0 Voturi pozitive0 Voturi negative

40 (de) vizualizări42 paginiRISK-RETURN ANALYSIS OF DYNAMIC INVESTMENT STRATEGIES

Apr 13, 2014

© © All Rights Reserved

PDF, TXT sau citiți online pe Scribd

RISK-RETURN ANALYSIS OF DYNAMIC INVESTMENT STRATEGIES

© All Rights Reserved

40 (de) vizualizări

RISK-RETURN ANALYSIS OF DYNAMIC INVESTMENT STRATEGIES

© All Rights Reserved

- Principles: Life and Work
- Principles: Life and Work
- The Intelligent Investor, Rev. Ed
- I Will Teach You to Be Rich, Second Edition: No Guilt. No Excuses. No BS. Just a 6-Week Program That Works
- Marrying Winterborne
- Rich Dad Poor Dad: What The Rich Teach Their Kids About Money - That the Poor and Middle Class Do Not!
- The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers
- The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers
- Secrets of the Millionaire Mind: Mastering the Inner Game of Wealth
- The Total Money Makeover: A Proven Plan for Financial Fitness
- The Intelligent Investor Rev Ed.
- The Intelligent Investor
- Alibaba: The House That Jack Ma Built
- Rich Dad's Guide to Investing: What the Rich Invest In, That the Poor and Middle Class Do Not!
- MONEY Master the Game: 7 Simple Steps to Financial Freedom
- Making Money
- Rich Dad's Increase your Financial IQ: Get Smarter with Your Money
- Rich Dad's Cashflow Quadrant: Guide to Financial Freedom

Sunteți pe pagina 1din 42

R I S K- R E T U R N

A N A LY S I S O F DY N A MI C

I N V E S T ME N T S T R AT E G I E S

JUNE 2011

I ssue #7

Benjamin Bruder

Research & Development

Lyxor Asset Management, Paris

benjamin.bruder@lyxor.com

Nicolas Gaussel

CIO Quantitative Management

Lyxor Asset Management, Paris

nicolas.gaussel@lyxor.com

QUANT RESEARCH BY LY XOR

1

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Foreword

The investment fund industry has changed dramatically over the last ten years, and we

are now seeing a convergence between hedge funds and traditional asset management. For

example, both institutional and retail investors now have easier access to absolute return

strategies in a mutual fund format. This convergence has accelerated recently with the

emergence of newcits and the increasing number of regulated hedge funds. The investment

decision-making process is now more complex as a result, with these dynamic investment

strategies and the number of underlyings and assets growing rapidly. Managing exposure to

risky assets is the main dierence between these investment styles and the traditional long-

only strategies. This dierence is highly signicant, however, and is not always understood

by investors and fund managers.

The traditional method for analysing and evaluating a strategy is to use risk-adjusted

performance measurement tools such as the Sharpe ratio (or the information ratio) and

Jensens alpha. These nancial models were developed to compare long-only strategies, and

are not really suitable for dynamic trading strategies, as they exhibit non-normal returns

and non-linear exposure to risk factors. In the 90s, practitioners and academics developed

alternative models to take these properties into account. Some extensions of the Sharpe ratio,

such as the Sortino, Kappa and Omega ratios, have now become very popular for analysing

the performance of hedge fund returns. Another way of understanding the risk-return prole

of dynamic strategies has been proposed by Fung and Hsieh (1997) by incorporating non-

linear risk factors in Sharpe-style analysis. These various measures dene the empirical

approach in the sense that they are computed on an ex-post basis but are not really suitable

for ex-ante analysis.

All these models are relevant, however they provide only a partial answer to understand-

ing the true nature of a dynamic strategy. Let us consider for example a long exposure

on a call option. From the seminal work of Black and Scholes (1973), we know that this

investment prole is equivalent to a delta-hedging strategy. A long position on a call op-

tion is therefore a trend-following strategy with dynamic exposure to the underlying risky

asset. Computing risk-adjusted performance or performing a style regression are certainly

not the obvious tools for analysing this dynamic investment strategy. A better way of un-

derstanding the risk and return of such a strategy is to use option theory. In this case, the

strategys performance is analysed by investigating both the payo function and the pre-

mium of the option. The latter of these two is split into an intrinsic value component and

a time-value component. Moreover, one generally computes the sensitivity of the premium

to various factors such as volatility, time decay or the price of the underlying asset. This

analytical approach gives a better understanding of the option strategy than the empirical

approach, which consists in analysing the ex-post risk-return prole of the option strategy

by computing some statistics on a real-life investment or on some backtests

1

.

1

For example, with the analytical approach, we may show that the trend of the underlying asset only

concerns the payo function and has no eect on the option premium. Such property could not be derived

from the empirical approach.

2

Another interesting example of a dynamic trading strategy is constant proportion port-

folio insurance (CPPI), developed by Hayne Leland and Mark Rubinstein in 1976. The

extensive literature on this subject

2

is mainly related to the analytical approach, and anal-

ysis of CPPI strategies is closer to option theory than the models developed to compute the

performance of traditional mutual funds. However, the CPPI technique is certainly one of

the better-known dynamic strategies used in asset management.

In view of this, we believe that the analytical method could be extended to a large class

of dynamic trading strategies, and not limited to options and CPPI. This seventh white

paper explores this approach. In this white paper, we develop a nancial model to better

understand the risk-return proles of a number of dynamic investment strategies such as

stop-loss, start-gain, doubling, mean-reverting or trend-following strategies. We show that

dynamic trading strategies can be broken down into an option prole and trading impact.

To a certain extent, the option prole can be seen as the payo function of the strategy,

whereas trading impact can represent the premium for buying such a strategy. In this

context, implementing a trading strategy generally implies a positive cost, which has to be

paid, as explained by Jacobs (2000):

Momentum traders buy stock (often on margin) as prices rise and sell as prices

fall. In essence, they are trying to obtain the benets of a call option upside

participation with limited risk on the downside without any payment of an

option premium. The strategy appears to oer a chance of huge gains with little

risk and minimal cost, but its real risks and costs become known only when its

too late.

Using this framework, we are also able to answer some interesting questions that are

not addressed by the empirical approach. For example: in which cases is the proportion

of winning bets (or hit ratio) a pertinent measure of the eciency of a dynamic strategy?

Which dynamic strategies like (or dont like) volatility? What are the best and the worst

congurations for a given dynamic strategy? What is the impact of the length of a moving

average in a trend-following strategy? What is the theoretical distribution of a strategys

returns? Why is long-term CTA dierent from short-term CTA? What are the risks of a

mean-reverting strategy? By answering all these questions, we provide some insights that

explain why and when some strategies perform or dont perform, and which metrics should

be used to evaluate their performance. We hope that you will nd the results of this paper

both interesting and useful.

Thierry Roncalli

Head of Research and Development

2

in particular the works of Leland and Rubinstein of course but also those of Black, Brennan, Grossman,

Perold, Schwartz, Solanki, Zhou, etc.

QUANT RESEARCH BY LY XOR

3

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Executive Summary

Introduction

When building a portfolio, investors have to choose from a wide range of investment styles.

Value investors, trend followers, global-macro or volatility arbitragers, to name just a few,

each oer a dierent way of generating returns.

Under the reasonable yet controversial assumption that markets do work, any extra

return is earned in exchange for a certain degree of risk. Hence, before even measuring

it, it is essential to identify and understand that risk in order to analyse the returns from

certain strategies. Unfortunately, this is a dicult task, especially in the case of dynamic

investment strategies, which are known to generate asymmetric returns. So how should we

proceed?

Since it is well established that options can be replicated using dynamic strategies, the

approach developed in this white paper consists in exploring the extent to which an option

prole can be associated with a given dynamic strategy. To keep things simple, we focus on

strategies running on a single asset. Excluding classical analysis of constant-mix strategies,

some of this papers key ndings are:

(1) Many dynamic strategies returns can be broken down into an option prole and some

trading impact,

(2) Contrarian strategies on a single asset tend to generate frequent limited gains, in

exchange for infrequent larger losses,

(3) Trend-following strategies on a single asset will perform if the absolute value of the

realised Sharpe ratio is above a certain threshold. The shorter-term the investment

style, the higher this threshold.

Dynamic strategies returns can be broken down into an

option prole and some trading impact

As a prerequisite to our analysis, investment strategies must be described as a function of

the underlying price only. This covers those situations in which a manager can decide how

much he has to invest simply by looking at the price and certain xed parameters. We show

in this paper that many popular strategies t into this framework.

In such situations, the holding function can be regarded as a traders delta. Hence, its

integral at some point in time corresponds exactly to the option prole associated with this

strategy.

4

Qualitatively speaking, the option hedging paradigm can be represented as follows:

C

l

v

l

v

1

v

1

v

whereas the fund management situation can be represented as follows:

1

l

1

l

Technically speaking, the option prole is the integral of the holding function while the

trading impact is related to the derivative of the holding function. These observations are

summed up in the following table:

Strategy Type Option Prole Trading Impact Hit Ratio

Average Gain /

Loss Comparison

Buying when

Convex Negative 50%

Average Gain >

market goes up Average Loss

Buying when

Concave Positive 50%

Average Loss >

market goes down Average Gain

The impact of volatility on directional funds depends on

the leverage of the strategy

Retail networks are used to distribute funds that maintain a constant exposure, typically

investing 20%, 50% or 80% of their wealth in risky markets. On the other hand, some

nancial products such as CPPI portfolios implement constant leverage on a given risky

asset. These two strategies belong to the constant-mix category.

