribution

In

in

is

may

such

be

based

For

observed

function

may

technique

the

tail

returns.

observations

limitations.

for

distribution

bootstrapping

simulates

its

sample

extreme

parametric

observed
as

in

it has

if the

tail

actual

distribution.

works

useful,

fills

the

the

bootstrap

of the

is

observations;

of

known

Parametric

approach

not

cases,

applicable.

on

the

properties

unobserved

returns

parametric

bootstrap

follows:

distribution

of

is decomposed

into

observed
three

returns

delivered

components

using

by

a basic

the

market

model:
1. The
2.

manager's

The

with

manager's

the

3. The
The

skill,

or alpha.

return

due

to

the

manager's

portfolio

correlation

benchmark.
manager's

idiosyncratic

decomposition

is

error.

performed

using

the

return

in

standard

market

model

regression:
(2)
where

R i,t

return

on

the

manager's
of

the

is

the

manager's

chosen

money

raw

benchmark

in period

management

dependency

of

the

skill

manager's

or

period
t,

i is the

alpha,

raw

t , R x,t

and

returns

is

the

measure
i,x

on

is a

the

raw

of

the

measure

benchmark

returns.
4. Once

parameters

pools

data

of

are

and

are

generated:

one

benchmark,

and

and

estimated

were

component
returns
5. Next,

pools

return

for

sample),
to

for

estimated

be

look

as

shown

the

data

is resampled

(constant
, and

0.002

sample

may

using

and
of

in Table
as

raw
14.1

follows:

returns
.

for

i,t . 1

0.05,

equation

For

given

three

manager,

example,

respectively,
and

(2),

then
benchmarked

if
the

a.

value

errors,
b.

is

drawn

at

random

from

the

pool

of

idiosyncratic

{ i,t }.

Similarly,

a value

is drawn

at

random

from

the

pool

of

i,x R

x,t }
c. A new

sample

value

is created

as

follows:

(3)
The

sampled

eliminated
Table

variables
from

14.1

The

the

number

on

the

of Generated

process

observations
bootstrap

be
in

tails.

to be
differ

expanding

the

sample

data
original

properties

of the

6.

The

distribution

are

incorporated

now

at

from

the

the

ac

rule

of

as

sample.

their

pools

(not

is

with

times

not

extra

resampling

as

there

uncommon

sample

perspective

the

of times.

observed

repeated

a better

many

It

then

thumb,

thousands

set

is

to gain

least

the

were
for

The

the

resampled

distribution,

observations

thus

conforming

to

sample.
values

treated

into

repeated

can

process

As

original

values

new

to

Components

steps

sufficient

repeated

the

process

in

deemed
of

should

returned

Bootstrap

outlined

of times

distribution

process

are

sample).

Examples

resampling

large

and

obtained

as

tail

were

index,

through

other

VaR,

the

sample

and

parametric

values

other

risk

and

are

management

calculations.
The

parametric

bootstrap

returns'

dependence

remain

constant

Managers
are

likely

Despite

given

the

dynamic

to have

to

time.

distribution

the

time-varying

a fuller

as

This

strategies

assumption
well

does
spanning

the

of

observed

the

the

have

that

on

the

raw

manager's

alpha

to

case.

different

parametric

notion

of returns

as

not

dependencies

shortcoming,
glean

on

a benchmark

through

with

this

managers

on

relies

several

bootstrap
true

distribution

in the

sample.

be

the

asset

classes

benchmarks.
allows
of

risk
returns

To

incorporate

framework,
the

portfolio
Suleiman,

tracking

of

on

et

manager's

Tepla

al.

into

(2005)

manager's

difference

the

benchmarks

and

the

Suleiman

contemporaneous
return

Shapiro,

error

benchmark.

managers'

propose

return

(2005)

in

define

between

the

benchmark

index:

return

at time

the

VaR

analyzing

excess

tracking

of

error

manager's

his

as

return

and

the

(4)
where

R i , t is the

manager's

benchmark,

estimated
In

manager's

on

addition

the
to

tracking
VaR,

simulationbased
capital

at

risk.

of

market

qualitative

Risk

objective

returns
also

be

Hedging

can

the
then

can,

as

often

well.

in

causal

These

auxiliary

excessively

result,

be

are

of

used

and

rely

which

Carlo

values

analyses

risk

as

Monte

are

however,

on

misleading

based

on

in

realized

be

negative

returns
trading

in USD/CAD
his

is

of as

manager

risk

strategy

exposure

characteristics

dynamic.

involves

want

to go

to USD/CAD.
of

by

enters

operation.

the

two

in

For

positive

As

Passive

risk

into

a position

securities

Hedging
the

is

in

the
manager

times

for

futures

detailed

is required

financial

whose

being

long

contract

analysis
to

most

long-term

USD/CAD

negative

of another.

hedging

offset

example,

always,

maximizes

particular.

payoffs

fortuitous
the

that

matching:

that

finding
short

portfolio

risk

characteristics

the

payoff

neutralized

The

of

create

a successful

or

the

may

to

passive

insurance.
with

Hedging

riskdownside

security

security

risk

hedging

thought

of one

offset

on
are

market

Scenario

estimates,

minimizing

payoffs

main

future

market

Management:

of

while

to

include

simulations

estimation,

VaR

is return

performance.

Higher-Order

akin

Carlo

and

parameters

may

estimate

estimate

risk

with

historical

can

to

assessment

comparison

The

to

exposure.

used

VaR

t . The

models

Monte

R x,t

observations.

statistical

The

be

time

error

derivatives
can

types

at

methods

determining
models

also

t and

make

of
such

to
the
a

decision.
Dynamic

hedging

is most

potentially

overlapping,

short-term

insurance

characteristics
dynamic

hedging

repeat

themselves.

instruments

or

market

the

the

to

rise

bond

prices

Depending

upon

may

sense

make

likely

to

such

in

this

Hedging

can

Delta

hedging

Portfolio
Delta

offset

help

in

of

financial

would

offset
particular

announcement

rates,

same

rate
while

is
U.S.

announcement.
and

negative
of

may

exchange

during

the

recurring

set

Fed

USD/CAD

together

of

these

interest
the

hedging,

U.S.

bonds,

the

U.S.

tail

risk

returns

determining

interest
in

as

it

either.

described

the

details

of

operation.
management

instruments

further

can

used
by

broken

in

HFT

carrying
down

be

strategies

it over

into

applied

the

to
as

to other

following

manage

well

as

to

instruments.
categories:

hedging

Hedging

In delta

hedging

system

enters

related

instrument

near-term
hedging

U.S.

risk

which

USD/CAD

for

two

U.S.

the

systems

during

following

to

of strategies
be

the

would

portfolio

capacity

in

the

chapter

of

of

distributions

hedging

risks

returns

the

distributions
trade

find

of

short-term

set

trading

to

during

fall

extensive

High-frequency

extend

to

in order

a dynamic

market

rise

the

short-term,

the

of market

strategy

rates,

return

out

previously

example,

interest

announcements

Mapping

trading

of

objective

a particular

of

the

The

case

possible

strategies

following
are

be

a series

manage

for

behaviors

primary

of

to

In the

developed

may

For

level

likely

rate

It

conditions.

about

be

when

of

is

returns.

trading

downside

through
contracts.

contracts

may

conditions,

done

insurance-like

of trading

market

the

often

HFT
and

futures
instruments

in a particular

closes
for

positions

a single

contract
related

financial
in a liquid

stock

written
to

instrument,

on
stocks,

or

a spot

that

related

the
instrument.

commodity

stock

commodities

or

portfolio

can

commodity.
or

futures

The
be
Delta
may

be

a liquid

option.

dates

and

price

of the

In

delta

liquid

options

at-the-money,

hedging,
purchases

hedging

quantity

prices

with

underlying

system

the

Most

strike

to be

prices

with

near

close

expiration

to

the

present

instrument.
for

every

unit

a specific
is

of the

tend

of

the

quantity

determined

HF-traded

HF-traded

of

by

the

the

hedging

average

instrument

instrument,

the

instrument.

This

relative

and

the

hedging

on

the

HF-traded

changes

in

instrument:

(5)
where

P HFT,t

computed
return
unit

per
on

of

To

hedging,
be

In

quantity

after

the

trades

situation

becomes

non-execution

in either
A

instrument

using

principal

that

such

were

and

In

on

does

of

new

dynamic
,

needs

limit

to

cases

The

the

the

profitability

the
risk

hedging

Care

destroy

arises:

when
as

trading

after

hedging

instruments.

orders,

always

the

challenge

hedging

executed.
not

tick

captured.

in

only

and

specification

compromise

and

same
hedging

Q hedging,t

quantity

the

involves

solution

and

application.

the

the

volume

accurately

latest

completely

over

moving-window

may

orders,

represents

the

instrument,

relies

solution

instrument

HF-traded

demanding

instrument

market

instrument

to ensure

primary

strategy

possible

hedging

estimated,

particularly

trading

of

time-based;

are

the
is

in the

high-frequency

altogether.

units

in the

changes

hedging,t

executing

computed

hedging

continuously

instrument

the

the

hedging,

instrument,

P hedging,t

to be

for

price

and

the

need

of

average

dynamic

time,

hedging

returns

recalculated
the

of

inappropriate

the

ensure

unit

return

standardize

both

are

average

selected

time.

clocks

the

chosen

the

instruments,

to

is

of

activity

the

hedging

trades

in

should

the

be

taken

of the

HFT

strategy.
Portfolio
The
is

Hedging

main
speed:

during

challenge

of dynamic

computation
which

the

of
markets

hedging

risk-minimizing
move

and

of high-frequency
allocations
render

the

strategies
takes

time,

just-computed

allocations

stale.

fast-paced

markets,

Figure

14.5

The

hedging
as

The

problem

illustrated

becomes

a moving

in Figure

14.5

Hedging

as

High-Frequency

target

in

.
a

Perpetually

Moving

Target

To

overcome

can

deploy

the
fast

challenge

described

portfolio

in Figure

optimization

14.5

algorithms

, trading

discussed

systems
in

detail

below.
A classic
solves

portfolio
the

hedging

following

strategy,

optimization

developed

by

Markowitz

(1952),

problem:

(6)
where

x i is

vector

of

the

portfolio

expected

the

assumed

matrix
risk

to

equation

(6),

of

0.5

to

hedging
but

only

the

i, i

and

trading

simplify

the

I ],

is

operation.

repeat

returns

[1,,

securities,

returns,

would
for

security

of
of

aversion

be

state-dependent

of

returns

variance-covariance
reflecting

weight

is

the
A

solution.
the

to

an

II

coefficient
is

commonly

process

pertaining

E [ R ] is

dynamic

outlined
specific

in

market

state.
The

solution

to

equation

variance-covariance
operation

the

(6)

matrix
execution

square

of number

Several

classes

time

of financial
of

algorithms

calls

V ,
of

for

inversion

computationally

which

has

instruments
have

an

been

been

of

the

demanding

shown

to

grow

as

considered.
proposed

to

simplify

and

speed

up

setting

the

Simultaneous

optimizing

Discrete

pairwise

Genetic

algorithms

following

are

the

Markowitz
for

the

be

to

the

trading
that

updating

of

20
the

exponentially

with

the

this

of

assets.

Tsagaris,

speed

improvement

method

using

portfolio

more

are

afresh

errors
computed.
in

estimation

this

errors.

estimation

number

time

of

trading

suitable

Jasra,

and

can

eigenvalue

weights

assets

later

hardly

be
and

estimation

the

making
many

or

input

errors,

to

strategies,

variances

of the

directly

shown

discussed

forecast

weights

of recalculating

and

some

computational

portfolio

to

framework,

issues

grows

high-frequency

instead

returns

10

when

due

that

been

exceeds

are

to alleviate

involved,

show

errors

average

used

algorithm

strategies

by

in detail.

algorithm

It has

forecasts

error-correction

in addition

(2010)

algorithms

is the

portfolio

erroneous

forecast

when

can

this

of the

specification.

if the

highly

Bayesian

Still,

(1952)

The

occur

chapter,

each

framework

optimization

produce

The

describe

equations

involved.

that

optimization

Equations

inefficient

of

(DPW)

simultaneous

it may

algorithms

sections

Simultaneous

follows

weights:

programming

Critical-line

The

portfolio

equations

Nonlinear

The

optimal

be

for
Adams

improved

decomposition,

with

each

new

tick

of

data.
Nonlinear

Programming

Nonlinear

programming

commercial

is

software.

techniques

with

the

The

optimization

portfolio

allocation

weights.

gradient

technique

whereby

function

at

any

function

given

point

class

nonlinear

objective

portfolio

of

of

optimizers

algorithms

maximizing
given

Some
they
and

or

specified
of

these

popular

in

employ

a variety

of

minimizing

the

parameters
algorithms

analyze

the

slope

select

the

fastest

of

the

target

such

as

employ

objective

increasing

or

decreasing
The

path

nonlinear

estimation
Most

to

often,

of

the

in

the

nonlinear

target

maximum

programming
errors

feasible

the

algorithms

the

input

algorithms

are

is

minimum,

are

means

high-frequency

optimizer

or

equally

and

too

sensitive

variances

of

computationally

environments.

provided

respectively.

by

to

the

returns.

complex

A recent

Steuer,

Qi,

was

developed

the

to

example

and

be
of

Hirschberger

(2006).
The

Critical

LineOptimizing

The

critical

lineoptimizing

(1959)

to

facilitate

algorithm

is

fast

portfolio

allocation,

many

by

issues.

by

10,000

when

Discrete

existing

their

portfolio

Steuer,

Qi,

at least

2,000

delay

of

optimization
time

and

of

the

Trades

of

blocks

incur

pertinent

environment

their

the
in

larger-than-normal

can

put

the

set

of

some

recent
of

the

(2007),
the
by

the

algorithm
a

factor

of

simultaneously.

transaction
a serious

complexity

First,

in

environments

million-dollar

in
form

these

or

sizes
costs

on

suited

algorithms

ignore

trading

well
that
the

of

to

too
the

the

system's

of

much
liquidity
most
size.
smaller

high-frequency
profitability.

a
the

settings;

trades

the

where

a prespecified

as
in

of

loss,

consume

clips
as

perfectly

still

contemporary

strain

accuracy

be

current

blocks

and

not

result

Second,
to

occur

higher

may

can
in

power.

transactions

in

precluded

method.

(2006)

considered

environment.

algorithms

considerations

has

outperforms

the

outputs,

microsecond

system

this

Hirschberger

whatever

allocation

one

that

of

Optimization

algorithms,

trading

considered

Niedermayer

algorithm

The

Instead

addresses

and

assets

theory.

delivers

drawback

(2000)

Markowitz

implement.

security

adapting

Todd

and

(DPW)

high-frequency

from

(2000)

to

by

portfolio

optimizer

Niedermayer

Todd

Pairwise

The

frontier,

and

to

and

designed

line

companies

own

individual

critical

Markowitz

his

easy

each

efficient

According

Markowitz

for
the

the

of

comparatively

weights

commercial

algorithm

computation

and

point

on

algorithm

the

providing

portfolios

Algorithm

simple

high-frequency

solutions

is

Aldridge
the

discrete

(2010).

equally

The

that

prespecified

outputs

is

algorithm

a fast

and

optimization
developed

compromise

in

weights

optimization

discrete

are

by

between

a full-fledged

weights

fractional

complex

optimization

is

setting

portfolio

No

the

clips

allowed.

of

The

the

algorithm

follows:

1. Candidates

This

DPW

to

(DPW)

portfolio

sizes.

as

Sharpe

pairwise

weighted

machine

works

alternative

ratios
step

for

selection

and

sorted

of the

a measure

into
from

estimation

of

the

overall

the

highest

utilizes

where

each

portfolio

the

Sharpe

fact

individual

are

that

ranked

ratio
the

strategy

to the

Sharpe

lies

using

on

lowest.

ratio
the

itself

efficient

frontier.
2.

An

even

selected

number

for

of

inclusion

strategies

into

the

with

the

portfolio.

highest

Half

Sharpe

of the

selected

should

have

historically

positive

should

have

historically

negative

correlations

with

universe

of financial

instruments

is selected

ratio

characteristics,

3. After
of the

the
Sharpe

according

to

measured

as

recorded
trading
4.

all

the

number

of

the

past

fixed

the

pairs

selected
The

quotes

number

formed

each

pair

of

the

strategies

market,

the

and

on

the

are

that

seconds

or

basis
ranked

liquidity

trades

half

market.

strategies

current

or

matched

strategy

with

the

negatively
liquid

strategy

the

strategies
the

rank.

market

can
have

even

be
been

minutes

correlated
positively

with

historical

historically

most

the

correlated

basis

correlation
correlated

should
liquid

matched
market,
with

the

of
the

of

and

so

market

with
the

according

strategy
the

two

correlated

occur

with

their

with

negatively

matching

be

the

process:

positively

The

should

on

following

opposite

historically

with

liquidity

ranked

through

Furthermore,

strategy

been

have

strategies

market.

correlated

have

are

Thus,

are
the

strategies

within

market.

market

least

liquidity.

the

the

strategies

to

current

with

are

activity.

liquidity,

with

all

their

over

After

the

correlations

ratios

positively
most
on
is

until

liquid
the

matched

with

the

The

least

strategy

liquidity-based

dynamic
the
5.

liquid

Next,

common

pair

just

the

either

the
of

two

million,

for
$1

minus

0, $1

indicates

portfolio

strategies,

is

$1

pairwise

million,

sign

various

the

the
is

the

the

short

positions

and
Once

of

discrete

discrete

million,

position.

with

strategy.

where

position

be

$3

$3

million,

the

is

selected

within

the

lowest

portfolio

position

exchange,

may

of

volatility

for

foreign

optimization

combinations

the

in

$2

movements

of one

computed

market.

high-frequency

high-frequency

million,

million,

the

to idiosyncratic

example,

clip

the

conditions

strategies,

For

transactional

considered

illiquidity

with

that

is due

strategies

strategy.

correlated

ensures

correlation

than

each

of
in

by

rather

for

portfolio
sizes

matching

captured

strategy

negatively

sizes

million,

$2

where

the

volatility

for

the

each

pair

of

volatility

are

selected.
6.

The

resulting

maximum

long

follows:

exceed

the

another,

a degree
The

allocation

cumulative

gross

for

errors

each

result

in

highest

long
in
returns

The

the

and

net

the
The

constrained

strategy

position

than

given

strategy.

each

net

smaller

well
following

the

reducing

cannot

cannot

the

exceed

aggregate

position

of the

clause

ensures

suited

high-frequency

properties:

brunt

the

to

of

the

number

of

impact

of

input

strategies

in

each

decision.

of

correlation

matched

positions
the

each

in

is smaller

particularly

historical
pair

position

cumulative

avoids
by

allocation

within

for

is predetermined

that

it has

algorithm

negative

positions

is

because

estimation
portfolio

limit

executed

neutrality.

algorithm

DPW

the

all strategies.

environments
The

and

set,

of market

DWP

The

size,

subsequently

constraints

short

separately
limits

are

allocation

and

a certain

gross

portfolios

allowable

maximum
as

pair

in

strategies
per

unit

input

strategies,
both
are

of

of

risk,

the

strategies
shown
as

securities
minimum
most

to

ensures

of

variance
the

historically

is determined

during

time.
produce
the

that
will
Long
the
Sharpe

ratio

ranking

positions

phase.

for

market

one

or

times

more

algorithm

is

optimization

strategy

very

fast

algorithms.

following

savings

the

total

ranking

The

the
are

system
likely

phase

of

is

2K ,

Most

other

securities,

requiring

2 K (K

search

results

due

to

in

short

idiosyncratic

within

strategies,

requires

the

grid

values.

In

the

allowable

portfolio

MM,

and

resulting

sufficient

to

volatility

are

partially

reflect

Instead,

optimization

more

can

analysis

to

be

particularly
outperforming

well

specific

the
pair

the

2 K

instead.
portfolio

optimizes
A

compute

among

optimal

into

size

only

for

between

standard

discrete

algorithm

$2

MM,

$1

short-term

both

paid

securities,
portfolio

mix

financial

benchmarks.

not

correlation

and

which

that

only

portfolio
in

portfolio

cointegration

of

such

as

options

based

on

cointegration

characteristics

cointegration-enhanced
operations

and

are

frequent

to

risk-return

trading

MM,

iterations

returns,

suggests
be

0, $1

volatility

necessitate

(1999)

seven

to 7 2 = 49.

and

using

weight

are

MM,

of

infinity,

stability;

the

there

number

portfolio

and

portfolio

here,

the

should

the

a few

correlation

in prices

strengthen

in

only

computations

potentially,

Auxiliary

The

the

ratio

computes

strategies

MM,

limits

that

attention

added

Sharpe

algorithms

presented

Alexander

further

operation.

only

computed

strategies.

futures,

This
from,

dynamics

of

$3

long-term

typically

rebalancing.

constituent

allows

notes

ensure

from

optimization.

MM.

(1999)

the

algorithm

dimensions.

weights:

$3

portfolio

comes

in

optimization

example

computations

Alexander

trading

two

search
main

DPW

portfolio

a 2 K -dimensional

Finally,

selected

in seeking

in

other

algorithm

1) correlation

each

or

with

time:

pair

employed

strategy

two

of the

portfolio

every

each

speed

the

among

grid

comparison

strategies

correlation

The

in

in computational

number

correlations.

$2

that

events.

The

If

The

portfolios
that

are

and

of
can

tasked

the
work
with

Genetic

Algorithms

Genetic

algorithms

Bayesian

approach.

compares

the

values,

and

from

recalculates
updates

realized

security

is

All

multiple

the

Bayesian

of

confidence

on

securities.

