To What Extent did Stock Index Futures Contribute to the October 1987 Stock Market Crash?

Author(s): Antonios Antoniou and Ian Garrett Reviewed work(s): Source: The Economic Journal, Vol. 103, No. 421 (Nov., 1993), pp. 1444-1461 Published by: Blackwell Publishing for the Royal Economic Society Stable URL: http://www.jstor.org/stable/2234476 . Accessed: 07/11/2011 07:27
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing and Royal Economic Society are collaborating with JSTOR to digitize, preserve and extend access to The Economic Journal.

http://www.jstor.org

( Royal Economic Society I993. Published by Blackwell TheEconomic Journal,103 (November), I444-I46I. Publishers, io8 Cowley Road, Oxford OX4 iJF, UK and 238 Main Street, Cambridge, MA 02I42, USA

TO WHAT EXTENT DID STOCK INDEX TO THE OCTOBER FUTURES CONTRIBUTE STOCK MARKET CRASH?*
Antoniou Ian Garrett and Antonios

1987

The October I987 worldwide stock market crash has been labelled variously a panic, a debacle, a long overdue price correction and the burst of a speculative bubble to name but a few. As noted in the Presidential Task Force Report (I988) establishing the cause(s) of the crash is important given the profound effect it had on confidence (see Roll (i 988) for one explanation of the cause of the crash). Of at least equal, if not greater, importance, however, is why the initial downward pressureon prices was converted into the rather bewildering collapse that followed. This issue has received quite a substantial amount of alia attention in the United States (see inter Blume et al. (i 989), Furbush (i 989) and especially Harris (I989)). A number of these studies of the crash in the United States have focused on various aspects of the relationship between stock index futures and the underlying stock index (see especially Harris (I989)) to determine whether there was a breakdown between the two markets, although none have examined the crash in quite the same way as we do here. The issue of market breakdown was also considered quite extensively in the Presidential Task Force Report. Essentially the argument is that the two markets, that is, the underlying spot market and the derivative futures market, should effectively function as one market if futures are to serve their designated role as, amongst other things, a means for hedging stock market risk and a vehicle for price discovery. During the crash, however, the question arises as to whether the two markets functioned as one or whether the link between the two markets broke such that they effectively functioned as two separate entities. The point here is that if a cascade effect is observed, the two markets did not operate as one and futures could not effectively perform their prescribed role. The evidence in Harris (i 989) suggests that this was the case in the United States. Unfortunately there has, to our knowledge, been little systematic empirical investigation of the crash in the United Kingdom. We aim to rectify this here. Using minute by minute values of the FTSE ioo Index and minute by minute transaction prices for the December I 987 stock index futures contract, we investigate the pricing relationship between these two markets on October igth and 20th, I987 to try and determine whether the link between the two markets broke, once the effects of non-synchronous trading are accounted for. To anticipate the results,
* We would like to thank two anonymous referees and the Editor, John Hey, for their very helpful and insightful comments which have substantially improved the paper. We would also like to thankJohn Hunter, Deborah Mabbett and seminar participants at Manchester University for helpful comments on earlier drafts of the paper. Finally, we would like to thank George Constantinides and especially Merton Miller for some very illuminating discussions. The usual disclaimer applies.
[ 1444

[NOV.

I993]

I987

STOCK

MARKET

CRASH

I445

we find that the link between the markets did indeed break on October igth, manifesting itself in a cascade effect in both markets, although it was restored on the 20th. The evidence seems to suggest that whilst the futures market exacerbated the decline, the cause of the breakdown lies with the stock market. The rest of the paper is organised as follows. Section I provides a brief overview of events surrounding and during the crash. Section II discusses the nature of the pricing relationship between the two markets. In Section III, we propose a method for removing the effects of non-synchronous trading from the FTSE IOOIndex and analyse the pricing relationship between the adjusted index and the stock index futures price. Section IV concludes.
I. THE CRASH

The downturn in share prices commenced on October 6th and they fell almost continuously over the next two trading weeks. The most telling evidence of what was to come can be found by examining the New York market on the I4th, I5th and I 6th October, where the Dow Jones Index fell by 95 points, 58 points and io8 points on successive days (Bank of England (I988)). This substantial downturn signalled the worldwide collapse that was to follow, the FTSE Ioo Index opening I 38 points down and closing 250 points down on Monday October igth' (Bank of England, I988). There is a surprising difference in attitudes between the United States and United Kingdom authorities as to the role derivative markets, and stock index futures in particular, played in the decline. The Presidential Task Force Report (I988) pays considerable attention to the importance of stock index futures in the decline. This singularly contrasts with the view taken by the Bank of England that ' ... the interaction of the cash and derivative products markets seems to have played a very limited direct role in the crash in London.' (Bank of
England (I988) p. 57).

Whilst this may be true, figures for the daily trading volume of the December futures contract on the Igth and 20th show that approximately i O,OOO contracts traded on each day, nearly double that of any other near-maturity contract in i 987.2 This reinforces the fact that we cannot overlook the importance of stock index futures in the market decline and in particular the change (if any) in the pricing relationship on these two crucial days in stock market history.
II. THE PRICING RELATIONSHIP

In considering the pricing relationship between stock index futures markets and the underlying stock market, there are two quite distinct strands that have emerged in the extant literature: those studies that analyse mispricing by
1 The London Stock Exchange did not open on the i6th October due to severe storms in the south of England. Consequently the collapse in share prices seems all the more bewildering. 2 At the time of the crash, the December contract was nearest to maturity. The other expiration months for the FTSE IOOstock index futures contract are March, June and September.