When exposure is less than 100%, these strategies will benet from trading impact.

These strategies will gain even if the return of the underlying asset is zero. On the contrary,

when their leverage is above 100%, constant-mix strategies can be hit badly by trading

impacts. For example, a 3 times leveraged strategy on an equity index with 20% volatility

would loose 12% per annum if the underlying performance is equal to zero, making the

performance attribution dicult in such situation.

Contrarian strategies tend to generate frequent limited

gains in exchange for infrequent larger losses

Some investors base their investment on the principle that they are able to identify an

intrinsic value for certain securities and that markets should eventually converge with their

QUANT RESEARCH BY LY XOR

5

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

forecasts. For equities, that value is based on forecasts such as the companys expected

future earnings, their growth rate or the degree of uncertainty surrounding those forecasts.

Since this intrinsic target value changes slowly over time, it can be viewed as an exogenous

parameter of the strategy. As a result, the only remaining variable is the price of the security

itself. These strategies thus fall within the scope of our study. If investors invest in opposite

proportion of the dierence between the price and the intrinsic value, the following graph

describes the typical option prole of this strategy.

C

M

?

?

As expected, this strategy exhibits a concave prole. The positive trading impact is

illustrated by the fact that if the market continues to quote approximately the same price,

the portfolio value increases. This is due to the numerous buy-at-low sell-at-high trades that

have been made in order to maintain the target proportion of holdings.

Trend-following strategies performances are related to the

square of the realised Sharpe ratio

Trend-following strategies are a specic example of an investment style that emerged as an

industry in its own right. So-called Commodity Trading Advisors are the largest sector of the

Hedge Fund industry. Surprisingly, despite its importance in the investment industry, this

investment style is largely overlooked by standard nance textbooks. Some attempts have

been made to benchmark trend-following strategies against systematic buying of straddles.

This makes sense qualitatively, as the essence of trend-following is to benet from trends

while accepting that returns will not be generated if markets do not trend enough.

In this white paper, we propose a simple model for trend-following in which we are able to

show that returns can be represented as an option on the square of the realised returns. This

shares some qualitative similarities with the straddle benchmark, but takes the analysis a

step further. For example, we are able to derive a necessary condition on the realised Sharpe

ratio of the underlying asset to obtain positive returns:

|Sharpe ratio| >

1

2

6

where is the average duration (in years) of the trend estimator. Moreover, the maximum

annual losses due to trading impact are proportional to this threshold multiplied by the

average volatility of the strategy.

Conclusion

In this paper, we review three classes of strategies (directional, contrarian and trend-

following) and obtain a quantitative breakdown for each of them. This breakdown provides

us with an accurate risk-return analysis of each of these strategies. More specically it il-

lustrates how the probabilities of making gains and losses have to be analysed together with

the corresponding average amount of gain or losses. High hit ratios are not necessarily a

sign of good strategies, but can reveal exposure to extreme risks.

The analysis of real-life situations would require extending the single asset case to the

multi-asset situations to better understand the result of adding such strategies. It would

also be interesting to build econometric tests to assess whether this model is capable of

providing accurate predictions of the behaviour of specic hedge fund strategies. This has

been left for further research.

QUANT RESEARCH BY LY XOR

7

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Table of Contents

1 Introduction 9

2 Breaking investment strategies down into an option prole and

some trading impact 11

2.1 Model and results . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 Trading impact associated with the stop-loss overlay . . . . . . . 13

3 Directional strategies: balanced and leveraged funds 15

3.1 Option prole and trading impact . . . . . . . . . . . . . . . . . 16

3.2 Predictions compared to actual backtests . . . . . . . . . . . . . 17

4 Contrarian strategies 19

4.1 Mean-reversion strategies . . . . . . . . . . . . . . . . . . . . . . 19

4.2 Averaging down . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5 Trend-following strategies 25

5.1 Analysis of trend-following strategies in a toy model . . . . . . . 26

5.2 Trend-following strategies as functions of the observed trend . . 28

5.3 Asymmetrical return distribution . . . . . . . . . . . . . . . . . 32

6 Concluding remarks 34

8

QUANT RESEARCH BY LY XOR

9

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Risk-Return Analysis of

Dynamic Investment Strategies

Benjamin Bruder

Research & Development

Lyxor Asset Management, Paris

benjamin.bruder@lyxor.com

Nicolas Gaussel

CIO - Quantitative Management

Lyxor Asset Management, Paris

nicolas.gaussel@lyxor.com

June 2011

Abstract

The investment management industry has developed such a wide range of trading

strategies, that many investors feel lost when they have to choose the investment style

that meets their requirements. Comparing these on a like-for-like basis is a dicult task

about which much has been written. The scope of this paper is restricted to strategies

investing in a single asset, and which are driven by the price of this asset. We show

how those strategies can be fully characterised by two components: an option prole

and some trading impact. The option prole depends solely on the nal asset value,

whereas trading impact is driven by the realised volatility. From this analysis, most

of these investment strategies can be categorised in one of three families: directional,

contrarian and trend-following. While directional strategies exhibit the same kind of

behaviour as the underlying, contrarian and trend-following strategies exhibit asym-

metric return distributions. Those asymmetric behaviours can be misleading at rst

sight, as a seemingly stable strategy may hide large potential losses.

Keywords: Dynamic strategies, option payo, asymmetric returns, trend-following strat-

egy, contrarian strategy, volatility.

JEL classication: G11, G17, C63.

1 Introduction

We are unable to list here the immense variety of investment styles that are used in the

nancial industry to generate returns. Value investors, growth investors, trend followers,

global macro analysts, long-short equity managers, special situations specialists or volatility

arbitragers, to name but a few, all rely on dierent and sometimes opposing views of how

markets work. However, within each style, some people succeed and some do not, preserving

the mystery of what are the determining factors of successful investment strategies. It is

very likely that none of these strategies is able to predominate suciently to eliminate the

others, as market forces would disable this very strategy in due course.

We are grateful to Philippe Dumont, Guillaume Lasserre and Thierry Roncalli for their helpful comments.

10

As heterogeneous as those strategies might be, there is a need for nal investors to

compare them in order to understand which are the determining factors in their performance,

how they compare and whether it makes sense to pay fees to a professional investment

manager.

Some metrics such as Sharpe ratio, factor analysis and specialised index benchmarking

have become widely accepted tools for analysing investment performance in the mutual fund

universe. However, this multi-factor analysis fails to account for more complex strategies

such as the ones used by hedge funds. Those funds use dynamic strategies that generate

returns that are dicult to link to the behaviour of standard factors such as the main equity

indexes. Imagine a stylised situation in which one has to compare two investment strategies.

The rst would systematically sell short puts on the S&P index while investing the proceeds

in short-term bonds. The second would consist in investing eighty percent of its assets in

short-term bonds and using the remaining cash to invest in quarterly call options on the

S&P index. Which is the better strategy? How can they be compared? It is quite clear

that a Sharpe ratio or linear factor analysis would not provide enough information to assess

their quality.

Some attempts have been made to provide answers to those questions, some of which

we will now review. Addressing non-linearities in returns both in pricing models and in

fund performance is not a new topic. To quote only a few, Harvey and Siddique (2000)

propose a factor model incorporating not only the returns but the square of the returns

to explain hedge fund returns. Elsewhere, Agarwal and Naik (2004) propose a similar

approach for analysing equity-oriented hedge fund performance. However, instead of using

the square of returns, they create synthetic factors that mimic the performance of call

options, introducing dierent kinds of non-linearities. Fung and Hsieh (2001) focus on trend-

following strategies. Roughly, trend-followers invest in proportion to the past performance

of a specic market, whether it is positive or negative. As a result, those strategies resemble

the delta of a straddle option in qualitative terms. Fung and Hsieh (2001) are thus testing

the hypothesis that the performance of trend-following strategies can reliably be compared

to the simulated performance of a strategy rolling lookback straddles on the MSCI World

index. In the professional world, many analysts classify strategies depending on whether

they are convergent or divergent, depending on their tendency to go against the market or

to follow it (see Chung et al. (2004), for instance).

The common feature of these approaches is that the design of the non-linear relevant

factor is mostly based on qualitative considerations and is used to provide econometric tests

of those assumptions. In this paper, we follow a slightly dierent route. Instead of trying

to analyse real-life hedge fund returns, we aim to constructively identify the exact payo

generated by popular dynamic strategies.