The

estimates,
the

the

those

of the

After

the

The

traditional

optimization
variance
sample

are

An

and

is

viewed

information,
with

Multiple

or

respect

priors,

increase

to

potentially

the

accuracy

of

the

of

of

(1986),
a portfolio
lessons

previous

of

weight
zero,

measures
has

the

is

has
scaled

of

for

weights

wide

forecast

intervals

that

tight

security

for

has

the

intervals

these

parameter

security.
are

When
similar

to

optimization.

approach,

by

forecast

confidence

portfolio

that

forecast

confidence

approach

variance-covariance

interval

weight

of

is the

Jorion

and

accurate

accuracy

the

Bayesian

mean

inaccurate

mean-variance

adjusted

weights,

of a particular

obtained

confidence

portfolio

intervals

estimates

estimate

analysts,

the

width

smaller

by

genetic

portfolio

developed

all

the

wider

classic

and

cases,

variable

estimate.

forecast.

the

confidence

retrieved

instruments

many

return

previously

or

while

determined,

depending

the

with

the

interval.

errors

continuously

recalculate

random

specification,

interval,

confidence

forecasted

estimate.

associated

accuracy

been

the

of

portfolio

subsequently

investors

for

are

a
of

for

model

with

basis

In

average

be

are

so-called

time.

context

obtained

distribution

fully

the

expectations

estimates
the

the

of

weights.
not

to

portfolio

methodology

computational

in

distribution

Under

do

considered

representing
the

but

the

self-correction

the

prices

portfolio

approach,

probabilistically
priors.

on

of

trajectory

considerable
Bayesian

forecasts

the

via

Bayesian
of

Bayesian

optimal

In the

the

forecasts

performance

future

adjust,

saving

past

comparison.

the

algorithms

from

Specifically,

adjusts

the

the

learn

works

applied
as

computed
gleaned

to

follows:
on

mean-variance
both

mean

a contemporary
from

historical

and
data
(prior)

observations.
The

dispersion

the

distributions

analyzed

of

time.

mean-variance
is

an

the

optimization

the

is

the

next

number

of

true

the

and

are

variance

and

following

equation

(7)

from

time

t 1 to

estimate

for

security

for
be

expected
used

t + 1, are

mean

observations

time
,

and

the

computed

t ,
the

variance

mean-variance
as

follows:

estimates:

in the

the

i,
return

in

of

collected

return

to

of the

mean

portfolio

period

precision

the

observations

estimators

security

for

v is the

sample

at

time

t , and

BS

is

0)

shrinkage

factor

for

the

corresponds

diffuse

Despite
HFT

mean:

the

can

high-frequency

Figure

correlations

observed

Trade

data

higher,
quote

frequency,

information

effective

low

correlations

on

of

zero

precision

(v

correlations
of tick
unavailable

45

in

200
when

decrease

percent

data:

500

to

also

of

to

quote
quote
trade

the

the
ETF

with

data.

point

low,
and
ms.

two

just

to zero

Quote

correlations
every

45
with

sampling.

may

daily

close

when
as

data
are

seconds.
sampling
Relatively

illuminate
reflects

Index

percent

decrease

200-ms

The

empirical
MSCI

of

financial

with

dramatically

likely

feature

iShares

sampled

data
data

any

and

hedging,

following

between

seconds,

percent

due

high-frequency

particularly

increases

correlations
about

S&P

are

30

of

illustrates

the

every

around

to

informativeness

very

14.6

frequency

higher

case

complexities

correlations

is sampled

sampling

The

be

data:

instruments.

much

The

estimates.

computational

hedging

(EFA).

to

completely

the

R p,t

return

shrinkage

N is the

more

of

average

individual

where

If

of

as

optimization

Bayes-Stein
of

distributions

shrinks

with

and

the

market

relative
makers'

correlation

of

the

Standard

&

Poor's

(EFA)

often

reaches

Figure

14.6

Correlations

Source

: Aldridge

Liquidity

Risk

Liquidity

risk

intraday

trading

may

a timely

manner
is

specific

firm's
at

Levels
on

the

the

cases,

delay

the

liquidity

of liquidity

systems
Capital

in

trade
its

from

market

the

normal

the

the

latter

either

being

the

can

execution
For

(LTCM)

result
and

cause

example,
in 1998

the
can

out

relative

to

for
with

to

that

in

suggests

the

to finance

and

transact

further
and

balance

the

shortfall

or borrowing.

in minor
can

close

available

willing

risk

liquidation

in

to instrument

(2006)

inability

to

associated

instrument

liquidity

through

risk

risk

risk

positions

liquidity

liquidity

Bervas

trading

Liquidity

inability

market

participants

consideration.

extreme.

Management

vary

of

liquidity

in

Index

or hedge

The

of

higher

between

sheet

levels

the

distinction

balance

low

market

the

in the

to

prices.

the

under

risk,

to unwind

lower

instrument

In mild

MSCI

during

liquidation.

The

the

liquidity

end-of-day

market

due

traders

inability

current

number

sheet

the

potential

instrument,

depend

iShares

Data.

high-frequency

during

size.

instrument.

and

in High-Frequency

normally

position

ETF

percent.

affect
or

the

the

65

500

(2010).

measures

positions

(S&P)

be

price

slippages

collapses
collapse
attributed

due
of

of

to

market

Long-Term
to

the

firm's

inability
To

to promptly

properly

to take

including

the

execution.
estimate
wildly

in

(2002)

estimates

risk

selected
(2006)

risk

account

little

high-volatility

long

liquidity

that

securities

the

any

volatility,

the

liquidation

Bangia

et

in

can

in CAC40

proposes

with
and

accounted

for

May

reach

it

liquidation

stable

risk

risk

portfolio,

are

positions

liquidity

portfolio

regimes.

USD/THB

of

associated

costs

with

that

exposure

all potential
costs

periods

document
risk

holdings.

liquidation

during

example,

Bervas

liquidity

opportunity

during

on

the
into

While

market

its

assess

necessary

vary

offload

delays

al.

to

costs

can

(1999),

for

and

50

in

easy

percent

1997,

over

costs,

are

17

is

of

Le

the

Saout

percent

of

total

stocks.

following

measure

of liquidity

risk:

(7)
where

VaR

this

is

chapter,

standard

is

spread

risk

the
of

Both
data

or from

Kyle's

computed

the

trade

expected

bid-ask
to

and
the

can

VaR

estimation

of

bid-ask

be

previously

spread,

and

desired

(1984)

the

discussed

spread,

the
S

Roll

measure,
through

value-at-risk

mean

corresponding

estimation.

the

market

deviation

coefficient

Using

the

is

the

percent
estimated

in
is

the

confidence
of

the

either

VaR

from

raw

model.
liquidity

adjustment

the

mean

using

OLS

and

can

standard

be

similarly

deviation

of

volume:

(8)
where

and

are

estimated

regression

following

Kyle

(1985):
(9)
P
and

t is

the

NVOL

in period

change
t is the

indicates

adjustment

market

price

difference

due

between

to

the

market
buy

and

impact
sell

of

market

orders,
depths

t.

Hasbrouck
best

in

(2005)
the
to VaR,

finds
impact
the

that
of
Amihud

the

Amihud

volume
(2002)

on

(2002)
prices.

adjustment

illiquidity
Similar
can

measure
to

be

Kyle's

applied

as

follows:
(10)
where

and

(2002)

are

illiquidity

executed

trade

executed

within

The

time

multiple

the

cases,

the

executed,

is

the

the

t , and

in

of the

number

relative

v d , t is the

for

they

for

the

liquid

HFT

for

simultaneous

the

less

orders,

as

the

multiasset

calls

limit

orders

orders

period

strategy

strategy

limit

rd ,t

applies

via

the

deviation

Amihud

of

trades

price

change

trade

quantity

d.

instruments

compromise

t,

trade

also

When

standard
, D t is

period

trade
risk

and

d during

liquidity

positions.

mean

measure

during

following

the

may

be

entering

acquisition

liquid

difficult

upon

instruments
to acquire.

illiquid

instruments

instruments

are

of

are

may
In

sent

such

first;

if

placed.

Summary
Competent

risk

management

and

often

enhances

The

risk

management

aspects

of

protects

overall

performance

framework

HFT

deployed

reduces

of high-frequency

of HFT

operation,

capital,

should

including

take

HFT

risk

strategies.
into

account

suppliers

and

all
the

government.
End-of-Chapter
1.

What

are

Questions
the

key

types

of

risk

faced

by

a high-frequency

trading

operation?
2. How
3.

to measure

What

is

trading
4.

the

and
credit

mitigate
and

market

risk?

counterparty

risk

from

high-frequency

perspective?

What

are

the

key

problems

in

high-frequency

portfolio

optimization?
5. What
1 The

is liquidity
hat

parameters
to comprising

risk?

notation
were
the

How

on
estimated

true

to measure

variables,
from

distribution

as

it?
in

a sample
values.

and

, denotes

distribution,

as

that

the

opposed

CHAPTER

15

Minimizing

Market

Algorithmic

execution,

routing,

refers

used
An

to

order

and

transmit

at

certain

the

highs

conditions.
aversion,

range
this

in-house
algorithm

the

trader's

by

Algo

to some

promising

unique

negotiate

preferred

on

human

From

the

portfolio
natural
Why

point

terms

by

detail

the

can

be
from

basis

for

the

for

off-the-shelf

savings

from

commission

delivered

by

best

execution

human-driven
for

market
clients.

practice

chosen

algorithms

algos

competed

the

trader's

the

benchmark.

from

pinpoint

for

broker

of

management,

Execution

execution

to

in

client

highs

order

and

Algorithmic

flow

lows,

and

execution

developing

by
to

builds

an

automated

the

quantitative

approach.

market

Execution

ability

cost

price

the

execution

provide

executional

best

produces

on

in

execution

time

optimality

benchmark

a one-off

the

of

the

algo

based

purchased

may

period
delivering

discussed

The

order.

customer's

prespecified

the

on

brokers

the

peak,

be

order

an

pinpointing

to

trader,

naturally

execute

a good

may

book.

of

its

time,

features,

portion

decades,

at

smart

methodologies

a given

precisely

state,

brokers

evolved

For

human-free

other

The
a

is

conditions

or

execute

according

the

and

during

of

or licensed

for

execution

the

a period

a buy-side

broker.

relative

practice.

of

of

provider,

exchange

algorithm

difficulties

market

part

parcel

available
price

of

execution
computer

to

the

optimality

a
of

the

when

concurrent

and

remainder

price

within

algo

consistently

improvement

The

trader,

way

would

the

as

programmatic

optimal

lowest
sell

known
of

algo

the

price

in

set

the

Given

and

built

the

execution.

or

to

execution

buy

risk

also

determine

ideal

lows

Impact

of

best

classical
execution

finance

and

algorithms

exist

to smooth

out

imperfections.
Algorithms?

algorithms
algos

view

help

have
traders

become
accumulate

essential
or

for
liquidate

all

investors,
large

positions

as

by

breaking

visibility

up

of

deploying
to

orders.
limit

execute.

often

associated
to

(2007),
level

save

of

execution

be

ways.
fees,
of

order,

fail

immediately,

the

avoiding

algo
the

timing

of the

market

by

and

an

and

depend

may

Other
parcels

also

impact

passive

yet

to execute.

cost

Ferstenberg

algorithm

spread,

depth,

comprise

market

produces:

order

lower

opportunity

Russell,

delivered

as

the

orders

costs

spread,

Engle,

fail

to

algos

substantially

Execution

limit

to

costs

to

bid-ask

orders

the

the

almost

shown

passive

to

net

with
largest

minimized

with

costs,

resulting

costs,

and

include

execution

the

the

risk

market

on.

impact

on
orders

aggravate

design
and

(VaR)

below

maximum

compare

Chriss

the

(2000)
the

aspects

size

of

the

each

obtained

execution

risk

in

price

designed

of

higher

used

but

can

execution

MI,

transaction

execution

order

contain

costs

nonexecution

higher

algo

of

to

of

execution,

expense

spread,

the

may

slices,

and

execution

costs

limit.

the

markets

convenient

the

execution

performance

pricing

at

metrics

measure

The

consider

in algorithmic

the

risk
the

proposed

efficient
asset

crossing

of

may

algorithms.

orders

Other

variability

certain

the

traders

component

from

so

costs,

of

risk

value-at-risk

various

been

by

and

markets,

when

resubmitted

According

capital

in the

impact

market.

different

the

market

costs.

associated

provides

then

aggressiveness

such

addition

capital

are

have

order

relative

Like

orders

to the

example,

algo,

parcel

To

of the

a few.

investor

of the

be

many

nonexecution

whenever

can

cancel

broker-dealer

name
for

costs

off

in

with

reducing

picked

closer

and

and

being

orders

costs,

exchange

pieces,

avoid

algorithms

execution

In

orders

at a price

(MI),

into

To

The

Different

the

orders

of
concept

frontier

model

algorithms,

of

an

developed

(CAPM),

graphical
algorithms

several

the

efficient

unit

of

trading

in the

framework

efficient

trading

representation
per

Almgren

of
risk

frontier.
of

the

frontier

performance

incurred

and

by

of
each

algorithm.

A sample

efficient

Analytically,

it can

Figure

Illustration

15.1

be

trading

described
of the

as

frontier

is shown

in equation

(1):

Efficient

Trading

in Figure

15.1

Frontier

(1)
where
is a measure
number
Cost

of ticks

is

expected

) is

the

the

the

from

the

aggregate
impact,

for

=0

As

order

child

is placed.

cost,

orders

the

of

including

the

strategy

associated

with

all

the

child

orders

aversion,

specific

a trader
aversion

who
level

to

the

does

not

=0.5

trader.
care

A
about

indicates

risk
the
fairly

Algorithms
algorithms

differences

cost-effective

risk

child

counting

trader.

Order-Routing
Order-routing

risk

example,

execution

level
risk

indicates

each

all

level

of

risk.

risk-averse

market

aggressiveness

degree

for

expected

cumulative

level

execution

profile.

aggressiveness,

at aggressiveness

aversion

issues

away

at the

placed
is

order

market

executed
Risk

of

are

between
execution

such,

the

designed

to seamlessly

securities
schedule

order-routing

markets
that

would

algorithms

navigate

and
fit

deliver
the

target

various

investors

investors'
the

a
risk

following

objectives:
Minimize

execution

Obtain

best

costs

price

Maximize

execution

Maximize

trading

Minimize
To

spreads,
market

to

and

help

vice

(more

on

maximize

the

order

into

according

to

preferences.

series

the

Large

large

trade

ensures

helps

thwart

the
and

minimize
of

trading
or

minimal

can

strategy

traders

attempting

clients

desiring

or

be

of

the

order

to

infer

the

slice

to funds

footprint

outside

all

investor's

important

by

To

orders,

and

Minimal

with

algos

child

particularly

to

impact.
the

advances

the
and

venues

footprint,

capacity.

detection

prices

trade

parcels

price,

to ensure

seek

minimal

scientific
size

best

market

for

algos

trade

smaller

latest

positions

speed

conditions,

and

obtain

higher

bid-ask

subsequent

forecasting

with

picking

low

and

To

market

and
of

slippage

execution

and

structure,

periods

later).

participation
size

pick

eliminate

times

venues

fee

short-term

To

trading

the

can

perform
during

market

reduce

these

processed

appropriate

with

or

market

appropriate

trade

are

present

maximize

reduce

algorithms

versa.

capture

can

execute

impact

orders

select

selection

time

sophisticated

the

algorithms

Venue

right

sell

footprint

costs,

conditions.
the

size

trade

minimize

speed

of

parties,

informational

the
and

content

of

orders.
Performance
some
at

the

low,

of

benchmarks.
end
and

metrics,

of

such

the

as

trading

The
his

is typically
may

day,

an

prices,

the

several

commonly

benchmarks

benchmarks.

algorithms

A benchmark

close

Algorithmic

building

execution

daily

forecasting

daily

and

the

close

is

models

be

the

average

daily

closing
of

open,
used

or

easy

on

daily

the

are

daily

open,

high,
complex

algorithms.
most

as

to

observed

more

reference
data,

relative

price

other,

execution

close

an

measured

common
for

is

any

often

algo
investor

the

case

with

low-frequency

nearly

always

forecasts
that

performed

usually

of their

execution

the

the

the

closing

the

only

way

is

algorithm

price

is

thorough

HFT

modeling.

difficult

other

common

algorithms

The

following

sections

as

models.

algo
each

the

portion

difficult
not

to

at all known

ahead

of

time

is

forecasting
in

Chapters

the

complexities

traders

often

performance
of

by

provider.

discussed

consequence,

discuss

to pay

are

of

algos

demand

Short-term

models

suitable

high

them

understanding
As

developed

notoriously

prices

is

a result,

willing

approximate

trading

in

analysis
the

As

to the

price-forecasting

requires

are

are

closing
to

and

too.

close
who

gains

data

prices,

closes,

daily

models,

high-frequency

and

execution

the

benchmarks,

short-term

utilizes
8-11,

other

and

deploy

of

Daily

closing
daily

algo-induced

Unlike

in advance,

future

closebased

betterment

achieve.

daily

outperform

of daily

Yet

modelers.

on

predict

consistently

traders

to

quantitative

of
deploy

benchmarks.

objectives

of

algo

in detail.

Minimize

Execution

Trading

costs

Broker

Costs

comprise

several

commissions,

Exchange

both

major

fixed

and

components:
variable

fees

Taxes
Bid-ask

spread

Slippage
Opportunity
Market
Obtain

impact
Best

The

core

Due

to

difficult

Price
principle

the
to

required
The

cost

best

Consider

of

natural
predict,

for

the

price
the

the

price
and

best

price

moves,
advanced

trading

the

direction

short-term

is

buy
of

low,
the

sell

price

forecasting

high.
can

models

be
are

purpose.
execution

following

is further
example.

complicated
It is 9:30

a.m.,

by
and

additional
a client

factors.
wants

to

buy

10,000

4:00

p.m.

shares
later

following

that

IBM

at

day.

most

The

at

the

closing

naturally

arising

price

recorded

issues

create

at
the

questions:

Given

the

market

4:00

p.m.

The

client's

that

normal

uncertainty,

desired

client's

how

buy

offer

side

trade
(bid

for

market

impact,

Other

and

price

in

the

the

child

side).

The

child

execution

price

be

at

be

bigger

broken

than

into

smaller

split?

is broken

into

smaller

child

orders,

executed?

deplete

some

liquidity

orders.

may

to trade

adverse

considerably

of

gaps

Can

the

liquidity

will

this

lead

on

to

effect,

the

adverse

known

as

or minimized?

participants

an

be

will

eliminated

decide

order

resulting

subsequent
be

order

orders

trade)

is

the

order

the

(sell

market

client,

should

are

prices

will

size

Should

10,000-share

frequently

Each

execution

size.

If so,

If the

what

day?

trading

parcels?

how

of

observe

in the

same

direction.

trading

footprint

direction,

Can

further

the

trading

of

the

moving

the

footprint

be

minimized?
Maximize
Fast

Execution

execution

orders

can

maximize
investors
send
orders,

process
selecting
Figure

executed

the

speed
poll

As
fewest
of
the

15.2

most
of

various
to

however,

the

capture

be

orders

conditions.
with

helps

may

their

Speed

the

are
a result,
limit

polling
proper
Maximizing

current

market

rapidly

in the

most

execution

of

market

exchanges

for

their

exchange
executed
limit

orders
multiple

with

the

most

orders

are

available.

best
Figure

Execution

for

a given
Speed

for

liquid

markets.

available
most

liquidity
in

executed
15.2
their

order

Market

orders,

rapidly

exchanges

exchange

conditions.

the
on

therefore,
liquidity,

and

first.

Limit

least
the

illustrates
liquidity
or order

To

liquid
markets

a sample
levels
slice.

and

In

the

example

Exchange

(shares,

liquidity

500

units

market

ensures
the

or

top-of-the

on

equal

Exchange
trader

place

2,000

aggregate

exchange

size
next

of

would
with

the

1,

or

the

Exchange

place
next

with

units

limit

limit

lowest

or

to

an
or

at the
only

off

placing

place

equal

buy
3, as

limit

on
most

the

order
smaller

exchange

slightly

moves

number

less,

then
and

orders,
that

order

place

his

order

proceed

to

there.
would

exchange

has

at

best

aggregate

in

bid-side

however,

at

the

exchange:

then

liquidity

of

2,

top-of-the-book

available
buy

Exchange
liquid

would

available

orders
a

move

next

trader

execute

buy

the

The

against

with

trader

2 and

liquidity

to

units.

trade

order

turn

2,000

liquidity

footprint.

Exchange

for

available

available

not

units

Exchange

order

liquidity

trading

bid);

footprint,

an

exchanges:

2,000

best

Exchange

does

or no

would

to

trading
to

three

3 has

Placing

order

of

at the

top-of-the-book

to

desiring

a limit

trader

matching

is

3 and

his
go

are

liquidity

Exchange

book

little

and

there.

market

, there

available

units

this

Exchange

on,

first

trader

example,

so

would

the

15.2

bid-side

minimize

leaving

order

liquidity

the

exhausting

market
our

fewer

that

Figure

units;

To

orders

market,

After

and

of 3,000

sell

the

in

available

only.

3,000

than

has

contracts,

has

for

shown

the

the

lowest

bid.

Exchange
limit

first

The

1,
buy

the

orders

available
may

at

or

the

may

currently

not

the

underlying

limit

orders

The

In

the

the

to

a buy

limit

order

for
of
are

are

most

limit

order

at

limit

2,

orders.
is this:

and

place

place

numerous.

market

Such

probability

impact

trader

Exchange

exchange

fewest,

highest

minimal

buy

the

orders

the

the

maximizing

minimize

the

markets

are

discussed

trading

following

explained

as

the

trader's

true

every

order

Other

or

process

of fast

execution.

algorithm

can

algorithm.

of

comparable

at different

exchanges

reveals

the

an

the

the

In

the

such

sizes

minimizes
information

as
As

views
on

can

it

reveals

result,

of the

these

trader.

views

content

the

placing

at

as

beyond

situations,

available
the

disturbance

capital.

information

in

disturbance

signal

trade

be

registered

of the

current
to

the
order.

least

causes

trader's

desire

to

disturbance

is a credible

about

knowing

MI

behind

to

may

placing

the

intuition

order

information

the

exact

committed

necessarily

and

The

every

sizes

quotes,

footprint,

5. The

beliefs

action
of

speed,

participants

without

orders

execution

order.

follows:

carries

market

observed

an

in Chapter

be

offer

the

Footprint

to

well,

have
as

point,

selection

orders

orders

this

exchange

limit

limit

is known

addition

used

the

At

place

for

wherever

process

book.
to

rationale

that

Minimize

the

competitive

wherever

ensures

of

proceed

most

The

orders

top

the

child

best

resulting

change

associated

with

bid

or

in market
each

order

slice.
Maximize
The

Trading

ability

to

deploying

Size

process

sizable

pension

fund

securities

capital

needs

without
reallocate

trading

the

limit

orders

seeking

in
to

trading

in

to

be

their

large

execute

to

trader

large

is

may

For

and

use

sell

order

may

investors
a

large

large

quantities

of

cost

in

to

To

a combination
order.

to

example,

positions.

individual
buy

critical

additional

fund's

each
a

buy

much

pension

processing

volume
strategies.

able

incurring

successfully
size,

large

maximize
of

To
first

order

do

market
so,

exhaust

the
and
trader
the

top-of-the-book
polling

ask

for

orders

the

best

matching
beginning

trader

switch

placing

the

beginning
the

with

all

smaller
with

least

than
the

the

rotating

bid-liquidity.

the

ask

liquidity

at

market

buy

at

each

bid-side
all

among
Figure

sequentially

Subsequently,

the

orders

by

placing

one.

increase

buy
and

best

liquid

and

liquid

markets

and

most

orders

increasing

accessible

available,

top-of-the-book

the
of

on

size

to limit

best

direction

ask
or

exchange,
may

liquidity

liquidity

the
the

15.3

the
by

exchanges,
exchanges

illustrates

in
such

strategy.
Figure

15.3

Maximizing

Trading

Size

Implementation

of

Execution

Algorithms

Most

researchers

develop

execution

algorithms

in

the

following

sequence:
1.