(C Royal EconomicSocietyI993
49-2

I446

THE

ECONOMIC

JOURNAL

[NOVEMBER

comparing the actual futures price with its fair, or theoretically correct, value to determine whether profitable arbitrage opportunities are available (see inter and those that alia MacKinlay and Ramaswamy (I988) and Chung (i99i)) analyse the lead-lag relationship between the two markets (Kawaller et al. (i 987), Harris (i 989) and Stoll and Whaley (i 990)). Most studies tend to focus on either the former or the latter issue, but not both. Whilst we focus more on the latter issue here, the former does provide some valuable insights and indicates that results from studies of the lead-lag relationship must be viewed with some caution as the models typically used to analyse lead-lag relationships may well be misspecified, as we shall see. A. Mispricing Studies that analyse mispricing and the existence of arbitrage opportunities typically compare the differential between the actual futures price quoted at with the fair futures price, F*T. The fair time t for delivery at time T, of futures price is given by eitherF,T,two commonly used fair pricing formulae. First we have (MacKinlay and Ramaswamy, I988)

F*

= Ste(r-d)

(T-t)

(I)

where St is the value of the underlying index and (r- d) ( T- t) is the cost of carrying the portfolio to maturity, (r-d) being the differential between the yield on the risk-free interest rate and the dividends from the portfolio and (T-t) being the time to expiration. Alternatively we have (Cornell and French, I983 a, b), F*T
=

Ster(T-t)_Dker(T-k) k

(k = t+

I,...,

T),

(I')

where D is the dividend inflow from the underlying portfolio and all other variables are as defined above. For ease of exposition, we will work with (i), though similar arguments follow for (i'). The theoretical basis,3 , -F*T) is compared with transactions costs to determine if arbitrage opportunities are present.4 If the theoretical basis falls outside of the no arbitrage window determined by transactions costs, then dependent on whether the futures contract is undervalued (overvalued) due to, say, bearish (bullish) speculation in the stock index futures market, arbitrageurs will buy (sell) futures and sell (buy) stocks. It is clear that the theoretical basis is very important in the pricing relationship given that index arbitrage links the two markets and the theoretical basis determines whether arbitrage opportunities are available. The basis itself, however, is equally important in the pricing relationship. To see
3One must be careful in talking about the basis for there are several definitions. Where there may be confusion, we will refer to the futures to cash price differential as the simple basis. When there is no risk of confusion, we will refer to it as the basis. The futures to fair price differential will be referred to as the theoretical basis. 4 Note that since the yield on dividends is typically less than the yield on the risklessasset, the basis should be positive.

( Royal Economic Society

I993

I993]

I987
(i)

STOCK

MARKET

CRASH

I447

this, take natural logs of logarithms):

(lower case letters denote variables in natural

St+ (r-

X*,T =

d (T- t).

(2)

Clearly, if the futures market is pricing the stock index futures contract correctly then
ft, T
XJ T *t, =?O'

(3)

Now, to see the importance of the basis in the pricing relationship, substitute
(2)

into (3) and rearrange to obtain

ft, T-St

= (r-d) (T-t).

(4)

It is clear from (4) that the simple basis also has an important role to play in the arbitrage process. It is also apparent that upon expiration (t = T), the basis will equal zero whilst before expiration it will, theoretically, equal the cost of carry, though if the contract is near to maturity carrying costs become trivial. From a theoretical viewpoint, the basis is crucial given that arbitrage provides an important link between the two markets. From an econometric point of view, the basis also has the rather appealing interpretation as the error correction mechanism which prevents prices in the two markets drifting apart without bound. The importance of the basis cannot be understated, for as Harris (i989, p. 77) points out, 'The (simple) basis is studied because it is a key determinant of whether arbitrage opportunities exist, because variance in the basis is a measure of how well integrated the markets are, and because the basis is related to tests for causality among the prices in the two markets.' B. Lead-lag Relationships The second component of the pricing relationship is the lead-lag relationship between the two markets. In general terms, the argument that underlies the analysis of lead-lag relationships between indices and index futures is predicated on the observation that, because of the existence of market imperfections, most notably transactions costs and short sale restrictions, the stock index futures price will lead the underlying spot index. This is because taking a position in the futures contract requires little capital outlay (in fact, margins can be posted in the form of interest-bearing securities so that the opportunity cost is effectively zero) and the trade can be effected quickly given the highly liquid nature of futures markets. Moreover, since the stock index futures contract represents a claim on the shares comprising the underlying index proportionate to their allocated weights, its price will reflect the equilibrium (true) value of the Index with each trade, whereas each share within the underlying index would have to trade before it reaches its equilibrium value. Thus, the futures contract will act as a vehicle for price discovery in the stock market. Of course, this relationship may in principle be two-sided. If the stock index futures price reacts to economy-wide information rather than security-specific, then information about a group of securities may cause the index to lead the futures price. Thus a feedback relationship may exist. These propositions concerning
(C Royal Economic Society
I993

I448

THE

ECONOMIC

JOURNAL

[NOVEMBER

the lead-lag relationship between the two markets are typically tested within the context of the following model :' =
i

+ai ft-i+ali
j

ft+j + t(5)

where Ast is the percentage change in the index value, Aft is the percentage change in the futures price6 and summations run from i,j = I,..., k. If the a. are statistically zero while the ai are not, then the futures leads the cash and vice versa. If both the ai and a, (or at least some of them) are non-zero then a feedback relationship exists. There are, however, problems with this approach to testing lead-lag relationships, particularly with the specification of (5). Essentially, (5) is a VAR, specified in first differences with st andft being I(i) in the terminology of Engle and Granger (i 987), such that their first differences are stationary (i.e. I (o)). However, it is well known from the Granger Representation Theorem (Engle and Granger (I987)) that if two I(i) series cointegrate, then there will exist a linear combination of these series which is 1(o) and an error correction representation of the VAR will exist. From the preceding discussion, it is clear that the error correction term in this case is the arbitrage link, that is, the basis. Thus, any analysis of lead-lag relationships must necessarily be tied to the arbitrage link between the two markets and as such tests of this sort should be conducted within the framework of the error correction representation of the VAR, with any restrictions tested. Failure to do so may well result in invalid inference because the arbitrage link is effectively ignored.
III. MODELLING THE PRICING RELATIONSHIP