The main idea for identifying this payo is borrowed from option pricing literature. It is

well known that the strategy for obtaining a call option payo consists in investing its delta

in the market on a day-to-day basis, this delta being the derivative of the price of the call. In

so doing, one obtains the payo of a call, less the so-called gamma costs. Imagine now that

a portfolio strategy can be dened as a function of a given underlying asset price. It is very

likely that the payo generated by such a strategy is the primitive of this strategy exposure,

plus or minus some trading impact. This covers those situations in which a manager can

decide how much to invest merely by looking at the price and some xed parameters. We

show in this paper that many popular strategies t this description.

QUANT RESEARCH BY LY XOR

11

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

This paper is thus structured as follows. In the second section, we show how simple

strategy performances can be broken down into an option prole and some trading impact.

Section 3 is dedicated to studying the properties of simple directional strategies consisting

in holding a constant fraction of ones wealth in a given asset. In section 4, we develop

an analysis of two popular contrarian strategies: the return to average strategy and the

averaging down strategy and we list their common features. Lastly, section 5 is devoted to

trend-following strategies, where an original result illustrates their typical convex behaviour.

2 Breaking investment strategies down into an option

prole and some trading impact

First of all, let us emphasise that the aim of this white paper is not to insist on the mathe-

matics of the results but on the nancial messages that are obtained. As a result we often

omit mathematical aspects such as ltrations, continuity of functions or detailed properties

of processes that would be necessary for a rigorous presentation. We hope that what has

been lost in terms of accuracy and rigour will be oset by the gain in simplicity and legibility.

2.1 Model and results

Consider a simplied situation in which an asset can be traded at each date t at price S

t

,

without any friction of any kind. S

t

is supposed to be governed by an ordinary diusion

model:

dS

t

= S

t

(

t

dt +

t

dW

t

)

S

0

= s

An investor is running an investment strategy consisting in holding a number f (S

t

) of

securities at any time. This strategy is supposed to depend only on the price itself and some

parameters, but not on time or on other state variables. For the sake of simplicity, it is

assumed that interest rates are zero

1

. The wealth of the investor at each date t is denoted

X

t

. On a day-to-day basis, or between t and t + dt, variation in wealth is written as:

dX

t

= f (S

t

) dS

t

(1)

If S

t

was deterministic and non-stochastic, it would be clear that:

X

T

X

0

=

_

S

T

S

0

f (S

t

) dS

t

= F (S

T

) F (S

0

)

where:

F (S)

_

S

a

f (x) dx,

whatever the a chosen. If it is not, Itos lemma applied to the function F yields the important

property we want to emphasise.

1

To recover results in situation with interest rates, S

t

and X

t

will have to be replaced respectively by

S

t

e

rt

and X

t

e

rt

in the dierent equations.

12

Proposition 1 Any portfolio strategy of the type (1) described above can be broken down

into an option prole plus some trading impact as follows:

X

T

X

0

= F (S

T

) F (S

0

)

. .

option prole

+

1

2

_

T

0

f

(S

t

) S

2

t

2

t

dt

. .

trading impact

(2)

While simple, the above proposition yields some interesting qualitative properties. Re-

garding model assumptions, it is worth noticing the robustness of this property, which can

be obtained whenever Itos lemma can be used with continuous price processes. This covers

a wide variety of models and does not rely on special probabilistic assumptions. These in-

clude Black-Scholes, of course, but also local and stochastic volatility models. On the other

hand, the option prole obtained is European. If the strategy is richer in terms of variable

states, the option prole will be a function of those dierent states.

Let us now elaborate on these two terms. Trading impact depends on the variation in

the number of holdings in relation to market uctuations. If the strategy involves buying

when the market goes up (f

illustrating the buy high sell low curse of trend followers. Conversely, strategies that play

against the market will always have positive trading impact, beneting from the opposite

eect. This trading impact increases with the volatility.

Interestingly, the sign of trading impact is directly related to the convexity of the option

prole. Positive trading impact is necessarily associated with concave prole. This is no

surprise to those familiar with option hedging, where it is well known that hedging a convex

prole will generate positive gamma gains, which explains the dierence between option

prices and their intrinsic value.

All strategies with positive trading impact share similar characteristics in terms of win-

ning probability. When trading impact is positive, using Proposition 1 leads to:

Pr {X

T

X

0

} > Pr {F (S

T

) F (S

0

)}

In a situation where F is non-decreasing, we get:

Pr {X

T

X

0

} > Pr {S

T

S

0

}

The probability of showing a prot is therefore higher than the probability of the underlying

asset going up. On a weekly basis, most nancial assets have a near 50/50 probability of

going up or down, which means that those strategies have more chance of showing a prot

than a loss. The higher the trading impact or the more concave the option prole, the

higher the probability of showing a prot. This eect is oset by higher potential losses. A

concave prole will therefore always exhibit negative skewness. Using a plain realised Sharpe

ratio to assess future fund performance is very likely to be awed, as it would be inated

articially by the frequent positive gains. Following this analysis, rather than indicating a

good portfolio manager, frequent positive gains may be symptomatic of strategies with high

possible losses. The reverse holds for strategies with negative trading impact.

This breakdown may be worth bearing in mind from a qualitative point of view. In some

cases, it might be tempting to focus on one term and to neglect the other, but in general

they are of equal importance. A typical example of such bias is the stop-loss overlay, which

is commonly used to protect against losses in a portfolio.

QUANT RESEARCH BY LY XOR

13

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

2.2 Trading impact associated with the stop-loss overlay

Imagine a situation where an investor runs an investment strategy which value is denoted S

t

.

In order to limit its losses, this investor wants to add a separate overlay to the strategy, which

would consist in doing nothing while the strategy is above a certain threshold level S

stop

, and

going short when it is below. In the following example, we will consider the initial strategy

as an underlying asset, and focus on the stop-loss overlay analysis. This overlay is purely

price dependent and should satisfy the assumptions of proposition 1. The corresponding

function f, representing the number of risky asset shares in the overlay portfolio, would be:

f (S

t

) =

_

0 when S

t

> S

stop

1 when S

t

S

stop

Along the lines of Proposition 1, the option prole is a put option of strike S

stop

. Trading

impact is more dicult to assess since f cannot be dierentiated in a traditional manner.

Technically, one should use another version of Itos formula, sometimes referred as Tanakas

formula. Interested readers can refer to Carr and Jarrow (1990) for a detailed analysis of

that strategy. Qualitatively, f can somehow be dierentiated and its dierential is equal

to zero everywhere except at S

stop

, where it has an innite positive value. Hence, trading

impact will necessarily be negative, proportional to the volatility and to the time spent by

the underlying asset around S

stop

. In a risk-neutral world, the average trading impact of

this stop-loss strategy is equal to the cost of the put. Interestingly, the stop-loss strategy

which could appear to be a free lunch as compared to buying a put option, generates some

trading impact, which is equal in average to the price of the put itself.

To conrm this eect, this strategy is simulated in Figure 1, with a stop loss level S

stop

equal to 90% of the original price. The resulting wealth is well below the asset price. Indeed,

this policy has a very strong trading impact around the strike level S

stop

. Each time this level

is crossed, the strategy suers from signicant trading costs (see Figure 2 for a description).

These costs cause the wealth level to deviate from the target prole. Each time the stop

Figure 1: Stop loss/start gain strategy trajectory

u

A

W

14

Figure 2: Trading impact appearance when crossing the strike

Asset move

Hedged

profile

Wealth

Losses

Stop loss level

Asset Price

loss level is crossed, the loss with respect to the target prole is proportional to the size of

the price movement. Total trading cost is therefore proportional to the volatility and to the

number of times the asset price crosses the strike S

stop

. In the following, we calculate the

average trading cost, supposing that the asset price has a trend equal to the risk-free rate.

In that case, the average trading cost is exactly equal to the price of the put option.

2.2.1 A tree-based approach

One might wonder whether this is an artefact of continuous time or not. To answer this

legitimate question, we provide a discrete time analysis and show how a similar result can

be obtained. We use a discrete tree-based approach to estimate trading costs, encountered

each time the asset price crosses the stop-loss barrier. In this tree model, the asset price can

increase or decrease by h at every time step. Each time step typically represents a business

day. Thus the typical time step is t =

1

/260 years and the asset price variation size should

be h =

representation of Figure 3, where the stop loss level S

stop

= 1

h

2

is represented by the red

line, just below the initial price. Each time the price crosses this red line (from above or

below), the investor loses

h

2

with respect to the target payo max (S

t

, S

stop

). Therefore, the

trading costs, i.e. the losses with respect to the target payo, will be

h

2

multiplied by the

number of times that the level S

stop

is crossed. The average number of times the barrier is

crossed behaves like

_

2

T

t

, where T is the total investment period, and t is the time step

2

.

Thus the average trading cost is given by:

L

h

2

_

2

T

t

_

T

2

This is exactly the price of the put option, in a Gaussian framework. This equality between

the average trading cost and the price of the put option holds in any risk neutral model.

2

This is the central limit probability for the asset value to quote at that value.

QUANT RESEARCH BY LY XOR

15

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 3: Tree based representation of the asset price

Compared to the protection based on the put option, the stop loss/start gain strategy

has approximately the same average cost. But this cost is very uncertain, while the cost of

the put option is known at inception. Even if the price of a put option is too high compared

to the expected future volatility, a hedging policy can be applied to replicate this option

(with the Black-Scholes formula). In this case, the cost of protection will depend on the

realised volatility alone, irrespective of the number of times the strike is crossed.

