the volatility effect of futures trading: evidence from lse traded stocks listed as individual...

International Review of Financial Analysis 15 (2006) 1–20

The volatility effect of futures trading: Evidence from

LSE traded stocks listed as individual equity

futures contracts on LIFFE

Khelifa Mazouz a,*, Michael Bowe b,1

a Finance, Accounting and Law Group, Aston Business School, Birmingham, UKb Division of Accounting and Finance at Manchester Business School, Manchester, UK

Received 4 October 2004; accepted 14 July 2005

Available online 4 November 2005

Abstract

This study investigates the impact of LIFFE’s introduction of individual equity futures contracts on the

risk characteristics of the underlying stocks trading on the LSE. We employ the Fama and French three-

factor model (TFM) to measure the change in the systematic risk of the underlying stocks which arises

subsequent to the introduction of futures contracts. A GJR-GARCH(1,1) specification is used to test

whether the futures contract listing affects the permanent and/or the transitory component of the residual

variance of returns, and a control sample methodology isolates changes in the risk components that may be

caused by factors other than futures contract innovation. The observed increase (decrease) in the impact of

current (old) news on the residual variance implies that futures contract listing enhances stock market

efficiency. There is no evidence that futures innovation impacts on either the systematic risk or the

permanent component of the residual variance of returns.

D 2005 Elsevier Inc. All rights reserved.

JEL classification: C22; G12; G14

Keywords: Individual equity futures; Three-factor model; GJR-GARCH(1,1)

1. Introduction

The impact of innovation in equity derivatives contracts on the volatility of the underlying

stocks has become a particular focus of attention for observers of the stock market. Although the

1057-5219/$ -

doi:10.1016/j.i

* Correspond

E-mail add1 Tel.: +44 1

see front matter D 2005 Elsevier Inc. All rights reserved.

rfa.2005.07.001

ing author. Tel.: +44 121 359 3611x3041.

resses: [email protected] (K. Mazouz), [email protected] (M. Bowe).

61 200 3407.

K. Mazouz, M. Bowe / International Review of Financial Analysis 15 (2006) 1–202

hedging activities of large institutional investors account for a major component of the market

volume, stock index futures trading has also attracted speculators and small traders. Some

commentators (see, for instance, Chatrath, Ramchander, & Song 1998) argue the participation of

such traders may enhance underlying volatility in the stock market.2 One result of this concern

has been the imposition of regulations designed to control potentially destabilising trader

behaviour. These include increasing the level of margin requirements, halting trading subsequent

to large movement in the value of index futures contracts and prohibition of program trading

when market movements exceed some pre-determined daily interval.

This study examines the market stability and efficiency impact of introducing individual

equity futures trading based upon a number of different volatility estimators for the securities

trading in the underlying market.3 The data sample consists of the equity futures contracts which

were introduced on the LIFFE (London International Financial Futures and Options Exchange)

beginning in January 2001. The analysis makes the following central contributions. First, our

empirical specifications explicitly account for (i) the fact that firm systematic risk is time varying

and (ii) high frequency financial return series are characterised by conditional heteroscedasticity.

Failure to correct for these phenomena is likely to generate inconsistencies when measuring the

impact of futures listing on the riskiness of the underlying stocks (Mazouz, 2004; Skinner,

1989). Our use of both the single factor CAPM and the three factor model of Fama and French

(1992) in conjunction with the GJR-GARCH (1,1) process to model total (time varying) risk

helps ensure that our results are robust across alternative model specifications. Second, our use

of a carefully constructed control sample helps alleviate the potential endogeneity bias which

characterises much existing research. Futures exchanges act to maximise member revenue from

futures trading, and choose to innovate contracts on underlying instruments that rate highly on

attributes such as price volatility. In turn, these attributes may reflect changes in market and

industry-wide conditions. As such, volatility in the underlying stocks may be measured to

coincide with futures listing even if the listing event itself has no impact upon volatility in the

underlying market. We would argue that the paper’s use of a control sample, complimented by

an individual stock approach focussing on the sign and statistical significance of any risk

changes, helps to mitigate the potentially severe empirical impact of such endogeneity bias.4

A review of both the theoretical arguments and existing empirical evidence on the impact of

futures listing on the volatility of the underlying instrument generates no clear consensus.

Theoretical analyses often focus upon the relationship between the observed price and intrinsic

value of a traded security. Any manifest distinction between the two is customarily attributed to

both trading noise and market-microstructure related noise, the latter manifest in the bid–ask

spread (Amihud & Mendelson, 1987; Black, 1986). In the context of the present paper the issue

is how the introduction of derivatives trading on LIFFE is likely to impact on these two bnoiseQcomponents of the return variance of the underlying LSE traded stocks.

Figlewski (1981) adopts a liquidity perspective on this issue, claiming that listing futures

contracts contract should enhance the underlying instrument’s liquidity, and that the information

released by new derivative traders may stabilise prices on the underlying market. The

2 Arguments along these lines were especially popular in the media after the 1987 stock market crash.3 Only a few papers, such as Lee and Tong (1998) and Chatrath et al. (1998), examine the volatility impact associated

with the introduction of individual equity futures contracts which currently trade in only four countries, namely Australia,

Sweden, Hong Kong and the United Kingdom.4 A similar bcontrol sampleQ methodology is also used by Mazouz (2004), but that is in the context of option not futures

listing on a different (US) dataset.

K. Mazouz, M. Bowe / International Review of Financial Analysis 15 (2006) 1–20 3

implication for the bid–ask spread is that the introduction of futures provides an opportunity for

market makers in the underlying to both hedge their exposure, and achieve faster turnaround on

their inventory, allowing them to reduce the bid–ask spread (Demsetz, 1968; Madhavan, 2000).

The bid–ask spread may also narrow as a result of the migration of informed traders from the

stock to the futures market. This migration of informed traders is explained by the inherent

leverage and lower transaction costs associated with futures and the potential to use futures to

construct portfolios which effectively circumvent restrictions imposed on selling stock short.

Furthermore, the availability of futures as hedging vehicles may make investment in riskier

stocks more attractive, thus increasing the demand for such stocks.

In contrast, some authors adopt an alternative perspective to the liquidity impact of

introducing derivatives. Stein (1987) and Dennis and Sim (1999) argue that the information flow

occasioned by new, less informed, derivative traders may have a destabilising impact on the

market, impairing the ability of informed traders in the underlying to make information

inferences from prices. This increase in trading noise may be sufficient to more than offset the

impact of any increased liquidity, ultimately manifesting itself in an increase in the bid–ask

spread. In similar vein, Gorton and Pennacchi (1993) argue that the existence of derivative

contracts may enhance the attractiveness of the underlying market to uninformed traders.5 Lower

transaction costs in the derivatives market enable investors to hedge their positions by trading in

futures in preference to making equivalent stock transaction. This increase in the proportion of

uninformed traders in both markets may lead to an increase in the information asymmetry

component of the bid–ask spread. One may also add that the enhanced amount of trading noise

in the stock market resulting from uninformed trading activity may delay the speed at which new

information is incorporated into the stock price. This is an issue we explore further in Section 3

and test in Section 5. A contrasting perspective on this trader composition phenomenon is

provided by Glosten and Milgrom (1985) and Easley and O’Hara (1987). They maintain that the

fact market makers are now trading more with uninformed traders means that the former are

better able to reduce the implicit binformation rentQ which they charge in order to protect

themselves when trading with informed traders. The consequent reduction in the bid–ask spread

will result in reduced bid–ask bounce in the stock’s price and a lower return variance.

Reflecting the ambiguities inherent in the theoretical discussion, the evidence provided by

existing empirical studies is also mixed. Several studies examine the impact of futures trading on

the volatility of prices in the underlying cash instrument.6 These include Froewiss (1978) and

Figlewski (1981) for GNMA securities; Edwards (1988) and Harris (1989) for the Standard and

Poors 500 Index; Antoniou and Holmes (1995) for the FTSE 100 Index; Simpson and Ireland

(1985) for the Treasury bill market. A consensus based on their findings remains elusive. Some

studies maintain there is no discernible volatility effect, others find evidence of a volatility

increases and some evidence of a decrease. Thus, we believe that further evidence to inform the

subject matter using a different methodological approach is highly desirable.