Researchers

research

in

explore
the

implementation.
research,

arbitraged

in the

algorithm

of

Some

available

the

area

may

published
optimal

traders
fearing

markets.
result

and

execution
may

that
In

in an

not-yet-published

reality,
algo

be
all

skeptical
known

a change
with

algorithm

an

of

academic
design
using

research

and
publicly

has

been

in parameterization
entirely

different,

of
yet

still

valuable,
2.

output.

The

such

researchers

as

MatLab

programming
3. The

algorithm

predictions

own

orders.
If

the

and

or

and

FAST,

and

Slicing

large

orders

(1995),

for

example,

the

the

all

trading

expensive

and

are

then

to

execution

can

enabled

shows

the

to

to

faster

by

assumptions
the

algorithms,

execution

algorithm

communicate

in

such

at

once,

difficult,

Gatheral,
be

. The

broken
first

making

as

slice

of

typical

account

schedule

is

moved

real

time

FIX,

ITCH,

at a time
Schied,

down

answer

the

following

questions:

Figure

15.4

Layers

of Algorithmic
and

for

and
into

three

called

the

Slynko

about

into
using
OUCH,

Slynko
distinct
macro

(2012)

Lakonishok
trade

60

smaller

a certain

Execution

and

execution
The

over

Chan

institutional

simultaneous
impossible.

layer,

Schied

if

it would

if not

one

research

that

15.4

Gatheral,

utilizing

satisfactory

languages

Figure

Source:

data

generated

outcome,

code

Java.

tick

is imperative:

volume,

executed

According

in

languages

their

or optimized

historical

-sending

econometric

transition

movement

risk
is

in

like.

executed
daily

C++

results

and
it

algorithm
a result,

on

price

step

where

quote-receiving

were

like

about

cost,

production,

as

is tested

previous

price,

the

R and,

languages

and

4.

model

percent
of the

child
time

trader,

of
order

orders

period.

(2012),
layers,

size

algorithmic
as

shown

allows

us

in
to

How

to slice

order

slicing

When

the

algo

should

be

day
In

what

order

order:

What

is the

general

rule

behind

the

algo's

at what

time

of the

should

each

child

mechanism?

size

should
algo
the

trade:

initiate
algo

How
its

frequently

child

and

trades?

should

trade:

How

should

trade:

large

be?

For

how

algo?
The

the

long

the

When

does

second

defines

algo
the

layer,

trader

order

as

can

properties

or

be

for

is responsible

a limit

is

the

horizon

of

the

stop?

which

additional

micro

algo

What

for

market

described

each

child

deciding

order,

as

the

order.

In

whether

and,

for

decides

to

to

limit

micro

trader,

particular,

execute

orders,

the

the

what

child

price

to

set.
Finally,

the

child

orders.

Over

the

smart

past

developing

two

macro

all

tailored

strategy

groups.

trading:

it

average

price

is

can

(VWAP)

execution.

As

At

is

such,

conditions.

outperform

solutions

on

first
static

market
static

is an

example

determined

given

delta

strategies,

strategies

since
and

perform

is an

dynamic

static
well,

strategies
but

only

router

and

can

dynamic
ahead

of

volume-weighted

strategy.
during

example
may

In contrast,
the

on

course

of a

strategies
do

of

contemporary

seem

under

in

conditions.

static

depend

the

decisions

order

refined

hedge

The

determined

of a static

strategies

send

made

market

into

conditions.

dynamic

conditions

the

and

to

been

smart

is completely

market

glance,

has

the

classified

past

dynamic

strategies

be

venues

execution.

and
to

strategy

A simple

best

trader,

precision

which

progress
for

great

based

strategy

strategy.

micro

A static

a dynamic

certain

significant
solutions

the

with

execution

current

years,

trader,

Optimal

market

router

mathematical

of the
be

order

dynamic

to

always

respond
not.

In

to

reality,

specific

market

assumptions.
The

performance

of

both

static

and

dynamic

strategies

is

often

compared

using

benchmarks,

for

example,

the

following

simple

prices

per

traded

metrics:
Average

realized

received

under

price

compares

different

the

actual

unit

algos:

(2)
where

P j is the

size

of child

order

Pretrade
order

realized

P 0

price

temporary

is the

total

size

V =

trading

trading

time-weighted
VWAP,

be

Table

discussed
Kissell

grouped

category

Pretrade

V j is

the

time

the

other

Table

Order
Intratrade

15.1

benchmarks
Execution
Posttrade

this

known
Benchmarks

such

others.
V Daily

is

the
of

liquidity.
performance
algos,
of

order

execution

of

such

volume

various
of this

pretrade,
classification.

ahead

In

to

comparison

sections

categories:

summarizes

allows

percentage

(2005),

broad

than

at t

used

liquidity.

evaluating

and

P recorded

execution

(TWAP),

P t

P t on

algorithms

of available
for

regress

, where

day,

the
have

of

price

of

V/V Daily

common

Glantz

three

the

better

shortfall,

and

(2005)

dependency

perform
size

after

stream,

to available

in subsequent

into

includes
15.1

the

security

et al.

particular

implementations

to

posttrade.

relative

advantage

price

trading

when

benchmarks

include

by

comparison

trade
a

the

Then,

may

to take

average

benchmarks,

can

V j allows

on

common
algos

According

j , and

at

of

, Almgren

significant.

volumes

volume

addition,

MI,

order

prevailing

price

t post

algorithms

ability

execution

child

induced

time

volume-adjusted

algorithms'
In

the

some

Similarly,

or

price

the

P j,post

statistically

large

conditions,

is

identify

P j,post

trade

process

market

effects,

pinpoint

to be

Total

the

P j,post

To

t , and

post

is

liquidity

disappeared.

ceases

slice

placed.

Posttrade

on

for

j.

price
j was

price

of execution,

as

(POV),

intraday

price

chapter.
benchmarks
intratrade,
The

and

pretrade
such

as:

Decision

price

Previous

close

Opening

price

Arrival

VWAP
TWAP
OHLC

Kissel

Trading

and

Glantz

decision
decided

Previous

day's

close

working

with

traders
Daily

open

the

was

price

at

which

advantageous
price,

for

which

can

the

trader

or

portfolio

trading.

be

used

as

a benchmark

for

daily.

price.

Arrival

price,

broker

received

the

intratrade

price

the

determined

TWAP,

also

that

was

includes

the

using

on

of daily

posttrade

when

open,

category

following

intraday

determined

average

prevalent

the

executing

order.

category

VWAP,

The

(2005)

price,

manager

The

close

price

Source:

The

Future

prices.

the

basis

high,

of prices

low,

includes

benchmarks:

the

and

future

throughout

close

prices

close,

the

the

day.

(OHLC)
price

not

known

a large

order

in advance.
TWAP
TWAP
into

attempts
equally

time

is

sized

intervals.

the

order
the

to

conceal

parcels,

the

order

which

are

Mathematically,

every

then

TWAP

predetermined

arithmetic

flow

average

of

by

breaking

sent

out

executes

at equally

a fixed

unit

of time.

The

prices

sampled

at

portion

resulting
the

spaced
1/ T of

TWAP

price

regular

unit

time

. When

trader

the

trader

intervals:
(3)
The

TWAP

chooses
also
to

algorithm

to

needs

size
order

S/N
of

execute
to

execute,
is
size

is

a large

decide

and

the

sent

to

S is

illustrated

on

the

order

the

total

in
of

total

processed.

size

The

N of

time

every
total

15.5

S using

number

execution
market

Figure

TWAP,

child

T . Next,

T/N

an

seconds,

number

orders

of

or

order
until

slices

slices

slice
the

and

of

entire
the

execution
the

time

traded

T are

security.

variation

in

beginning

volume
of

child

does

These

not

enough

so

reasonable

represented

as

in

to

Figure

of

large

include

market
other

depth

the

market,

yet

can

each

be

the
The

that

each

large

executed

order

child

at

variables.
so

order

to

historical

enough

TWAP

, with

specific

small

move

resulting

15.6

day,

slices

entire

The

may

host

select

the

T.

characteristics

trading

significantly

that

time

the
and

is

using

characteristics

execution,

objective

order

determined

throughout

overarching

frequent

best

flow

order

or

within
can

drawn

be

as

an

arrow.
Figure

15.5

TWAP

Process

VWAP
The

VWAP

algorithm

methodologies.
large

order

the

trading

trading
larger

in

such

pool

determine

principle
a way

volume

volume

cost-effective
To

The

is currently

is
of

is

matching

of

of VWAP
that

higher,

lower.

one

Higher
orders

most

popular

execution

is straightforward:

VWAP
and

the

child
child

trading
and

break

orders

orders

are

volume
result

are

is
in

larger
smaller

likely
faster

to
and

up

when
when
provide
more

execution.
the

execution

schedule,

the

VWAP

algorithm

uses

map
one

of historical
shown

using

for

volume

hand,

the

every

trading

scaled

by

during

that

sizes

of trading

data:

for

the

of

the

the

time

Figure

15.7

Map

VWAP

past

day,

are

the

as

day,
of

With

in

often

map
the

determined
the

such

total

order

historically

equation

(3).

algorithm.
of Resulting

of Historical
and

Chriss

Process

TWAP

Volume
(2000)

Order

Averages

Flow

for

(or

Futures

the

other

shows

VWAP
as

as

computed

15-minute

VWAP

volume

shown

is

every

month.

orders

proportion

variations,
map

the

trading

child

period,

VWAP

Almgren

trading

throughout

VWAP

Diagram

15.8

. The

over

15.6

Figure

15.7

the

Figure

Source:

Figure

of

period

the

volume

in

month

interval

average

of intraday

equities

preceding

duration)

diagrams

averages

the
map

follows:
size

in
for

is

observed
Figure

15.8

(4)

The

resulting

benchmark

VWAP

price

can

be

determined

as

follows:

(5)
The

VWAP

map

accurately

reflect

average,
that
VWAP

is

utilizes

specific

markets

change

little

data

trading

hourly
April

volumes

in

2010,

the

uptick

in

their
month

2009
the

and

the

VWAP

volume

at

the

open

of activity

at the

Figure

15.9

futures

(FGBL)

by
market

Persistence
POV

April

a lull

of

the

Such

of

2010.

volume

have

the

remained

largely

the

European

and

lunchtime,

of

variations

that

Figure

15.9

computed

average

grown

map

orders

patterns:

futures,

While

on

success

example,

Eurobund

not

so,

child

relative

volume
For

does

Even

intraday

next.

futures

postU.S.

and

allocation

intraday

for

Eurobund

of

followed

the

map

shape

sessions,

own
to

VWAP

an

liquidity.
of

data

conditions.

deliver

persistence

one

historical

market

intraday

possess

for

on

can

the

on

from

the

solely

algorithm

based

illustrates

based
concurrent

a VWAP

efficiently

using

is

hourly

from
the

2009

to

same:

the

U.S.

followed

by

an

trading
a

spike

close.
of

Intraday

Volume

Distribution,

Eurobund

According

to

(CFTC)
the

the

and

causes

the

the

first

to

started
of

over

the

Figure
and

$4.1

previous
15.10

the

determined

for
entire
used

E-minis

the

POV

large
with

(SEC)

the
on

significant

POV
May

POV

volatility

set

report

that

2010.

The

in market
trader

at

on

algorithm
6,

fundamental

TWAP

and

10

large

order

calculation

VWAP,

during

minutes.

POV

the
set

prices

initiated

9 percent

to

The

child

size

of each

fixed

at

of

volume

Like

TWAP

next

should

regular

POV

exclude

order
of

period

continues

period's

time

child

percentage

predefined

previous

order

POV.

orders

execution
The

the

child

a previous

is processed.
of

behind

sends

and

recorded

example,

algorithm

algorithm

dynamically

volume

in

of

markets

the

when

it was

Commission

minute.

Unlike

trading

that

illustrates

VWAP,

intervals.
is

billion

the

Trading

Commission

(2010),

in

occur

Futures

Exchange

crash

mayhem
discovered

Commodity

and

flash

examination

trade

U.S.

Securities
of

created

joint

of
until

the
time,
the

trading

volume

the

volume

generated
Figure

by

the

15.10

POV

POV

trader

himself:

Process

(6)
While

the

POV

joint

algorithm

the

volume

own

and

used

by

generated

volume

and

SEC

have

When

properly

TWAP

and

the

caused

the

report

large

programmed,

himself,

mention

has

trader
failure

crash
one

dynamically

the

accounted

for

to account

to

child

for

his

orders,

proportions.

distinct

adjusts

responding

whether

increasing

of flash

POV

instantaneously

not

exponentially
crisis

POV

did

fundamental

trader

generate

VWAP:

conditions,

the

by

would

could

CFTC

to

such

advantage
present

events

over
market

as

shifts

in

liquidity.
Issues

with

TWAP,

Basic

VWAP,

sections

were

but

suffer

1.

These

from

Models
and

Optimality

limited

pricing

or

be

are

shown

can

be

to

be

the

1990s

discussed

and

are

still

in the
widely

previous
popular,

easy
for

to spot

to

with

Earlier

Brownian
optimal.

shown

be

optimal

only

in

specific,

conditions.

assumptions

arithmetic

models

shortcomings:

market

Conditions

Under

in

serious

models

models

execution

developed

not-very-common
2. The

POV

advanced

mathematical

tools.

Models

about

market

motion
Both

(ABM),

martingales

dynamics,
TWAP,
and

like
and
ABM,

martingale
VWAP

can

however,

assume
One

that

can

where

the

also
the

market

show

does

that

trend

not

VWAP

trend,

a hardly

is optimal

completely

realistic

in rapidly

dominates

condition.

trending

short-term

markets,

noise-induced

volatility.
In most

market

sizable,

conditions

however,

sections

of

models

this

models

models'

primary

general

markets.

strategies

the

trend

lose

describe

to most

their

the

market

and

the

volatility

optimality.

latest

are

The

advanced

later

execution

conditions.

Models
like

TWAP,

VWAP,

mission

send,

simple

both

models

chapter

of Earlier

Popular

with

these

applicable

Security

where

tools

is

Due

to

their

child

like

to

the

and

break

POV

also

lack

and

hide

the

the

child

up

regular

nature

orders

can

autocorrelation

of

be

and

security.
order

flow

orders

surprisingly

advanced

The

these

easy
tools

in

to

like

spot

Fourier

analysis.
TWAP,

for

familiar

with

electrical
CDs.

As

orders

Tag

all

3.

for

and

from

to

thereby

VWAP
instead,

may
the

to

digital
is

identical

hide

15.6

market

trade

order

TWAP

To
one

to

from

a core
remedy

comprises

detect

tick

flow

processing,

deployed

size.
therefore,

the

signal

often

data,

needs

data

as

study

of

scratched
the

market

anyone

regularly

TWAP

orders

in

sells

as

size;

do

to:
either

buys

and

4.

buy

sell

each

process

systems

of

all

Within

This

of

Figure

in Chapter

same

intervals

in

recent

2. Separate

little

that

of tick

discussed

the

basics

shown

stream

1.

the

does

engineering

spaced
the

example,

trade

ticks

into

virtual

buckets

by

trade

ticks.

bucket,
one
can

identify

trades

continuously

the

time

and

eliminating

the

original

seem

secure

VWAP

occurred

at

identical

time

another.
be

predict

that

more

trades

repeated
size

as
are

of

the

purpose
the
scaled

in
next

real

time,

TWAP

of TWAP

allowing
installment,

orders.

trades

are

not

uniform

by

the

time-specific

in

size;
trade

volume

or

volatility

averaged

over

the

appear

to prevent

VWAP

flow

To

see

process
observed
used

the

the

VWAP

be

Figure

Sample

scaling

as

by

the

computed.

scaling

security

scaling

volume

and

Order

used

in averaging

width

of

time

functions

will

or

scaling
flow,

may

in reality,

consider

an

equity

Descaling

all

or volatility

function

of

the

Figure

VWAP

trade
as

15.8

ticks
the

one

transforms

enables

TWAP-like

flow.

days

flow

order

subsequently

used

Flow

by

bars

so,

such

day

TWAP.

process

VWAP

order

as

trading

While

VWAP,

functions

Even

complete

month.

transparent

15.11

order

previous

of VWAP

Figure

TWAP,

15.11

the

or

VWAP-generating

of the

of

of
in

identification

number

week

as

same

into

VWAP

just

shown

by
in

previous

limitations

as

during

reverse-engineering

can

the

observed

different

repeating

may

volume

which
the

the

or

using

orders

differ

as

analysis

in

the
well

averages

are

over

different

sent

by

volatility,

intraday

descaling

times

identify

traders

either

over

several

in Equities

the

precomputed

with

given

scaling

function.
The

order

Regular
order

flow
spacing

sizes,

functional
during
In

sent

the

response

broker-dealers
transparency

via

POV

of orders,

gives

away

form

of

time

elapsed

to

order

such

coupled
the
flow

is
the

security

basic

algos.

In

dependent

While

similarly

some

of

form
POV,

volume

of
the

executed

order.

market

timing
the

case

the

POV

identified.

functional

the
on

previous

and

be

predictable

flow.

issues,
sizes

can

with

order

since

randomize
of the

algorithms

of

participants
orders

randomization

and

to

reduce

may

restrict

other

market

basic

methodology

with

participants'

ability

described

advanced

digital

to

observe

earlier,

signal

the

processing

the

order

orders

will

flow

still

techniques,

with

be

such

the

traceable
as

Fourier

analysis.
Fourier
the

analysis

noise.

is often

Digital

used

Fourier

restore

scratched

music

of

singers.

Likewise,

pop

slightly
The

randomized
key

mathematical

analysis
CDs

Fourier

construct

Continuous

(as

specified

as

in

used

to

slightly
be

off-key
used

to

voice
detect

algorithms.
is

time

to digital)

buried

routinely

can

analysis

connecting

opposed

the

analysis

of basic

signals

are

or to correct

flow

in

repetitive

models

Fourier

order

concept

to identify

Fourier

and

forms

transform,

frequency

of the

Fourier

domains.
transform

are

follows:

(7)
(8)
where

represents

frequency-domain
of

continuous

cyclical

cycles
domain.
the

time

cycles

domain

Furthermore,

cycles:

Figure

50

transformed
in

analysis.
process

sinusoid

of

15.12

50

variable,

Figures

Fourier

time-dependent

process

time-based

function.

capabilities

repetition

per
with

through

Figure

15.12

can

with

second).
Fourier

become
the

15.12

that

function

spike

analysis.
a

single

falls

Recurrent

Process

be

frequency
Figure

Hz.
Sample

and

directly

15.15

The
clear
onto

Hz

shows

spike
the

the

by
(hertz,
the

perfectly

simple

generated
50

is

illustrate

shows

of

15.13

F ( k)

a
or

same

repeating
in

frequency

frequency

of

Figure

15.13

Fourier

15.14

Signal

Representation

of the

Process

Shown

in

15.12

Figure

Corrupted

with

Zero-Mean

Random

Noise

Figure

Figure

15.15

Single-Sided

Figure

15.14

generated

at

represent

shows
50

Hz

random

Amplitude

different

and

120

stream

Spectrum

of y ( t )

time-based
Hz
of

corrupted
data,

such

function:
by
as

noise.
other

two
The

sinusoids
noise

traders'

may
orders

mixed

in

with

hard

to identify

the

Fourier

the

the
the

of

realistic

flow

the

of

one

that

order

example
Figure

by

have

market

the

clear

and

in

in

15.14

a pass

representation
at 50

example.
the

VWAP
models

in TWAP,

Figure

However,

peaks

flow

discussed

market

in the

Hz

Similar
sea

of

have

been

next

of

the

120

Hz

noise

question.

and

and

are

through

ideas

into

VWAP,

the

induces

with
be

extend
orders,

developed

POV

to

algorithms.

section.

Replenishment

to

avoid

VWAP,

and

advanced

models

a mixture

of trend

that
rate

replenishment

following

and

TWAP,

shown

a constant

of replenishment
15.16

basic

developed

it can

book

book

assumptions

conditions

conditions,

The

chart.

advanced

are

induced

researchers

these

the

of this

of TWAP

embedded

cycles

Models

use

normal

The

: clear

order

years,

algorithms

Advanced

periodic

few

The

latest

15.15

domain

usefulness

past

VWAP.

delivers

frequency

issues

order

eyeballing

in Figure

overcome

To

by

just

identification

Over

or

shown

dominate

placing

TWAP

transform

periodicity,

to

the

the

refers
market
in a limit

to

order.

and

Order

Book

algorithms,
under

volatility.

Under

strategy

is

replenishment.

process

book.

of

work

trading

book

Figure

order

in a Limit

the

POV
that

optimal

of order

transparency

15.16

of

repopulation
illustrates

an

Stylized

replenishment

possesses

after

liquidity

some

assumed

to

book

slides

price.

The

shadow
The

shadow

exist
up

form,

has

and

order

taken

the
liquidity

to

away.
the

which

the

price

the

order

in

book

book
reverts

order

book

levelthe

the

to

is

shadow

movement

book

of the
is

order

shadow

with

h ( E s ),

as

the

axis

resilience

resilience,

that

The

current

price

to as

is specified

structure

of

down

is referred
book's

and

been

of

assumes

independent

reversion

form

process

function

of

the

the

book's

book.

function

of

the

trading

follows:

(9)
where

E t is

the

aggregate

away

from

prevailing

order

flow,

E 0 = 0,

measures

how

recovers

following

size

market
and

fast

the
an

price

X t=
order

order

of

orders

at

time

E t for

book
of

limit

size

ticks

X t , and

available

at

t , X t is

the

T . The

function

away

from

satisfies

ticks

aggregate

the
the

h (E s )
market
following

properties:
The

function

is strictly

The

function

is a locally

y | for
and

all

x and

the

left

to

The

has

strategy
be

in

process

of the

the

first

[0,):

| h ( x )- h ( y )|

independent

of

derivative

the

amount

market.

of trading

on

a constant

measures

processed
rate

is

and

function

a bounded

trader's

in X,

Lipschitz

y , where

function

execution

increasing

As

is defined

x and

. The
of

such,

the

total

x0 = X

C |x y ,

trader's
order

and

still

xT =

0.

as

(10)
Price
any

continuous

process
can

S t , the

always

be

traded

instrument,

process.