A prot-taking strategy is the exact opposite of the stop loss strategy. Diametrically

opposite results thus apply. As for the stop loss, investor can choose between selling a call

option and a prot-taking strategy. The call option seller abandons returns above S

stop

in exchange for receiving a xed premium P. The denitive prot-taking strategy also

abandons returns above the same threshold, but is not exposed to market drawdowns once

the prots have been locked in. The prot-taking strategy that re-invests in the asset when

the price is below the threshold abandons high asset performance, but receives trading gains

each time the strike is crossed.

3 Directional strategies: balanced and leveraged funds

A common way to capture the risk premia yielded by equity markets consists in running

investment strategies that invest a stable proportion of ones assets in risky markets. Merton

(1971) shows that this strategy is indeed optimal for investors having a logarithmic utility

and constant assumptions on expected returns and risks. Those strategies are sometimes

referred to as constant-mix strategies. Retail networks are used to distribute products tagged

as conservative, balanced or agressive. They often implement constant-mix type of strategies,

typically investing 20%, 50% or 80% of their wealth in risky markets. The popular 130/30

leveraged strategies are another example of constant-mix. Leveraged strategies whereby an

investor maintains a constant leverage of 2 or 3 on some asset also belong to the constant-

mix category. Constant proportion portfolio insurance (CPPI) strategies are a combination

of constant mix strategies plus some zero-coupon bond. Eventually, a strategy consisting in

16

being short the market is a constant mix strategy with a leverage equal to 1.

A widespread proxy to understand the behaviour of constant-mix is the equivalent buy

and hold strategy. We will show in the following that, even if satisfactory for a short-term

horizon, this approximation may turn to be misleading over an horizon of more than a few

months.

3.1 Option prole and trading impact

For the sake of simplicity, we shall focus on a single asset case, similar to the one described

in section 2. The constant proportion of wealth invested in the risky asset, exposure, is

denoted by e. Locally, the relative variation of wealth is proportional to the return of the

risky asset:

dX

t

X

t

= e

dS

t

S

t

(3)

where

dX

t

X

t

is the return of the strategy between t and t + dt and

dS

t

S

t

is the return of the

risky asset. This equation is slightly dierent from equation (1) as it involves exposure rather

than number of shares. However, we can follow exactly the same route, and by applying

Itos lemma to ln X

t

we get a similar proposition.

Proposition 2 In the case of constant exposure e, the portfolio strategy can be broken down

into an option prole multiplied by the exponential of the trading impact in the following way:

X

T

= X

0

_

S

T

S

0

_

e

. .

option prole

exp

_

_

_

_

_

1

2

_

e e

2

_

_

T

0

2

t

dt

. .

trading impact

_

_

_

_

_

(4)

where

_

T

0

2

t

dt is the cumulated variance of the risky asset between times 0 and T. If

volatility is constant over time, this quantity is equal to

2

T.

As in Proposition 1, the wealth can be split at a certain date into an option prole

and some trading impact. However, here the two terms are multiplied rather than added

together. The option prole is a power option, whose power is exposure e, while the trading

impact depends on realised volatility alone. Since both terms are positive, constant-mix

strategies always ensure positive wealth in a market that trades continuously without any

gap.

The e e

2

term is a variation of holdings as in Proposition 1. For conrmation of this,

let us consider the discrete case where a fraction of the wealth, e, is invested in a risky asset.

Let f denote the number of securities held in the portfolio. Initially f

t

=

eX

t

S

t

. If the risky

asset moves by x% then we have:

S

t+t

= S

t

(1 +x)

X

t+t

= X

t

(1 +ex)

f = e

X

t+t

S

t+t

e

X

t

S

t

The variation of holdings can then be computed as

f

S

= X

t

e (e 1)

S

2

t

QUANT RESEARCH BY LY XOR

17

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 4: One year option prole adjusted from trading impact, with 20% volatility

C M

t

&

The trading impact in equation (4) appears to be of the same kind as in equation (1) of

Proposition 1 and related to the variation of number of holdings with respect to the variation

of the risky asset. The option prole is not linear with respect to the asset price, except in

the obvious cases of a delta one product (e = 1) or a portfolio fully invested in cash (e = 0).

In the case of balanced funds, with positive exposure and no leverage, the option prole

is concave and trading impact is positive (see Figure 4). In terms of indexation to the

underlying market, the strategy is less indexed to the risky asset when its value is high, and

more indexed to this asset when its value is low. On the other hand, leveraged strategies

(e > 100%) and short selling strategies (e < 0) exhibit a convex prole. Those proles oer

potentially very high returns at the cost of more frequent losses. Trading impact increases

rapidly as e grows (see Figure 5). For example, the inuence of volatility is 3 times larger

3

for e = 3 than for e = 2. Figure 4 describes the eet of those strategies on nal wealth,

taking both target payo and trading impact into account for 20% volatility and a 5-year

horizon.

3.2 Predictions compared to actual backtests

All these formulas are derived from continuous time mathematical models, but they are quite

accurate in practical situations. We compare backtested results of constant mix strategies

combining the DJ Eurostoxx 50 index and cash. For each simulation, constant mix strategies

start with an initial value equal to 1. Then, the value of the strategy after one year is com-

pared to the relative value of the Eurostoxx 50 with respect to the starting date. Simulations

are started on each business day between January 1987 and December 2010. Figures 6 and

7 plot the values of each backtest with respect to the relative value of the Eurostoxx 50.

These values are compared to the prediction obtained with formula (4), where the volatility

parameter is set equal to 20%. Figure 6 illustrates the balanced strategy (50% invested

3

The expression

1

2

`

e e

2

2

T

remains unchanged for both exposures.

18

Figure 5: Annualised trading impact as a proportion of the initial wealth

s

8

in Eurostoxx and 50% invested in cash), while Figure 7 illustrates the 4 times leveraged

Eurostoxx 50 strategy.

The cash rate is still taken to be 0. As shown in Figure 6, the prediction is very accurate

for the 50/50 constant mix strategy. On the other hand, the 4 times leveraged strategy has

a larger prediction error. This is because the prediction is computed with xed volatility of

20%. The eective 1-year volatility of the Eurostoxx can be very dierent from this value.

When e = 50%, sensitivity to the realised variance is very low and equal to 12.5% of the

Figure 6: 50/50 constant mix strategy on the Eurostoxx 50

l

l

QUANT RESEARCH BY LY XOR

19

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 7: 4 times leveraged strategy on the Eurostoxx 50

l

l

variance

4

.

It is much higher for leveraged strategies. When e = 4 for instance, it is equal to 600%

of the variance. The nal value of the leveraged strategy thus depends heavily on the realised

variance. This explains the dierences with respect to the 20% volatility formula. As bullish

markets are generally less volatile than bearish markets, backtested portfolios are generally

higher than the formula for high returns of the Eurostoxx 50 index (right side of Figure 7),

and lower than the formula for most negative Eurostoxx returns (left side of Figure 7).

4 Contrarian strategies

Let us now focus on contrarian strategies. From Proposition 1, we expect that, going against

the market, those strategies will have a tendency to exhibit frequent small gains and less

frequent large losses. In particular, this section is devoted to the study of two popular

strategies: the mean-reversion strategy and the averaging down strategy.

4.1 Mean-reversion strategies

4.1.1 Strategy denition

In some situations, investors state that an underlying should quote close to a price denoted

S

target

in the sequel. S

target

can be obtained as the fundamental value of a stock obtained

using nancial analysis. It can also be a kind of average value around which an asset is

supposed to exhibit some mean-reversion behaviour. Mean-reversion rationales are frequent

in nancial markets, as certain ratios are supposed to remain within a certain absolute range,

outside which the situation is deemed abnormal. Price/earnings, volatility levels, spreads

between stocks or indexes among others are indicators that are commonly used as a basis

for mean-reversion analysis.

4

Sensitivity to the cumulated variance

2

T is equal to

1

2

`

e e

2

e = 50%.

20

Figure 8: Mean-reverting investment policy

l

l

All these strategies can be summarised in a simple guideline: buy the asset when its price

is below its target S

target

, and sell it when the price is above. The simplest corresponding

rule consists in holding an amount proportional to the distance between the price and its

target level, S

target

. In the same framework as Proposition 1 this can be written as:

f (S

t

) =

m(S

target

S

t

)

S

t

(5)

f (S

t

)S

t

is the total amount of the risky asset bought at time t and m is a scaling coecient.

This investment rule is illustrated in Figure 8. Obviously, this investor has a short position

when the asset price is above the average, and a long one if the asset price is below the

average. The number of risky asset shares in the portfolio decreases with respect to the

asset price. f is eectively a decreasing function of S, as shown in Figure 8. The investor

thus takes advantage of volatility when the price oscillates around a given level (even if this

level is not the target level S

target

).

In terms of risks, this exposure policy would be unlimited if the asset price rose to innity.