The procedure adopted in the paper is as follows. First, we decompose volatility into both

systematic and diversifiable risk components (the latter measured by the variance of the

5 Ma and Rao (1988) argue that uninformed traders are more likely to use derivatives market for hedging purposes.

Their reasoning is that uninformed traders, characterised by higher prediction errors, tend to exhibit more risk averse

behaviour. Cox (1976) proposes that derivatives traders use different trading information sets than those who confine

their activity to the underlying market.6 This is in addition to the fairly extensive literature which examines the impact of options listing on the underlying

market. For recent analysis relevant to this paper see Sahlstrom (2001) and Mazouz (2004).


prediction error) using the conventional single factor capital asset pricing model (CAPM). This

is an approach to total risk decomposition commonly used in the option listing literature (Bowe

& Mazouz, 2004; Chaudhury & Elfakhani, 1995; Damodaran & Lim, 1991; Draper, Yadav, &

Watt, 1992; Sahlstrom, 2001; Skinner, 1989; Whiteside, Dukes, & Dunne, 1983).7 Second, in

contrast to other studies in the derivatives listing literature, we also employ the Fama and French

(1992) three-factor model (TFM) to distinguish between systematic and diversifiable risk. The

initial analytical rationale for this choice is provided by Fama and French (1992), which suggests

that the asset returns are better explained by the TFM than the conventional CAPM

representation, and this result is confirmed by a number of other studies, including Fama and

French (1993, 1996) and Liew and Vassalou (2000). The empirical rationale is twofold: (i) using

the alternative TFM specification provides a useful robustness check on the results we obtain

from the CAPM, and (ii) goodness-of-fit tests suggest that the TFM fits the present dataset better

the conventional CAPM. Third, the GJR-GARCH (1,1) process of Glosten, Jagannathan, and

Runkle (1993) is used to test whether the CAPM and TFM estimates of the change in the variance

of the prediction errors between pre-and post-futures periods are permanent, and whether futures

trading affects the speed at which information is incorporated into the stock price.

The central results of the analysis can be briefly summarised. The findings of the initial

CAPM-GJR-GARCH(1,1) specification suggest that futures listing increases the systematic risk

and decreases the diversifiable risk of the underlying stocks. The implication is that the presence

of futures trading enhances the firm’s cost of capital and reduces the amount of noise in the stock

market. These results contrast with those generated by the TFM-GJR-GARCH(1,1) approach.

After modelling stock returns using the TFM and the variance of the prediction errors by the

GJR-GARCH(1,1) process, we observe a decrease in both the systematic risk and the

diversifiable components of the variance of returns.

Moreover, the observed risk changes are not unique to the futures listed-stocks. Regardless of

the return generating process, the change in the systematic risk and the variance of the prediction

errors of the control sample stocks8 are not significantly different from the risk changes observed

in the sample of futures-listed stocks. Thus, overall we interpret the findings as supporting the

hypothesis that the introduction of futures trading has no impact on the volatility of the

underlying stocks. However, there is evidence that futures trading innovations increase

(decrease) the effect of current (old) news on the conditional variance. This indicates that the

introduction of individual equity futures contracts appears to enhance market efficiency. This

indicates that futures contracts should not be regarded as redundant securities.

The remainder of the paper is organised as follows. Section 2 presents the empirical

procedures, indicating how the paper measures systematic and diversifiable risk. The data is

presented in Section 3. Section 4 reports and analyses the results of the empirical analysis.

Finally, Section 5 summarises and briefly concludes.

2. Empirical procedures and testing methodology

To measure the changes in the volatility of the underlying stocks, we employ a variety of

volatility estimators. This ensures the results are robust to alternative empirical specifications.

First, we compare measures of systematic risk of the underlying stock derived from both the beta

7 The change in the systematic risk between pre- and post-futures periods may be caused by futures listing or by other

factors. Ignoring this fact may lead to misspecification of the noise component of the variance.8 We discuss the composition and construction of the control sample in the data section.


of the CAPM and the TFM of Fama and French (1993, 1996). Second, the variances of the

residual series generated from both the estimated CAPM and TFM are used to estimate the

diversifiable risk of the underlying stocks. Third, we employ the GJR-GARCH (1,1)

specification to test whether any observed changes in the variance of the prediction error are

permanent, and if the existence of a related futures contract contributes to the speed at which

information is incorporated into the stock price.

2.1. Tests using the capital asset pricing model (CAPM)

In the context of the CAPM, the prediction errors and the systematic risk of the underlying

stocks can be expressed by means of the following equation:

Ri;t � Rf ;t ¼ ai þ bi;b þ bi;cDFutures

� �Rm;t þ ei;t ð1Þ

where ei,t is the prediction error in relation to stock i’s return at time t, and Rm,t is the return on

the market in excess of the risk-free rate, Rf,t also at time t. DFutures is a binary variable taking a

value of unity following the introduction of futures trading in stock i and zero otherwise. ai is a

stock-specific constant term. The coefficients bi,b and bi,c represent the beta before futures and

the beta change across pre- and post-futures trading periods for each stock i, respectively. The t-

statistic associated with the coefficient bi,c is used to infer the statistical significance of any

perceived systematic risk change subsequent to futures listing.

If the introduction of futures trading reduces the amount of trading bnoiseQ in the stock

market, the variance of the prediction error, eit, is expected to decrease subsequent to contract

innovation. This prediction error is modelled using the following set of equations:

ei;t ¼ ci;thi;t where ci;t ~ N 0; 1ð Þ ð2Þ

and

h2i;t ¼ ui;b þ ui;cDFutures þ di;1e2i;t�1 þ ki;1h

2i;t�1 þ Bi;1e

2i;t�1It�1 ð3Þ

The statistical significance of the change in the permanent component of the variance is

measured by the t-statistic associated with the dummy coefficient in the GJR-GARCH(1,1)

specification given by Eq. (3). Specifically, we interpret the coefficients ui,b and ui,b+ui,c as the

permanent components of the prediction error variance in the pre- and post-futures periods,

respectively. Eq. (3) implies that the permanent components of the variance are not affected by:

(i) the forecast variance from the previous period (the coefficient ki,1); (ii) any information about

the variance observed in the previous period (the coefficient di,1) and (iii) any asymmetric news

effect (the coefficient Bi,1 of the dummy variable It�1, which takes a value of unity if ei,t�1 is

positive and a value of zero otherwise).9 Alternatively stated, we interpret the coefficient di,1 as

the amount of recent news, or news that arrived in the preceding period, which is incorporated in

current prices, while ki,1 reflects the cumulative amount of old news the price contains.

Another indication as to whether futures trading stabilises the market for the underlying

stocks can be ascertained by the significance of any change in the speed at which new

information is incorporated into the stock price when one compares pre- and post-futures trading

periods. The impact of futures trading on the information adjustment coefficient is measured by

9 See Engle and Ng (1993) for a discussion of issues relating to measuring the impact of news on volatility.


estimating the GJR-GARCH(1,1) Eq. (3) for each stock i separately, in both the pre- and post-

futures trading periods:

h2i;t ¼ ui þ di;1e2it�1 þ ki;1h

2i;t�1 þ Bi;1e

2i;t�1It�1 ð4Þ

If futures trading increases the speed at which new information is incorporated into the stock

price, we would expect an increase in the coefficient di,1 and a decrease in the coefficient ki,1

following the listing of a new contract.10 The non-parametric Wilcoxon-Signed Ranked test

(WSRT) can then be used to check whether the pre- and post-listing series of di,1(ki,1) belong to

the same distribution, without assuming any particular functional form for that distribution.