A. The Data The data we use to model the pricing relationship are minute by minute values of the FTSE I 00 Stock Index and minute by minute transactions prices7 for the December I987 FTSE IOOStock Index futures contract for the Igth and 20th October I987. The data used are for the period 09.IO to I 6.05 on both days. The data were kindly provided by the London Stock Exchange and LIFFE and are plotted in Figs. I and 2. One of the interesting features of the data is the fact that the futures appears to have traded at a discount which was at times substantial. This is indicative of the presence of arbitrage opportunities (see fn. 4). An interesting issue to be analysed is why these apparent opportunities for arbitrage persisted. However, one must be careful in uncritically using the data for the FTSE IOO Index since the recorded value is unlikely to reflect its true value. This arises because not all shares within the Index will necessarily trade in any one given minute. Some will react to new information with a time lag, leading to the so-called problem of non-synchronous trading whereby the
From now on, as there is no risk of confusion, t, T will be referred to as ft. Usually, Aft is included to make (5) 'structural', in which case instrumental variables should be used in its estimation. The omission of Aft does not alter the substance of the arguments that follow. 7There are a few minutes during both days where transactions never took place. In this case, we use an average of the bid-ask quotes for that minute. These periods of no trading are, however, very few and far between. (C Royal Economic Society
I993

I993] 2185

I987

STOCK

MARKET

CRASH

I449

2100

2015
....

1931 9:10 Futures Fig. 1985

i.

10:55

12:40 Index --

14:25

16:05

Minute by minute FTSE ioo futures and index prices i9 October

I987.

1840

1695

1550 9:10 Futures Fig.

2.

10:55 Index--

12:40

14:25

16:05

Minute by minute FTSE ioo futures and index prices 20 October i987.

reported value of the Index contains old, or stale, prices. The implications of non-synchronous trading should be fairly obvious, particularly with regard to the perceived presence or otherwise of arbitrage opportunities given that the existence of arbitrage opportunities is determined by comparing the basis (adjusted for the cost of carrying) with transaction costs. If the basis itself is not
C)Royal EconomicSocietyI993

I450

THE

ECONOMIC

JOURNAL

[NOVEMBER

correct, as will be the case if non-synchronous trading is severe, then arbitrage We opportunities may simply be statistical illusions (see Miller et al. (i99i)). consider this issue in more depth in the next section. Trading Non-synchronous B. Removing Effects In recent years attention has re-focused on non-synchronous trading and its effects on prices and returns. The issue of non-synchronous trading is not new (see, for example, Fisher (I966)) but there has been a shift in emphasis away from its effect on the empirical application of asset pricing models (Dimson (I979), for example) to its effect on prices and returns (MacKinlay and Ramaswamy (I988), Blume et al. (I989), Harris (I989), Lo and MacKinlay (i99o) and Stoll and Whaley (I990)). The problem non-synchronous trading induces is that if the Index does contain stale prices, spurious autocorrelation in prices and returns will be observed. There is also another aspect to this argument. The FTSE IOO Index is constructed by taking a weighted average of the mid-quotes, at which market makers are forced to trade, of the prices of the securities that comprise the Index. Therefore, the issue of non-synchronous trading is also essentially one of whether all previous information is incorporated in the current price quotes. If information arrivesrandomly, then in the presence of non-synchronous trading, one would expect to find a moving average error. To demonstrate, suppose that returnscomprise two components: the expected return and the unexpected return, which is a random process. If information also arrives randomly, then by definition, its effects will be felt through the unexpected return component. If this information is not incorporated immediately, however, then the unexpected return component will be correlated since previous information will also have an effect. Therefore, non-synchronous trading will generate a moving average error structure in returns and as such, any model of non-synchronous trading must capture these features. The problem of removing non-synchronous trading effects can be thought of as a signal extraction problem where the signal to be extracted is the true value of the Index. Thus, we can formulate the problem as (6) + St = St* ut, where St is the observed value of the Index, St*is the signal to be extracted, thereby representingthe true value of the Index, and utis the non-synchronous (6) can be viewed as an trading adjustment. Following Garrett (i99i), unobserved components model and we can extract S* using the Kalman Filter. The advantage of using the Kalman Filter in this situation is that it only makes use of past information and as such, can be applied in real time. In the terminology of the Kalman Filter, (6) is known as the measurement equation. To make the model complete, we also require what are known as transition equations which describe the evolution of the unobserved component. We specify the transition equations for the evolution of St*as
+ St*= St*-, I?t- + et)
Alt= At-l + Cts(7b)
K Royal Economic Society
1993

(7a)

I993]

I987

STOCK

MARKET

CRASH

145I

Specifying the transition equations in this way means that S* evolves as a random walk, with the trend in St*also being allowed to evolve in a stochastic fashion. The intuition behind this specification of the problem stems from the arguments in Ross (I989). Ross (I989), for example, shows theoretically that prices, and hence the rate of change of prices, will move in response to new information. Given the assumption that information arrives in a stochastic fashion, prices, and the trend in prices, will evolve stochastically. This is captured by the transition equations given by (7). Further, the system given by (6) and (7) can be written as an ARIMA(o, 2, 2) process. Therefore, the model takes account of the moving average component that is present when there are non-synchronous trading effects. In addition, if observed returns are highly autoregressive because of non-synchronous trading, this model will provide a good approximation to the true value of the Index. A final advantage of this approach is that, if there is no apparent trend in the observed price series, then , can be constrained to zero. If fi is constrained to zero, then St can be written as an ARIMA(o, i, i) process. This legitimises the apparently ad hocapproach of including an MA( i) error in models of returns to capture non-synchronous trading (see, for example, Hamao et al. (I990)). To demonstrate the workings of the Kalman Filter in estimating (6) and (7), using the notation in Harvey (I987), we can set up the model in state space form as follows. Define the following vectors and matrices
Z=
?)a