Moreover, the number of asset shares is unlimited when the asset price goes to 0. Note that

this framework makes it possible to set limits on the amount and/or number of shares, by

using a dierent denition of f. This would lead to more acceptable maximum risks for this

strategy. However, to keep things simple, this is not done here. In this situation, proposition

1 applies straightforwardly. The option prole and the trading impact are equal to:

Option prole = m(S

target

ln (S

T

) S

T

)

Trading impact =

1

2

mS

target

_

T

0

2

t

dt

As expected, the trading impact is always positive and proportional to the realised variance

of the asset during the investment period. Thus, for a given nal value of the asset, the

nal wealth increases with the realised variance. Conversely, the option prole is concave

and can potentially lead to unlimited losses.

QUANT RESEARCH BY LY XOR

21

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 9: Investor wealth (S

0

= 80%, S

target

= 100%)

C

M

?

?

4.1.2 Analysis of the option prole

Suppose that the asset price at initial date is not equal to the average price. A signicant

prot can be made if the asset price is closer to the average at maturity than it was at

inception date. Figure 9 shows the nal wealth of an investor as a percentage of the initial

wealth. Suppose that the initial price of the asset is equal to 80% of the long term aver-

age S

target

= 100%. Over a six-month time horizon, a signicant mid-term prot can be

generated if the asset price moves closer to the average. If the price increases from 80% to

100%, the realised gain is 3.3%. Conversely, signicant losses are incurred if the asset price

decreases. If the price falls from 80% to 60%, realised losses are equal to 7.7%. Nevertheless,

investors may accept this risk if they strongly believe that the asset price will converge to

S

target

in the near future.

4.1.3 Trading Impact

Now, suppose that the investor starts with initial wealth X

0

= 100%. Suppose also that

the initial price S

0

of the asset is equal to the average price S

target

. Figure 10 shows the

wealth of the investor at time T as a function of S

T

. We also assume that the annualised

volatility of the risky asset is a constant 20%. In this gure, the option prole is always

lower that the initial wealth of the investor. Mathematically, F (S

T

) is negative for all S

T

.

The option prole always makes a negative contribution to the performance. Naturally, this

loss increases when the asset price moves away from the average. Prots come from the

trading gains. These gains increase when the strategy is performed for a longer time.

This strategy thus delivers performance when, after a volatile trajectory, the asset price

nishes near the average. On the other hand, signicant short-term losses can occur if the

price moves away from the average. This is therefore an asymmetric strategy, involving

small and slow gains with high probability and large and quick losses with low probability.

22

Figure 10: Investor wealth (S

0

= S

target

= 100%)

C

M

?

?

4.2 Averaging down

In this section, we will show that this type of strategy is highly likely to deliver gains,

balanced by a signicant bankruptcy risk.

4.2.1 A miracle recipe for recovering losses?

Let us start with an example. Suppose that an investor buys a stock at $100 price, and

that the stock price drops to $90. The dierence between the average entry price and the

current price is $10. If the investor buys another share in the same stock (i.e. doubles his

position), the average buying price is now $95. The dierence between the current price

and the average entry price is now only $5. Of course this new gure does not correspond

to any actual loss reduction. The $10 losses are just diluted into a larger position. As the

exposure is larger, a small $5 increase of the stock price is now sucient to cancel out all

previous losses. But the larger exposure to the asset may also exacerbate future price falls.

A $10 asset price decrease will now lead to a $20 loss. Some investors may be tempted to

average down once again, in order to take advantage of a potential rebound. On the other

hand, the investor may face severe risks after doubling his exposure a few times.

This strategy can be related to the martingale gambling technique. It was originally

designed for a game in which a gambler doubles his stake if his bet is successful, or else

loses it. The martingale strategy is to double the bet after every loss, so that the rst win

would recover all previous losses, plus a prot equal to the original stake. The gambler will

almost surely win if he is allowed to bet an innite number of times (see Harrison and Kreps

(1979) for a detailed analysis). Unfortunately, he may go bust before his rst win in real life

situations, as his stakes double at each loss. Indeed, after 5 consecutive losses, the gambler

has to bet 32 times his original stake, which is unacceptable in most real life situations. The

losses in the most adverse scenario (going bust) are of several orders of magnitude over the

expected gains. For the same reasons, averaging down strategies can recover losses if a small

price increase occurs, but may result in the investor going bust if this increase happens too

late.

QUANT RESEARCH BY LY XOR

23

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 11: Exposure policy as a function of the wealth level

L

l

t

Clearly, averaging down could absorb losses provided the asset price stays above a certain

limit. Beyond this limit, the strategy may lead to huge losses due to the increasing exposure

to the asset. The goal of this section is to identify this threshold.

4.2.2 A ne line between objective achievement and severe losses

Let us formalise this strategy. As previously, X

t

stands for the investors wealth at time

t, and S

t

for the price of the risky asset. Suppose that an investor has an initial wealth

X

0

= 100%, and wants to obtain a target wealth of X

target

= 110% whenever the asset

price increases by R = 10%. Initially, this objective can be attained by investing 100% of

the wealth in the risky asset. Now, suppose this investor starts from a wealth level of X

t

.

The exposure needed to obtain target wealth of X

target

if the underlying asset moves by

10% is described by the following relationship:

X

target

= X

t

(1 +R e (X

t

))

or equivalently:

e (X

t

) =

1

R

(X

target

X

t

) (6)

The variation of wealth will be governed by equation (3) but with variable exposure. The

determining factor in this strategy is the distance between current wealth and targeted

wealth. It shares some similarities with contrarian strategies in which the determining

factor is the distance between the current asset value and a target asset value. This exposure

policy is decreasing with respect to current wealth, as shown in Figure 11. Intuitively, it

is an increasing function of the objective, as more risks must be taken to achieve higher

objectives.

24

Proposition 3 Consider a portfolio strategy that follows the exposure rule (6). The distance

to the target wealth can be broken down into an option prole and some exponential trading

impact:

X

target

X

T

= (X

target

X

0

)

_

S

T

S

0

_

1

R

. .

option prole

exp

_

_

_

_

_

1

2

_

1

R

+

1

R

2

__

T

0

2

t

dt

. .

trading impact

_

_

_

_

_

(7)

First of all, the objective is only attained asymptotically, as exposure converges to 0 as

the strategy approaches the objective. The option prole is always below the static strategy

with similar initial exposure, as shown on Figure 12. The averaging down overperformance

may only come from trading impact (i.e. volatility). The longer the investment period, the

higher this trading impact will be. In this example, volatility is presumed to be constant

and equal to 20%.

Figure 12: Averaging down strategy outcome

M

M

M

W

8

While the strategy prole stays close to the objective when performance is positive, it

decreases very quickly for negative returns. In this example, the prole leads to bankruptcy

when the asset price decreases by 20%. Indeed, the risky asset return objective R is typically

low compared to 100%. Therefore, the term

_

S

T

S

0

_

1/R

may be a very large negative power

of the spot price

5

. This large power explains the sudden loss behaviour of the option prole.

From a nancial point of view, this is explained by the increasing exposure when the

strategy moves away from the objective. Over a six month horizon, the strategy delivers a

positive P&L as soon as the risky asset has a return greater than 10%. If asset volatility

is equal to 20%, the probability of a positive P&L after 6 months is 76%. Unfortunately,

5

In Figure 11, we take R = 10%. Therefore the option prole involves a power

1

R

= 10 of the

underlying asset.

QUANT RESEARCH BY LY XOR

25

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 13: 1Y return of the average down strategy

/

this attractive hit ratio is counter-balanced by potentially very large losses. The probability

of going broke within six months is 2%. This is low, but not negligible, as this single event

could absorb years of prots plus the initial capital.

4.2.3 Backtesting

This asymmetry can be shown in a backtest of this strategy on the DJ Eurostoxx 50 index.

We provide a series of one-year simulations, with starting dates ranging from January 1987

to December 2010. The return objective is set at 10%. The resulting one-year returns

are displayed in Figure 13. One-year returns meet the objective for long time periods. For

example, these returns are very stable from 1991 to 2000, providing comforting results during

those nearly ten years. On the other hand, it leads to severe collapses, or even going broke

during the market downturns of 2002 and 2008.

5 Trend-following strategies

Trend-following strategies are a specic example of an investment style that emerged as an

industry in its own right. So-called Commodity Trading Advisors are the largest sector of

the Hedge Fund industry. According to BarclayHedge

6

, the total AUM managed by CTA as

of Q1 2011 was $290Bn AUM, equivalent to 15% of total Hedge Fund AUMs at that date.

This said, trend-following styles are not restricted to CTA funds, as they have been used

by many other investment managers for a long time. They were mentioned, for instance, in

Graham (1949):

In this respect the famous Dow Theory

7

for timing purchases and sales has

had an unusual history. Briey, this technique takes its signal to buy from

a special kind of break-through of the stock averages on the up side, and its

6

See www.barclayhedge.com/research/indices/cta/Money_Under_Management.html.

7

An outmoded denomination of momentum or trend-following techniques.