2.2. Tests using the three-factor model (TFM)

The extent to which the beta coefficient in Eq. (1) accurately reflects the systematic risk of the

underlying stock has been extensively scrutinised in the asset pricing literature. In proposing

their alternative, the TFM, Fama and French (1993, 1996) argue that incorporating variables

reflecting firm size, book-to-market (B/M) ratios and the return on the market portfolio into the

estimation equation improves the resulting empirical systematic risk (and other) estimates for the

underlying stocks.11 As indicated later in the empirical section (Section 4.1), the same is true for

the estimates from the dataset used in the present paper, a fact which justifies our choice of the

TFM. Moreover, using the TFM also ensures our findings on the impact of futures trading are

robust to alternative measures of the systematic and diversifiable risk components of the

underlying stocks. The modified version of the Fama and French TFM used in this study is:

Ri;t � Rf ;t ¼ ai þ bi;mb þ bi;mcDFutures

� �Rm;t þ bi;hmlb þ bi;hmlcDFutures

� �Rhml;t

þ bi;smbb þ bi;smbcDFutures

� �Rsmb;t þ 1i;t ð5Þ

where Ri,t is the observed return of security i at time t, and Rhml,t and Rsmb,t are the returns on the

familiar HML and SMB portfolios, respectively, in the Fama and French TFM. The details of the

precise manner in which this paper empirically constructs these HML and SMB portfolios is

presented in Section 3. ai is a constant term specific to security i. The coefficients bi,mb, bi,hmlb

and bi,smbb represent stock i’s pre-futures systematic risk as captured by the market portfolio,

HML portfolio and SMB portfolio, respectively. The coefficients bi,mc, bi,hmlc and bi,smbc

measure the change post-futures listing, in the stock’s systematic risk as captured by HML, SMB

and market portfolio, respectively. The t-statistics associated with the coefficients bi,mc, bi,hmlc

and bi,smbc can be used to ascertain the statistical significance of these systematic risk changes.

The term 1i,t is the error associated with predicting the return of security i at time t.

If futures trading stabilises the market for the underlying stocks, the variance of the error term

1i,t is expected to decrease after futures are introduced. The paper replaces the estimates of ei,t inEqs. (2) and (3) by those obtained for 1i,t from Eq. (5), and re-estimates Eq. (4). The significance

of the coefficient uic can then be used to examine whether any change in the variance of the

10 A similar methodology is used by Mazouz (2004) in the context of options listing and stock volatility, and by

Antoniou and Holmes (1995) when ascertaining the impact of introducing futures trading on the FTSE 100 index.11 The precise economic interpretation of these factors is still the subject of much extensive debate in the literature. One

interpretation is that they (somehow) better capture the impact of accounting variables such as firm size (MV), book-to-

market equity (B/M), earnings to price (E/P) and cash flow to price (C/P). Detailed discussion of this model is beyond the

scope of this paper, but it can be found in many sources, such as Cochrane (2001).


prediction error is permanent. In an analogous fashion to the previous sub-section, the impact of

current and old news on the variance of returns can be measured by coefficients di,1 and ki,1 in

Eq. (3), respectively, re-estimated after replacing ei,t by 1i,t. The non-parametric Wilcoxon-

Signed Ranked test (WSRT) is again used to check whether the pre-listing series of di,1(ki,1) andthe post-listing series of di,1(ki,1) belong to the same distribution, again without imposing any

particular functional form.

3. Data

Trading in individual stock futures on the London International Financial Futures and Options

Exchange (LIFFE) commenced in January 2001. A complete list of the dates on which each

individual stock futures contract was introduced, between January 2001 and November 2002, is

obtained from the LIFFE website. The list includes 22 listed contracts relating to UK firms that

are traded on the London Stock Exchange (LSE). The final data sample consists of 21 stocks12

which possess a complete set of 400 daily price observations (excluding holidays) available from

Datastream on either side of the futures listing dates. The 400 trading days on either side of the

futures listing date constitutes the sampling interval used to measure the long-term market

stability effect associated with the introduction of futures trading. To capture the medium- and

short-term impact of futures innovation, both 200 and 100 trading day periods on either side of

the futures listing date are used.

As a proxy for the risk-free rate, we utilise daily prices for the 3-month UK Treasury bill,

while the FTSE All Index proxies for the market portfolio. To construct the Fama/French SMB

and HML portfolio factors, the size and market-to-book ratio of the FTSE All Index constituent

stocks are also downloaded from Datastream. The SMB and HML portfolios of the TFM are

constructed utilising the approach outlined in Mun, Vasconellos, and Kish (1999). SMB is

defined as the time series of differences in average returns between the 10% of firms in the FTSE

All Index with the highest, and the 10% with the lowest market capitalisation. HML represents

the time series differences in average returns between the 10% lowest market-to-book values (the

inverse of book-to-market ratios) and the 10% highest market-to-book ratios.13

It is important to note that any observed volatility change across pre- and post-futures listing

periods has the potential to be caused by factors other than futures innovation. The paper also

uses a carefully selected control sample to account for the possibility that the futures listing

decision is endogenous, and the fact that changes in market- and industry-wide conditions which

are contemporaneous with, but independent of, futures listing can impact upon on the measured

volatility of returns in the underlying stock market. The control sample consists of 21 non-

futures listed stocks. These stocks are selected by first matching each futures-listed stock with a

potential control stock from the same industry sub-sector, based on LSE classification. If several

stocks are candidates for the control stock we select that with the closest market capitalisation.

This paper maintains that our attempt to eliminate the influence of potential biases that may arise

as a result of sample selection is a strength of the analysis, and the control samples generated

allow us to interpret our results with some confidence. Table 1 provides a summary of some

relevant pre-futures listing risk characteristics of both the sample of optioned stocks and the

12 MMO2 is excluded from the analysis, as it does not have the minimum set of 100 daily price observations prior to the

introduction of the futures contract.13 The 10% largest (smallest) companies, used to construct the SMB portfolio, and the 10% highest (lowest) book-to-

market ratios, used to calculate the HML portfolio are revised on the 1st January of every year.

Table 1

Pre-listing risk characteristics of the sample of futures listed stocks and the control sample

Mean (median) pre-listing

Sample of futures listed stocks Control sample

100 days

around listing

200 days

around listing

400 days

around listing

100 days

around listing

200 days

around listing

400 days

around listing

Mean (median) of the residual variance from CAPMa 34.4 (30.1) 38.4 (33.2) 46.8 (43.4) 67 (36.2) 57 (39.7) 60 (44.3)

Mean (median) of the residual variance from TFMb 33 (29.1) 36.5 (31.9) 46.2 (41.7) 64.8 (32.5) 54.4 (38.9) 57.5 (41)

Mean (median) beta from CAPM 0.985 (0.951) 0.901 (0.872) 0.918 (0.820) 0.808 (0.666) 0.745 (0.636) 0.783 (0.634)

Mean (median) beta from TFM 1.013 (0.864) 0.958 (0.880) 0.918 (0.820) 0.849 (0.671) 0.571 (0.692) 0.838 (0.760)

Mean (median) SMB factor from TFM �0.116 (�0.052) �0.072 (0.058) 0.091 (0.084) 0.153 (0.111) 0.078 (0.058) 0.05 (0.055)

Mean (median) HML factor from TFM 0.212 (0.193) 0.261 (0.261) 0.278 (0.337) 0.107 (0.127) 0.141 (0.145) 0.132 (0.178)

p-value

OLS regressionsc Mann-Whitney testd

100 days

around listing

200 days

around listing

400 days

around listing

100 days

around listing

200 days

around listing

400 days

around listing

The equality of CAPM residual variance 0.328 0.332 0.394 0.345 0.404 0.771

The equality of TFM residual variance 0.334 0.392 0.448 0.308 0.557 0.716

The equality of CAPM beta 0.288 0.363 0.572 0.076 0.163 0.232

The equality of TFM beta 0.228 0.037* 0.450 0.061 0.190 0.258

The equality of SMB factor from TFM 0.043* 0.199 0.589 0.040* 0.320 0.421

The equality of HML factor from TFM 0.438 0.194 0.351 0.678 0.222 0.421

a Residual variance from both the CAPM and the TFM are multiplied by 105.b The TFM is constructed adopting the procedure in Mun et al. (1999). SMB is defined as the time series of differences in average returns between the 10% of firms with the

highest market capitalisation and the 10% with the lowest market capitalisation. HML represents the time series differences in the average returns between the 10% lowest market-

to-book values (the inverse of book-to-market ratios) and the 10% highest market-to-book ratios.c The p-values are based on the dummy coefficient of OLS estimation of the following equation: pre-listing risk measurei ,j ,t =Intercept+Coef.�Di ,j ,t +E; where the variable

pre-listing risk measurei ,j ,t is made up of 2n observations, where n is the number of futures listing stocks. At each listing date t, the pre-listing risk measurei ,j ,t takes on two values.