0=

I]

[t

Then the measurement and transition equations can be written as

= St Zt/ t+8t)
at = O(3t-1 + Vt)

(8)
(9)

where uthas variance o2htand vt has variance o-2Qt. If we define as the best A&-1 estimate of 0tY1 and Pt_1 as the covariance matrix of &t-l then we have the following prediction equations:
o tlt-l= Ptit-l = e OPt-CZt-1) +Qt (Io) (I I)

As observations on St become available, so we can update the prediction equations and thereby update the estimate of St*.The updating equations are given by A -1 zt + htk (I 2) at= 0tit-l + Ptit- Zt(St - z P,) t-/z
Pt=tit-Ptt-1 zt zt Ptt-l/Zt Ptit-1 zt + ht. (I 3)

The prediction and updating equations define the Kalman filter. The likelihood function can then be expressed as a function of the one-step-ahead prediction errors and the model can be estimated by maximum likelihood. This seems a more natural and intuitive way to address the non-synchronous trading problem. The model is entirely compatible with the class of ARMA(p,
( Royal Economic Society
1993

1452

THE

ECONOMIC

JOURNAL

[NOVEMBER

q) models for returns proposed by Stoll and Whaley (I99O), without incurring the identification problems that occur, not only through identifying p and q, but through estimating over-parameterised versions of such models. In addition, the model is similar in spirit to that proposed by Harris (I989), who essentially uses factor analysis to extract a factor common to each security comprising the S & P 500 Index. However, this approach does not require the detailed data that Harris' (I989) method requires.8 The system (6) and (7) was estimated using (the log of the) minute-byminute recordedvalue of the FTSE i oo Index to generate the non-synchronous trading adjustment. Graphs of the non-synchronous trading adjustment to the Index9 are presented in Figs. 3 and 4. As can be seen, the adjustment is

0.01

-0 01

9:10 Adjustment

10:55

12:40

14:25

16:05

Fig. 3. Minute by minute non-synchronous trading adjustment I9 October

I987.

relatively small on both days (the means of the absolute value of the adjustment and the absolute value of the adjustment relative to the Index on the igth, for example,are oooo8 and ooooI respectively oo002and o0oo3for the 20th) and perhaps reflecting the fact that the Index only comprises iOO shares. Further, although very small, non-synchronicity is less of a problem at the opening than it is at the closing. One explanation for this observed behaviour is that, at the time of the crash, the stock market closed after the futures market, such that orders could still be processed. In addition, certainly for the igth, the Bank of England (I988) observed that
8 Specifically, to implement Harris' (I989) approach, transaction by transaction data on each individual security is required to construct a measure of St*. ' The figures are multiplied by IOOfor readability.

K Royal Economic Society 1993

I993]

I987 STOCK MARKET

CRASH

I453

001

-0-01

9:10 Adjustment

10:55

12:40

14:25

16:05

Fig. 4. Minute by minute non-synchronous trading adjustment 20 October

I987.

'The capital resources readily available to the London market makers allowed them, in the initial stages of the crash at least, to absorb substantial amounts of stock. The market imbalance was spread among all the major firms active in the equities market, three quarters of which are part of well-capitalised financial conglomerates.' (Bank of England (i 988) p. 56). This is a point we will return to later on. C. Did theLinkBreak? From the preceding discussion, any analysis of the pricing relationship should be conducted in the context of a VAR. However, we also argued that any attempt to analyse lead-lag relationships between the stock market and the stock index futures market must take into account the arbitrage link between the markets which, from the discussion in Section II, can be identified as the basis. The basis provides this link theoretically through its role in index arbitrage and econometrically through its role as the error correction mechanism which ensures that the two prices do not drift apart without bound, that is, it ensuresthat in the long-runf = s. To incorporate these features, begin by specifying the following unrestricted N-variable VAR:
x j+H1xt-1+ * I* +kxt-k+ t (I4)

where tt is a vector of intercept terms, x = [f, s*'] and ? is a vector of error terms. Following Johansen (i 988) and Johansen and Juselius (i 990),
reparameterise (I 4) to give
Axt = t+
K Royal Economic Society
I993

r1

AXt1

+***

rk-1

AXt

k+l + nxt-k +

?t*

(5)

I454

THE

ECONOMIC

JOURNAL

[NOVEMBER

Equation (I 5) is now a VAR reparameterised in error correction form where .. .-Hk) HI = - (I -Hll-. represents the long-run response matrix. Writing the long-run response matrix as H = op', then the linear combinations P'Xtk will be I (o) if there is cointegration, with a being the adjustment coefficients, and the long-run response matrix will be of reduced rank. The Johansen test for cointegration is then based on testing the rank of the matrix H. Denoting rank (H) by r, there are three possibilities. First, r = o in which case all of the variables are I (i) and there are no co-integrating vectors. Second, r = N in which case all of the variables are I(o) and there will be N co-integrating vectors given that any linear combination of stationary variables will also be stationary. Finally, o < r < N in which case there will be r linear combinations of the non-stationary variables that are stationary, that is, there will be r cointegrating vectors or, equivalently, N- r common stochastic trends. The advantage of using the Johansen procedure is that it is possible to test restrictions on the co-integrating vectors, the statistics being X2 distributed. This is particularly useful in this case since we know the form the co-integrating vector should take. For the basis to be the co-integrating vector, we require proportionality to hold, that is, in terms of the equationft = yo +y ls* + et, we require Yi to be equal to one. yo can be interpreted as the cost of carry in this case since on an intra-day basis it will be constant and if the futures contract is near maturity, it should be near zero. Table i reports the test statistics Table i Testsfor Numberof Co-integratingVectors
Igth October Null
Alternative
Amax Atrace