Independent

expected

impact

measured

as

S t , can
of

the

inflicted

cost

be

by

assumed

shape

to follow

of

strategy

the
on

price

price

C :

(11)
The

expected

parts

as

value

of

MI

cost

can

be

expressed

via

integration

by

follows:

(12)

The

most

on

developing

rigorous
1.

on

best

execution

algorithms

motion,

asset

pricing.

price

functions

observed

section

evolution

price

reviews

has

under

the

focused
following

the

the

that

evolutions
latest

specification

most

can

be

used

in an

MI

framework.

models

developed

to

commonly

describe

under

the

two

any

price

models.

When

Price

Most

security

Brownian

(13)

research

execution

Brownian

in modern

empirically

the

of

assumptions:

Generalized

This

stream
optimal

Geometric

used
2.

recent

Follows

motion

contemporary

Geometric

pricing
with
price

models
price
level

Brownian
assume
increments
S t , as

Motion
that

prices

dS t exhibiting
well

as

incurring

follow

geometric

dependency
drift

on

The

vanilla

execution

optimization
et al.,

cost

measures,

function,

can

then

be

not

incorporating

specified

as

any

risk

follows

(see

Forsyth

execution

is

. Under

the

motion,

the

and

the

2011):

(14)
where,

as

before,

assumption

of

resulting

(2011)

on

many

Euler-Lagrange
optimal

price

rate

of

Brownian

solution

the

show,

optimal

geometric

optimal

dependent

for

the

of
of the

strategies

equations

the

cost

minimization

price

path.

However,

lead

to the

produce

cost-minimizing

the

trading

costs

problem
as

almost

Forsyth

identical

following

are
et

al.

outcome.

closed-form

solution

strategy:

(15)

The

resulting

expected

minimal

cost,

becomes

(16)

See

Forsyth

et al.

Price

Follows

When
While

most

(2011)

a Generalized

assume
describing

models

derivation.

now-traditional

Black-Scholes,
accurately

for

proposes

empirical

roots.

to evolve

as

In

Geometric

model

such

follows

of

the

security

Gatheral,

such

motion
prices,

price
price

Function

models,

Brownian

short-term

models,

(see

ImpactBased

asset-pricing

evolution
to

Market

as
a

changes
level

at

the

as
model

new

breed

closer

to

time

t is

of
their

expected

2011):

(17)
where
.

the

The

impact

strategy
resiliency
next
(18)

risk

be

or noise
of

trading
of

the

expressed

component

prior

trading

rate

dynamics

order
as

book

is the
is

price-level

quantified

h ( E t ). The

and

independent

using
the

expected

the

function
execution

execution
measuring
cost

can

To

minimize

expected

cost

[C],

one

is required

to solve

the

following

equation:
(19)
which

can

requires

be

cost

dependent
that

interpreted

as

invariance

on

volume

with

volume

impact

the

trading

impact

stays

follows:
rate.

E t , the

optimal
Since

value

the

optimality

cost

of
is

cost

directly

condition

requires

constant:

(20)
See

Obizaeva

Gatheral
Case

and

Wang

for

details.

(2011)
1: Exponential

When

the

form,

(2005),

Market

market

and

Schied

(2010),

and

Resiliency

resiliency

h ( E t )= e -pt , the

Alfonsi

can

equation

be

(20)

assumed
then

to

can

be

follow

rewritten

exponential
as:

(21)
from
be

where

the

expressed

expected

execution

cost

of

a trading

strategy

can

as

(22)
To

derive

note

suitable

that

E t can

be

conditions

for

expressed

as

E t , Obizhaeva

and

Wang

trading

prior

(2005)

(23)
where

E 0

chosen

time

measures

the

0. Integration

residual
by

parts

1,

the

impact
then

of

to

the

yields:

(24)
Normalizing
volume

E 0
impact

by

E 0

optimality

condition

with

constant

becomes

(25)
Equation

(25)

then

translates

into

(26)
The

original

optimality

condition

for

the

execution

cost

[equation

(21)]

can

then

be

expanded

(27)
Substituting
into

the

volume

impact

equation

(26)

produces

optimal

trading

rate

component

the

from

following

equation

(27)

result:

(28)
The

u t can

then

be

determined

as

(29)
Equation

(29)

can

be

interpreted

resiliency

function

can

execution

strategy

is composed

of:

be

as

assumed

to

follows:
be

when

exponential,

A large

block

trade

of size

at the

beginning

A large

block

trade

of size

at the

end

of execution

orders

placed

continuum

where

is

function,
The
market
15.18

of

TWAP-like

the

parameter

h ( E t )= e -

resulting

the

execution

with

illustrates

in

market

the

optimal

of execution
horizon,
at

exponential

process.
T.

trading

rate

market

resiliency

t.

optimal

resiliency

small

the

optimal

0.1

strategy
is

for

illustrated

execution

strategies

T =1
in

and

Figure
for

exponential
15.17

different

Figure
trading

frequencies.
Figure

15.17

Market

Impact

Source:

Optimal

Gatheral,

and

Execution
Exponential

Shied

and

in

a Market

Decay
Slynko

with

of Temporary
(2011)

Linear
Impact

Permanent

Case

2: Power-Law

When

the

market

function,
via

h ( E t )= t -

the

constant

Market

Resiliency

resiliency

can

, the

optimal

volume

be

assumed

to

can

once

strategy

impact

requirement,

fit

the

power-law

again

be

equation

(30):

from

h ( E t )= t -

derived

(30)
The

optimal

trading

rate

is then

(31)
with

representing

analytically

a parameterized

determined

from

constant

equation

, and

(32):

(32)

where

gamma

continuous
end

of

illustrated
Figure

function

is

for

continuous

with

large

execution

15.18

15.19

Optimal

Different

Trading

Source:

Obizhaeva

z.

singular

times

in Figure

defined

as

The

T.

trades

The

optimal
at

optimal

discrete

the

strategy

beginning

execution

n,

and
schedule

and
is
the
is

.
Execution

Frequencies
and

resulting

block

0 and

for

Wang

(2005)

with

Exponential

Resiliency

for

Figure

15.19

Market

Impact

Source:

Case

and

Gatheral,

the

restored,
and

constant

the

a Market

Decay

and

with

Linear

of Temporary

Slynko

Permanent

Impact

(2011)

Resiliency

resiliency

fits

market

resiliency

optimal

strategy

volume

in

Power-Law

Market

market
the

Execution

Shied

3: Linear

When

+,

Optimal

impact

a straight
function
can

line

until

is described

yet

requirement,

again

equation

be

the

markets

as

h ( E t )=(1

deduced

are
-

t)

via

the

(33):

(33)
The

optimal

no-trading

trading
intervals

Figure

15.20

Market

Impact

Source:

strategy
between

Optimal

Gatheral,

and

comprises
the

Execution
Linear

Shied

in

Decay
and

blocks,

harmonic
as

a Market

of Temporary

Slynko

(2011)

block

shown
with

in Figure
Linear

Impact

trades

with

15.20
Permanent

The

aggregate

each

of the

execution

schedule

size

, so

that

is
the

broken

total

down

trading

into

size

2N

trades

X satisfies

(34)
Practical

Implementation

To

determine

the

previous

following

the

optimal

order

section,

the

Execution

slices

per

execution

Strategies

the

framework

trader

can

presented

go

in

through

the

steps:

1. Estimate
2.

of Optimal

Fit

the

empirical

distributions

MI

of

function.

temporary

and

permanent

MI

of

the

traded

security.
3. Derive

optimal

4. Back
5. Put
The

test
the

allocation

the

execution

strategy

resulting

on

basis

of step

1.

strategy.

to use

strategies

the

in real-life

perform

well

production

environment.

in chosen

market

conditions.

today's

markets.

Summary
Algorithmic

order

a necessary
large

function

and

today
execution

With

afford

to

algos,

previously

participants.

the
on

best

best

offer

exchange

buy

550

order

build

units

and

and

considerable

use

as

value

technology

to all

costs,

advanced

available

such

from

only

to

co-location

provide

A contains

300

investors

routing
select

is

investors,

most

order

It

and
few

added

best
market

benefits

of

speed.

End-of-Chapter
1. The

delivers

plummeting

Services
and

is inseparable

that

small.

can

security

execution

Questions
offer
on

on

exchange

exchange

C contains
on
send

his

B contains
just

behalf.

them

to

100
How

500

units.

Your

would

you

exchanges

units

units,

of instrument
and

customer
break

under

up

the

best

wants
the

the

minimal

VWAP,

and

X,
offer
you

customer's
impact

algorithm?
2. What

is TWAP?

3. What

are

the

VWAP?
main

POV?

shortcomings

Explain.
of TWAP,

POV?

to

4.

How

can

the

disadvantages

of

remedied?
5. What

is resilience

of the

order

book?

TWAP,

VWAP,

and

POV

be

CHAPTER

16

Implementation

of HFT

Systems

High-frequency

trading

(HFT)

applications,

comparable

launches

and

practices

of implementing

Model

having

HFT

systems,

by

making

and

typically

outperform

tasks,

are

to

typical

life

16.1

mission-critical

This

NASA

chapter

reliable

HFT

shuttle

describes

best

systems.

nature,

require

in these

treacherous

example.

hesitation-free

programmed

traders

under

rapid

As

traditional

a result,
human

decision

computer

systems

mission-critical

trading

market

conditionssee

computer

trading

traders

on

systems

trading

desks

world.

development

similar

be

piloting

error.
and

to

Cycle

human

for

for

Properly

replacing

the

room

execution.

(2009b),

rapidly

around
The

their

tend

software

accurate

Life

particularly

Aldridge

with

little

Development

systems

that

of
of

cycle

a fully

the
of

automated

standard
a

trading

software

development

system

follows

development

process

is

process.

illustrated

in

path
The

Figure

Figure

16.1

A sound
phases:
1. Planning
2. Analysis

Typical

development

Development

process

Cycle

normally

of a Trading

consists

System

of the

following

five

3. Design
4. Implementation
5. Maintenance
The

circular

system

nature

development.

complete,

of

as

well

project
completed

as

project

feasibility

may

study

operating

that

and

address

human

resources,

the

project,

phase

the

other

day-to-day

include

concrete

schedules,

of

of

its

project

from

points

of
and

the

sufficient

and

technical
compliance,

The

targets

budgets

the

process,

the

view.

estimated

by

economical

the

and

the

economics,

has

operational

goals

of

what

The

project

the

and

is accompanied

requirements.

of

be

goals

view

in terms

the

to

cycle.

planning

project

of

appears

determine

whereas

feasibility
and

established

the

quality

modifications

high-level

The

potential,

the

planning

like.

whether

(P&L)

issues

technical

explore

profit-and-loss

is to

generate
look

system

development

phase

evaluates

model,

considerations

of

to

the

continuous

advanced

to a new

planning

the

of

demand

lead

the

illustrates

a version

issues
that

purpose

process

When

new

enhancements
The

of the

outputs

set
for

for

the

the

entire

system.
During

the

analysis

requirements
project

for

(which

current

release),

the

development
the

ultimate

given

the

allocated

design

including
with

in

determines

and

solicits

analysis

ability

which

initial

stage

to

features

is arguably

shape

the

feedback

because

it
the

team

is

aggregates
scope

are

out

from

the

most

here

that

functionality

of

the

of

the

users

and

critical

stage

stakeholders
of

the

system

budget.

process

incorporates

detailed

diagrams,

output

formats

documents.

An

project

discrete

into

functionality,

process,

phase

other

of

are

and

The

have

The

system

features

management.
in

stage

objective

business
such
of the

components

as

specifications
rules,

those

design

and
of

stage

subsequently

daily

of functionality,
screenshots,
reports

is to separate
assigned

to

along
and

other

the

whole

teams

of

software

developers;

interfaces

that

designed

by

the
can

of

future

development

teams

the

forward.

The

and

to

individual

modules

themselves

is

work

begins.

together
While

may

have

various

to

that

system

developed,

modules

usually

dedicated

developed
of

each

design
is,
and

the

functionality
stage.

The

discrepancies
observed

development
the

code.

testing

The

team
teams.

for

project

casesthat
to

is,

verify

the

some

development

cases.

When

test

staff
to

the
The

resolution

and

little
still

documents

are

Scripts

have
has

The
the

then

subsequently

back

be

who

execution

defined

case

be

process

people
the

sent

to
to

test

any

test

or

among

may

procedures

are

variance

remains.

monitors

prespecified
bugs

putting

system.

than

diligently

then

to

system

other

been

integration

refers

operation.

testing

project

have

project

wrappers

personnel

the

communication

the

The
teams

modules

proper

proper

stage.

the

work

the
modules

design

a functional

installation

personnel

between

the

encounter

ensure

according
test

the

software

implies,

projects

to ensure

performance.

name

important,

tested

of

programming;

the

to create

to

comprehensively
involves

in

individual

stage,

most

software

actual

test

its

components,
and,

different

blueprints

specifications,

as

written

early
modules

test

as

by

the

the

integration

be

Such

computer

develop

tested

planned

the

used

defined

predefined

to

outlines

involves

the

individual

in

later

then

satisfied

successfully

problems

also

components

operation

programmers

Integration,

the

internal

smooth

are

according

other

among

specifications

against

management

of

finally,

or individual
the

software

well-specified

code.

phase,

according

developers.

are

completed

implementation
teams

of

phase

that

have

with

enables

design

will

seamlessly

packaging

paths

software

components

communication

of the

developed

teams

The

functionality

correctness

in

software

streamlines

going

lock

different

specification

discrete

in

the

bugsthat
performance
over
returned

to

the
to

Successful

implementation

subsequent

maintenance

phase

addresses

such

as

Key

16.2

of

system-wide

by

the

deviations

newly

the

system.
from

discovered

deployment
The

planned

and

maintenance
performance,

bugs.

Implementation

Steps

Most

followed

phase

troubleshooting

System

is

in Implementation

systematic
.

This

trading

section

of High-Frequency
platforms

discusses

are
each

Systems

organized

as

component

of

shown

in

the

process

Figure
in

detail.
Figure

Step

16.2

Typical

1: The

The

core

that

contain

following
Receive,
Perform

Core

the

Process

Engine

engine

is composed
core

logic

of
of the

one

or

several

trading

mechanism

incoming

quotes.

functions:
evaluate,
run-time

Implement
Initiate

High-Frequency

run-time
and

transmit

and

archive

econometric
portfolio
buy

and

analysis.
management.
sell

trading

signals.

run-time
and

processors
perform

the

Listen

for

and

Calculate

run-time

Dynamically
market

C++,

although

Java

and

the

P&L.
risk

based

high-frequency
some

Q,

on

current

portfolio

Nasdaq

high-frequency

distributed
OMX,

disables

for

allocations

distinguishes
considered

to

programs

do

as

of

Kx

Systems.

is said

collection,

it from

C++,
be

not

lighter

and

have

a result,

C++

the

than

processing

systems

work

yet
of

the

code
that

C++

is

meaning

overhead

faster

of

Java

systems.

Java,

power

often

engine

in Java,

the

use

database

matching

down

in

to

and

functionality

faster

written

known

syntax

written

slows

and

are

are

The

core

also

firms

language

to be

but

systems

trading

hybrid
by

example,

garbage

Java;

production-bound

a commercial

organization

by

of execution.

conditions.

of

C++

confirmation

manage

Most

often

receive

that

required

than

Java-based

systems.
The

core

initiate

engine

and

transmit

executing
and

and

an

the

price

and

the

P&L

and

feed

back

into

the

portfolio

The

design

inputs,
the

the
the

econometric

to

returns,

while

Incoming

model,
portfolio

communication
are

most

(FIX)
financial

protocol

run-time

with

In

position

The

addition

status

system

then

parameters

that

management

to

the

raw

incorporates

sizes,
and

and

quote

inputs

other

maximization

from

information
of

portfolio

risk.
outgoing

a broker-dealer
via

specifically

designed
Other

and

order

portfolio

framework

transmitted

information.

client.

receiving

the

management

engine.

portfolio

along

back

then

piece.

diversification

between
often

the

of

current

minimizing

quotes,

to
risk

management

Upon

sends

management

core

framework

broker-dealer.

assesses

implementation

econometric

the

size

and

portfolio

relevant

to

management

broker-dealer

calculates

reflects

portfolio

orders

order,

order-filling

the

the

orders
and

Financial

protocols,

for

a client

and

any

or an

Information

FAST

exchange,
eXchange

transmission
like

other

of
and

real-time
Nasdaq's

proprietary

ITCH

According

to

emerged

and

the

FIX

in 1992

trading

as

between

since

the

broker-dealers,

best
global

banks,

steering

around

the

process

dedicated

like

any
typical

FIX
The

the

characters

many

# 9),

is

types

that

sent

to

result

system

MsgType
the

communication
as

as

the

example,

type
are

acknowledgments

to ensure
For

the

the

of the

as

is typically

inactivity.

If either

message

from

sent

after

communication
the

other

party,

from
of

planning

expense,

field

much

the

(FIX

message

field

party

technical

a string

# 35).

of

header
Among

execution

directives

messages

designed

a prespecified

ensure

and

has

number

has

not
a

message

to

problems.

sends

number

messagea

operational

it

the

and

running.

Heartbeat

party

body,
fields:

# 8),

order

and

three

to follow

up

remains

message

associations,

careful

header,

housekeeping

connection

unforeseen

from

Implementation

following

and

remains
= 0 is the

(FIX

quotation

communication

any

of

overseen

providers

significant

message

other

of

for

percent

is

and

requires

of

message

well

used

process.

composed

of the

fact,

representatives

free.

demand

of a message

body

and

may

contains

beginning

message

and

is

and

however,

development

always

in the

field

and

message

header

identifying

(FIX

system

FIX,

80

technology

is open

via

resources

other

trailer.

standard

was

that

utilities

information

various

in 2006.

of

industry

has

In

FIX

language

consisting

equity

among

firms,

FIX

It

customers.

buy-side

a programming

and

Its

method

of exchanges

exchanges,

world.

and

percent

for

Brothers.

fixprotocol.org,
of

committee,

investors,

communication

75

broker-dealers,

institutional

percent

as

Salomon

transacting
by

),

framework

and

and

75

over

( www.fixprotocol.org

communication

conducted

described

site

Investments

by

and

used.

communications

dominant

trading
firms,

is

by

a bilateral

to a survey

systematic

also
web

exchanges,

according

sell-side

are

industry

Fidelity

become

FIX

OUCH,

received

TestRequest

not

The
of

that

the

been

lost

heartbeat
seconds
a

heartbeat
message

of

(MsgType

heartbeat
the

1)

poll

message

connection

MsgType

interest

is

is

in either

which

buying

normal

client

= R

of

S)

of

the

that

carry

steps

to restart

to

broker-dealer

quote

stream

quote

agency
with

stream.
to

of

their

message

responds

continuous

and

or an

Request
a

it.

transmit

a proprietary

requests

no

Exchanges

messages

a Quote

If

message,

taken

Interest.

in either

actual

party.

TestRequest
are

of

Interest

indicates

with

Indication

broker-dealer

message

(MsgType

and

or selling

circumstances,

Request

lost
as

communication

following

Indication

MsgType

other

received

known

use

capacity.

the

is considered

broker-dealers

ask

to

the

Quote

information,

Under
Quote

messages

such

as

bid

or

prices.

Other

message

orders,

day

requests,
the

limit

and

following

[Field
For

types

include
orders,

orders

such

multiday

acknowledgments.

as

single-name

orders,

All

fields

orders,

various

in the

list

cancellation

body

are

included

in

status

of

format:

#] = [data]
example,

an

order,

35

= 8|

All

field

to communicate

the

following

sequences

computer

value

sequence

are
of

that

the

carries

the

is used:

terminated

0x01.

message

The

with

a special

character

character

looks

like

that

has

when

seen

on-screen.
The

body

whether

of

the

is

it

information.
interest,
and
the

quote

The

other
body

necessary
are

included

contains

request,

message

a timestamp

[Field
and

message

that

body

quote
further

includes

transaction
in the

the

details

of

itself,

or

specifies

following

Like

order
the

milliseconds,
data.

the

message,
and

exchange

a security
the

header,

trade

symbol,
all

fields

format:

#] = [data]
each

character

field
0x01.

sequence

is

terminated

by

special

of

computer

in

Finally,

at

the

end

checksuma

sum

message

of

included

arrived

in full.

of this

of

as

An

the
digital

body

of

values

of

verification

example

of

of

every
all

message

the

characters

whether

a FIX

message

functionality

can

is

the

the

in

the

message

is shown

has

in Chapter

book.

The

risk

management

components:

tracking

performance

and

performance

limits

parameters

may

parameters,

the

basic

generating
breached.

include

message

other

the

components
warning

be

and

include
of

count

variables

the

messages

Appropriate

in

the

management

per

discussed

system

should

risk

limits

following

unit

detail

time,

P&L

later

in

this

chapter.
Step

2: Quote

Most

quotes

Archival
are

engines

that

are

next

archived

As

discussed

can

be

are

delivery.

in

As

will

be

tick

data

do

the

not

may
by

researcher

be

an

not
the

has
benefit

most

HFT

without

that

of

the

completely

data

buyer

data

fills

data,

most

best

from

representative

his

important
practices

the

internally

of the

data

quotes.

from

one

quote

entity

datastream

cases,

of the

purchased

quote

process

interface.

informational
that

archived

data
by

While

void

suggest

received

which

lose

recorded
some

own

networks

point-to-point
may

entity's
In

quotes

over

guarantee

representative

an

public

technology

entity.

from

The

(UDP)

co-location

each

next

the

Protocol
not

FIX

purposes.

over

in quote-receiving

be

no

delivery

does

proprietary

scenarios.

simulation

Datagram

broadcast

that

historical

will

User

using

message
and

quote

guarantee

to

archived

various

2,

variations

purchased

entity

Chapter

identical

archived

with

a result,

next

and

reconciliation

are

Furthermore,

will

for

quotes

the

tested

unreliable:

some

to

accepted

the

when

each
as

trading
such

system

data
during

production.
Most

firms

messages

with
in text

serious
or binary

HFT
file

needs
formats.

archive
The

all
text

received
file,

known

and
as

sent
a

flat

file

in

the

industry,

readable
can
or

by

be

humans

comma

opened
files,

BLOBs

, are

machines.

than

matching

continuous

weekly

New

terabyte

of

been

time.

is

Networks
Many

very

databases

seamless

have

attempted

to
and

data,

products.