26

selling signal from a similar break-through on the down side. The calculated-not

necessarily actual-results of using this method show an unbroken series of prots

in operations from 1897 to 1946. On the basis of this presentation the practical

value of the Dow Theory would appear rmly established; the doubt, if any,

would apply to the dependability of this published record as a picture of what a

Dow theorist would actually have done in the market.

Surprisingly, despite its importance in the investment industry, this investment style is

largely overlooked by standard nance textbooks. Most available documents about trend-

following techniques consist of a collection of testimonials of how successful traders managed

to make money out of their trading rules. It is dicult to identify a clear direction among

those publications. Of these, let us quote Turtles (2003), which oers a valuable introduction

to CTA systems

8

. This text summarises the trading principles advocated by Richard Dennis,

a pioneer of systematic CTA trading.

From a very dierent perspective, as mentioned in the introduction, Fung and Hsieh

(2001) propose to benchmark trend-following strategies against the returns of a lookback

straddle on the MSCI World. They follow a line of thought similar to Proposition 1 and

infer this prole from the qualitative properties of the investment process.

This section is devoted to providing a simplied analysis of those strategies. While diver-

sication across a large number of assets greatly improves the eciency of trend-following

strategies, to keep things simple we are going to focus on the single asset case here. First,

trend-following strategies are analysed within a simple tree model, which identies the qual-

itative properties of CTA strategies. In a second section, a more rened model is proposed,

which gives a precise representation of trend-following strategies in terms of option prole

and trading impact.

5.1 Analysis of trend-following strategies in a toy model

This section provides stylised facts about trend-following strategies in a binomial tree model.

It is borrowed from the rst part of Potters and Bouchaud (2005). In this model, the

risky asset starts with an initial price of $100. This price increases or decreases by $1 at

each period, with a similar probability. The period would typically be one day. Figure 14

describes the cumulated P&L of a corresponding long investment in this asset for the next

three periods.

In this framework, the simplest trend-following strategy would be the following. Suppose

that an investor observed a positive trend before the start date. His strategy is to invest $100

in the asset, and to sell this position after the rst negative return is observed (which can be

interpreted as a negative trend). Figure 15 represents the P&L of this strategy depending

on the risky asset trajectory. We also compute the probability of each outcome, assuming

that the positive and negative return probabilities are both equal to 50%. The nal P&L is

represented in light blue boxes.

In this simple model, the loss probability is 50%, while the gain probability is only 25%

(the remaining 25% corresponding to neither prots or losses). However, while losses are

limited to $1, the gains can go up to any value. The average gain is $2, which osets the

smaller gain probability. Thus, the returns distribution of this strategy is positively skewed:

small losses are frequent, while gains occur rarely, but with a larger amplitude. This is

8

The text The original Turtle Trading Rules may be found on the web site www.originalturtles.org.

QUANT RESEARCH BY LY XOR

27

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 14: P&L of a risky asset investment in the binomial model

2

1

4

1

8

1

16

1

32

1

28

Figure 16: P&L distribution of the trend-following strategy

W>

illustrated in Figure 16. With Proposition 1 in mind, this is very consistent with a strategy

exhibiting a convex option prole and negative trading impact. Let us now be more precise

and obtain the corresponding breakdown in a continuous time model.

5.2 Trend-following strategies as functions of the observed trend

5.2.1 A model of a trend-following strategy

Trend-following estimators use past returns to predict future price changes. In the previous

example, the trend was estimated using only the last return: the investor forecasts a positive

(resp. negative) return if the last one was positive (resp. negative). In reality, longer period

returns, such as one month or one year past returns, are used for estimation. Many trend

followers use moving averages of prices or returns in order to provide more stable predictions.

For example, some consider the dierence between a short-term (e.g. one month) moving

average of the asset price, and a long-term average (e.g. six month). This dierence is

positive when past returns are mostly positive during the last six month period. It is

therefore considered to be a valid estimator of past trends. In real life situations, investors

use more complex combinations of such indicators.

Our purpose here is not to nd the best estimate, but to provide clear analysis. We will

therefore consider the simple example of a moving average of past daily returns. We choose

a moving average with exponential weights:

t

=

1

i=0

e

it

S

tit

S

t(i+1)t

S

t(i+1)t

Such an estimator has interesting properties. It depends on a single parameter , which

represents the average duration of estimation. Due to exponential weighting, recent returns

have a larger impact than older ones. It does not suer from threshold eect, that is,

changes of regimes due to a past observation that exits the averaging window. Moreover,

this estimator has theoretical foundations, as it can be interpreted as the Kalman lter for an

QUANT RESEARCH BY LY XOR

29

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

unobservable trend. Lastly, it produces simple formulas for the performance of the related

strategy. In particular, one can derive the dynamics of the moving average depending on

the asset returns:

t+t

=

_

1

t

_

t

+

1

S

t+t

S

t

S

t

or, in a continuous time framework:

d

t

=

1

t

dt +

1

dS

t

S

t

(8)

In the following, we suppose that the investor considers that the best returns estimation

between times t and t + dt is

t

dt. Based on this assumption, the investor can simply apply

the optimal Markowitz/Merton strategy

9

. Here, we will still consider that the risk-free rate

is 0. In this case, its exposure to the risky asset will be:

e

t

= m

t

2

(9)

where is the trend estimator, is the annualised volatility of the underlying, and m is a risk

tolerance parameter. Interestingly, this exposure strategy is qualitatively consistent with the

principles described in the Turtles Rules. The exposure to the risky asset is proportional

to this risk tolerance. This parameter depends on the investors risk prole. This exposure

is also proportional to the trend estimator value

t

and inversely proportional to the risk.

Note that it is not capped and is almost never 0.

Following this strategy, the investors wealth evolves as equation (3):

dX

t

X

t

= e

t

dS

t

S

t

= m

t

2

dS

t

S

t

(10)

Now, by inverting equation (8), the dynamic of the wealth can be written as a function of

the trend

t

. We can then follow the steps of Proposition 2 where the variable is no longer

the asset price S

t

but the asset trend

t

.

Proposition 4 The logarithmic return of a trend-following strategy that follows the exposure

function (9) can be broken down into an option prole and some trading impact:

ln

X

T

X

0

= m

_

_

_

_

_

2

2

_

2

T

2

0

_

. .

option prole

+

_

T

0

_

2

t

2

_

1

1

2

m

_

1

2

_

dt

. .

trading impact

_

_

_

_

_

(11)

It is worth remembering that the trend

t

is not a model assumption but the actual

measured trend.

5.2.2 Option prole

The option prole is related to the square of the observed trend:

Option prole = m

2

2

_

2

T

2

0

_

9

See Merton (1971) for details.

30

This is similar in concept to the straddle prole suggested by Fung and Hsieh (2001). It is

also perfectly in line with regressions including squared returns factors performed by Harvey

and Siddique (2000). This option prole part evolves quickly, and has a short term memory.

It vanishes along with the measured trend . Interestingly, the worst case scenario is that

the measured trend falls from its initial value

0

to 0, which would lead to an upper limit

of

m

2

2

2

0

. This quantity depends largely on the value of the measured trend at the start.

This is natural, as the strategy is more exposed to the risky asset when the trend is high.

This option prole is displayed in Figure 17, where we consider a six month moving average,

with 15% risky asset volatility, and a risk tolerance parameter m = 5% (see the next section

for details about this set of parameters). Furthermore, we assume that the initial measured

trend is equal to

0

= 0. As a consequence, the option prole is always positive.

Figure 17: Short term option prole, as a function of the risky asset return

5.2.3 Trading impact

On the other hand, trading impact is obtained by a cumulated function of the realised trend:

Trading impact =

_

T

0

m

_

2

t

2

_

1

1

2

m

_

1

2

_

dt (12)

This trading impact changes slowly, and has a long-run memory due to its cumulative nature.

The evolution of this long term P&L depends only on the ratio

2

t

2

which is the square of

the measured short term Sharpe ratio. Trading impact is negative on the long term if this

ratio is below

1

(2m)

, and otherwise positive.

The risk tolerance m can be related to the target volatility

target

by setting m =

target

a good estimate for long term trading impact per annum is given by

2

t

2

1

2

. This provides

a necessary condition for the trend-following strategy to generate prots:

|Sharpe ratio| >

1

2

QUANT RESEARCH BY LY XOR

31

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

For a 6 month moving average (i.e. =

1

2

), the long-term prot increases if the annualised

measured Sharpe ratio is above one in absolute terms.

Eventually, the worst-case scenario for the long term P&L is that

t

= 0, which is no

surprise! In that case, annual losses due to trading impact are capped at

m

2

=

target

2

. In

the case of a six-month moving average, annual maximum long-term losses are equal to the

target volatility of the strategy.

5.2.4 Backtested example

In this section, the trend-following strategy is backtested on the DJ Eurostoxx 50 index.

Instead of performing the trend-following strategy directly on the index, it is run on a

volatility targeted index which constantly adjusts its exposure to the Eurostoxx 50 in order

to keep its volatility equal to = 15%. The strategy itself uses a six-month moving average

( =

1

2

), with target volatility of

target

= 5%. This involves a parameter m = 5%, with an

average absolute exposure above 33% (the average backtested absolute exposure is around

50%).