20th October
Amax Atrace

r=

r< r=
= =

2I-7I

25s3I

i8-sI

23-83

3-606

3-606

5-38

5-38

Restrictions
Yo
X' %: I%

(I) (I)

I4-I7

2-339

Yi

I4.I3

2-233

discussed in Johansen and Juselius (i990) for the number of co-integrating vectors and also tests the restrictions on the co-integrating vectors. The null hypothesis of zero co-integrating vectors is rejected at the 5%0 level on both days whilst the null of one co-integrating vector cannot be rejected.10 It is clear, then, that both variables are I(i), with the linear combination being I(o). What is interesting from these results is the form of the co-integrating vector for
10 For critical values seejohansen andJuselius (I990), table A3. Given that the robustnessof thejohansen procedure to relaxation of the Gaussian assumptionsis unknown, and we find evidence of significant ARCH effects (see the GARCH models in table 2), we also tested for co-integration and tested for a unit root in the basis using the Phillips-Ouliaris ( I990) and Phillips-Perron (i 988) tests which allow for such heterogeneity. Whilst, in the interest of brevity, we do not report these results here, they confirm the conclusions reached in the main text. These results are available from the authors upon request.

? Royal Economic Society I993

I993]

I987

STOCK

MARKET

CRASH

I455

the two days and this provides us with a first idea as to what happened on these two days. The important restriction here is the proportionality restriction, though the cost of carry restriction does have minor interest.11 The proportionality restriction is strongly rejected for the igth October, as, less importantly, is the zero cost of carry restriction. The implication of this is that whilst the two markets were linked on the igth, that link was not the basis (adjusted for the cost of carry) and therefore, by implication, the link was not the one provided by index arbitrage. Indeed, for the igth, the co-integrating s*. The vector is given byft = I27045* - 2~o765, which clearly shows thatft evidence, then, suggests that the arbitrage link did not operate effectively on the i gth October: the important link between the two markets broke down. The implication of this is that the mechanism that serves to stabilise prices in both markets, index arbitrage, would not serve its purpose. If stock index futures prices were falling such that the futures price was below its fair value and outside of the no-arbitrage window then arbitrageurs would buy futures and sell stock. Thus, initial selling pressure would be transmitted from the stock index futures market to the stock market. If the futures price then rose so that it lay outside the upper no-arbitrage window, then the reverse trade would be initiated and buying pressure would be transmitted from the stock index futures market. Thus, the futures price would fluctuate around its equilibrium value and the basis would be stationary. However, the basis on the igth has a unit root12 and thus of little guide in determining the existence of arbitrage opportunities. This was not the case on the 20th, when the link between the two markets was the basis and again, by implication, the arbitrage link was restored. We will return to the question of how this might have occurred later. What we appear to observe, then, is different behaviour by the markets on the two different days. On the 20th, the error correction term was the basis whilst on the i gth it was not. This suggests that the important link, index arbitrage, did not function effectively. It must be noted, however, that some link did still exist on the igth because both prices continued to fall in unison. The implication of this is that we should observe differences in the behaviour of the pricing relationship between the two markets on both days. From the earlier discussion and the results of the Johansen procedure, we know that the pricing relationship should be modelled in the context of the error correction representation of the VAR. We began by estimating the VAR with io lags of each variable (except the error correction term, which is lagged once),"3 with each equation showing the presence of significant ARCH effects. This is perhaps not surprising given the rather volatile events on those two days. To capture this volatility, we reestimate the
" The zero restriction is not so important for if the homogeneity restriction were valid but the zero cost of carry restriction were not, the error correction term would simply be the basis adjusted for the cost of carrying and this is what arbitrageurswould compare with transactionscosts to see if arbitrage opportunities were available. 12 For example, ADF tests on (ft-S*) and A(ft-st*) yield values of - 245 and -I 438 respectively. The 2'87. 5% critical value is approximately13 We realise of course that the choice of lag length is somewhat ad hoc.However, i o minutes does not seem an unreasonable starting point for our analysis given the extraordinary events that took place on those two fateful days. Full results for the VAR without GARCH are available from the authors upon request. ? Royal Economic Society
I993

I456

THE

ECONOMIC

JOURNAL

[NOVEMBER

models using the GARCH (i, i) specification (see Bollerslev (i986)), whereby + the conditional variance evolves according to C2 = j + cxe62 3o,21. Parsimonious models for both markets on both days are reported in table 2. The Table
2

Modelling the Pricing Relationship

I9

October I987
(-2.5IO) tO_ t_1+o 605
(IO 99)

Futures

Aft =

I II 98Ast* - 0-o5oeCmt,

(30o83)

G,t=

0-132+0o2I8
(5 294)
=

(3 973)

Index

Q%c(I6)

I5

86, QH(IO) = 5-o6

AS*= o?oIoAft-l+o?oI4Aft-2+00OII

0OI ft-5 ft-3+0ooI5Aft-4+0 (2-760) (4423) (5-02I) (3 367) (2-768) + oo7Aft-6+ o003ft7 + 001 2Aft-8 + 0022Aft9 +O-I94AS*1 (8 I46) (3-6I3) (4-IOI) (2-934) (2-045) + 0-2 I I ASt-2 + o I43ASt_3+ O I46ASt-4 + ?0049ASt-5