Most

are

time-sensitive

operation

of the

simulation

Most
and

execution

level

databases,
not

has

where

it into

been

the

shown

crack

the

replace

data

the

trading

Retrieval

execution

flat

optimized

system.
to be

KDB,

much

as

could

have

today

such

files

Area

their

in HFT

the

input/output
compromising

data

occurs

a critical

efficiently

tool,

archiving
with

Slow

distributed

a promising

Friday

for

engine,

to

on

because

is not

in

data.

of tick

time

foreign

Storage

market

archival.

are

yet

to stored

is

system.

as

of

much

markets,

ends

ago,

to HFT

of

however,

to enter

however,

HFT

occupy

unsuitable

functionality

delays

performance
the

databases

can

or
by

ticks

requirements

access

high-frequency

and

and

all

incarnations

allow

faster

exchange

years

Various

real-time

most

10

LINUX.

ICAP's

night

storage

just

reasonable.

(SANs)

BLOB

Such

expensive

foreign

file

readable

are

records

is

Excel

OBjects

characters

example,

Sunday

on

Binary

BLOBs

with

into

editors

Large

file

in the

loaded

broker

such

space.

prohibitively

storage

Each

text

flat

fields

easily

Interdealer

Along
on

be

as

The

The

(Hex)

for

begins

storage.

plain

to

files.

engine,

storage

or

of readability,

flat

BLOB

can

hexadecimal

BLOB.

York

PCs

referred
in

of

translation.

and

on

expense

exchange

night,

separated,

recorded
the

form

special

Notepad

compact

ICAP's

simplest

without

commonly

At

each

the

or tab

with

Binary

more

is

only

at

parameter.
retrieve

by

Kx

adopted

data,

Systems,
by

several

firms.
Step

3: Posttrade

The

best

engine
the

HFT

reconciles
same

returns,

code
trading

Analytics
systems

do

production
on
costs,

the
and

not

stop

results
same
risk

data

there.
with
and

management

simulation
updates

posttrade

analysis

results

run

distributions

parameters

to

with
of

be

fed

back

into

the

main

management
Step

engine,

portfolio

optimization,

and

risk

components.

4: Simulation

Simulation

refers

Simulation

is

test
A

processing

within
that

make-belief
for

in production.

will

hard-pressed

execution

one

a short

works

money,
be

important

a strategy
strategy

to

key

time

well

in

reason:

span,

without

trading

has

positive

actual

chance

that

fails

results

strategy.

researchers

risking

a strategy

deliver

it allows

simulation

Conversely,
to

of

in

to
capital.

on
in

actual

simulation

real

trading

environment.
The

simulation

trading
the

engine

ideas

on

trades.

past

Unlike
that

tested

simulation

language

(C++

Simulation
the

the

best

than

the

order
The

can

Thus,

size

of

the

most

resulting

in

worse

offer

more

buy

best

even

new

executing

early

or

stages
Excel,

in the

of
ideas

production

average

than

on
the
MI

per

unit

by

be

order

is

impact,
results,

size,

the
and

is

filled

at

smaller

market

market
of

incur

please

order

sell

bid.
of

the

assumed

to

performance

predicted

trade

be

a small
best

if

different

market

market

execution

will

trading

can
the

only

correctly.

assumed

at the

orders

market

live

be
the

overstate

and

requires

time

Similarly,

executed

market

prices

details

of

queue.

time

orders

simulation

can

size

trading,

programmed

at the

about

the

trading,

order

the

ask

to be

in

quote

that

of

approximate
the

the

real-life

limit

submitted

assumptions

live

estimate

tests

actually

coded

is

and

prevailing

assumed

In

better

typically

of

market

a market

system.

(for

are

in

MatLab

simulation

order

providing

will

in

that

without

still

approximation

of

ask,

be

are

coded

of

at the

best-quote

however,

an

market

executed

submitted.

module

data

that

engine

part

run-time

often

execution

treatment.

independent

or Java).

execution

to be

and

are

produces

Specifically,

an

ideas

development
in the

is

MI
best

see
HFT
then

orders,
the

or
bid

trading
slippage,

and

Chapter

best
5).

researcher
adjust

To
can

the

best

quotes
then

by
be

the

estimated

considered

values

executed

size-dependent

MI,

applying

degree

a great

market

sell

of the

orders

at the

closer

be

The

prevailing

reflecting

of

will

MI.

ask

buy

plus

real-market

conservatism

expected

market

to

the

will

expected

conditions

simulation.

to execute

orders

and

Similarly,

at the

best

bid

the
less

MI

estimation.
Simulation
can

of

be

considered

crosses
price
be

execution

the

to

order

equals

the

executed

in

market

price

drops

market

price

rises

In-sample

sample

of

data

issues

dog

the

To

make

on

may

result,

When
or

simulations
be

the

price

of the

buy

limit

the

price

of the

limit

sell

can

be

order

market

may

to

that

needs

involves

running
data.

simulation

which

not

sure

following

order

order

the

in-sample

strategy

unused

10.1.

limit

A limit

the

limit

strategies

out-of-sample.

results

when

in Figure
the

As

complex.

of

the

market

may

not

limit

executed
order,
order,

price

to

order

when

the

or when

the

and

the

limit

is guaranteed.
of

the

trading.

above

is also
only

shown
price,

consider

execution

Simulation

as

below

orders

executed

order

live

generally

limit

be

price,

limit

trading

order

of

to

strategy

strategy

be

tested

the

on

was
the

in normal

the

the

The

up

in-sample

runs

process:

hold

performed

first

data

on

a chance

out-of-sample

Natural

overfitted,

in real-life

is

and

trading,

Out-of-sample

copious

testing

be

same

conditions.

out-of-sample.

strategy

the

developed.

can

market

has

strategy

and

amount

the
testing

of

previously

usually

performed

in

the

a back-test

. The

back-test

order:

1. Back-test
2. Paper

trading

3. Production
Back-Test
A strategy
typically
two-year

run

on

utilizes
minimum

historical
at

data

is called

least

two

years

of

for

the

test

is

most
likely

recent
an

tick

data.

overhang

The
from

low-frequency

monthly

provides

number

making

statistically

theorem

in statistics.

In

theory,

number

reserve

data)
a

condition

nonrecurring
of

historical

making
A

system.

on

The
risk

live

from
be

is

used

in risk

distribution
for

the

system

To

mitigate

flash

stress

extreme

the

made-up
crash

event.

of

May

However,

two

the

it

is

however,

the

smaller

set

the

years

more

the

of

the

more

continuous

data-snooping

model
the

model

overfits

back-test

to

on

development

of

of

are

returns

a fresh

is known

least

used

two

years

running

as

results:

tick

the

costs

low

probability

2007),

the

system

can

also

back-test
may

that

into

obtained

the

back-tests

hidden

inputs

data

though

and

trading

returns

of

even

given
as

of trade

applications,

returns

of

from

distribution
at

risk

obtained

misleading

occur

fail

to

when

live.

swans

running

ratio

and

minimizes

estimation

unexpected

black

representing
the

generate
the

the

in

over

least

Running

in the

management

the

data.

a back-test

is trading

as

consider

all

to

limit

prove

Sharpe

performance,

model

distributions

model

may

account

known

the

to

observations,

when

models

Still,

central

6. In practice,

of

(at

occurs

useful

quantification

running

the

unused

preferred

capital.

data

inferences

also

the

preferred

strategy

in the

out-of-sample

back-test

HFT

that

data

is

the

for

results.

that

aberration

by

months

sufficient

sufficient

in Chapter

the

historical

ensures

days

number

strategy
of

also

the

under

determined

observations

evaluate

in the

large

is

of

24

nearly

inferences

discussed

larger
to

confidence

set

of

the

opportunities

bias,

as

days:

observations

number

strategy

to produce,

larger

tick

evaluation

monthly

minimum

of

because

of

significant

the

performance
thought

performance

and
(Taleb,

HFT

code

stress-test
6,

2010,
data

through

corresponding

HFT

historical

scenarios.
represents

extreme

The

researcher
or

simulated

data

a historical
to

returns,

some

may
data

surrounding
example

of

hypothetical

simultaneous
The

failure

out-of-sample

minimum,

evaluation
of

the

average

returns,

explained

in Chapter

Once

Paper

the

markets

results

should

idea's

Sharpe

can

need

process

trading

to

be

be

evaluated.

compute

and

At

basic

performance:

ratio,

simulated.
a

statistical

cumulative

maximum

and

drawdown,

as

6.

strategy

back-test,

financial

back-test

the

parameters

is

determined

strategy

to

is moved

perform

into

paper

satisfactorily

trading,

discussed

without

placing

in

the

next.

Trading

A strategy
trades,
all

of global

run
is

known

orders

should

in real

time

as

in a text

on

paper

file.

live

data,

trading.

The

but

The

orders

paper-trading

and

trades

the

stage

records

actual
records

at a

minimum

include:

A granular

timestamp

of the

order,

with

a minimum

of 1 ms

or

finer

price,

and

best

precision.
A code
Last

of the

observed

ask

size,

Order

model

price,

end-of-day

execution

best

bid

size,

reconciliation

best

ask

of orders

and

data.

module

between

be

the

origin

a historical

from
main

price.

difference

should

includes

the

bid

instrument.

quantity.

main

data

financial

best

for

Assumed
The

traded

the
of

live

the

trading

quote

quote-streaming

archives

and

functionality.

feeds
In

receives

the

real-time

model

data;

module

paper
tick

to

trading
data

the
that

it sequentially

and

back-test

back-test
reads

the

system

historical

module

system,
from

the

that

a different

trading

tick
has
quote

venues

and

broker-dealers.
Except

for

back-test

differences
systems

simultaneously
chapter
assumption

that

receiving

should

and

reviews

in

the
both

use

the

be
same

systems
back-testing

quotes,
identical;
code

for

paper-trading
they
core

implementation
and

paper-trading

can

and
be

functionality.
process
engines

built
This

under
are

the
built

and

tested

in parallel.

the

process

practices

including

for

sections

high-frequency

process,

best

following

of developing

development
the

The

testing

summarize
systems,

common

steps

in

detail

the

system

well

as

discuss

pitfalls,as

developed

key

trading

systems.

capital

is

Production
A

strategy

live

running

trading

or

in

as

venues

via

system

running

in

as

records

for

The

Step

by

the

most

system

virus
of

not

the

that

specific

bounds.

Once

the

the

authority

causing

the

Common

Pitfalls

Message

Acknowledgment

A market

order

client

The

venue

the

reliable

live
HFT

archive

The

as

to
An

locally

help

well

paper

all

trading

generate

assess

strategy

order

executional

and

shortfall
measures

of

observation

of

paper

Implementation

slippage

prices

trading

and

other

obtained

in

real

counterparts.

the

victim

system

to

malicious
for

however,

are
down

have

required

unaccounted

performance

bounds

is

some

event
trader,

the

breach

been

such

in the

model

normally

system

breached,

the
for

the

ensure

activity

should

of the

trading

to

falls

human
day

be
within
trader

or until

the

resolved.

Implementation
Loops

communication

receives

fall

to shut

in Systems

sends

of

human

sure

The

live

or a market

to making

conditions

of

supervision
does

role

have

as

direct

limited

should

will

implementation

simulated

human

The

to

above.

analysis,

sent

Supervision

a computer

itself.

as

allowing

their

are

to

strategy.

the

5: Human

that
as

records

referred

protocols.

need

trading

in performance

and

still

live

known

Continuous

will

the

of the

costs

markets

production

usually

orders

messaging

discussed

provides

latent

other

reconciliation

formally

shortfall

or

Production

trading

difference

is

live

paper

and

performance

FIX

on

the

and

fulfillment

time

production

trading

orders,

real

a market
the

process
order
order/the

includes

to the
broker

broker

the

following

or a trading

forwards

the

steps:
venue.

order

to

the

exchange.
The

trading

The

client

receives

The

order

is executed.

The

trading

The

client

Time

venue

the

and

client

ms

Regardless

between

is sufficient
A

and

When

position

the

until

order

quantity
is

position

and

is

counters:

positions.

The

lacking

The

is being

HFT

the

easy
one

for

idea,

in

keeps

however,

two.

finite

time

that

of

execution

upon

One

solution

orders,
does

position

not

during

extreme

The

mistake

limit.

the

occur

the

in an

involves

and

limit.

execution

orders

position

market

receiving
out

an

logic:

specific

resulting

set

is

following
to

spewing

the

sent-in

ms,

or

a certain

potentially

fix.

to 100

condition

response

only

to

order

minute

the

the

is adjusted

above

to 50

runaway

reaches

well

grow
a whole

Consider

executed,

the
In

a market

confirmation

portfolio

system

of executions,

common

Time

the

to

algos.

orders

total

of

process,
and

for

counter

acknowledgments,
the

execution

origination

out

take

order

acknowledgment.

may

still

runaway

an

speed

number

counter.

the

execution

may

cause

sends

sends

execution

of the

position

acknowledgment.

client

order

time

wildly

rookie

conditions

execution

acknowledgment.

this

order

acknowledgment.

the

an

speed

algorithm

the

round-trip

the

ill-programmed

time

instance

to produce

frequent

trading

the

round-trip

of the

elapses

execution

in Europe,

the

order

acknowledgment.

the

the

or less;

in Asia

order

receives

States,

an

out

from

the

out

sends

receives

United

is 10

the

venue

elapses

moment

sends

keeping

two

other

for

executed

easily

to

someone

experience.

Distortion
simulation

during

a run

recorded

by

data

vary

can

1. The

number

runs
time
the

in its
of

another

process

greatly,

own

that
mostly

of financial

time

using

quotes

process.

The

collected

the

because

of the

instruments

for

collected

frequency
data

that

following

which

the

and
of

is now
two
original

the

stored
quotes

historical
factors:
process

collected
2.

quotes.

The

speed

of

the

computer

is

due

system

on

which

the

original

process

ran.
Their

impact

realization
(the

in most

quote

receive
the

The

locally:

on

control.

The

the

a remote

over
the

a.

the

following

Client

the

b. Names
2. The

are

server

will

quote

client

supplying
that

trader
a

communicate
the

runs

has

always

full

remote
with

Internet.

To

to

perform

the

or

a series

of

by

also

begin

indicate

the

the

receive
following

messages

broker-dealer

a quote
Some

For

example,

the
per

same

to

the

have

not

that

of

quotes

It

any

as

as

obscure
is
while

requested.

message.

client

is not

The
allowed

reason.
to

the

client.

soon

as

a new

higher-frequency

for
as

option

may

designing

to

the

quote

surrounding

the

EUR/USD

300

keep

is,

quotes

times

many

important

The

mannerthat

unusual

by

some

day.

the

high-volatility

accompanied
time,

client's

quotes

client

during
is

are

asynchronous

securities

it

trading

frequency

the

in an

quotes

the

for

stream

streamed

be

the

whether

requested
to

available.

to

which

acknowledging

quotes

send

rate
At

the

for

announcements,

exchange

expected

to

respond,

typically

others.

second.

a message

securities

will

will

economic

as

to

process:

(given

of the

server

becomes

one

will

any

The

geared

application
the

client

information:

response

to receive

than

server

of financial

server's

the

the

to

has

is

its

server).

server

quotes

server

identification

houses

3.

sends

such

application

almost

has

usually

the

that

which

is

client

application

with

client

server

and

comprise

a local

over

connection,

process

systems

broker-dealer
often

the

quote

application)

hardware

that

client

communication

with

(a

most

the

Most

trading

broker-dealer

server

The

is

computer

meaning

of

systems.

server

client

application,

1.

nature

and/or

and

quotes).

the

trading

collecting

quotes

quotes,

to

quotes

generate
in

mind

quote-receiving

per
only
the

part
4.

of the

application.

Quote

client
the

distortion

to

collect

client's

machine,

process

computer.

Here,

all

incoming

arrival,

with

This

queue

can

airport

discarded,
client

hence

to

client

all other
the

quote

arrival

previously,
process

the

quotes

may

stored

negligible

by

computer

clock

A client

system
from

from

on

human

its

alone.

If

generate

queue.

securities
any

significantly
the
trading

considerable.

the

quotes
signals,

as

for
given
as
are
the

length

or

full

are

vary

varying

from

lengths,

the

are

time

by

the

result

further
distortions

of
of

the

As
is

system

to

to

the

are
the

discard

and

on

taking

While
thus
by

the
differ

collected

and

on
the

distortion

quote-processing

in timestamps

or

Depending

timestamp

mathematically

quotes.

therefore

server.

to

significant

time
may

able

a file

of incoming
arrival

noted

time.

be

in

Even

a second

can

quotes
day,

quotes

client

computer
of

timestamp

earliest

processed.

takes

an

the

a quote

processing

time

all

the

engine.

fraction

quote

which

of

Unlike

queue

picks

the

copying

quote

the

client's

processor.

may

of

than

the

The

to

the

then

quotes

the

at

order

finite

series

leading

a tiny

down

find

system

system

be

arrive
the

in the

processing

standards,

given

at

older

assign

arrival

of

differ

slow

may

number

volatility

may

the

and

computer

the

check-in.

queues

faster

such

time

and

timestamp

may

the

time

the

market's

more

operation

of

equal.

the

queue

On

to the

has

processing;

to

the

occur.

airport

time

have

arrive

the

until

simple

that

for

closer

up

arrivals

queue,

queue

shifted

quote-storing

quote

the

are

filling

often

they

a queue

for

quote

as

closest

queue

being

soon

into

a line

systems

in

the

arrivals

database

the

are

from

quote

the

client

responsibility

can

located

issue:

characteristics

them,

seemingly

quote

as

placed

as

the

first

if the

quotes

queue

new

the

system

Once

the

however,
any

quotes

quotes
of

It is the

issues

are

thought

therefore,

the

several

earliest

be

next.

all

quotes

the

line,

capacity;

happens

and

their

the

often

delay

manipulated
may

be

to
even

5.

Naturally,

power
on

systems

will

encounter

faster

more

machines.

sequential

and

differences

streams

that

may

with

distortion

machines
fewer

system

of quote

computers

timestamp

drop

in

in turn

reliability

on

Faster

quotes

slightest

The

running

produce

as

can

can

be

at

in

running
processing

result.

result
trading

processing

systems

quicker

quotes

different

delivery

than

are

power

slower

Even

the

different

quote

signals.

improved

in the

following

four

ways:
1.

Time-stamping

before

putting

the

2. Increasing
3.

quotes

the

Increasing

collected

the

on

any

four

fairly

easy

built

from

quote

to

scratch,

arrives

queue.

to

the

largest

size

securities

for

which

ask

the

many

executing

addressed

the

greater
client

in particular

However,

firms

to

establishing
when

and

For

feasible

the

when

brokers,
planning

software
in the

using

shelf

may

to use

quote

application

off-the

be

an

given

quotes

are

the

reliability

are

is designed

and

FIX

clients,
difficult

protocol

or

those

impossible

client,

how

preceding

the

for

including

off-the-shelf

manufacturer

to

it may

be

issues

client.

of Execution

Duration

of

execution

can

strategies

for

arbitraging

temporary

depend

on

the

ability

Whoever

detects

exchange

first
of

trading

platforms:
speed

the

to

execution

get

make

the

mispricing

is likely

Speed

The

quote

client.

implement

customize.

Speed

quote

of

toward

by

be

each

queue.

memory

given

distributed

can

of the

number

steps

delivery.

prudent

the

when

analysis.

Reducing

These

size

into

system

cost/benefit
4.

quote

immediately

or

break

market
orders
and

HFT

mispricings,

posted
gets

his

with
order

to generate

the

most

profit.

controlled

by

the

following

is

of applications

generating

trading

signals.

models.
for

Most
example,

lightning
posted

components

speed.
on

the

of

The

proximity

executing
The

of

applications

generating

trading

signals

to

the

broker.

speed

of

the

executing

broker's

platform

in

routing

execution

requests.
The

proximity

The

speed

Figure

16.3

illustrates

the

Figure

16.3

Execution

Process

To

of the
of the

enhance

physical
broker

message

the

and

administration
of

and

clients

proximity

that
or

work

at

and

trading

are

hosting

execution

the

speed

percent

discussed

in

to

the

and

the

of execution

in proximity

centers.

employ

systems

of providing

99.999

due

clients

typically

making

process.

delays

between
on

orders.

of execution

servers

services

failure,

are

the

alleviate

their

capable

power
least

to

dependent

hosting

exchange.

flow

signals

or house

proximity

systems

applications

in processing

of

staff

to the

time-dependent

to co-locate

Co-location

broker

security

exchange,

choose

case

exchange

transmission
or

often

executing

recovery
sure

of

the

detail

in

services

that

the

time.

in
client

Co-location

Chapter

of

the

book.
Testing
The
or

Trading

costs
bugs,

of
can

Systems
rolling
be

out

a system

substantial.

that

contains

Thorough

programmatic
testing

of

the

errors,
system,

therefore,
the

is essential

following

Data

Testing

has

testing

Regression

testing

Automation

testing

Set

Testing

Data

set

testing

historical

data

streaming
that

the

the

data

Data

and

in a back-test
The

minimizes
to

is

built

data

past

on

time.

to change

popular

consistency

are

time,

data

but

caused

the
by

is

whether

obtained

testing

from

is to

and

analysis

and

should
at

ascertain

distortions
trading

for

testing

observed

testing

also

in
signal

is

at

given

is

consistent

be

consistent
observed

allow

for

should

for

distributions

not

be

drastic,

disruption.
based

It is implemented

sampled

frequencies:

USD/CAD

market

for
that

exhibit

should

changes

data

received

different

should

a large-scale

for

data

distribution

example,

distribution
set

all

statistical

sampled
for

data

procedure

set

data,

influences

that

data

USD/CAD,

of autocorrelations.

data

into

when

Naturally,

with

they

for

premise

The

one-minute

year.

the

data

of data

run-time

the
fall

properties

historical

the

should

throughout

one-minute

real-time

objective

that

of

smoothly.

security

distributional

or

validity

undesirable

ensure

testing

consistent

to

provider.

work

particular

used

system

set

unless

refers

data

generation

1.

the

model.

testing

Data

testing

of the

testing

Integration

the

roll-out

testing

System

with

to wide

stages:

set

Unit

prior

on
as

testing

for

follows:

frequencysay,

10-second

intervals.
2. Autocorrelations

are

estimated

for

a moving

window

of 30

to 1,000

observations.
3. The
outliers

obtained
are

autocorrelations

identified,

and

their

are
origin

then

mapped

is examined.

into
The

distribution;
distributional

properties

can

Have

the

month,
Are

quarter,

Testing

Unit

testing

system

no

due

be

verifies

the
a

analysis

engine.

ensures

that

avoiding

expensive

each

can

Testing

any

errors
system

code

the

or to the

sampling

software

the

of

are

part

code

past

addition

frequencies

to

component
of an

for

the

small

caught

lowest

of

blocks

early

in

the

function

the

from

the

at later

the

or

componentfor

component
in

of

application;

medium-level

disruptions

posttrade

the

ground

integration

up

process,

stages.