The first value is the pre-listing risk measure of the future listed stock i, and the second term is the pre-listing variance of non-future listed (control) stock j. Di ,j ,t takes on a value

of unity if the pre-listing risk measure observation is taken from the future listed stock i, and a value of zero if the pre-listing risk measure is taken from the non-future listed stock

j. Coef. indicates whether the pre-listing risk measures of the future listed stocks and those of non-futures listed stock are the same, and E is the error term.d Mann-Whitney is a non-parametric test, which tests whether the pre-listing risk measurei ,t of the future listing stocks and the pre-listing risk measurej ,t of the control stocks

belong to the same distribution.

* Indicates significance at the 5% level.

K.Mazouz,M.Bowe/Intern

atio

nalReview

ofFinancia

lAnalysis

15(2006)1–20

8


control sample, selected according to the above criteria. The tests we conduct give us confidence

that the listed stock sample and the control stocks have risk characteristics which are drawn from

the same distribution.

4. Empirical results

4.1. The systematic risk estimates

Panel A of Table 2 reports the results of the impact of futures trading on the beta coefficient of

the CAPM. The average (median) systematic risk component of the variance of returns is higher

after the introduction of individual equity futures. This increase appears to be related to the

length of study interval. The average (median) beta increases are 0.006 (0.058), 0.122 (0.154)

and 0.282 (0.827) in the 100-, 200- and 400-day intervals, respectively. However, the non-

parametric Wilcoxon Signed Rank test (WSRT) suggests that the systematic risk increase is

significantly different from zero only when measured over the 400-day period around futures

contract listing.

To obtain a more complete picture of changes in the beta coefficient, we disaggregate further

and observe the behaviour of the systematic risk of each individual stock across pre-and post-

Table 2

The effect of the futures trading on systematic risk: The single-factor CAPM approach (Eq. (1))

Estimation period in days

100 around

futures listing

200 around

futures listing

400 around

futures listing

Panel A: The sample of futures-listed stocks

Average (median)

bi ,b 0.985 (0.951) 0.901 (0.872) 0.912 (0.894)

bi ,c 0.006 (0.058) 0.122 (0.154) 0.282 (0.827)

Wicoxon Signed Rank Testa

Z-score ( p-value) �0.261 (0.794) �2.450 (0.117) �3.493 (0.000)**

Number of stocks with

(5%) Significant bi ,b 19 20 21

Positive bi ,c (sign) 11 16 18

(5%) Significantly positive (negative) bi ,c 02 (04) 06 (01) 18 (01)

Panel B: The control sample

Average (median) of

bi ,b 0.8077 (0.6658) 0.7449 (0.6363) 0.7827 (0.6338)

bi ,c 0.0025 (0.0942) 0.0642 (0.1603) 0.0741 (0.2099)

Mann-Whitney Testb

Z-score ( p-value) �0.818 (0.414) �0.138 (0.890) �2.151 (0.031)*


(5%) Significant bi ,b 19 17 21

Positive bi ,c (sign) 13 16 15

(5%) Significantly positive (negative)bi ,c 05 (03) 07 (04) 11 (03)

* and ** indicate significance at the 5% and 1% levels, respectively.a The non-parametric Wicoxon Signed Rank test (WSRT) is used to ascertain whether the pre-futures beta (bi ,b in Eq

(1)) series and the post-futures beta (the sum of bi ,b and bi ,c) series belong to the same distribution without assuming any

particular functional form for the distribution.b The non-parametric Mann-Whitney test is used to test the null hypothesis that the sample of the future-listed stocks

and the control sample exhibit the same beta change after futures contract listing.

.


futures trading periods. Panel A of Table 2 shows that the coefficient bi,c in Eq. (1) is positive in

52.4%, 76.2% and 85.7% of the cases in 100-, 200- and 400-day intervals, respectively. This

suggests that most listed stocks experience an increase in systematic risk subsequent to futures

listing. However, the increase in systematic risk is less pronounced over both the 100- and 200-

daily intervals when considering the lack of general statistical significance of the coefficient bi,c

in Eq. (1). In fact, stocks with insignificant systematic risk change dominate both the 100- and

200-day intervals (15/21 and 14/21 stocks, respectively). Stocks with significant increase in

systematic risk dominate only the 400-day interval (18/21 stocks), and over this interval the

effect is quite pronounced.

The finding that the introduction of futures trading induces a statistically significant beta

change only over the long-term 400-day interval raises the question as to whether such a change

may be induced by factors other than contract listing. One can argue that other causal factors,

such as alterations in market- and industry-wide conditions, are more likely to be manifest in

systematic risk changes in the longer-than in the shorter-term. Such changes should be manifest

in the control sample. Thus, to distinguish between the effect of futures trading and that of the

other factors, we compare the post-listing beta change in the sample of futures-listed stocks and

that observed in the control sample.

The results are reported in Panel B of Table 2. It is apparent that over the 100- and 200-day

intervals the mean (median) of the change in the beta coefficient of CAPM, measured by bi,c in

Eq. (1), in the control sample is essentially comparable to the beta change observed in the sample

of futures-listed stocks. Indeed, this casual inference is confirmed by the non-parametric Mann–

Whitney test (MWT) which indicates no significant differences in the beta change when

comparing the sample of future-listed stocks with the control sample. The picture alters in the

longer term. The MWT in the 400-day study interval suggests reports a Z-score of �2.151,

indicating that the observed long-term beta increase is indeed significantly higher for the sample

of future-listed stocks than for the control sample. This result is consistent with the interpretation

that the long-term beta increase observed in the sample of futures-listed stocks may at least be

partly explained by the introduction of individual equity futures trading. This view is also

confirmed by the fact that the sample of futures listed stocks has more stocks with significant

beta increase than the control sample.

To this point the analysis provides some qualified support for the perspective that futures

trading increases the systematic risk of the underlying stocks, particularly in the long run.

However, this finding could simply be the result of a misspecification bias in relation to the

systematic risk estimator. As such, it is important to verify the robustness of the estimates using

alternative model specifications. To gain further insight to the issue, we choose to employ the

familiar Fama/French TFM. Selected goodness-of-fit tests indicate that the (GARCH

augmented) TFM produces an improved fit to the data set relative to the (GARCH augmented)

CAPM. For example, we employ the BDS test14 to compare the degree of non-linearity in the

standardised residual series generated from the CAPM and TFM specifications. The BDS test is

designed to determine the extent of the presence of the non-linearity beyond that captured by a

given model. For model comparison purposes, we use the absolute mean of the BDS test statistic

over m-dimensions. In general, we observe non-linearity in most of the residual series generated

14 The BDS statistic (for details see Brock, Dechert, & Scheinkman, 1987; Brock, Hsieh, & LeBaron, 1991) has been

demonstrated to have high power against specific non-linear alternatives for model comparison purposes. Following

Joseph (2003), we undertake our analysis with m and l set to m =2, 3, 4. . .10 and l =0.25, 0.50 and 0.75. Joseph (2003, p.798) also suggests that bthe BDS statistic has a good power against several types of deviations from the iid assumptionQ.