(3-6I3)

(4 446)
0t2 =0-002

(3-497)
+
(I2

(2-022)

o-988t2
27)

i + oo46y2_
(I *02 I)

(6-765)
Qsc(io)
=

I6-35,

QH(IO)

"IO

20

October I987
(-4-60)

Futures
Ct2 =

Aft =-0-294ft2-490 + 0-464e2

1+0-272
02

(4-076) (6-o87)
QSC(IO)
=

I5525,

(2-328) = QH(IO) IO-38

+

Index

AS

ooo4f

+ o484S*

1 + o293ASt*2

oo004ecm,

(2-708)
at2 =

(9-I90)
0-002
(2-969)

(5 I54)
+ 0760oTtL
(I I -93)

(3.574)

+ O- I 054-1
(3-557)

QSC(IO)

I2 7I,

QH(IO) =

9-46

Notes:

ecmis the error correction term.

Figures in parentheses are t ratios.

Constants in the variance equation are multiplied by

QSC and QH are portmanteau

Io5

for readability.
distributed X2(.).

tests for serial correlation and heteroscedasticity,

models seem to be adequately specified with none of the diagnostic tests being significant at the i 00 level. Turning our attention to the results, an interesting scenario emerges. If both the equity and futures markets were effectively functioning then the basis should be significant in explaining price movements in both markets. However, we observe something very different on both the igth and the 20th October I 987. The first indication of breakdown is the fact that, as was discussed earlier, the basis was not the error correction term on the igth. The results also tend to support those found in studies of the United States markets (for example, Kawaller et al. (I987), Stoll and Whaley (I990)) in the sense that the evidence suggests that the stock index futures market tended to lead the stock market with some evidence of a weaker feedback relationship. However, what is of interest here is the extent of the feedback from the stock market to the stock
t Royal Economic Society
I993

I993]

I987 STOCK MARKET

CRASH

1457

index futures market on the i gth October for whilst some of the feedback occurs through the error correction term the interesting point to note is the coefficient on As* 1: it is approximately I 2. Thus, whilst the stock market was reacting to declines in the stock index futures market, the stock index futures market was reacting (indeed one could argue overreacting) to declines in the stock market. Therefore, a vicious downward spiral in prices, or cascade effect, ensued. Moreover, given the evidence presented earlier that the arbitrage link effectively broke on the i gth, there was nothing to counteract the fall (see the earlier discussion in section II.A and III.C). This conclusion of market breakdown is reinforced when we consider the behaviour of the conditional variance for the markets on both days. An interesting aspect of the interaction between conditional variances is the notion of co-integration in variance (for a brief discussion, see Bollerslev et al. (I 992)). There is a great deal of evidence (see the review and bibliography in Bollerslev et al. (I992)) showing that for many financial time series the restriction that a+ fl = i in the conditional variance equation cannot be rejected such that the conditional variance has a unit root (Integrated GARCH, or IGARCH, in the terminology of Engle and Bollerslev (I986)) such that shocks to the variance persist indefinitely. This obviously raises the question about whether the conditional variances of two similar series co-integrate such that a linear combination of them shows no persistence. Through the similarity of the stock index and stock index futures prices one would expect their conditional variances to co-integrate, and this is indeed the case on the 20th since shocks to the conditional variances for both markets are not persistent and therefore a linear combination would not be persistent. However, the i gth again shows an altogether different state of affairs, with the conditional variance for the index exhibiting I(i) behaviour (a test of the restriction cz+fl = i yields x2(I) = O0I93) as opposed to the apparent I(o) type behaviour of the conditional variance for the futures. Given that co-integration requires the same order of integration in the individual series, it is apparent that the two are not cointegrated in variance and this is a further indication of market breakdown. Thus, there is clear evidence that there was a breakdown. Moreover, the overreaction of the futures price would appear to provide prima facie evidence that the anti-futures lobby is correct: futures destabilise. However, this is not the case, as we shall see. D. WhyDid The LinkBreakand Why Was,It Restored? It would seem that the initial downward pressureon prices manifested itself in the decline that followed because the arbitrage link between the two markets broke down. The important question is why this should be the case. Given that the appropriate arbitrage strategy to undertake was to buy futures and short stock, any restrictions on short sales may provide at least some explanation of why the link broke. This certainly goes some way to explaining why the cascade effect occurred in the United States, for stock cannot be sold short unless the previous movement was a price increase (the so-called up tick rule). However, there are very few, if any, restrictions (certainly at the time of the crash there
C Royal EconomicSocietyI993