Testing

integration

testing

is

components;

the

aggregates

of

to

its

code

ensures

any

Testing

System

testing

system

test
as

completed

System

testing.

test
is

the

of

is

Testing
defects

a postintegration

testing

the

system

code

As

incorporates

its

name

to
being

of

code

increasingly
built

modular

up

larger

from

modular

interoperability

are

caught

and

test

of

system

several

implies,

interoperability

administered

state.

that

is

unit

the

testing

fixed

once
early.

as

processes

whole.

described

follows.

Graphical

user

human

interface

responsible
testing
on

during

occur.

from

code

follows

as

different

measurement

testing

The

questions:

box?

is a testable

range

latency

of the

individual

A unit

Integration

again

following

changed

production
at

functionality

example,

pieces

the

deviations

properly.

to

Integration

version

repeated

that

of a unit

method

to the
on

systemic

works

definition

distribution

the

or year?

should

that

to answer

the

of programs

testing

Unit

further

of

changes

or removal

ensure

analyzed

properties

these

The

be

for

typically

screen

are

interface
of

the

monitoring
ensures
connected

(GUI)
system

software
enables

trading
that

activity)

all the

with

the

buttons
proper

testing
the

ensures

user

(e.g.,

to perform
and

displays

functionality

her

that
the

person

tasks.
that

the

GUI
appear

according

to

the

specifications

development
and

is

not

speed

of

system

performance

limited

take

functionality.

is a critical

if the

price
time?

worst-case
performance

testing

if an

that
to

is often

solution

breach-detection
occurs.

HFT

coming

from

hijack

account

the

to

the

such

trading

an

many
without

incurring

where

or

system
must

can

subsequent

to

the

significant

wreak
tested
to

testing

system

the

the
timing

percent

that

shuts

the

not

of

and

to

to

the

create

capacity
process

event

attempt

to

confidential
However,
employees

improper

and

threats

capital.

having

is

provide

the

may

workers

impact?

testing

other

into

testing

security

underestimated;

performance

in

users

considerable

the

or

plan

trading

profitably

other

either

breaches

and

the

the

affect

Security

vulnerable

taken

down

they

be

and

within

P&L?

breaches

steal

system

What

component

be

and,

the

10

organizations.

disgruntled

refers

does

scenarios

does

will

unscrupulous

should

be

the

of high-frequency

positions?

how

passwords,

can

testing

to document

a contingency

can

attempt

intent

testing

securities

and

threats

possibilities

Scalability

the

as

hypothetical

its

overcoming

Internet,

intraorganizational
malicious

the

systems

Is

occurs

and

by

numbers,

in

God

security

for

request?

how

indispensable

mechanism

breach

information

and

possible

long

drops

holding

overlooked

identify

GUI

how

testing

extreme

there,

another

concerns

example,

of

system

system

such

attempts

security

act

are

is

of
For

What
the

of the

impact

a particular

to

example,

process

of

of the

software

with

the

scenarios

designed

of

perspective?

component

performance.

leaving

process

phase

in nature

shutdown

A stress-testing

system's

Security

design

include

For

management

quantify

exchange,

may

a risk

subsequently,

short

and

from

systems.

GUIs

a system

trading

very

the

is similar

process

testing

react

testing

to

Stress

the

to

a particular

acceptable

on

during

process.

Usability
but

developed

costly

access
havoc.

All

account.
of

the
at

The

system.
the

How

same

answer

time
to

this

question

may

reality.

appear

Every

requires
large

trivial,

incremental

an

allocation

number

of

considerably

quotes,

trading

signals,

maximum

slow
and

the

trading

anything

but

added
and

to

Internet

system

P&L

as

of

in

system

bandwidth.
on

the

performance,

a result.

same

distorting

A determination

securities

platform,

trivial

the

simultaneously

down

number

of each

is

power

processed

permissible

characteristics

matter
measure

computer

securities

may

the

security

of

machine

the

but

will

including

be

based

available

of
on

the

computing

power.
Reliability

testing

system.

Reliability

What
can

determines

are
we

the

testing

these

include

unexpected

memory

space,

operating.

system

should

be

Recovery

for

should

not

exceed

whether

an

act

process

ensures

operational

of God

ensures

that

terminations

of

computer

system

data

system.
When

is suddenly

upon

restart.

normally

if

unplugged

and

plugged

Use-Case

Testing

The
system

term

use-case
according

the

testing
to

the

that
crash,

the

time.
is

The

Recovery
the

the

network
back

refers
system

cable

and

it

testing

is

also

unexpected
include

running

application

should

event,
recovery

should

application

system

adverse

restored

is

the

stop

of time).

through

application

restarted,

the

percent

recovery

testing

to

high-frequency

documented

maintained

may

system

an

is

often

insufficient

(i.e.,

in

integrity

Similarly,

operating

then

verification

system's

integrity

the

scenarios:

99.99

How

to

well-designed

operational

a system

the

the

percent

the

conditions

due

leads

of
questions:

fails?

failure

0.01

prespecified

data

following

valid

or

that

within

to

failure

following

system
The

that

any

to remain

of

shutdowns

else

rate

refers

the

crashes,

failure

testing

the

to occur?

anything

guaranteed

rate

answer

which

conditions

and

probable
to

under

system

The

trading

seeks

conditions

expect

the

and
should

should
be

the
the
have

continue

unexpectedly

in.

to

the

performance

process

of
guidelines

testing

the
defined

during

the

testing,

design

stage

a dedicated

documents

any

the

behavior

that

the

tester

system

is

not

deployment

are

trained

scenario

and

actual

are

impartial

not

levels:

is

issues

impair

and

and

may

the

ranks

involved

coded

the

and
ensures

software
system.

The

reasons:
between

system

given

module.

in code

the

development

be

need

to

the

than

and

are

that

of

organization.

testers

usually

span

inconsequential.

system

Moderate

addressed

bugs

following

bugs

are

addressed

the

more

until

three

Critical

performance

priority.

be

expensive

by

and

highest

to

for

reported

Inconsequential
not

professional

less

in savings

intended

need

functionality.

testing

by

several

of the

moderate,

with

and

behavior

Use-case

discrepancies

considerably

bugs

critical,

addressed

observed

system

parameters.

for

document

the

use-case

errors.

or

significantly

the

who

is important

resulting

Discrepancies

its

In

of using

performed

emotionally

labor

programmers,

within

to

steps

occur.

performance

to found

Testers'

to

programmers

Testers

development.

between

typically

of testers

Testers

the

supposed

is

the

system

follows

is operating

testing

testers,

the

discrepancies

that

Use-case

of

and

need

are

considerable

bugs

of

bugs

in

to

the

critical

cosmetic

a downtime

be

variety

occurs

within

of coders.

Summary
Implementation
process,

in

noncritical
Testing,

of

systems?
2. What

of

is the

to best

systems

mistakes

however,

End-of-Chapter
What

which

components

according

1.

high-frequency
can
the

system

paramount

practices

be

very
may

activity

established

is

for

time-consuming

costly.
be

that

prudent

has

software

Outsourcing

to be

strategy.
conducted

development.

Questions

are
What

the

stages
are

is a back-test?

the

in
stages
What

development
in HFT
are

the

of

high-frequency

implementation?
peculiarities

of back-testing?

trading

3.

Suppose

125.13
can

the

when

the

the

its

market

researcher

4. Suppose
How

back-testing
price

assume

a system

much

system

paper-trading

is

placing

is 125.14.

that

produces

the

At

limit

order

a Sharpe

ratio

testing

does

this

a limit

what

market

was

methodologies

6. What

is use-case

can
testing?

be

deployed

Why

in data

is it valuable?

price

in the

need

performance?

5. What

order

at
level

executed?

of 12

system

buy

testing?

to

back-test.
ascertain

ABOUT

THE

IRENE

AUTHOR

ALDRIDGE

recognized

is an

expert

high-frequency

on

Industry

Finance

and

Risk

LTD,

vehicle

specializing

strategies.

an

At

advises

and

trading

in

venues

and

of

trading

runs
the

continuously

updated

hundreds

of

has

Committee

High-Frequency
holds

Columbia

called

panels,

Futures

for

Goldman

been

for

has

Future

Trading

regulation
is

also

the

latest

institutional

market

developed

been

broker-dealers.

and

is

well-received

upcoming

various

by

schedule

institutions

Sachs
to

and

the

Computer

Commission's

on

CIBC.

contribute

including
of

design,

making
and

on

of

at HFTtraining.com.

worked

including

pools)

and

for

The

is available

Aldridge

regulatory

Commodity

from

Alpha,

in Toronto,

government

Aldridge

professionals.

entities

strategy

was

and

and
Aldridge

Aldridge

program

program

trading

dark

investors

training

Aldridge

worldwide

Aldridge

Foresight

by

industry

to ABLE

years,

HFT

first-of-it-kind

sessions

and

an

and

resource

institutional

trading

duties,

management

online

Alpha

government

operations.
an

for

funds,

risk

ABLE

implements,

high-frequency

low-frequency

participants;

Street

optimal

at

of her

exchanges

systems,

research

Aldridge

Prior

and

as

well

high-frequency

part

hedge

(including

AbleMarkets.com,

high-frequency

As

as

a proprietary

designs,

strategies.
large

research

high-

and

Aldridge

broker-dealers,

of

training

Alpha,
trading

of

Finally,

quantitative

of

Manager
and

is

Department

Institute,

firm

and

Aldridge

University,

Portfolio

investing

educator.

York

consulting

manager,

quantitative

seasoned

New

portfolio

Polytechnic

investment

implementation

founder

at

of

Quantitative

ABLE

high-frequency

both

Engineering,

proprietary

also

and

and

Trading,

consultant,

subjects

Professor

Partner

deploys

the

trading,

currently

Managing

investment

Over

to
U.K.

Trading

Wall
the

numerous
Government

and

the

U.S.

Subcommittee

on

Trading.
an

MBA

University,

from

INSEAD,

and

a BE

an

MS

in electric

in financial
engineering

engineering
from

the

Cooper

Union

completing

her

speaker

at

Trading,

Post

to

(NPR)

and

by

New

the

Aldridge

York

with

strongly

presently

and

FINalternatives,

the

Aldridge

process
is

contributor

Reuters

to

including

with

of

frequent
academic,

the

HedgeWorld,

Dealing

on

serves

major

CBC,

current

BNN,

Journal

Advanced

Technology,

In

, the

issues

Wall

Stuart,

believes
as

Aldridge

Street

other

giving

treasurer

Journal,

major
back

at

organization

charged

the

area

York

in New

National

and

City.

with

Carnegie
continuous

BBC,

and

been

has

Radio

been

quoted

Associated
LP,

business
to the

including

Public

has

Bloomberg

and
into

ZDF,
on

addition,

Thomson/Reuters,
Jon

networks,

German

economic
radio.

Times

television

not-for-profit
jewel-like

in

publications,

Magazine,

Bloomberg

Times,

Show

presently

University.

media

Business,

discuss

Financial

York

events

appears

FOX

invited

New

is

often

CNBC,

and

mainstream

FXWeek,

Aldridge

Daily

at

Futures

Huffington

York,

industry

and

Trading,

New

PhD

top

practitioner
of

in

Press,

Forbes,
news

outlets.

community,
Hill

The

and

Neighbors,
improvement

she
a
of

ABOUT

THE

This

book

The

web

materials

WEB

is

accompanied

site

supplements

like

PowerPoint

related

materials.

register.

Once

latest
To

activity
receive

The

by

in the
these

high-frequency.

the

web

site,

web

site

in

the

book

in-class

requires

subscribers

HFT

www.hftradingbook.com

materials

presentations,

registered,

www.hftradingbook.com
enter

SITE

valid

also

with

.
teaching

project

ideas,

e-mail

address

receive

and
to

updates

on

the

web

site

at

space.

free

benefits,
.

When

visit
prompted

the
for

book's
a

password,

please

REFERENCES
Agarwal,

V.

and

Involving

Hedge

Aggarwal,

N.

and

Announcements
Rates,

and

Finance

Ahn,

H.,

11,
K.

Volatility:

Ajayi,

Asset

and

Interest

R.

Prices

A.

to

Applied

and

K.

from

Portfolio

Studies

17

1992.

Balance

Prices:

Influence

on

Rates.

Journal

of

Decisions
(1),

6398.
of

Trade

Equity

Prices,

International

Money

Chan,

2001.

Stock

Limit

Exchange

Orders,

Depth

of Hong

and

Kong.

Journal

of

Security

767788.

and

S.

M.

Mehdian,

Financial

Economics

I. E.,

2009a.

High-Frequency
E.,

1995.

An

Reactions
Empirical

Investigation.

5, 203218.

Discrete

Pair-Wise

Working

2009b.

Global

Surprises:

Trading.

I.

Schirm,

the

US-Induced

Aldridge,

and

8095.

Major

Aldridge,

Risk

of Financial

C.

and

Bae

56,

2004.

Review
D.

Evidence

of Finance

Naik,

Funds.

R.

Exchange

Y.

Portfolio

Management

in

paper.

Systematic

Funds

Outpeform.

Working

paper.
Aldridge,

I.

E.,

Markowitz

2010.

and

Management

F.

Fabozzi,

, Wiley

Aldridge,

I.

switching

models.

Aldridge,

I. E.,

Financial

Working

E.,

eds.,

& Sons,

2011,

2012a.

High-Frequency
Equity

Hoboken,

Estimating

Working

Markets.

with

Data.

Valuation

and

In

H.

Portfolio

NJ.
volatility

cut-outs

using

Markov

paper.

Determining
U.S.

Patent

Percent
and

of

HFT

Trademarks

Participation

in

Office,

Patent

Transaction

Tax.

Pending.
Aldridge,
Working
Aldridge,

I.

E.,

The

Consequences

of

paper.
I. E.,

Trademarks

2012c.

Office,

Aldridge,

I. E.,

Aldridge,

I.

Futures?

2012b.

Patent

2012d.
E.,

Journal

Estimating

of Trading

Impact,

U.S.

Patent

and

Pending.

Predicting

2012e.

Market

Can

Liquidity,

working

High-Frequency

, 7, 7582.

paper.
Traders

Game

Aldridge,
U.S.

I.

Patent

E.,

2012f.

and

Trademarks

Alexander,

C.,

Philosophical

Estimating

Probability

Office,

1999.

Patent

Optimal

Transactions

of

Royal

Market

Crash.

Pending.

Hedging

of the

Using

Society

Cointegration.

, Vol.

357,

No.

1758,

20392058.
Alfonsi,

A.

and

Absence

of

Working

paper.

Almeida,
Effect

A.

Price

Alvaro,
of

Schied,

Charles

R.

Goodhart

of Financial

and

Transactions.

N.

R.

C.

Impact.

Risk

18,

Amihud,

Y.,

2002.

time-series

E.

Book

Payne,

Frequency

and
Models.

1998.

The

Exchange

Quantitative

Analysis

Optimal

, 12,

Execution

Order

Richard

High

2000.

of Risk

Thum,

on

Trade

Limit

and

and

Chriss,

Journal

Almgren,

in

News'

Journal

Almgren,

Optimal

Manipulations

Macroeconomic

Behaviour.

2010.

Rate

33,

Execution

of

147.
Portfolio

6163.

Hauptmann

and

H.

Li,

2005.

Equity

Market

5762.
Illiquidity

and

stock

returns:

Cross-section

and

effects.

Journal

of Financial

Amihud,

Y.

spread.

Journal

of Financial

Amihud,

Y.

H.

Spread,

Residual

Finance

and

Markets

and

44,

Anand,

A.,

on

the

by

Informed

H.

5, 3156.

Mendelson,

1986.
Economics

Mendelson,
Risk

Asset

and

17,

1989.
Size

pricing

the

bid-ask

223249.

The
on

and

Effects

Stock

of

Beta,

Returns.

Bid-Ask

Journal

of

479486.
S.

Chakravarty

Evolution

of

and

and

Liquidity:

T.

Martell,

Choice

Uninformed

Of

Traders.

2005.

Empirical

Market
Journal

versus
of

Evidence
Limit

Orders

Financial

Markets

8, 289309.
Andersen,
Micro

T.
Effects

G.,

T.

of

Macro

in Foreign

Exchange.

Andersen,

T.

Errors:

Volatility

G.,

T.

Bollerslev,

X.

Diebold

Announcements:
American

Bollerslev

Forecast

F.

Evaluation

N.

C.

Real-Time

Economic
and

and

Review

Meddahi,
Using

Vega,

Price
93,

2005.

High-Frequency

2003.

Discovery

3862.
Correcting
Data

the
and

Realized

Volatilities.

Andritzky,

J.

R.,

Impact

of

Bonds.

Emerging

Angel,

G.

73,

Bannister

279296.

and

N.

T.

Tamirisa,

Announcements

Markets

1992.

Georgetown

Review

Limit

on

2007.

The

Emerging

Market

8, 2037.

versus

Market

Orders.

Working

paper,

University.

Arnuk,

S.

Saddle

River,

Artzner,

P.,

and

Measures

J.

Saluzzi,

Asquith,

F.

P.,

Delbaen,

R.

Classification
Atiase,

J.

Broken

Oman

K.

security

and

Eber,

Markets

and

C.

Safaya,

. FT

Press,

Upper

price

behavior

around

Research

(Spring),

in a Limit

Short
Paper

No.

and

Trade

w14158.

firm

earnings

capitalization,

announcements.

2136.

Stoikov,

Book.

Coherent

Sales

information,

Sasha

Order

1999.

9, 203228.

Working

Predisclosure

and

Heath,

2008.

1985.

Marco

D.

Finance

NBER

of Accounting

Avellaneda,

M.

Mathematical

Algorithms.

R.

Trading

2012.

NJ.

of Risk.

Journal

J.

Macroeconomic

J.,

and

Econometrica

Quantitative

2008.

High-Frequency

Finance

, Vol.

8,

No.

3.,

217224.
Bagehot,

Walter

Analysts

Journal

Bailey,

W.,

Rim

(pseudonym).
, (Mar/April

1990.

Stock

P.,

and

Prices:

Bond

E.

of Financial

and

Ball,

P.

Income
Bangia,
1999.

and

A.,
Liquidity

F.

Finance

Elton

and

Evidence

Brown,

From

Journal
X.

Diebold,

Risk,
and

with

An

and

Green,

the

U.S.

Treasury

36,

523543.

2001.

Empirical

Pacific

Journal

Wharton

Economic
Market.

Evaluation

of

Research

Schuermann

Implications

Management.

and

Implications.

C.

of Accounting
T.

Financial

22.
Announcements

Analysis

1968.

in Town.

of

9, 344356.
T.

Quantitative

Numbers.

Measurement
9906.

J.

Supply

and

Game

12-14,

Evidence

Money

Balduzzi,

Only

1971),

Money

Markets:

International

R.

US

The

and
for

Journal

Accounting

6, 159178.
J.

Traditional
School,

News

D.

Stroughair,
Market

Working

Risk
Paper

Barany,

E.,

Detecting

M.

P.

Market

Data.

Kent

G.,

Markets.

Berber,

A.

Finnerty

and

News

and

the

Efficiency

C.

Caglio,

Empirical

Bernanke,

Ben

60,

Bervas,

Arnaud,

limit

B.,

P.

order

Finance

2006.

book

Market

and

and

Borrowing.

C.

the

1972.

Journal

Boscaljon,

B.,

Spatt,

order

International

16,

131145.

1996.
Bond

Strategies

and

Working

2005.

Reserve

and

paper,

What

Explains

the

Policy?

Journal

of

Its

Review

1995.
flow

in

the

Incorporation

into

8, 6379.

An

A.

Securities

45,

Regulatory

T.
are

and

Market

empirical
Paris

analysis

Bourse.

J.

of

the

Journal

of

George,

with

in
10,

2008.
Working

T. J. George,

2011.

with

Restricted

444455.

of Business

informed.

Markets

Equilibrium

Changes

Journal

and

providers

Boulatov,

of

NYSE.

Kuttner,

Stability

of Business

2005.

A.,

liquidity

the

Liquidity

Capital

International

Boulatov,

J. Kopecky,

16551689.

Fisher,

Industry.

from

Using

623634.

Submission

Federal

Financial

Hillion,

50,

Black,

Markets

N.

to

12,

2012.

Effects

Kenneth

Order

Kenneth

Sengupta,

12211257.

Management.

Biais,

in

and

Reaction

Finance

Risk

S.

2004.

Evidence

of Lausanne.

Market's

Finance

of Futures

I.

Long-Memory

E.

University

Stock

and

Analysing

Quantitative

Journal

and

Information:

by

Florescu

Joseph

Macroeconomic
Futures

I.

Crashes

High-Frequency
Becker,

Varela,

Pharmaceutical

151164.

Securities

trading

when

paper.

Hidden

Informed

the

and

Liquidity

Displayed

Liquidity

Providers.

Working

paper.
Boyd,

John

Market's

H.,
Reaction

Usually

Good

Bredin,

Don,

Shocks

and

Economics

Jian

for

to

35,

and

O'Reilly
Returns.

315331.

Ravi

Jagannathan,

Unemployment

Stocks.

Gerard
REIT

Hu

Journal
and
The

2005.

News:
of Finance
Simon

Journal

Why
60,

Stevenson,
of Real

The

Bad

Stock

News

Is

649672.
2007.
Estate

Monetary

Finance

and

Breen,

W.

equity

J.,

L.

S.

liquidity.

Hodrick,

Management

Brennan,

M.

J.,

Factor

Specifications,
of
49,

Brennan,
and

M.

Journal

Brodsky,

E.,

Fund

Impact

and

Burke,

G.,

of

C.,

an

38,
H.