Table 3

The goodness of fit statistics for the three factor model (TFM) and GARCH(1,1)

The sample of future listed stocks and the control samplea

100-day window 200-day window 400-day window

Number (%) of significant LM(1) from CAPMb 16 (40%) 27 (67.5%) 33 (82.5%)

Number (%) of significant LM(1) from TFMc 13 (32.5%) 26 (65%) 30 (75%)

Number (%) of significant ARCH(1) from CAPMd 8 (20%) 16 (40%) 18 (45%)

Number (%) of significant GARCH(1) from CAPMe 25 (62.5%) 31 (77.5%) 40 (100%)

Number (%) of significant ARCH (1) from TFMf 10 (25%) 13 (32.5%) 34 (85%)

Number (%) of significant GARCH(1) from TFMg 27 (67.5%) 29 (72.5%) 35 (87.5%)

a There are 40 stocks in total. Eq. (5) always has a larger log-likelihood and a smaller sum of squared residuals than Eq.

(1). This suggests that the TFM is an improved specification of the return model than the CAPM.b The LM test is based on the residual of Eq. (1).c The LM test is based on the residual of Eq. (5).d The ARCH(1) is represented by the coefficient d i ,1 in Eq. (3) when Eq. (3) is used to model the variance of the

residual from Eq. (1).e The GARCH(1) is represented by the coefficient k i ,1 in Eq. (3) when Eq. (3) is used to model the variance of the

residual from Eq. (1).f The ARCH(1) is captured by the coefficient d i ,1 in Eq. (3) when Eq. (3) is used to model the variance of the residual

from Eq. (5).g The GARCH(1) is represented by the coefficient k i ,1 in Eq. (3) when Eq. (3) is used to model the variance of the

residual from Eq. (5).


from both CAPM and TFM.15 However, the magnitude of the BDS statistic tends to be

consistently smaller under the TFM than under the CAPM, suggesting the TFM consistently

outperforms the CAPM specification. Moreover, the TFM model specifications have a greater

log-likelihood and a smaller sum of squared residuals than those found under the CAPM. We

believe this justifies use of the TFM as an alternative systematic risk estimator for the data in the

present sample. The statistics from some other common diagnostic tests, including LM test, of

the model specification are reproduced in Table 3.16

The summary results illustrating the impact of futures contract innovation on each of the

Fama and French systematic risk factors are presented in Table 4. The table indicates that the

average (median) pre-listing systematic risk factor relating to the market portfolio (bi,mb in Eq.

(5)) is always the largest in magnitude of the three risk factors. This result holds both across the

sample of futures-listed stocks and the control sample, and is a consistent finding across all

of the three time intervals studied. The risk factor relating to SMB (bi,smbb in Eq. (5)) is

found to be the smallest of the three risk explanatory factors, which is expected given that

futures contracts are listed only on stocks with large market capitalisation. Panel A of Table

4 summarises the results of the impact of futures trading upon the three systematic risk

factors. Over the 100- and 200-day intervals, the non-parametric WSRT indicates that none

of the three risk factors exhibits statistically significant changes after futures are introduced.

Once again, it is in the 400-day event window where significant results are found. However,

they suggest a somewhat different interpretation to that provided by the CAPM results in

16 Note that the LM test suggests that the ARCH effect disappears when the residual series are modelled using GJR-

GARCH(1,1).

15 Note that the presence of the non-linearity in the residual series also justifies and supports the papers use of GARCH-

type models.

Table 4

The impact of futures trading on systematic risk factors: The TFM approach (Eq. (5))


100 around

futures listing

200 around

futures listing

400 around

futures listing


Average (median) of pre-futures risk factor

bi ,mb 1.013 (0.864) 0.958 (0.880) 0.918 (0.820)

bi ,smbb �0.116 (�0.052) �0.072 (0.058) 0.091 (0.084)

bI ,hmlb 0.212 (0.193) 0.261 (0.267) 0.278 (0.337)

Average (median) change in risk factor

bi ,mc �0.003 (�0.065) 0.086 (0.099) 0.153 (0.208)

bi ,smbc 0.113 (0.227) �0.084 (�0.039) �0.223 (�0.213)

bi ,hmlc �0.090 (0.020) �0.177 (�0.134) �0.213 (�0.342)

Z-score ( p-value) from WSRT for the risk factor changea

bi ,mc �0.156 (0.876) �1.790 (0.073) �2.589 (0.01)**

bi ,smbc �1.338 (0.181) �0.852 (0.394) �3.597 (0.000)**

bi ,hmlc �0.365 (0.715) �1.651 (0.099) �2.172 (0.030)*

Number of significant pre-listing risk factors

bi ,mb 19 20 21

bi ,smbb 03 04 05

bi ,hmlb 04 07 16

Number of positive (negative) risk factor change-sign

bi ,mc 12 (09) 13 (08) 16 (05)

bi ,smbc 14 (07) 10 (11) 02 (19)

bi ,hmlc 12 (09) 10 (11) 07 (14)

Number of significantly positive (negative)

bi ,mc 02 (04) 05 (01) 10 (02)

bi ,smbc 01 (00) 01 (02) 00 (04)

bi ,hmlc 00 (01) 01 (03) 01 (07)


Average (median) of pre-futures risk factor

bi ,mb 0.849 (0.671) 0.571 (0.692) 0.838 (0.760)

bi ,smbb 0.153 (0.111) 0.078 (0.058) 0.050 (0.055)

bi ,hmlb 0.107 (0.127) 0.141 (0.145) 0.132 (0.178)

Average (median) change in risk factor

bi ,mc �0.040 (0.115) 0.153 (0.181) 0.042 (0.145)

bi ,smbc �0.120 (�0.230) �0.225 (�0.149) �0.058 (�0.144)

bi ,hmlc �0.088 (�0.051) �0.038 (�0.028) �0.142 (�0.072)

Z-score ( p-value) from Mann-Whitney testb

bi ,mc �0.918 (0.358) �0.868 (0.385) �1.434 (0.152)

bi ,smbc �1.925 (0.054) �0.893 (0.372) �1.811 (0.070)

bi ,hmlc �0.063 (0.948) �0.818 (0.414) �2.402 (0.016)*

Number of significant pre-listing risk factors

bi ,mb 21 21 21

bi ,smbb 04 03 04

bi ,hmlb 04 09 11

Number of positive (negative) risk factor change-sign

bi ,mc 13 (08) 16 (05) 16 (05)

bi ,smbc 07 (14) 04 (17) 12 (09)

bi ,hmlc 10 (11) 08 (13) 14 (07)



100 around

futures listing

200 around

futures listing

400 around

futures listing

Number of significantly positive (negative)

bi ,mc 02 (03) 05 (01) 07 (02)

bi ,smbc 00 (02) 01 (02) 00 (02)

bi ,hmlc 00 (00) 02 (00) 03 (07)

* and ** represent significance at 5% and 1% levels, respectively.a The non-parametric Wicoxon Signed Rank test (WSRT) is used to ascertain whether there is a change in any of the

TFM risk factors in Eq. (5) between pre- and post-futures listing periods.b The non-parametric Mann-Whitney test is used to determine whether any change in the three TFM systematic risk

factors in Eq. (5) is the same for both the sample of the future-listed stocks and the control sample across pre-and post-

futures listing periods. The null hypothesis in each case is that there is no change in the relevant systematic risk factor.

Table 4 (continued)


Table 2. In the TFM equation estimates all the three risk factors, relating to the market,

SMB and HML portfolios, experience a statistically significant decrease after the initiation of

futures trading.

Further insights are gained by examining panel B of Table 4 which reports the results of

utilising the MWT to test the null hypotheses that there is no change in any of the three

systematic risk factors across the sample of futures listed stocks and the control sample, in any of

the three selected time intervals. No significant difference is detected in any of the risk factors in

either the 100- or 200-day intervals around the futures introduction date. Notwithstanding the

WSRT results in panel A, the same finding holds over the 400-day interval for changes in the

systematic risk relating to both the market and SMB portfolios. However, the systematic risk

related to the HML portfolio decreases by an average of 0.213 (0.142) for the sample of futures-

listed stocks (control stocks) after futures listings, and this difference is significant at the 5%

level, again supporting the inference that futures innovation is associated with a decrease in

systematic risk.

