Brooks et al.

, 2001, A trading strategy based

on the lead-lag relationship between the spot
index and futures contract for the FTSE 100

陳韋翔1 , 陳奕卲2

NCNU Graduate School of International Business Studies

05/23

1 99212501
2 99212509
. . . . . .
Introduction

Research motivations:
Traders frequently take coincident positions in both the cash and
futures markets
Any lead-lag relationship do not last for more than half an hour;
this study uses high frequency (10 min data)
Emphasis on forecasting accuracy and development of trading
strategy for market practitioners to gain trading profits

. . . . . .
Theoretical Relationship between Spots and Futures

Consider a cost of carry model,

Ft = St e[(r−d)(T −t)] , ft = st + (r − d)

, where ft = ln(Ft /Ft−1 ), st = ln(St /St−1 ).

Futures and spot returns perfectly contemporaneously related
Existence of lead-lag relationships
Market sediment
Arbitrage trading

. . . . . .
Theoretical Relationship between Spots and Futures

. . . . . .
Theoretical Relationship between Spots and Futures

Transaction preference for futures (sentiment indicator)

Highly liquid market
Easily available short positions
Low margins
Leveraged positions
Rapid execution

. . . . . .
The Data

13,035 10-min observations

All trading days from June 1996 to April 1997
FTSE 100
Spot prices calculated every 1 min
Futures prices taking average of last bid/ask prices during 10 min
period

. . . . . .
Methodology

The spot and futures prices should never drift too far apart,
suggesting that a cointergrating relationship might be
appropriate.
We employ the Engle and Granger (1987) single equation
technique rather than the Johansen (1988) for simplicity.

ln St = γ0 + γ1 ln Ft

. . . . . .
Methodology

We do indeed find, as expected, that the log-price series for the

spot and the futures market are I(1).
If cointergration exists between the two series, then the Granger
representation theorem states the there is a corresponding error
correction model (ECM) as following,
∑
r ∑
s
∆ ln St = β0 + δẑt−1 + βi ∆ ln St−i + αi ∆ ln Ft + ϵt
i=1 j=1

where ẑ = ln St − γ̂0 − γ̂1 ln Ft .

. . . . . .
Methodology

. . . . . .
Methodology

. . . . . .
Methodology

According to the DF test result of ẑt , there is a clear evidence of

rejection of the null hypothesis of a unit root in these residuals
and we therefore conclude that there indeed exists a
cointergrating relationship.
By using SBIC, we select one lag of each of ft and st for inclusion
in the ECM.

. . . . . .
Methodology

The positive coefficient of ft−1 implies that the price discovery

role of the futures market for the spot market.
The coefficient of ẑt−1 is negative, suggesting that if st is large
relative to the equilibrium relationship at time t-1, then it is
expected to adjust downwards during the next period.

. . . . . .
Methodology

Consider a ECM-COC, the cost of carry theory model.

ẑt = ln St − γ̂0 − γ̂1 ln Ft − γ̂2 (r − d)(T − t)

The coefficient estimates are extremely similar to those observed

in the previous case, and the cointergrating regression are indeed
stationary.

. . . . . .
Methodology

. . . . . .
Methodology

Consider an ARMA model.

∑
p ∑
q
st = α0 + αi st−i + βj ut−j + µt
i=1 j=1

Again the SBIC criterion, it suggests that only one autoregressive

lag and no moving average lags are optimal. So, we utilize AR(1)
in following context.

. . . . . .
Methodology

. . . . . .
Methodology

Consider a VAR model.

∑
p ∑
q
st = θ0 + θi st−i + ϕi ft−j + vt
i=1 j=1

A multivariate extension of SBIC is used. Once again selects a

lag length of one for the variables.

. . . . . .
Methodology

. . . . . .
Methodology

Conduct a forecast by using 1040 10-min observations for May 1997, a

bullish month, which were not included in the original sample. We
then compare to the actual return by criteria such as RMSE, MAE.
1 VAR is better than ARMA model.
2 ECM is better than VAR and ARMA models. ARMA and VAR
will lose any long-term properties of the data.

. . . . . .
Methodology

. . . . . .
Trading Strategy

Liquid trading strategy

Buy and hold strategy
Filter strategy
Better predicted return than average
Better predicted return than first decile, 10%
High arbitrary cut-off, by a rigorous standard

. . . . . .
Trading Strategy

. . . . . .
Trading Strategy

Useful applications in financial markets.

Active equity market makers (significantly lower transaction
costs)
For traders interested in high frequency transacting
In future may generate average returns in excess of transaction
costs
Form a more appropriate proxy for the index to reduce
transaction costs
Trading becoming increasingly automated (slippage time
reduced)
Nevertheless, there are potential profitable circumstances for market
makers.
. . . . . .