J.

Alternative

Characteristics,
Returns.

and

Journal

of

Their

Limit

for

of

the

Financial

Microstructure

Illiquidity

in

Stock

441464.

foreshocks.

The

, 5 (Fall),
Jr.,

Geophysical

Order

of Hedge

Investors.

Journal

of

2644.
and

L.

Sharpe

X.

Properties

for

Lei,

of Trading

Sharper
and

Statistical

Implications

Journal

Hansch

Market

41,

density

2002.

and

1996.

Compensation

spatial

Hanweck

1994.

Open

1998.

470483.

Economics

Kat,

Liquidity.

O.

Predicting

L10305

M.

Returns

G.,

the

The

Investments

Burgardt,

2002.

Subrahmanyam,

Subrahmanyam,

of Financial

and

Alternative

Cao,

A.

2011.

Index

48,

Stock

On

Letters
C.

A.

Korajczyk,

345373.

Returns.

Brooks,

and

Expected

Pricing:

Research

A.

Security

J. and

Asset

R.

Science

T. Chordia

Cross-Section
Economics

and

Measuring

Market

1, 7084.

Ratio.

Wang,

Book.

2006.

2004.

Futures
The

Working

23,

56.

Informational

paper,

Content

Pennsylvania

State

University.
Chaboud,
of

A.,

the

E.

Hjalmarsson,

Machines:

Market.

FRB

Chaboud,

Algorithmic

International

Alain

Interest

Parity:

Economics

P.

Challe,

Edouard,

2003.

Journal

of Economic

Return
33,

L.

K.

C.,

from

but

J. Karceski

Factors.

Not

Journal

115,
and

B.

in

Chiquoine,

the

Discussion
H.
for

Sunspots

Theory

and

Trading

Jonathan

It Works,
349362.

Vega

Finance

and

66,

Chan,

C.

and

Wright,

Predictable

Rise

Exchange

No.

980.

2005.

Journal

Uncovered

of

International

Asset

Returns.

182190.
J. Lakonishok,

of

Foreign
Paper

Long.

2011.

Financial

1998.

and

The

Risk

Quantitative

and

Analysis

159188.

Chan,

L.

and

J.

Lakonishok,

1995.

The

Behavior

of

Stock

Price

around

Institutional

Cohen,

K.,

S.

Transaction

A.

Turnover,

Journal

of Financial

Cont,

R.,

A.

order

book

spread.
Corsi,

D.

1989.

Notes

David

M.,

What

1983.
38,

30,

James

Moves

Existence

of

287305.

Macroeconomic
in

the

Daily

News,

Stock

Returns.

235259.
2011,

The

Information

price

impact

of

Ulrich

Mller

and

Volatility

No.

M.

Stock

effects

on

the

bid-ask

14571469.

High-Precision

Economics

Cutler,

2005.

Stoikov,

Zumbach,

Consistent

Data.

Galai,

Gilles

89,

1981.

paper.

of Finance

Fulvio,

2001.

Working

and

Journal

S.

Whitcomb,

and

Economy

28,

11471174.

D.

Strategy,

Clustering

and

50,

and

Stivers,

Volatility

Kukanov,

T.,

Chris

Research

events.

Copeland,

Schwartz

of Political

and

and

of Finance

Placement

Journal

Robert

Stock

R.

Order

Spread.

Connolly,

Journal

Maier,

Costs,

Bid-Ask

15,

Trades.

Michel

from

Dacorogna,

High-Frequency

2, 183204.

Poterba

Prices?

and

Lawrence

Journal

of

H.

Portfolio

Summers,

Management

412.

Dacorogna,

M.

Pictet,

2001.

Press:

San

Datar,

M.,

An

T.,

Liquidity

and

Financial

Markets

Demsetz,

H.,

of Economics,
Dennis,

P.

of Stock

Short-Selling
Journal
Ding,
Approach

Muller,

R.

High-Frequency

Y.

Naik

Returns:

The

82,

1, 33-53.

W.
and

B.

Narayan

1968.

of Financial
X.,

to

A.

Olsen

and

Finance

O.

V.

Academic

and

An

Robert

Alternative

Radcliffe,
Test.

1998.

Journal

of

1, 203219.

J. and

D.

U.

CA.

Asset

Ownership

Diamond,

Gencay,

Introduction

Diego,

Vinay

R.

Liu

to Opinion

J. P.
and
and

Cost

Weston,

2001.

Informed
R.

Asset

E.

Price

Economics
and

of Transacting.

P.
Mining.

S.

The

Who's

Informed?

Trading.

Working

Verrecciha,

1987.

Adjustment
18,
Yu.

Quarterly

to

Private

An

Journal

Analysis

paper.
Constraints
Information.

277311.
2008.

Proceedings

Holistic

of WSDM

Lexicon-Based
2008.

on

Dowd,

K.,

2000.

International

Adjusting

Review
and

A.

Trade.

Journal

of Finance

Easley,

David,

Nicholas

Journal

Easley,

F.

and

in

Improved

Finance

Time

Sharpe

9 (3),

and

the

Ratio.

209222.

Price

Impact

of

24672498.

Kiefer,

Maureen

O'Hara

Information,
51,

and

Marcos

in

and

Joseph

Infrequently

B.

Traded

14051436.

Prado,

Liquidity
25,

David,

information

M.

de

and

Studies

Easley,

55,

An

and

2000.

Liquidity,

Lopez

Toxicity

Financial

Engle,

of Finance

David,

Flow

19,

R.

1996.

Stocks.

Risk:

of Economics

Dufour,

Paperman,

for

and

a High

O'Hara,

Frequency

Maureen,
World.

2012.

Review

of

14571493.
Maureen

securities

O'Hara,

markets.

1987.

Price,

Journal

of

trade

size,

Financial

and

Economics

6990.

Easley,

David,

security

price

Ederington,

and

O'Hara,

adjustment.
Louis

Information:

Maureen

H.

News

1992.

Journal

and

Jae

of Finance

Ha

Releases

Time

Lee,

and

and

47,

1993.

Volatility.

the

process

of

577606.

How

Markets

Journal

of

Process

Finance

48,

11611191.
Edison,

Hali

Rates

to

Reserve

J.,

1996.

News

The

Reaction

Releases.

System,

of Exchange

Board

International

of

Finance

Rates

and

Interest

the

Federal

Governors

of

Discussion

Paper

No.

570

(October).
Edwards,
New

Sebastian,

Information.

Evidence
Rates,
Eisler,

Martin
on

the

NBER
Z.,

Cancellations.
Eleswarapu,
in the

Effects

Working

Book

Nasdaq

of

R.,

1997.

Market.

Evans,

Monetary

Paper
and

Rates,

No.

Market

1993.

Policy

and

Some

Shocks

Empirical

on

Exchange

4271.

J. Kockelkoren,
Market

Efficiency

9, 377382.

Charles

Events:

Working
V.

Letters

and

J. Bouchaud,

Order

Exchange

Economics

Eichenbaum,

of

1982.

2009.

Orders,

The
Limit

Price
Orders

Impact
and

Paper.
Cost
Journal

of Transacting
of Finance

and
52,

Expected

21132127.

Returns

Eling,

M.

and

Performance
Journal

K.,

R.

and

A.,

Order

C.

Holden,

Indiana

on

31,
M.

P.

and

New

the
of

Choice

Hedge

of

Funds?

26322647.
The

Accuracy

Nasdaq.

of

Journal

of

Trade

Financial

529551

Jain

the

Evaluation

from

35,

Does

O'Hara,

Evidence

Analysis

Choice

the

Finance
and

Rules:

Quantitative

2007.

Influence

Michaely

Classification

Ellul,

Schuhmacher,

Measure

of Banking

Ellis,

and

F.

R.

York

Jennings,

Stock

2007.

Determinants

Exchange.

Working

of
paper,

University.

Engle,

R.

Portfolio

and

R.

Ferstenberg,

Management

Errunza,

V.

European

and

33,

K.

Stock

Execution

Risk.

Journal

of

3445.

Hogan,

Market

2007.

1998.

Macroeconomic

Volatility.

European

Determinants

Financial

of

Management

4, 361377.
Evans,

M.

and

Dynamics.
F.
NJ:

ed.,

John

and

Wiley

The

L.

S.

Impact

Farmer,

J.

Efficiency
Favre,

Deutsche

Banking

D.,

A.

Exchange

Rate

170180.

Financial

Models

Hoboken,

C.

25:

Jensen,
to

Nasir,
on

F.

A Review

of

Theory

383417.

and

R.

Roll,

New

Information.

2008.

Earnings

Share

Lillo

Impact.

Galeano,

Hedge

AG.

Markets:

1969.

The

International

Prices.

Announcements:
Journal

of

Money,

, 3646.

Market
J.-A.

A.

Size

Gerig,

2008.

Bank

and

121.

Firm

with
A.

of

of Finance

M.

and

and

and

Optimization
Ferraris,

Mohd

Shapes
L.

110,

Capital

Prices

10,

of

Investment

Journal

Stock

Review
C.,

Flow

Economy

Efficient

Fisher,

of

Fan-fah,

Order

Encyclopedia

1970.

F,

Economic

of Political

Work.

Adjustment

2002.

& Sons.

Empirical
E.

Lyons,

2012.

Eugene,

Fama,

K.

Journal

Fabozzi,

Fama,

R.

Equity

and

H.

Working
2002.

Funds.
Market

Waelbroeck,

Impact

How

paper.

Mean-Modified

Journal

2011.

of Alternative
Models.

Value-at-Risk
Investments.
Working

paper,

Flannery,

M.

J.

and

Factors

Do

Influence

Studies

15,

751782.

Fleming,
Bond
Review

Protopapadakis,

J.

and

Federal

Stock

Eli

M.

2002.
Returns.

Remolona,

Reserve

Bank

Macroeconomic

Review

1997.

of New

What

York

of

Financial

Moves

Economic

the
Policy

3, 3150.

Fleming,
and

A.

Aggregate

Michael
Market?

A.

Michael

Liquidity

in

Information.

the

and

Eli

U.S.

Treasury

Journal

Fleming,

Michael

Structure

Of

M.

P.

J.

A.,

Optimal

Trade

Working

paper,

Foster,

F.

J.

and

S.

Kennedy,

University
S.

19011915.

M.

Remolona,
BIS

S.

the

T.,

Kadan

T.

Price
Response

1999b.

Working

Tse

Formation

H.

Public

The

paper

and

to

Term

No.

71.

Windcliff,

Mean-Quadratic-Variation

2011.

Approach.

of Waterloo.

Viswanathan,

Forecast

The

54,

Effects.

1999a.

Market:

Eli

Execution:

and

Remolona,

of Finance

Announcement

Forsyth,

Agents

J.

1996.

Forecasts

of

Strategic

Others.

Trading

Journal

of

When

Finance

51,

14371478.
Foucault,
Market

for

Thierry

O.

Liquidity.

Foucault

Anonymity

&

Freeman,

C.

Lessons

from

Frenkel,

Jacob,

of News':

Sophie
in

Studies

E.

Review

Matter

Financial

and

20,

1987.

Kandel,

of Financial
Moinas

Electronic

Limit

Studies
Erik

Order
18,

As

11711217.

Theissen,

Order

Book

2007.

Markets?

Does

Review

of

17071747.

London,

1981.

Lessons

&
Limit

Technology

Japan.

2005.

Flexible

from

the

A.

Hiseh,

Policy

and

Frances

Performance:

Pinter.

Exchange
1970s.

Economic

Rates,

Journal

Prices

of Political

and

the

Economy

Role
89,

665705.
Fung,

W.

and

D.

Dynamic

Trading

Strategies:

Financial

Studies

10,

275302.

Garman,

Mark,

1976.

Market

Economics

3, 257275.

1997.
The

Case

Empirical
of

Microstructure.

Hedge

Characteristics

of

Funds.

of

Journal

Review

of

Financial

Garman,
Price

M.

B.

and

Volatilities

M.

J. Klass,

from

1980.

Historical

On

the

Data.

Estimation

Journal

of

of Security
Business

53:

6778.
Gatev,

E.,

Trading:

W.

N.

Goetzmann,

Performance

Financial

Studies

Gatheral,

J.,

Quantitative

K.

G.

Rouwenhorst,

of a Relative-Value

2006.

Arbitrage

Rule,

Pairs

Review

of

, 797827.
2010.

No-Dynamic

Finance

Gatheral,

J.

and

geometric

Brownian

International

&

10,
A.

and

Market

Optimal

trade

Impact.

749759.

Schied,

2011.

motion

Journal

Arbitrage

in

of

the

Almgren

Theoretical

and

and

execution

Chriss

under

framework.

Applied

Finance

14,

353368.
Gatheral,
impact

J.,
and

A.

Schied,

and

Fredholm

A.

integral

Slynko,

2012.

equations.

Transient

linear

Mathematical

price

Finance

22,

445474.
George,

T. J.,

Bid-Ask

Spread

Financial

Kaul,
and

Studies

Getmansky,
of

G.

A.

Glosten,

L.

R,

properties

Glosten,

L.,

1994.

L. R.

Bid-Ask

C.

New

Estimation

Approach.

B.

of the

Review

of

Is

Financial

Markets:

Finance

4, 73114.

E.

the

hedge

Arndt,

Issues

fund

of
prices.

49,

econometric
returns.

model
Journal

of

Journal
Open

Estimating

of Financial
Lutat,

Working

and

the

spread
of Finance.
Limit

Order

Book

the

Components

Economics
and

T.

21,
E.

of

123142.
Uhle,

2011.

paper.

O'Hara,
and

bid-ask

11271161.

1988.

M.

M.

the

Electronic

Harris,

Journal

and

An

529610.

Trading.
A.

in

2004.

Components

L. E.

Spread.
P.,

illiquidity

of Finance

and

High-Frequency
Goodhart,

1991.

I. Makarov.

of transaction

Journal

Glosten,

and

74,

1987.

statistical

Inevitable?

Nimalendran,

Components:

and

Economics

Gomber,

Lo,

correlation

Financial

the

its

M.

4, 623656.

M.,

serial

and

1997.

Applications.

High

Frequency
Journal

of

Data
Empirical

in

Gorton,

G.

about

B.

and

Commodity

Gregoriou,

Ratio.

N.

of

and

B.

Smith,

Determinants

Grilli,

Gueyie,

J.-P.,
Funds

Order

and

Journal

62,

2003.
Using

Investments
Turnbull

Facts

Fantasies
47-68.

Risk-Adjusted

Modified

Sharpe

6, 7783.

and

R.

White,

Aggressiveness.

2000.

The

Journal

of

Costs
Financial

6588.

and

Puzzling

Analysts

Hedge

A.

of

56,

V.

of

2006.

Financial

of Alternative

M.,

Economics

Rouwenhorst,

and

Funds

Journal

Griffiths,

G.

Futures.

G.

Performance

K.

N.

Roubini,

Evidence

from

1993.
the

Liquidity
G-7

and

Countries.

Exchange

Rates:

Mimeo,

Birkbeck

College.
Handa,
Price

P.,

R.

Schwartz,

Formation

Markets

in

and

an

Order

G.

Prices.

Journal

Harris,

R.,

A.,

1987.

Driven

2003.

Quote

Market.

Setting

Journal

1997.

Harris,

Optimal

1998.

Stylized

Trading

Instruments

7, 176.

Hasbrouck,

Joel,

Trades.

Journal

Hasbrouck,

of

Oxford
Hasbrouck,
Hybrid

2005.
from

Hasbrouck,
institutions,

1991.

J.,

2007.

Economics,
University
J.

39,

Markets

Dynamic

and

Financial

Review
Order
Financial

Measuring

the

46,

Data.

Stock

Development:

41,

13946.

Submission

Strategies

in

Markets,

Institutions

&

Information

Content

of

Stock

179207.
Costs

and

Working

Empirical
and

and

131140.

and

Problems.

Trading
Daily

Information

Business

Economic

of Finance

J.,

Evidence

and

Stock

European

L.,

Macroeconomic

of Economics

Re-assessment.

The

Tiwari,

6, 461489.

Hardouvelis,

Some

A.

Returns

for

US

Equities:

paper.
Market

Microstructure:

Econometrics

of

Securities

The
Trading

Press.
and

Market:

G.
The

Saar,
Island

2002.
ECN.

Limit

Orders

Working

and

Volatility

paper,

New

in
York

University.
Hasbrouck,

J.

and

G.

Saar,

2011.

Low-Latency

Trading.

Working

paper,

New

Hautsch,
and

York

N.

&

Optimal

Data,

University.
R.

Huang,

Order

Sizes:

SFB

649

2011.

Limit

Evidence

from

Discussion

Sonderforschungsbereich

Order

Flow,

NASDAQ

Papers,

649,

Market

Impact

TotalView-ITCH

SFB649DP2011-056,

Humboldt

University,

Berlin,

Germany.
Hedvall,

K.,

Sellers

Behave

Data

J.

Niemeyer
Similarly

Examination

Empirical

Finance

Hollifield,

B.,

Limit

Order

Horner,

and

of

Evidence

the

Rosenqvist,

Limit

1997.

Order

Finnish

Book?

Stock

Do
A

Buyers

and

High-Frequency

Exchange.

Journal

of

Analysis

of

4, 279293.

R.

Miller

and

Markets.

M.

in

G.

R.,

Review

1988.

from

P.

Sandas,

2004.

of Economic

The

Value

Switzerland,

of

Studies
the

Journal

Empirical

of

71,

10271063.

Corporate

Voting

Banking

and

Right:

Finance

12,

6984.
Huang,

R.,

Spread:

and

H.

Stoll,

General

1997.

The

Approach.

Components

Review

of

of

Financial

the

Bid-Ask

Studies

10,

9951034.
Huberman,

G.

and

Quasi-Arbitrage.
Hvidkjaer,

ITG

Stanzl,

Econometrica

Soeren,

Review

W.

2006.

of Financial
Global

Trading

72,

Studies

2004.

Manipulation

and

12471275.

Trade-Based
19,

Cost

Price

Analysis

of

Momentum.

457491.

Review,

2010.

Investment

Technology

Group.
Jegadeesh,

N.

strategies:

An

Finance

, 56,

Jobson,

J.

testing
36,

with

and

S.

Titman,

evaluation

of

2001.

Profitability

alternative

of

explanations.

momentum
Journal

of

699720.
D.

and

the

B.

Sharpe

M.
and

Korkie,

1981.

Treynor

Performance

measures.

Journal

hypothesis
of

Finance

889908.

Jones,
Volatility.

C.,

G.

Kaul

Review

and

M.

of Financial

Lipson,

1994.

Studies

Transactions,

7, 631651.

Volume

and

Jones,

C.

M.,

Macroeconomic
Financial

O.

Lamont

News

and

Economics

Jorion,

47,

Philippe,

Analysis.

Bond

R.

L.

Market

Lumsdaine,

1998.

Volatility.

Journal

of

315337.

1986.

Journal

and

Bayes-Stein

of

Financial

Estimation

and

for

Quantitative

Portfolio

Analysis

21,

279292.
Jorion,

Philippe,

Capital
Issue

2000.

Risk

Management.

Management

European

Lessons

Financial

from

Long-Term

Management

Vol.

6,

3, 277300.

Kandir,

Serkan

Yilmaz,

Characteristics

and

International

Research

Kaniel,

R.

Use?

Journal

Kaplan,

and

H.

D.

Downside

Evidence

Journal

of Finance

and

2006.

J.

A.

K.

and

Liquidity

Provision.

Kawaller,

I. G,

Behavior

and
and

the

the

Informed

Kappa:

3545.
Traders

Generalized

Measure.

Koch

and
of

500

2004.

of Financial

Extent

S&P

2004.

Odders-White,

D.

Do

16,

Journal

of

8, 4254.

Review

P.

Turkey.

18671913.

Performance

E.

Firm

from

Economics

Orders

Knowles,

Measurement

Kavajecz,

What

79,

Risk-Adjusted

Performance

Variables,

Returns:

Liu,

and

Macroeconomic

Stock

of Business

P.

Prices

2008.

Studies

T.W.

Koch,

Feedback

Index.

Technical
17,
1993.

of

and

10431071.
Intraday

Between

Journal

Analysis

S&P

Market

500

Financial

Futures

Research

16,

107121.
Kestner,

L.

Futures

25,

Kirilenko,
Flash

N.,

A.,

Crash:

Kissell,

Kissell,
Distribution

A.

The

Robert

R.

Handle

on

True

Performance.

and
of

Kyle,

M.

of High

Samadi,

and

Frequency

T.

Tuzun,

Trading

2011.

on

an

The

Electronic

paper.
and

New

S.

Impact

Working

AMACOM,

Getting

4446.

A.

Market.

1996.

Morton

Glantz,

2003.

Optimal

Trading

Strategies

York.
R.

Malamut,

Trading

2005.

Understanding

Algorithms.

Institutional

the

Profit

Investor

and
, Guide

Loss
to

Algorithmic

Trading,

Krueger,

Anne

Labor

Market

paper

5769.

Kyle,

A.,

B.,

Modles

of

Respond

Market

More

Rationality.

Auctions

to

Reliable

NBER

and

working

Insider

Trading,

13151335.

2002.

de

Test

Markets

Continuous

53,
E.,

Do

1985.

Saout,

2005.

1996.

Data?

Econometrica
Le

Spring

Intgration

Valeur

en

du

Risqu.

Risque

de

Banque

et

Liquidit

dans

les

Marchs,

No.

61,

NovemberDecember.
Lee,

C.

Data.

and

Ready,

Journal

Lhabitant,
funds.
Li,

M.

of Finance

F.-S.,

and

Zuliu

F.

Lillo

J.

of

John,

Lo,

1965.

W.

and

Profits

Due

Studies

3, 175208.

to

A.

Business

Conditions.

to

R.

and

and

R.

Lyons,
Foreign
321351.

and

and
A.

421,

from

future
46,

Intraday

the

is

hedge

111.

Stock

Economic

Market

to

States.

IMF

47,

Quantitative

Exchange

K.,

1995.

Market.

for

Selection
Budgets.

Are

Review

The

Adjustment

The

Role

of

Financial

Bonds

and

7, 137146.
of

Order

Contrarian
of

Stocks,

Economics

43,

the

Capital

When

On

Financial

Exchange
Flow.

Rates

Journal

of

467488.

of Microstructural
Journal

and

and

1990.

1997.

Analysis
Tests

Assets

Overreaction?

Nummelin,

2008.

curve

1337.

MacKinlay,

Information:

Richard

of Risk

Statistics

Applied

Master

129130.