I458

THE

ECONOMIC

JOURNAL

[NOVEMBER

was no up tick rule) on the short sale of stock within the account period14 on the London market. Moreover, given that the crash occurred during an account period (the account began on I 2th October, with settlement of any trades during the account to take place on 2nd November), the answer to the question of why arbitrage could not function effectively must lie elsewhere. Conversations with specialists have revealed that a major problem, certainly on the i gth, was the drying up of liquidity in the stock market to the extent that arbitrage transactions became impossible to undertake. This same point is made by the Bank of England (I988) who note that the response of market makers to the increased selling pressure, which they were able to absorb initially, was a widening of the bid-ask spread in order to compensate first for long positions they had accumulated by absorbing the selling pressure but were not themselves in a position to liquidate fully and second to the perceived possibility of default by investors come the end of the account. Indeed, as is documented by the Bank of England ((I988), p. 57, table D) the average October I987 was some 3300 higher than the spread for the period I9-23 average spread for the whole of September I987. This widening of the spread is indicative of a liquidity problem. Since the structure of the London market emphasises the provision of liquidity through increased competition, and hence narrow spreads, the implication is that market makers either significantly reduced their bid prices and/or significantly increased their ask prices, making trading unattractive. This action then removes liquidity from the market. In addition, during the crash market makers reduced the average quote size by half and finally, some market makers even refused to answer their telephones, thereby failing to fulfil their obligation to make markets in all trading conditions (Bank of England, I 988). Given that, with the futures being undervalued the appropriate arbitrage strategy would be to buy futures and simultaneously sell stock, it is hardly surprising that the arbitrage link broke down. To illustrate, consider the following scenario. An investor perceives the existence of an arbitrage opportunity. We have already seen that market makers widened their spread, increasing the cost of the stock market side of the transaction. Further, the quote size was halved, again increasing the cost of the stock market side of the transaction. Finally, the investor had to find a market maker willing to transact in the first place. Taken together, arbitrage was effectively prohibited. The cause of the breakdown, then, appears to be a liquidity problem. Moreover, because this liquidity problem occurred in the stock market, the stock market was the cause of the break. Consider now the 20th. We see a partial reversal of what occurred on the I gth: we see -that the stock index futures market leads the stock market, both through Aft-l and through the error correction term, with no feedback from the stock market to the futures, even though the futures sold at a discount. This also coincides with the restoration of the basis as the arbitrage link. It would appear, then, that the reaction of participants in the stock index futures market on the 20th was effectively to ignore price movements in the stock market whilst
14 The settlement system in the London market is based on the notion of an account period, whereby any trades are not settled until the end of the account. The account period is typically two to three weeks.

? Royal Economic Society

I993

I993]

I987

STOCK

MARKET

CRASH

I459

the stock market utilised the information provided by price movements in the index futures market. There are several points to note about the behaviour of the markets which may explain why this apparent 'reversal' took place. Consider first of all the conclusions reached in the Quality of Markets Report (see the discussion in Kleidon and Whaley (I992)) about why sellers were willing to trade futures at a discount. They argue that two factorswere at work: sellers did not believe they could transact immediately in the stock market at quoted prices and second, sellers may not have believed that the prices quoted were the correct ones. Consider now the non-synchronous trading adjustment plotted in Figs. 3 and 4. These show that the non-synchronoustrading problem was more severe on the 20th and therefore, by implication, there was less trading in the stock market on the 20th relative to the Igth. Add to this the argument that liquidity dried up and the results confirm the conclusions reached in the Quality of Markets Report. This situation helps to explain why the basis was restored as the link. The reason for the break in the link on the igth was the absence of liquidity. The Quality of Markets Report suggest that this liquidity problem drove sellers away from the stock market to the futures market. By implication, there was less trading in the stock market on the 20th relative to the Igth, thereby alleviating the liquidity problem and in turn, alleviating the original source of the breakdown. Drawing all these points together, the evidence points to the stock market as the cause of the breakdown.
IV. CONCLUSIONS

In this paper, we have set out to analyse the pricing relationship between the FTSE IOOIndex, a representative measure of stock market performance, and the FTSE IOO stock index futures contract on the igth and 20th October i987, the period of the stock market crash. In particular we have set out to investigate the extent to which the FTSE IOOfutures contract contributed to the crash. To address this question, we examine the pricing relationship between the stock market and stock index futures market on these two fateful days. In doing this, we synthesis two apparently diverse strands of the literature on modelling pricing relationships between stock and stock index futures markets into an error correction framework. The advantage of this framework is that it immediately yields testable propositionswith regard to the functioning of these markets. Before modelling the pricing relationship, however, we address the non-synchronous trading problem prevalent in high frequency price data on indices. Utilising a new method for removing these effects, we find that nonsynchronicity explains little of the observed behaviour of the markets, a result consistent with Harris' (i989) findings for the United States. Despite the fact that the futures traded at a discount, which is indicative of arbitrage opportunities, we find that the link between the two markets on the I gth was not the link provided by arbitrage. We also find that the futures price strongly leads the Index with some weaker evidence of feedback from the Index to the futures on the I gth, a result apparently consistent with evidence from the
( Royal Economic SocietyI993

I460

THE

ECONOMIC

JOURNAL

[NOVEMBER

United States for stable time periods. However, in this turbulent period we observe the futures price on the i gth overreacting to information contained in the previous minute's Index price. On the basis of this evidence, it is tempting to conclude that the futures market was to blame. This is not the case. What we observe on the i gth is a situation where arbitrage trades could not be executed effectively because of liquidity problems in the stock market. As a result, the arbitrage link broke down, the outcome being a vicious downward spiral in prices in both markets. For the 20th, the futures continued to trade at a discount, again pointing to the presence of unexploited arbitrage opportunities. In addition, we also observe a change in the nature of the pricing relationship with the restoration of the basis as the link between the two markets, the futures still leading the spot but this time with no feedback from the spot to the futures. As the Quality of Markets Report suggest, the liquidity problem drove sellers away from the stock market to the futures market. This is precisely what restored the basis as the link. What seems clear, then, is that the futures market did not serve its purpose on the igth. Indeed, it helped exacerbate the downward movement in prices. However, the blame for this does not lie with the futures market, for the initial source of the problem was the stock market, in particular the drying up of liquidity. The message from this is clear. There is no need to look towards further regulating the futures market as a separate entity because the futures market was not the source of the problem. To regulate the futures market further is to alleviate the symptoms without curing the illness. Regulating the two markets as a single entity, as recommended in the Presidential Task Force Report (i 988) is only part of the solution. In addition, it is necessary to consider the trading practices in both markets. Kleidon and Whaley (I992) suggest that the solution for the United States is more efficient trading systems for the stock market. Similar conclusions must apply for the United Kingdom, with reforms of trading practices bringing trading systems in both markets closer together."5 We cannot know for sure, but we suspect that had this been the case the crash might never have taken hold in the way it did. Brunel University Date of receiptoffinal typescript: March I993