Figure 18: Backtest of a trend-following strategy on the vol-target Eurostoxx 50

S nAv

1 l

The backtest is performed between January 1987 and May 2011. Figure 18 shows the

breakdown of the strategy value between the option prole and the trading impact. Long

term trading impact is computed from formula (12), independently of the strategy value

calculations. This component is smoother than the strategy NAV. The short-term option

prole is the dierence between the strategy NAV and the long term P&L. This component

is always positive, as the measured trend at the start is set at 0. The strategy NAV is

therefore always above the long term value. By construction, this long term cannot decrease

more than 5% p.a. (the instantaneous slope of the long term P&L, i.e. the daily trading

impact, is given in Figure 19). The worst-case scenario thus appears fairly acceptable.

32

Figure 19: Annualised daily contribution to trading impact

The return distribution of the strategy depends heavily on the distribution on the trend

estimator , as shown in equation (11). Of course, the behaviour of depends of the model

assumptions on the asset price dynamics. Let us suppose a standard Gaussian dynamic for

S

t

with xed volatility and a xed trend . In this case, the long term distribution of the

trend estimation is Gaussian, with an average of and volatility of

2

.

In the case of the six-month exponential moving average, the standard deviation of the

measured trend is equal to the standard deviation of the yearly return of the underlying asset.

The long term returns of the strategy are driven by the square of the measured trend

2

and are then highly asymmetric. The stationary distribution of the annualised contribution

to the long term P&L is given in Figure 20, in the same conditions as in the backtest of the

previous section. This distribution is computed under two sets of assumptions: in the rst

one, the underlying asset has no drift (risk neutral probability), while in the second one the

Sharpe ratio is constant and equal to 1 (this involves a constant drift = 15%) .

This distribution looks very similar to the toy model version of Figure 16. The general

result remains unchanged. The probability of losing is higher than the probability of gaining,

but the average gain is much higher than the average loss. As expected, the trend-following

strategy works poorly in trendless models (the average P&L is equal to 0), and much better

in models with a positive trend. For an easier comparison of the winning and losing proba-

bilities, the cumulative version of the distribution is illustrated in Figure 21. When there is

no trend, the positive P&L probability is around 30% as opposed to 50% in the case where

the Sharpe ratio is equal to 1.

QUANT RESEARCH BY LY XOR

33

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Figure 20: Stationary distribution of the annualised daily trading impact

S

S

Figure 21: Cumulative distribution of the annualised daily trading impact

S

S

S

34

6 Concluding remarks

In this paper, we have shown how the returns on many popular strategies can be broken down

into an option prole and some trading impact, both being of similar importance, on average.

We reviewed three classes of strategies (directional, contrarian and trend-following), and

obtained a quantitative breakdown for each of these. This breakdown allows for an accurate

risk-return analysis of each of those strategies. More specically, it illustrates how the

probabilities of making gains or losses have to be analysed together with the corresponding

average amount of gain or losses. High hit ratios are not necessarily signs of good strategies,

in fact they can reveal exposure to extreme risks.

Analysis of real-life situations would involve extending the single asset scenario to multi-

asset situations, for a better understanding of the impact of adding such strategies. It

would also be interesting to build econometric tests to assess whether our model can provide

accurate predictions of the behaviour of specic hedge fund strategies. This has been left

for further research.

QUANT RESEARCH BY LY XOR

35

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

References

[1] Agarwal V. and Naik N.Y. (2004), Risks and Portfolio Decisions Involving Hedge

Funds, Review of Financial studies, 17(1), pp. 63-98.

[2] Black F. and Perold A.F (1992), Theory of Constant Proportion Portfolio Insurance,

Journal of Economic Dynamic and Control, 16(3-4), pp. 403-426.

[3] Carr P. and Jarrow R. (1990), The Stop-Loss Start-Gain Paradox and Option Valua-

tion: A New Decomposition into Intrinsic and Time Value, Review of Financial Studies,

3(3), pp. 469-492.

[4] Chan L.K.C., Jegadeesh N. and Lakonishok J. (1998), Momentum Strategies, Jour-

nal of Finance, 51(5), pp. 1681-1713.

[5] Chung S.Y., Rosenberg M. and Tomeo J.F. (2004), Hedge Fund of Fund Alloca-

tions Using a Convergent and Divergent Strategy Approach, Journal of Alternative

Investment, 7(1), pp. 44-53.

[6] Conrad J. and Kaul G. (1998), Anatomy of Trading Strategies, Review of Financial

Studies, 11(3), pp. 489-519.

[7] Dybvig P.H. (1988), Distributional Analysis of Portfolio Choice, Journal of Business,

61(3), pp. 369-393.

[8] Dybvig P.H. (1988), Inecient Dynamic Portfolio Strategies or How to Throw Away

a Million Dollars in the Stock Market, Review of Financial Studies, 1(1), pp. 67-88.

[9] Fung W. and Hsieh D.A. (1997), Empirical Characteristics of Dynamic Trading Strate-

gies: The Case of Hedge Funds, Review of Financial Studies, 10(2), pp. 275-302.

[10] Fung W. and Hsieh D.A. (2001), The Risk in Hedge Fund Strategies: Theory and

Evidence from Trend Followers, Review of Financial studies, 14(2), pp. 313-341.

[11] Goetzmann W., Ingersoll J., Spiegel M. and Welch I. (2002), Sharpening Sharpe

Ratios, NBER, 9116.

[12] Gollier C. (1997), On the Ineciency of Bang-Bang and Stop-Loss Portfolio Strate-

gies, Journal of Risk and Uncertainty, 14(2), pp. 143-154.

[13] Graham B. (1949), The Intelligent Investor, 2003 revisited edition and updated by

Jason Zweig, Harper & Row, Publishers.

[14] Harrison J.M. and Kreps D.M. (1979), Martingales and Arbitrage in Multiperiod

Securities Markets, Journal of Economic Theory, 20(3), pp. 381-408.

[15] Harvey C.R. and Siddique A. (2000), Conditional Skewness in Asset Pricing Tests,

Journal of Finance, 55(3), pp. 1263-1295.

[16] Jacobs B.I. (2000), Momentum Trading: The New Alchemy, Journal of Investing, 9(4),

pp. 6-8.

[17] Kaminski K.M. and Lo A.W. (2007), When Do Stop-Loss Rules Stop Losses?, Working

paper, ssrn.com/abstract=968338.

[18] Lehmann B. (1990), Fads, Martingales and Market Eciency, Quarterly Journal of

Economics, 105(1), pp. 1-28.

36

[19] Lei A. and Li H. (2009), The Value of Stop Loss Strategies, Financial Services Review,

18, pp. 23-51.

[20] Leland H.E. and Rubinstein R. (1988), The Evolution of Portfolio Insurance, in D.L.

Luskin (eds), Portfolio Insurance: A Guide to Dynamic Hedging, Wiley.

[21] Lo A.W. and MacKinlay A.C. (1990), When are Contrarian Prots due to Stock

Market Overreaction, Review of Financial Studies, 3(2), pp. 175-205.

[22] Merton R.C. (1971), Optimal Consumption and Portfolio Rules in a Continuous-Time

Model, Journal of Economic Theory, 3(4), pp. 373-413.

[23] Merton R.C. (1981), Timing and Investment Performance: I. An Equilibrium Theory

of Value for Market Forecast, Journal of Business, 54(3), pp. 363-406.

[24] Perold A.F. (1986), Constant Proportion Portfolio Insurance, Harvard Business

School, Manuscript.

[25] Potters M. and Bouchaud J-P. (2005), Trend Followers Lose More Often Than They

Gain, arxiv.org/abs/physics/0508104v1.

QUANT RESEARCH BY LY XOR

37

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Lyxor White Paper Series

List of Issues

Issue #1 Risk-Based Indexation.

Paul Demey, Sbastien Maillard and Thierry Roncalli, March 2010.

Issue #2 Beyond Liability-Driven Investment: New Perspectives on

Dened Benet Pension Fund Management.

Benjamin Bruder, Guillaume Jamet and Guillaume Lasserre, March 2010.

Issue #3 Mutual Fund Ratings and Performance Persistence.

Pierre Hereil, Philippe Mitaine, Nicolas Moussavi and Thierry Roncalli, June 2010.

Issue #4 Time Varying Risk Premiums & Business Cycles: A Survey.

Serge Darolles, Karl Eychenne and Stphane Martinetti, September 2010.

Issue #5 Portfolio Allocation of Hedge Funds.

Benjamin Bruder, Serge Darolles, Abdul Koudiraty and Thierry Roncalli, January

2011.

Issue #6 Strategic Asset Allocation.

Karl Eychenne, Stphane Martinetti and Thierry Roncalli, March 2011.

38

QUANT RESEARCH BY LY XOR

39

RI SK- RETURN ANALYSI S OF DYNAMI C I NVESTMENT STRATEGI ES Issue # 7

Disclaimer

Each of this material and its content is condential and may not be reproduced or provided

to others without the express written permission of Lyxor Asset Management (Lyxor AM).