Portfolios

Craig

Payne,

2003.

Stock

Market

K.

Macroeconomic

Financial

of

Mantegna,

Valuation
in

Stock

Lflund,

Love,

the

Review

across

R.

The

of Economics

Andrew

bright,

Responses

Nature

Investments

Review

is

Business

1998.

and

function.

Risky

Direction

98/79.
Farmer

price-impact
Lintner,

future

Announcements

Paper
F.,

The

Hu,

Trade

733747.

International

Macroeconomic
Working

Inferring

46,

2004.

Thunderbird

Li

1991.

of

Financial

Hypotheses
Economics

in

the
39,

Lyons,

Richard

Rates

. MIT

Mahdavi,

2004.

Harry

Approach

to

Exchange

Return
Sharpe

When

Ratio.

Returns

Journal

Are

of

Not

Alternative

4757.

M.,

1952.

Portfolio

Selection,

Journal

of

Finance

7791.

Markowitz,

Harry,

Investments

Blackwell,

John

H.

Portfolio
New

Portfolio

Wiley,

M.

Choice

Hope,

Business
J.

and

P.

York.

Todd,

Capital

F.,

and

V.

Second

2000.

Markets

Vance

Conditions.
1961.

Movements.

Diversification

Edition,

of

1991,

Mean-Variance

. Frank

F.,

R.

Control

Regulation

Business

Rational

Bowman,

J.

1993.

Basil

Analysis

Fabozzi

in

Associates,

315-335.

D.

Emanuel,

Equity

Prices,

Studies
and

29,

Firms'

Stock

of Financial

Expectations

and

on

51,

Roley,

Review

Econometrica,

Navissi,

Values.

6, 683707.

the

1999.

News,

Theory

The

Effect

Journal

of

of

Price

of

Price

Economics

3347.

Tatiana,

Control:

New

Efficient

MA.

and

Grant

Nenova,

Selection:

Pennsylvania.

McQueen,

Muth,

1959.

Cambridge,

Markowitz,

68,

Microstructure

Adjusted

6 (Spring),

Markowitz,

and

The

Risk-Adjusted

Distributed:

Investments

and

2001.

Press.
M.,

Normally

7 (1),

K.,

2003.

The

A Cross-Country

Value

Analysis.

of

Corporate

Journal

Voting

of

Rights

Financial

and

Economics

325351.

Niedermayer,

Andras

Markowitz's

and

Critical

Line

Daniel

Niedermayer,

Algorithm.

2007.

Working

paper,

Applying

University

of

Bern.
Nikkinen,
Stock

J.,

M.

Market

Omran,

Reactions

Announcements.
Obizhaeva,
Supply/Demand
Odders-White,

Global
A.

and

J.

R.

Sahlstrm

and

J.

to

Scheduled

US

Macroeconomic

17,

92104.

Finance
Wang,

Dynamics.
H.

P.

and

Journal
2005.

Working
K.

J.

Optimal
paper,

Ready,

ij,

Trading

2006.

Global
News

Strategy

and

Ratings

and

MIT.
2006.

Credit

Stock

Liquidity.

Review

Orphanides,

Athanasios,

Macroeconomic
the

Reserve

Parlour,

C.

A.W.A.

Boot

and

L.

and

and

stock

returns.

Economic

R.

K.

W.

in the

and

Securities,
Routing.
A.,

Markets.

Journal

Anomalies.
Roll,

Ross,

in an
S.

Bicksler

Vitale,

eds.,

Risk

Note

58,

of

Financial

and

expected

642685.
Stock

Prices

and

4967.

Interdealer

Trade

In

J.

and

Frankel,
of

Information

G.

Galli,

Foreign

and

Exchange

Destinations

and

Smart

Order

Book

and

Price

Series.
Aggressiveness
Markets

The

in

Limit

7, 5374.

Specialist's

Order

Book

Harvard.

Simple

Implicit

Market,

Measure

Journal

Return,
and

, 113,

Inverted-Price

paper,

In

Press.

of Financial

Efficient

risk

1985.

Market.

Order

Liquidity

Microstructure

2011.

1977.

of

Amsterdam.

Roley,

of Chicago

A.,

News:

A Survey.

Handbook

Economy

1996.

The

1996.

1984.

Bad

of Governors

Markets:

eds,

2003.

Exchange

Working

R.,

Spread

P.

2004.

Kevin,

Thakor,

Vance

Research

Ranaldo,

Is

Board

Order

of Business

eds.,

Pragma

Rock,

V.

Foreign

. University

119157.
News

Market.

, North-Holland,

Journal

Markets

Order

V.

of Political

and

Giovannini,

19,

Good

Limit

Stambaugh,

News.

Perraudin,

Stock

2008.

A.

Journal

D.

When

the

Seppi,

& Banking

Pstor

A.

and

Studies

System.

D.

Intermediation

Flows

1992.

News

Federal

Pearce,

of Financial

Risk

Return

and

of

the

of Finance
Arbitage.

in Finance

Effective
, 39,

Bid-Ask

11271139.

in I. Friend

, Massachusetts:

and

J.I.

Ballinger,

189218.
Rosu,

I., 2005.

paper,

University

Rosu,

I.,

2010.

Working

paper.

Sadeghi,

M.,

Macroeconomic

Dynamic

Model

of the

Limit

Order

Book.

Working

of Chicago.
Liquidity

1992.
News:

and

Stock
The

Information

Market
Australian

in

Order

Response
Evidence.

Driven

to

Markets.

Unexpected
International

Monetary

Fund

Sandas,

P.,

Working
2001.

Making:

Empirical

Financial

Studies

Savor,

P.

14,

and

Schwert,
over

Shadwick,

W.

Measure.

Journal
M.,

and

Equilibrium

Does

Market.

Returns

of

Scheduled
paper,

The

Provision

with

of Financial
C.,

1964.

Limit

Studies

2002.

6 (3),

Prices:

Risk.

Performance

RAPMs

Asset

and

103150.

Universal

Management

of

Change

Orders

10,

Measurement

Capital

Conditions

Volatility

1115-53.

of Investment

under

and

Market

A.I.R.A.P.Alternative

F.,

Market
Review

Working

Stock

, 44,

Keating,

Journal

William

Order

Asset

of Performance

2004.

Investments.
Sharpe,

Why

Review

F.

Competitive

of Pennsylvania.

Liquidity

Specialist.

Sharma,

2008.

of Finance

1997.

a Limit

and

Announcements.

1989.

Journal

D.,

Strategic

Wilson,

University

W.,

Time?,

Seppi,

from

News

G.

Selection

705734.

M.

School,

92/61.

Adverse
Evidence

Macroeconomic
Wharton

Paper.

5984.
for

Alternative

2 (4),

106129.

A Theory

of

Journal

of

Market

Finance

19,

425442.
Sharpe,

William

Business

39,

Sharpe,

William

of

the

Treasury
Simpson,
An

F.,

Marc
Impact

1992.

and

of

Marc

Asset

Class

Shares.

Performance.

Allocation:
Journal

Sanjay

Journal

Management
of

Portfolio

F.

Ramchander,

Macroeconomic

W.,

of Futures

Sanjay

Ramchander

Estate
and

News

Journal

Response

of Real
Brian

223239.

W.

Markets.

Smith,

Fund

of

Style

and

Management

719.

Asymmetric

Journal

Mutual

Measurement.

1992,

Simpson,

1966.

119138.

Performance
Winter

F.,

of

Finance
Ben

Journal

2004.
on

Markets
and

Equity

REIT

and

Economics

Amoako-Adu,

1995.

of

Financial

and

the

An
Spot

24,

and

Futures

453478.

James

R.

Returns
34,
Relative

Quantitative

Examination

Webb,
to

2007.

Inflation.

513529.
Prices
Analysis

of Dual
30,

Sortino,

F. A.

Portfolio

and

R.

van

Management

Sortino

F.

A.,

Triangle.

R.

Sotiropoulos,

(Spring),

der

Meer

of Portfolio

M.

G.,

paper,

Algorithmic

America

Merrill

Lynch.

Steuer,

R.

Optimization:
BWL

Y.

New

and

A.

Plantinga,

Capabilities

Journal

1999.

18,

Execution

Trading

and

Risk.

of

2731.

Order

Qi

Downside

Management

2012.

working

E.,

1991.

, 17
van

Journal

derMeer,

The

Dutch

5059.

Using

Quantitative

Trading

Signals,

Research,

M.

Hirschberger,

and

Future

Bank

2006.

Methods.

of

Portfolio

Zeitschrift

fr

, 2.

Stoikov,

S.

Order

and

Book

Stoll,

H.

Theory

1989.

and
H.

and

Stock

R.

and

A.

Improbable

GPUs,

Field

Theodoros
Adaptive

Good
Financial

P.,
Times:

and

and

44,

Spread:

115-134.

Dynamics
Journal

2005.

of
of

Risk

Swan:

The

for

Luk,

2009.

Online

Stock

Index

Financial

and

Management

with

Impact

of

Highly

& Niall

Proc.

of

Arrays

for

ACM/SIGDA

2010.

Selection.

, 20,

Order

Int'l

Robust

Working
Flow

in

and

paper.
the

Bond

19752019.

Market

Overreaction

Expectations

Equilibrium

9751007.

Comparison

Processor

Adams,

Strategic

Studies

6372.

Portfolio
and

Stock

09:

Arrays,

Jasra

of Financial

12,

Parallel

FPGA

Informed

A Rational

W.

Massively

Gate

& Ajay

1999.

Studies

Black

Generation,

Algorithms

Veronesi,

Bid-Ask

Paper.

L.,

Programmable

Review

The

L. Tepla,

The

FPGAs,

2007.

Limit

House.

Tsagaris

C.,

and

Howes,

Number

Using

441468.

2007.

B.,

of the

of Finance,

Returns.

Working

. Random

Random

Markets,

LBS

D.

Symp.

25,

Liquidation

Components

1990.

Futures

Asset

Paper.

Journal

Whaley,

Shapiro

Nassim,

CPUs,

E.

Analysis
B.,

Optimal

the

Tests.

R.

Benchmarking.

Vega,

Inferring

Index

Suleiman,

Thomas,

2012.
Working

Empirical

Quantitative

Taleb,

Rolf,

Information.

R.,

Stoll,

W.

to

Bad

Model.

News

in

Review

of

Valeri

Voev

using

High-frequency

Financial

& Asger

in

Integrated

the

Covariance

Presence

of

Estimation

Noise.

Journal

of

5 , 68104.

W.,

Evidence

2007.

Data

Econometrics

Wasserfallen,
Market:

Lunde,

1989.

Macroeconomic

from

Europe.

and

S.

News

Journal

of Banking

and
and

the

Stock

Finance

. 13,

613626.
Wongbanpo,

P.

Macroeconomic
Countries.
Young,
(1),

Sharma,

Fundamental
Journal

T.

C.

W.,

of Asian

1991.

2002.

Dynamic
Economics

Calmar

Market

Interactions:
13,

Ratio:

Stock

and

ASEAN-5

2751.

Smoother

Tool.

Futures

20

40.

Zigrand,
Financial
Government's
in Financial

Jean-Pierre,
stability

Dave

Cliff

and

computer

Foresight

Project

Markets.

and
based
on

The

Terrence

Hendershott,

trading.
Future

Working
of

Computer

2011.
paper,

UK

Trading

INDEX
Administrative

and

legal

costs

Algorithms
common

elements

critical

of

lineoptimizing

execution
genetic
order-routing
maximize

execution

maximize

trading

minimize

execution

minimize

footprint

obtain

best

size
costs

price

time-weighted

average

volume-weighted
sample

speed

price

average

(TWAP)

price

(VWAP)

market-making

Algorithmic

trading

Alpha
Alternative

investments

Arbitrage

strategies

Automated
Average

market-making

strategies

gain/loss

Back-test
Basis

trading

Bayesian

approach

Benchmarking.

See

Beta
Bid-ask

spread

tightness

of

Bollinger

bands

Broker
Burke

commissions
ratio

Performance

attribution

Calmar

ratio

Cancel

orders

Capacity

on

system

disconnect

evaluation

Capital

asset

pricing

Central

processing

model
unit

(CAPM)

(CPU)

Cointegration
Commodity

Futures

Trading

Commodity

markets,

event

Comparative

arbitration

ratios

Complex

orders

Conditional

Sharpe

ratio

Connectivity,

cost

of

Convergence

and

divergence,

Core

Commission

market

engine

Correlation
Counterparty
Credit

risk

Cuban,

Mark

Custody
Dark

risk

and

clearing,

cost

pools

Data,

costs

of

administrative

and

legal

connectivity
custody

and

electronic

clearing

execution

hardware
software
staffing
Data

set

Direct
DirectEdge
Directional

testing

access

to markets

exchange
traders

of

(CFTC)
in

Directional
event

trading

arbitrage,

commodity

around

events

application

of

markets

emerging

economies

equity

markets

fixed-income

markets

foreign

exchange

futures

markets

real

estate

markets

investment

trusts

(REITs)

events
forecasting

methodologies

strategies,

developing

tradable

news

corporate
industry
macroeconomic
Discrete

pairwise

(DPW)

optimization

Drawdown
Electronic
cost

trading

of

Emerging

economies,

event

arbitration

in

Equities
arbitraging

different

classes

fragmentation

in

large-to-small

information

Equities

( continued

liquidity
pairs
risk

issuer

spillovers

arbitrage
trading

arbitrage

Equity

markets,

Evaluation
Event

of same

period,

arbitrage.

event

arbitrage

length
See

also

in

of
Directional

trading

around

events

Excess

return

Exchange

on

value

at risk

fees

Execution
aggressive

vs.

passive

process
speed

of

Exponential
Field

market

resiliency

programmable

gate

Fill-or-kill

orders

Financial

Industry

Financial

Information

First-in-first-out
Fixed

Foreign
event

Authority

eXchange

(FIX)

execution

fragmentation

Fixed-income

(FPGA)

Regulatory

(FIFO)

income,

array

markets,

triangular

in

event

arbitration

Forex,

in

arbitrage

uncovered

interest

parity

fragmentation

in

Fournier

arbitrage

analysis

Front-running
Fundamental

analysis

Futures
arbitrage
fragmentation

in

Futures

Industry

Futures

markets,

event

Gambler's

Ruin

Problem

Geometric

Brownian

Graphics
Hardware
cost

of

Association

processing

(FIA)

arbitration

motion
unit

protocol

schedule

exchange
arbitration

(FINRA)

(GPU)

in

in

history

of

Hedging
delta
portfolio
High-frequency

data

bid-ask

and

bounce

buy-and-sell

identifiers,

lack

of

defined
deviation

from

irregular

spacing

properties

normality

of

recording
volume
High-frequency

trading

capitalization

(HFT)

of

defined
economics

of

costs

of doing

costs

of data

event

arbitrage

as

evolution

financial
key

business

of trading

markets

processes

leverage
number

methodology

suitable

for

of

of
of traders

participants

in

competitors
government
investors
service
risk

and

technology

management

strategies,

of.

performance

providers
See

Risk
and

management
capacity

of

alpha

decay

basic

performance

capacity

measures

evaluation

comparative

ratios

evaluation

period,

measurement,

length

principles

performance

implementation

common

pitfalls

key

steps

High

life

cycle

in

trading
water

Hurst

of

in

development

testing

of

attribution

systems,

model

of

systems

mark

exponent

Iceberg

orders

Ignition
Integration

testing

Interactive

Brokers

International

Organization

Internet

communication

Interval

price

Investor

of Securities
models,

limits

Commissions

common

(IPLs)

protection

front-running
intentional
market

market
crashes,

manipulation,
predicting

based

on

abnormal

based

on

asymmetry

Jensen's
Kappa
Kill

alpha
3

switches

Knight
Kurtosis

Capital

detecting

Group

trading

patterns

of liquitidy

in limit

order

books

(IOSCO)

Kyle's

lambda

Latency

arbitrage

strategies
Latency
Law

costs

of One

Price

Layering
Lee-Ready
Legal

rule

entity

Limit

identifiers

order

Linear

(LEIs)

books

market

resiliency

Liquidity
Liquidity

arbitrage

Liquidity

risk

Lit

pools

Low-latency
Lower

trading

partial

Machine

moments

(LPMs)

learning

Market

crashes,

predicting

based

on

abnormal

based

on

asymmetry

Market

impact

trading

patterns

of liquidity

in limit

order

books

(MI)

costs
estimation

of

minimizing
advanced
basic

models

models,

execution
optimal

issues

algorithms
execution

order-routing
permanent,
basic
data

with

strategies,

practical

algorithms
empirical

estimation
preparation

model

estimation

of

implementation

of

Market

makers

Market

making,

data,

automated

information

market

does

market

moves

quotes

widen

trade

not

information

evolution

in order

of order

of tick

data

of order

as
as

book

flow

flow

aggressiveness

shape
key

rebounds

markets

autocorrelation

order

move

and

moves

modeling

in

as

as

predictor

predictor

of market

predictor
predictor

of market

of market
of market

principles

nave
fixed

strategies
offset

order-arrival

rate,

offset

trend-dependent

as

function

of

offset

volatility-dependent

offset

profitable
as

a service

bid-ask

spread,

block

tightness

transactions,

market

depth

market

resilience

order

book,

order-flow
price

price

at best

shape

per

technical

support

strategy,

simulating

Market

manipulation,

Market

orders

Market

risk

bid

sensitivity
and

to

best

ask

of

imbalance,

change

of

price

sensitivity

unit

volume

and

resistance

detecting

levels

to

movement

movement
movement
movement

critical

lineoptimizing

discrete

pairwise

genetic

algorithms

algorithm
(DPW)

optimization

hedging
delta
portfolio
nonlinear

programming

simultaneous
stop

equations

losses

determining

parameters

value-at-risk
volatility

(VaR)
cutouts

determining
ex-ante
Market
lit

structure

and

swap

dark

pools

execution

facilities

Markets
microstructure

of

aggressive

vs.

passive

complex

orders

equities,

fragmentation

forex,

fragmentation

fixed

income,

futures,
limit
market
options,

convergence
fragmentation
fragmentation

trading

hours

modern

in
in

in

books

swaps,

types

in

fragmentation

fragmentation

order

execution

of markets
vs.

past

and
in
in

divergence

financial

landscape

HFT

as

HFT

participants

of the

evolution

media,

of trading

modern

of high-frequency

Markov

state

and

HFT

traders

dependency

quantity

Merrill

methodology

markets,

number

Maximum

1970s

limits

Lynch

Message

acknowledgment

Message

throttle

loops

limits

Messaging
core

architecture

network

throughput

protocols
Financial
User

Information

Datagram

Protocol

Transmission
speed

Protocol/Internet

Protocol

(TCP/IP)

security

Sharpe

Moving

average

National

Best

New

(FIX)

(UDP)

Control

and

Modified

eXchange

York

ratio
convergence
Bid

Stock

No-cancel

and

Offer

Exchange

divergence
(NBBO)

(MACD)

rule

(NYSE)

range

G
Omega
Opportunity
Options,
Order

costs
fragmentation

flow,

modeling

autocorrelation
evolution

in

of tick

adverse

selection

effective

bid-ask

information

of order
data

as

flow

predictor

components
spread

as

in
predictor
of market
of bid-ask

of market
movement
spread

movement

information-based
quoted

impact

bid-ask

order

spread

aggressiveness

shape

of order

Pairs

as

book

as

predictor
predictor

of market
of market

movement
movement

trading

Paper

trading

Parametric

bootstrapping

Performance

attribution

Performance

measures,

alpha

and

average

basic

beta
gain/loss

correlation
drawdown
return
skewness

and

kurtosis

volatility
win

ratio

Phishing
Pinging
Post-trade

analytics

Power-law

market

resiliency

Prehedging
Price

appreciation

Price

reasonability

Price-time

priority

and

timing

risk

costs

execution

Production
Protection

points

Pump-and-dump
Quote

archival

Quote

matching

Quote

stuffing

Real

estate

investment

trusts

(REITs),

event

arbitrage

in

Real-time

position

Rebate

validation

capture

Rebates
Regulation
global

key

initiatives

efficient

trade

matching

investor

protection

jurisdiction
market

structure

stability

of systems

Return
Risk

arbitrage

Risk

management

measuring

HFT

credit

counterparty

and

risk

market
liquidity
regulatory

and

legal

Securities

and

Exchange

Securities

Information

Sharpe

Commission
Processor

ratio

conditional
modified
Simulation
back-test
paper

trading

production
Simultaneous
Skewness
Slippage
Sniffing
Sniping

equations
and

kurtosis

or latency

costs

framework,

(SIP)

(SEC)

Software,

cost

Sortino

of

ratio

Spaghetti

Principle

of Modeling

Spoofing
Spread

scalping

Staffing,

cost

Statistical

of

arbitrage

models
practical

applications

of

cross-asset
equities
foreign

exchange

general

considerations

indices

and

ETFs

options
Sterling
Stop

ratio
losses

determining

parameters

Supervision,

human

Surveillance

measures,

kill

near-term

switches

legal

entity

Swap

identifiers

execution

facilities

Swaps,

fragmentation

System

testing

graphical

user

interface

recovery
reliability
scalability
security
stress
usability

and

(LEIs)

performance

in

(GUI)

software

Systematic

trading

Taxes
Technical

analysis

Technological
hardware,

innovations
history

of

messaging
core

architecture

network

throughput

protocols
speed

and

security

software
Tick

data.

Tick

rule

Time

See

High-frequency

distortion

Time-weighted
Trade

average

matching,

Trading

execution

(TWAP)

efficient

and

execution

bid-ask

spreads

market

impact

opportunity
price

definitions

costs,

implicit

overview

costs

costs
and

or latency

market

impact,

permanent

of

costs

appreciation

slippage

data

price

costs

background

basic

data

estimation
impact,
model

preparation

transparent
broker
exchange

execution

commissions
fees

risk

costs

costs

market

estimation

timing

costs

of
empirical

estimation

of

taxes
Trading

hours

Trading

systems,

data

set

testing

integration

testing

system
unit

testing

testing

testing

use-case

testing

Transmission
Treynor

Control

ratio

Triangular

arbitrage

Uncovered

interest

Unit

Protocol/Internet

parity

arbitrage

testing

Upside

Potential

Use-case
User

ratio

testing

Datagram

Value-at-risk

Protocol

(UDP)

(VaR)

Volatility
Volatility

cutouts

determining
ex-ante
Volume-weighted
Wash
Win

trades
ratio

average

price

(VWAP)

Protocol

(TCP/IP)

S-ar putea să vă placă și