This latter finding is in apparent contradiction to the previously reported results obtained

from the CAPM analysis. Not only does this confirm the need to perform robustness

checks, but it also provides a further indication of the problems associate with inferring

results using averages in this type of analysis.17 Given these qualifications, we believe that

observing the behaviour of the systematic risk factors associated with each individual stock

across the pre-and post-listing periods, may provide a more accurate representation of the

overall behaviour of the sample. Once again, Table 4 indicates that in all three study windows,

the majority of stocks do not exhibit statistically significant changes in any of the Fama/

French systematic risk factors after the introduction of futures trading. Table 4 also suggest

that for every systematic risk factor and event window, the number of stocks with statistically

significant increases (decreases) is broadly comparable, both in the sample of listed stocks and

the control sample. In contrast to the results obtained using the MWT, the sample of futures-

listed stocks and the control sample contain the same number of stocks (seven) with

statistically significant decreases in the HML portfolio systematic risk factors. Thus, after

examining individual stock behaviour, the results appear to indicate that overall, the

introduction of futures contracts has no impact on the systematic risk of the underlying

17 See, for example, Bowe and Mazouz (2004).


stocks. In the next sub-section we turn our attention to analysing its impact upon diversifiable

risk.

4.2. Diversifiable risk: the variance of the prediction errors

Table 5 summarises the results of the measured impact of futures trading on the variance of

the prediction errors of the underlying stocks. These findings are based on estimates generated

by both model’s residual series, namely ei,t of the CAPM model in Eq. (1) and 1i,t of the TFM in

Eq. (5)). The specific impact is captured by the dummy coefficient included in the GJR-

GARCH(1,1) process, ui,c in Eq. (3).18 Panel A of Table 5 indicates that regardless of whether

the residual series are generated by Eq. (1) or Eq. (5) the average (median) of the variance of the

prediction errors of the sample of futures-listed stocks decreases after futures trading. This result

holds consistently across all of the three study intervals. The WSRT indicates that the statistical

significance of this decrease in the variance of the prediction error depends both on the study

interval and whether ei,t or 1i,t is used to estimate Eq. (3). Specifically, when Eq. (3) is estimated

from the residual series ei,t (1i,t) the decrease in the variance of the prediction errors is

significantly different from zero only across the 100- and 400-day (200- and 400-day) intervals.

The observed decrease in the variance of the prediction errors across pre- and post-futures

trading periods can be attributed to several potential sources. It may be attributed to: (i) changes

in the market- and/or industry-wide conditions; (ii) measurement bias; or (iii) the existence of

futures contracts reducing trading bnoiseQ in the underlying stock markets, thereby improving the

ability of information concerning fundamental values to be incorporated efficiently into stock

prices. In order to distinguish between these three alternative explanations, we begin by

measuring the change in the variance of the prediction errors of the control sample stocks across

pre- and post-futures listing periods.

Panel B of Table 5 indicates that in contrast to the sample of futures-listed stocks in panel A,

the average of the change in the variance of the prediction errors is negative in the 100-day

interval and positive in both the 200- and 400-day intervals. This holds whether the prediction

error measurement is based on ei,t or 1i,t. However, again based on both ei,t or 1i,t, the median

change in the variance of the prediction errors in the control sample is negative in all three of the

study windows. This is in accordance with the results from the futures-listed stocks.

Furthermore, the non-parametric MWT indicates that the sample of futures-listed stocks and

the control sample exhibit significantly different measured changes in the variance of the

prediction error in the 200-day interval, and even here only when ei,t is used as the prediction

error measure.

The similarity in the behaviour of the variance of the prediction error in the futures-listed and

control stocks is even more pronounced upon examining the statistical significance of the change

in the prediction error variance of each individual stock. Even using Eq. (1) to predict returns in

the 200-day interval, the coefficient ui,c in Eq. (3) suggests that only 2/21 (3/21) futures-listed

(control) stocks exhibit a permanent variance change after introducing futures contracts. We

conclude that the above cross-sample difference in the variance of ei,t in the 200-day interval

suggested by the MWT is more likely to be due to measurement bias rather than the effect of

futures trading per se.

18 We also ran tests after adding a dummy variable, with value of zero before and a value of unity subsequent to futures

introduction, to an EGARCH(1,1) specification. The conclusions obtained remain qualitatively very similar. The specific

details are available upon request.

Table 5

Futures trading and the variance of the prediction errors: GJR-GARCH(1,1) approach (Eq. (3))a


100 around futures listing 200 around futures listing 400 around futures listing

CAPM TFM CAPM TFM CAPM TFM

Panel A: Sample of futures-listed stocks

Average (median)

u i ,b 1.10e�04 (9.73e�05) 1.03e�04 (5.94e�05) 1.14e�04 (8.34e�05) 1.24e�0.4 (9.08e�05) 4.86e�05 (4.65e�0.5) 5.60e�05 (4.69e�05)

u i ,c �2.23e�05(�1.45e�05) �1.20e�05 (�1.00e�05) �2.60e�05 (�1.30e�05) �2.54e�05(�1.39e�05) �1.21e�05 (�1.27e�05) �1.52e�05 (�1.25e�05)

Wicoxon Signed Rank Testb

Z-score ( p-value) �2.381 (0.017)* �1.790 (0.073) �1.164 (0.244) �2.450 (0.014)* �2.798 (0.005)** �2.816 (0.005)**


Positive (negative) u i ,c sign 06 (15) 07 (14) 03 (18) 05 (16) 03 (18) 04 (17)

Significantly positive

(negative) u i ,c

00 (03) 01 (03) 00 (02) 01 (01) 01 (08) 01 (09)

Panel B: Sample of control stocks

Average (median)

u i ,b 1.43e�04 (4.21e�05) 1.42e�04 (5.16e�05) 1.39 e�04 (6.38e�05) 1.42e�04 (9.36e�05) 1.49e�04 (3.53e�05) 1.39e�04 (2.48e�05)

u i ,c �1.60e�05 (�9.40e�06) �2.20e�05 (�1.70e�05) 1.40e�05 (�4.5e�06) 2.36e�06 (�6.4e�06) 1.33e�05 (�3.77e�06) 1.05e�05 (�2.76e�06)

Mann-Whitney Testc

Z-score ( p-value) �0.692 (0.489) �0.616 (0.538) �2.063 (0.039)* �1.044 (0.269) �1.120 (0.263) �0.818 (0.414)


Positive (negative) u i ,c sign 06 (15) 03 (18) 06 (15) 08 (13) 06 (15) 05 (16)

Significantly positive

(negative) u i ,c

01(02) 00 (00) 02 (01) 02 (04) 02 (07) 02 (06)

* and ** represent significance at the 5% and 1% levels, respectively.a Note that we repeat all analyses using an EGARCH(1,1) specification instead of GJR-GARCH(1,1) and the conclusion remains the same. Details of the EGARCH(1,1) tests are available upon request.b The non-parametric Wicoxon Signed Rank test (WSRT) is used to determine whether there is a change in the variance of the prediction error between pre-and post-futures periods.c The non-parametric Mann-Whitney test is used to ascertain whether the change in the variance of the prediction error is the same for both the sample of the future-listed stocks and the control sample

when measured across pre- and post-futures listing periods.


atio

nalReview

ofFinancia

lAnalysis

15(2006)1–20

15

Table 6

Futures trading and the price adjustment to information (Eq. (4))


100 around futures listing 200 around futures listing 400 around futures listing

CAPM TFM CAPM TFM CAPM TFM


Average (median)

Pre-listing ARCH factor 0.055 (0.024) �0.062 (�0.068) 0.040 (0.019) 0.106 (0.066) 0.084 (0.065) 0.094 (0.081)

Pre-listing GARCH factor 0.682 (0.716) 0.778 (0.989) 0.577 (0.782) 0.716 (0.852) 0.821 (0.865) 0.803 (0.862)

Change in the ARCH factor �0.087 (�0.086) 0.152 (0.112) 0.106 (0.061) 0.048 (0.018) 0.004 (0.021) 0.023 (�0.003)

Change in the GARCH factor 0.078 (0.168) �0.208 (�0.225) �0.024 (�0.155) �0.213 (�0.323) �0.132 (�0.060) �0.130 (�0.059)