This material has been prepared solely for informational purposes only and it is not intended

to be and should not be considered as an oer, or a solicitation of an oer, or an invitation

or a personal recommendation to buy or sell participating shares in any Lyxor Fund, or

any security or nancial instrument, or to participate in any investment strategy, directly

or indirectly.

It is intended for use only by those recipients to whom it is made directly available by Lyxor

AM. Lyxor AM will not treat recipients of this material as its clients by virtue of their

receiving this material.

This material reects the views and opinions of the individual authors at this date and in

no way the ocial position or advices of any kind of these authors or of Lyxor AM and thus

does not engage the responsibility of Lyxor AM nor of any of its ocers or employees. All

performance information set forth herein is based on historical data and, in some cases, hy-

pothetical data, and may reect certain assumptions with respect to fees, expenses, taxes,

capital charges, allocations and other factors that aect the computation of the returns.

Past performance is not necessarily a guide to future performance. While the information

(including any historical or hypothetical returns) in this material has been obtained from

external sources deemed reliable, neither Socit Gnrale (SG), Lyxor AM, nor their af-

liates, ocers employees guarantee its accuracy, timeliness or completeness. Any opinions

expressed herein are statements of our judgment on this date and are subject to change with-

out notice. SG, Lyxor AM and their aliates assume no duciary responsibility or liability

for any consequences, nancial or otherwise, arising from, an investment in any security or

nancial instrument described herein or in any other security, or from the implementation

of any investment strategy.

Lyxor AM and its aliates may from time to time deal in, prot from the trading of, hold,

have positions in, or act as market makers, advisers, brokers or otherwise in relation to the

securities and nancial instruments described herein.

Service marks appearing herein are the exclusive property of SG and its aliates, as the

case may be.

This material is communicated by Lyxor Asset Management, which is authorized and reg-

ulated in France by the Autorit des Marchs Financiers (French Financial Markets Au-

thority).

c 2011 LYXOR ASSET MANAGEMENT ALL RIGHTS RESERVED

Lyxor Asset Management

Tour Socit Gnrale 17, cours Valmy

92987 Paris La Dfense Cedex France

www.lyxor.com

PUBLISHING DIRECTORS

Alain Dubois, Chairman of the Board

Laurent Seyer, Chief Executive Ofkcer

Nicolas Gaussel, PhD, Head of Quantitative Asset Management

EDITORIAL BOARD

Serge Darolles, PhD, Managing Editor

Thierry Roncalli, PhD, Associate Editor

Guillaume Jamet, PhD, Associate Editor

The Lyxor White Paper Series is a quarterly publication providing

our clients access to intellectual capital, risk analytics and

quantitative research developed within Lyxor Asset Management.

The Series covers in depth studies of investment strategies, asset

allocation methodologies and risk management techniques.

We hope you will nd the Lyxor White Paper Series stimulating

and interesting.

R

f

.

7

1

1

6

3

5

S

t

u

d

io

S

o

c

i

t

r

a

le

+

3

3

(

0

)

1

4

2

1

4

2

7

0

5

0

6

/

2

0

1

1

- Hedging Performance of Protective Puts and Covered Calls Portfolio : A Study of NSE NIFTY OptionsÎncărcat deRahul Pinnamaneni
- How to Read an Options Chain.pdfÎncărcat describerone
- PROJECT REPORT on Risk and Return (1)Încărcat deNeha Rohilla
- RISK-RETURNÎncărcat dePrashanth Kulkarni
- Derivatives Report 10 Sep 2012Încărcat deAngel Broking
- Derivatives Report 2nd April 2012Încărcat deAngel Broking
- SS 17 - AnswersÎncărcat dejus
- PIYUSHÎncărcat dePiyush Chauhan
- Stock Futures and Options Reports for the Week (6th - 10th June '11)Încărcat deDasher_No_1
- Option BasicsÎncărcat debhavnesh_mutha
- Module 1Încărcat deRaghu Kasturi
- Microsoft Word - Step By Step LearningÎncărcat dePankaj Shukla
- Intro to DerivativesÎncărcat desandipkodgire1
- Final Project Report on Financial DerivativiesÎncărcat deMarketing Expert
- Step by Step LearningÎncărcat deSanjay Chatterjee
- Tutorial+2+(chapters+5,8,10)Încărcat demohitkapoor
- Derivaties-ValenduÎncărcat dePraveen Sharma
- Futures and OptionsÎncărcat denbhaskaran
- Market Making GuideÎncărcat deAlbert Tsou
- nullÎncărcat deswami_ratan
- OptionsÎncărcat desachinrema
- Handouts 9-10 - Financial Risk Management - IITU-unlockedÎncărcat deАсылбек Алимханов
- Intro to Options GuideÎncărcat dedaytrading_de
- Ifm Chapter 6 Financial OptionsÎncărcat deminhhien222
- Hughes 3ReportÎncărcat deMarlon Douglas
- eiteman_178963_im05Încărcat deJane Tito
- top 10 fixed return option trading strategies by kavita mehataniÎncărcat detarun gautam
- Intro to Options Webinar NotesÎncărcat deMallikarjun Reddy Chagamreddy
- Airline Fuel Hedging_ An Overview of Hedging Solutions Available.pdfÎncărcat desaif ur rehman shahid hussain (aviator)
- What is OptionÎncărcat deSumit Rathi

- Improving Care to Communication Vulnerable PatientsÎncărcat deSagar Tanna
- Safety FirstÎncărcat deSagar Tanna
- AnnouncementÎncărcat deSagar Tanna
- Slip, Trip, and Fall PreventionÎncărcat deSagar Tanna
- Learn English in 30 Days PDFÎncărcat demanojj65
- Organizational Structure and DesignÎncărcat deSagar Tanna
- Fire Life Safety Practices HospitalsÎncărcat dedelarosa12
- Me Mp 070812Încărcat deMichael
- Legal and Ethical Issues in Medical PracticeÎncărcat deSagar Tanna
- PATIENT SATISFACTION REGARDING HEALTHCARE SERVICESÎncărcat deSagar Tanna
- Patient Safety Walk-RoundsÎncărcat deSagar Tanna
- Slip Trip Fall PreventionÎncărcat deSagar Tanna
- In Case of FireÎncărcat deSagar Tanna
- Volunteer Department.pptÎncărcat deSagar Tanna
- Using Medications SafelyÎncărcat deSagar Tanna
- PCPNDT Act 1994Încărcat deSagar Tanna
- Absorption and marginal costingÎncărcat dehhmzz63
- 11583_Chapter -3 PlanningÎncărcat deSagar Tanna
- LG IndiaÎncărcat deSagar Tanna

- MicroCap Review Spring 2018Încărcat deMicroCap Review Magazine
- Market Tantrums and Monetary PolicyÎncărcat deSteven Hansen
- Elliott WaveÎncărcat dejrladdu
- VSA Anna CoulingÎncărcat deIstiaque Ahmed
- Weekly Flow ReportÎncărcat deFriedrich Hayek
- Technical Market Indicators - SSRN-id2449344Încărcat deAlexis Tocqueville
- Technical Analysis From a to ZÎncărcat deSwati Singh
- economics 2 PROJECT rajat.docxÎncărcat deZC
- Chapter 09 Test Bank - Static.docxÎncărcat deyoo
- MalaysiaÎncărcat deJackson Hole
- Pristine - Guerrilla Trading Tactics (Oliver Velez)Încărcat deMaurilioQRO
- Win by Not Losing a Disciplined Approach to Building and Protecting Your Wealth in the Stock Market by Managing Your RiskÎncărcat dePrathik Rai
- Equity Research & PortfolioÎncărcat deSidd Bhattacharya
- AAIIJournal_February2010Încărcat deramel yanuarta
- Third Quarter 2010 GTAA EquitiesÎncărcat deZerohedge
- Synergy PROÎncărcat demagicxel
- Tech Terms WordÎncărcat deSuprabha Hani
- Pristine Patterns 1.01aÎncărcat deANIL1964
- Trading ManualÎncărcat deBadrulhisam Salleh
- Avoiding Option Trading TrapÎncărcat deSudhakar Reddy
- Trader c0dchef — Trading Ideas & Charts — TradingView India1Încărcat deAbhi
- Cmt1 QuestionsÎncărcat dearjbak
- Brown-Cliff Computations 2002Încărcat deddkiller
- Bnp Fx WeeklyÎncărcat dePhillip Hsia
- Hedging Instruments Against Foreign Exchange Riskfinal Report1213Încărcat dePrachi Jain
- "Trading the 10 O'Clock Bulls" Geoff ByssheÎncărcat deapi-26247058
- Saxo Asset Allocation - OctoberÎncărcat deTrading Floor
- The Economist Corporate Network Asia Business Outlook Survey 2018Încărcat dejuanperez32
- 1Încărcat deSriranga G H
- Year Ahead Asia 2011Încărcat derezurekt

## Mult mai mult decât documente.

Descoperiți tot ce are Scribd de oferit, inclusiv cărți și cărți audio de la editori majori.

Anulați oricând.