REFERENCES

Bulletin,February, pp. 5 I-8. Bank of England (I988). 'The equity market crash.' Bankof EnglandQuarterly Blume, M. E., MacKinlay, A. C. and Terker, B. (I989). 'Order imbalances and stock price movements on October 19 and 20, I987.' Journalof Finance,vol. 44, pp. 827-48. Bollerslev, T. (I986). 'Generalised autoregressive conditional heteroscedasticity.'Journalof Econometrics,

vol. 3I,

pp. 307-27-

15 For example, in the United Kingdom there are two markets closely linked yet with different trading systems. The stock market is a purely screen-based system whilst the futures market has trading based on open outcry. Whilst in theory there is no reason as to why purely screen-based systems should not execute trades efficiently, in practice human and technical factors will ensure this is not the case in times of market turbulence.

? Royal Economic Society

I993

I993]

I987

STOCK

MARKET

CRASH

I46I

Bollerslev, T., Chou, R. Y. and Kroner, K. F. (I992). vol. empirical evidence.' Journalof Econometrics, 52, pp. 5-59. Chung, Y. P. (i 99i). 'A transactions data test of stock index futures market efficiency and index arbitrage profitability.' Journalof Finance,vol. 46, pp. I79I-809. Cornell, B. and French, K. R. (i983a). 'The pricing of stock index futures.' Journalof FuturesMarkets, vol. 3, pp.- I4Cornell, B. and French, K. R. (i 983 b). 'Taxes and the pricing of stock index futures.' Journalof Finance, vol. 38, pp. 675-94Dimson, E. (I979). 'Risk measurement when shares are subject to infrequent trading.' Journalof Financial Economics, 7, pp. I97-226. vol. Engle, R. F. and Bollerslev, T. (I986). 'Modelling the persistence of conditional variances.' Econometric vol. 5, pp. I-50. Reviews, Engle, R. F. and Granger, C. W. J. (i 987). 'Cointegration and error correction: representation, estimation and testing.' Econometrica, 55, pp. 25I-76. vol. Fisher, L. (I966). 'Some new stock market indexes.' Journalof Business,vol. 39, pp.I 9I-225. Furbush, D. (i 989). 'Program trading and price movement: evidence from the October i 987 market crash.' vol. FinancialManagement, I9, pp. 68-8I. Garrett, I. (I99I). 'Nonsynchronous trading and the stock market crash.' Mimeo, Centre for Empirical Research in Finance, Department of Economics, Brunel University, London. Hamao, Y., Masulis, R. W. and Ng, V. (I990). 'Correlations in price changes and volatility across vol. 3, pp. 28I-307. international stock markets.' Reviewof FinancialStudies, vol. 44, pp. 77-99. Harris, L. H. (I989). 'The October I987 S&P 500 stock-futuresbasis.' Journalof Finance, in Fifth Harvey, A. C. (I987). 'Applications of the Kalman filter in econometrics.' In Advances Econometrics: World Congress, vol. I (ed. T. F. Bewley). Econometric Society Monograph No. I3. Cambridge: Cambridge University Press. and Dynamics Control, Johansen, S. (i 988). 'Statistical analysis of cointegration vectors.' Journalof Economic vol. I2, pp. 23I-54. Johansen, S. and Juselius, K. (I990). 'Maximum likelihood estimation and inference and cointegration and with applications to the demand for money.' OxfordBulletinof Economics Statistics,vol. 52, pp.
I69-2I0.

'ARCH modelling in finance: a review of theory and

Kawaller, I. G., Koch, P. D. and Koch, T. W. (I987). 'The temporal price relationship between S&P 500 futures and the S&P 5oo Index.' Journalof Finance,vol. 42, pp. I309-29. Kleidon, A. W. and Whaley, R. E. (I992). 'One market? Stocks, futures and options during October I987.' Journalof Finance,vol. 47, pp. 85I-77. Lo, A. W. and MacKinlay, A. C. (i ggo). 'An econometric analysis of nonsynchronous trading.' Journalof Econometrics, 45, pp.I 8I-2II. vol. MacKinlay, A. C. and Ramaswamy, K. (I988). 'Index-futures arbitrage and the behaviour of stock index futures prices.' Reviewof FinancialStudies,vol. I, pp. I37-58. Miller, M. H., Muthuswamy,J. and Whaley, R. E. (I99I). 'Predictability of S&P 500 Index basis changes: School Business,University of paperno.33i, Graduate arbitrage-induced or statistical illusion?' CRSP Working of Chicago. vol. Phillips, P. C. B. and Perron, P. (I988). 'Testing for a unit root in time series regression.'Biometrika, 75, pp. 335-46. Phillips, P. C. B. and Ouliaris, S. (iggo). 'Asymptotic properties of residual-based tests for cointegration.' vol. Econometrica, 58, pp. I65-93. on submitted to the President, Report Market Mechanisms, Presidential Task Force (i 988). Presidential TaskForce The Secretary of the Treasury and Chairman of the Federal Reserve Board, January I988. AnalystsJournal,vol. 44, Roll, R. (i 988). 'The international stock market crash of October I987.' Financial pp. I9-35. Ross, S.A. (I989). 'Information and volatility: the no-arbitrage martingale approach to timing and resolution irrelevancy.' Journalof Finance,vol. 44, pp. I-I 7. Stoll, H. R. and Whaley, R. E. (i 990). 'The dynamics of stock index and stock index futuresreturns.' Journal Analysis,vol. 25, pp. 44I-68. of Financialand Quantitative

? Royal Economic Society I993