Wilcoxon Signed Rank Test (WSRT)a

Z-score ( p-value) for ARCH change �1.929 (0.054) �2.487 (0.013)* �1.234 (0.217) �0.400 (0.689) �0.089 (0.931) �0.156 (0.876)

Z-score ( p-value) for GARCH change �1.130 (0.259) �1.894 (0.058) �0.921 (0.357) �2.207 (0.027)* �1.894 (0.058) �2.138 (0.033)*

Number of stocks

Increase (decrease) in ARCH factor-sign 08 (13) 16 (05) 15 (06) 12 (09) 11 (10) 10 (11)

Increase (decrease) in GARCH factor-sign 13 (08) 06 (15) 10 (11) 06 (15) 05 (16) 08 (13)

Moved to significant ARCH (GARCH) 02 (03) 02 (01) 03 (02) 02 (04) 04 (00) 06 (00)

Moved to insignificant ARCH (GARCH) 02 (07) 04 (07) 03 (07) 04 (10) 07 (02) 04 (02)


Average (median)

Pre-listing ARCH factor 0.166 (0.076) 0.213 (0.053) 0.248 (0.071) 0.326 (0.092) 0.219 (0.121) 0.233 (0.078)

Pre-listing GARCH factor 0.476 (0.725) 0.538 (0.779) 0.585 (0.783) 0.571 (0.657) 0.541 (0.546) 0.691 (0.790)

Change in the ARCH factor �0.079 (�0.044) �0.077 (�0.079) �0.141 (0.007) �0.185 (0.005) �0.139 (�0.018) �0.157 (�0.229)

Change in the GARCH factor 0.634 (0.704) �0.022 (0.147) 0.043 (0.058) 0.013 (�0.070) 0.137 (0.043) 0.077 (0.049)

Mann-Whitney Testb

Z-score ( p-value) for ARCH change equality �1.497 (0.134) �0.824 (0.068) �1.094 (0.274) �0.566 (0571) �1.623 (0.105) �1.346 (0.178)

Z-score ( p-value) for GARCH change equality �1.623 (0.105) �1.421 (0.155) �0.893 (0.372) �1.824 (0.068) �2.428 (0.015)* �1.673 (0.094)

Number of stocks

Increase (decrease) in ARCH factor-sign 10 (11) 10 (11) 11 (10) 11 (10) 09 (12) 10 (11)

Increase (decrease) in GARCH factor-sign 13 (08) 12 (09) 11 (10) 10 (11) 14 (07) 11 (10)

Moved to significant ARCH (GARCH) 00 (04) 02 (05) 01 (04) 01 (02) 02 (03) 04 (03)

Moved to insignificant ARCH (GARCH) 05 (04) 02 (05) 05 (03) 07 (05) 09 (01) 09 (00)


atio

nalReview

ofFinancia

lAnalysis

15(2006)1–20

16


4.3. Futures innovation and the price adjustment to information

In this section we examine the ability of futures trading innovation to enhance market

efficiency by focussing upon its impact on the information adjustment coefficient of the GJR-

GARCH (1,1) specification given by Eq. (4). Panel A of Table 6 reports a summary of the results

obtained from estimating Eq. (4) separately for both pre- and post-futures trading periods for the

sample of futures-listed stocks. Both the CAPM and the TFM are used to generate the residual

series over the customary 100-, 200- and 400-day study intervals. The results indicate that in

both approaches over all three intervals, the GARCH-term (specifically the coefficient ki,1 in Eq.

(4)) in the GJR-GARCH(1,1) is the main determinant of the variance of the prediction error, with

the ARCH-term (the coefficient di,1 in Eq. (4)) having a much smaller effect.

However, once again model differences are apparent. In the CAPM-GJR-GARCH(1,1)

specification the ARCH-terms (GARCH-terms) exhibit an average post-listing increase

(decrease) of 265% (4.16%) measured over the 200-day interval and 4.76% (16.07%) over

the 400-day interval. In contrast, over the 100-day interval, we observe an average decrease

(increase) of 158.2% (11.43%) in the ARCH-terms (GARCH-terms) subsequent to futures

listing. Interpreting these findings purely in terms of the signed coefficient change of the

impact of current and old news on the variance of returns, one could argue that futures

contract innovation enhances market efficiency in both the medium- and long-term, while

having the reverse effect in the short-term. However, the non-parametric WSRT suggests that

in this CAPM specification all the pre- and post-futures trading first order ARCH (GARCH)

coefficients are insignificantly different from one another. As such, futures contract listing

appears to have a negligible impact on the speed at which information is incorporated into the

stock price.

The results from the TFM-GJR-GARCH(1,1) approach differ slightly. Table 6 indicates that

the news impact is not sensitive to the study interval selected, with current (old) news always

exhibiting a higher (lower) impact on the return variance following the initiation of futures

trading. The WSRT also indicates that the introduction of futures trading improves efficiency in

the underlying market measured over all three-study intervals. Although the average decrease in

the coefficient ki,1 is statistically significant only in the 100-day study interval, the Z-score from

the non-parametric WSRT suggests that, regardless of the length of the study interval, old news

(ki,1 in Eq. (4)) has a significantly lower impact on the return variance subsequent to futures

listing. We interpret this reduction in the first order autocorrelation exhibited by the variance of

prediction errors as indicating that futures trading innovation may contribute to improving the

efficiency of the underlying stock market.

However, before finally reaching such a conclusion, one must allow for the possibility that

the observed change in the ARCH (GARCH) elements of the TFM-GJR-GARCH(1,1) may be

due to contemporaneous changes in market- or industry-wide conditions, or simply be the result

of measurement bias. To account for the impact of market-and/or industry-wide conditions, we

Notes to Table 6:

* and ** represent significance at the 5% and 1% levels, respectively.a The non-parametric Wicoxon Signed Rank test (WSRT) is used to check whether there is a change in the first orde

ARCH (GARCH) factor between pre- and post-futures contract listing periods.b The non-parametric Mann-Whitney test is used to ascertain whether the change in the first order ARCH (GARCH) is

the same for both the sample of the future-listed stocks and the control sample across pre-and post-futures contract listing

periods.

r


again focus on the observed changes in the ARCH (GARCH) terms when the TFM-GJR-

GARCH(1,1) is applied to the sample of control stocks. The results are summarised in Panel B

of Table 6. With the exception of certain results in the 100-day interval, the qualitative results are

in marked contrast to the sample of futures-listed stocks. On average the control sample stocks

now exhibit a measured decrease (increase) in the impact of the current (old) news on the

variance of prediction errors. However, with one exception (the CAPM over 440 days) the non-

parametric MWT implies that in all of the three study windows, the sample of futures-listed

stocks and the control sample cannot be differentiated on the basis of changes in the current (old)

news components of the prediction error variance subsequent to the introduction of futures

trading.

The apparent similarity between the sample of futures-listed stocks and the control sample in

relation to the change in the news components of the variance of prediction errors is less

pronounced when the stocks are examined individually. Table 6 indicates that, regardless of the

return generating process or the length of the study window, the number of stocks for which the

ARCH-terms become significant (insignificant) after the introduction of futures trading is higher

(lower) for the sample of futures-listed stocks than for the control sample. Moreover, the number

of stocks for which the GARCH-term becomes significant (insignificant) following the initiation

of futures trading is lower (higher) for the sample of futures-listed stocks than for the control

sample. It would appear that relatively speaking, subsequent to futures listing, current (old) news

has more (less) impact on the variance of stocks in the sample of futures-listed stocks, whereas

the reverse holds for the control sample. Viewed collectively, these findings suggest that futures

trading improves the pricing efficiency in the underlying stock market by increasing the speed at

which information becomes incorporated into the stock price.

5. Summary and conclusion

The central contribution of this paper is the use of the Fama and French three factor model

(TFM) jointly with GJR-GARCH(1,1) to investigate the impact of individual equity futures

contract listings on both the systematic and diversifiable risk components of the variance of the

underlying stock. Several studies, including Fama and French (1992, 1993, 1996) and Liew and

Vassalou (2000), argue that the TFM is a better predictor of stock returns than the conventional

single factor capital asset pricing model (CAPM). Thus, ignoring the systematic risk factors that

are captured by the HML and SMB portfolios may bias the results of studies which examine the

impact of futures trading on the risk components of the underlying stocks. For example, a study

may attribute a measured change in the variance of the prediction errors generated from the

CAPM to the introduction of futures trading. However, this change may not be caused by futures

innovation but simply be a manifestation of changes in either, or both, of the systematic risk

factors reflected in the HML and SMB portfolios of the TFM.

With data taken from LIFFE listings of LSE traded stocks, our empirical findings imply that

the observed impact of futures contract innovation on the systematic risk of the underlying

stocks is indeed sensitive to whether the CAPM or the TFM is used to model the underlying

stock returns. Under the CAPM (TFM) specification, the sample of futures-listed stocks exhibits

an increase (decrease) in their systematic risk components after the introduction of individual

equity futures contracts. However, regardless of whether the CAPM or the TFM is used to model

stock returns, a GJR-GARCH(1,1) estimation indicates that the sample of futures-listed stocks

exhibits a decrease in the permanent component of the variance of their prediction errors after a

futures contract listing. Furthermore, the evidence also suggests that subsequent to having a


futures contract listed on LIFFE, current news is more efficiently incorporated into the

underlying stock price on the LSE.

Employing a control sample methodology to isolate the impact of futures trading from that of

other factors, the paper also investigates if this observed decrease in the listing firms’ cost of

capital, together with the increase in the speed at which information is incorporated into the

stock price post-listing, could potentially be attributed to factors extraneous to the listing event.

Analysis of the results suggests that the change in both the systematic risk and the permanent

component of the conditional variance observed from the sample of futures-listed stocks can be

completely explained by contemporaneous changes in market and industry-wide conditions, and

should not be attributed to the listing event, per se. However, the observed behaviour of the

conditional variance of the control sample is consistent with the interpretation that futures

contract innovation indeed improves the underlying market’s efficiency by increasing the speed

at which information is incorporated into the price of futures-listed stocks.

Acknowledgments

The comments provided by colleagues at Aston University, the University of Portsmouth and

the University of Manchester, in particular Patricia Chelley-Steeley and Nathan Joseph on earlier

versions of the paper, are gratefully appreciated. The paper has also benefited significantly from

the comments of an anonymous referee and the editor, Tom Fetherston.

References

Amihud, Y., & Mendelson, H. (1987). Trading mechanisms and stock returns: An empirical investigation. Journal of

Finance, 42, 533–553.

Antoniou, A., & Holmes, P. (1995). Futures trading, information and spot price volatility: Evidence from the FTSE100

Stock Index Futures contract using GARCH. Journal of Banking and Finance, 19, 117–129.

Black, F. (1986). Noise. Journal of Finance, 41, 529–530.

Bowe, M., Mazouz, K. (2004). Does options listing have any impact on the time varying volatility of the underlying

stocks? Evidence from NYSE stocks listed on the CBOE, European Financial Management Association 2004 Basel

meetings, available www.ssrn.com

Brock, W., Dechert, W., & Scheinkman, J. (1987). A test for independence based on the correlation dimension,

University of Wisconsin, Madison, University of Houston and University of Chicago.

Brock, W., Hsieh, D., & LeBaron, B. (1991). Nonlinear dynamics, chaos, and instability: Statistical theory and economic

evidence. Cambridge, MA7 MIT press.

Chatrath, A., Ramchander, S., & Song, F. (1998). Speculative activity and stock market volatility. Journal of Economics

and Business, 50, 323–337.

Chaudhury, M., & Elfakhani, S. (1995). The volatility effect of options listing: Some Canadian evidence. Quarterly

Review of Economics and Finance, 35, 97–115.

Cochrane, J. (2001). Asset pricing. Princeton, NJ7 Princeton University Press.

Cox, C. (1976). Futures trading and market information. Journal of Political Economy, 48, 1215–1237.

Damodaran, A., & Lim, J. (1991). The effects of options listing on the underlying stocks’ return processes. Journal of

Banking and Finance, 15, 647–664.

Demsetz, H. (1968). The cost of transacting. Quarterly Journal of Economics, 82, 33–55.

Dennis, S. A., & Sim, A. B. (1999). Share price volatility with the introduction of individual share futures on the Sydney

futures exchange. International Review of Financial Analysis, 8, 153–163.

Draper, P., Yadav, P. K., & Watt, W. H. (1992). The impact of option listing on underlying stock returns: The UK

evidence. Journal of Business Finance & Accounting, 19, 485–503.

Easley, D., & O’Hara, M. (1987). Price, trade size, and information in securities markets. Journal of Financial

Economics, 19, 69–90.

Edwards, F. R. (1988). Futures trading and cash volatility: Stock index and interest rate futures. Journal of Futures

Markets, 8, 421–439.

http://www.ssrn.com


Engle, R. F., & Ng, V. K. (1993). Measuring and testing the impact of news on volatility. Journal of Finance, 48,

1022–1082.

Fama, E. R., & French, K. R. (1992). The cross-section of expected stock returns. Journal of Finance, 47, 427–465.

Fama, E. R., & French, K. R. (1993). Common risk factors in the returns of stocks and bonds. Journal of Financial

Economics, 33, 3–56.

Fama, E. R., & French, K. R. (1996). Multifactor explanations of asset pricing anomalies. Journal of Finance, 51,

55–84.

Figlewski, S. (1981). Futures trading and volatility in the GNMA futures. Journal of Finance, 36, 445–456.

Froewiss, K. (1978). GNMA futures: Stabilising or destabilising? Economic Review (pp. 20–29). Federal Reserve Bank

of San Francisco.

Glosten, L. R., Jagannathan, R., & Runkle, D. (1993). On the relation between the expected value and the volatility of the

normal excess return on stocks. Journal of Finance, 48, 1779–1801.

Glosten, L. R., & Milgrom, P. (1985). Bid, ask, and transaction prices in specialist market with heterogeneously informed

agents. Journal of Financial Economics, 14, 71–100.

Gorton, G., & Pennacchi, G. (1993). Security baskets and index-linked securities. Journal of Business, 66, 1–28.

Harris, L. (1989). S&P 500 cash stock price volatilities. Journal of Finance, 44, 1155–1175.

Joseph, N. L. (2003). Using monthly returns to model conditional heteroscedasticity. Applied Economics, 35, 791–801.

Lee, I. C., & Tong, H. C. (1998). Stock futures: The effect of their trading on the underlying stocks in Australia. Journal

of Multinational Financial Management, 8, 285–301.

Liew, J., & Vassalou, M. (2000). Can book-to-market, size and momentum be risk factors that predict economic growth?

Journal of Financial Economics, 57, 221–245.

Ma, C., & Rao, R. (1988). Information asymmetry and option trading. The Finance Review, 23, 39–51.

Madhavan, A. (2000). Market microstructure: A survey. Journal of Financial Markets, 3, 205–258.

Mazouz, K. (2004). The effect of CBOE option listing on the volatility of NYSE traded stocks: A time-varying variance

approach. Journal of Empirical Finance, 11, 695–708.

Mun, J. C., Vasconellos, G. M., & Kish, R. (1999). Tests of the Contrarian Investment Strategy from the French and

German stock markets. International Review of Financial Analysis, 8, 215–234.

Sahlstrom, P. (2001). Impact of stock option listings on return and risk characteristics in Finland. International Review of

Financial Analysis, 10, 19–36.

Simpson, W., & Ireland, T. (1985). The impact of futures trading on the cash market for treasury bills. Journal of

Financial and Quantitative Analysis, 20, 371–379.

Skinner, D. J. (1989). Options markets and stock return volatility. Journal of Financial Economics, 23, 61–78.

Stein, J. (1987). Information externalities and welfare reducing speculation. Journal of Political Economy, 95,

1123–1145.

Whiteside, M. M., Dukes, W. P., & Dunne, P. M. (1983). Short term impact of futures trading on underlying securities.

Journal of Financial Research, 6, 313–321.

the volatility effect of futures trading: evidence from lse traded stocks listed as individual...

Documents