Volatility onthe J ohannesb urgStockE xchange and the Introd uc tionofFuturesContrac ts O w enB eeld ers¤ Departm ent ofE c onom ic s E m ory U niversity Atlanta,G a,30 32 9-34 89 e-m ail: ob eeld e@em ory.ed u J uly 31,2 0 0 0 Ab strac t W e analyz e the e®ec t ofthe introd uc tionofa f uturesm arket onthe spot volatility ofthe J ohannesb urg Stoc k E xc hange usingthe m ethod of B essimb ind er and Seguin(19 9 2 ).W e ¯nd that spot volatility ispositively related to the sec ular leveloftrad ing inthe spot and f utures m arket. Sec ond , using openinterest asa m easure ofm arket d epth w e ¯nd that it is negatively related to spot volatility, i.e. f utures trad ing d oes not increase spot volatility. F inally, the unexpec ted numb er oftransac tions inthe f uturesm arket isstrongly related to spot volatility - thisresult is c onsistent w ith J ones,K auland Lipson(19 9 4 ). ¤Iw oul d like to thank M ariz a G oosenofthe South Af ric anFuturesE xc hange, M ic helle Francisofthe J ohannesb urg Stock E xchange and StephenE vansofM c G regor B FA f or their generousand invaluab le assistance. Iw ould also like to thank my tw o c apab le researc h assistantsonthe projec t, J ing Xu and Ning Liu f or their hel p inpreparing the d ata ¯les.All rem ainingerrorsare m ine. 1 1 In trod uc tion T heintroductionofafutures marketis supposedtohavemanybene¯ts suchas risk-sharing(hedging),anincreaseinmarketdepthand adecreasein volatility due to the more rapid rate atwhich information is re° ected in prices. H owever,opponents offutures markets argue thatany increase in marketdepth is duetospeculation and speculation leads togreaterspotmarketvolatilitythat negates thebene¯ts.T heobjectiveofthis paperis to° eshouttherelationship betweenthespotvolatilityoftheJohannesburgStockExchange(JSE)andthe introductionofnewstockindexfutures contracts ontheSouthA fricanFutures Exchange (SA FEX )in 1 995.In particular,we investigate the relationship between futures volumeandspotpricevolatilityaftercontrollingforspotmarket volume. T herearemanytheories thatrelatevolatilityorpricechanges tothelevelof tradeandmanytheories purportingtoshowthecosts andbene¯ts oftheintroduction offutures markets.Karpo®(1 987)provides acomprehensivereviewof thetheories thatpredictapositivecontemporaneous relationship between volumeandpricechanges andtheempiricalstudies thatsupportthetheory.W ith regard totheempiricalanalysis oftheintroduction offutures markets,theresults are less supportive.B essembinderand Seguin (1 992)(hereafterB &S92) argue that many of the empirical studies are ° awed because they are event studies with asamplesizeofone.Forexample,manystudies thatcomparethe ex ante and ex post levels ofvolatility are unable to partialoutthe e®ectof a secularchange in volatility in the latterperiod;this may lead to erroneous conclusions.T oremedythesituation B &S92 proposeanewmethodology. T headvantageofusingtheB &S92 methodis thatwecantestforapositive relationship betweenspotvolatilityandfutures tradingactivityafterpartialling outthee®ectofspottradingactivity.W ithinthis frameworkweinvestigatethe following fourhypotheses: (i)spot volatility and futures trading volume are positively related;(ii)spot volatility and the numberoffutures transactions are positively related;(iii)the numberoffutures transactions has a stronger relationship withvolatilitythanfutures volume;(iv)openinterestis aproxyof marketdepth andis negativelyrelated tovolatility. B ecause ofthe trends in the spotand futures activity variables, they are decomposed intothree components.T he ¯rstcomponentis the seculartrend and the second and third components are the predicted and unexpected components ofthedeviation from the trend.A llthetrend components are highly correlated and areusefulin capturingthe relationship between overalllevelof tradingactivity and volatility.T hepredicted components areespeciallyuseful as ameasureofmarketdepthespeciallywhenappliedtoopeninterest.Finally, theunpredicted componentcaptures the relationship between newtradingactivityand volatility. T he outline ofthe paperis as follows.In section 2 we discuss the theories thatsupportand oppose the introduction offutures markets with the aim of summarizing the testable hypotheses. Similarly, in section 3 we discuss the theories thatrelatethevolumeoftradetopricechanges andvolatilitywiththe 2 aim ofsummarizingthetestablehypotheses.Section 4 contains abriefhistory ofthe South A frican futures exchange, section 5 contains a briefsynopsis of ourempiricalmethodologyand section 6 contains adiscussion ofthedata.W e reporttheresults in section 7and concludewith section 8. 2 T he R el ationship b etw eenSpot Vol atil ity and FuturesM arkets M anyofthebene¯tsoftheintroductionoffuturesmarkets° owfrom theirtransactions costadvantages.Futures markets reducesearch,liquidity,moralhazard andbrokeragecosts (Cagan(1 981 ),Ja®e(1 984),V eljanovski (1 986)).T hebene¯ts ofcheapertrading opportunities leads to quickerinformation revelation, less mispricing, greaterliquidity and newrisk-return combinations.H owever, theimpacton spotvolatilityis unclear. B lack (1 976), Cox (1 976), P eck (1 976), T urnovsky (1 979), Campbell and T urnovsky(1 985)showthatfutures markets areclearinghouses formarketinformationandprovideadditionalinformationforspotmarketparticipants.D ue to the more rapid dissemination of market-wide in¯rmation, D anthine(1 978) shows thattheintroductionoffutures tradingleads toincreasedmarketdepth. A dditionalliquiditymayalsobeprovidedbyspeculatorswho° ocktothefutures marketwhereitis cheapertotrade.T hus speculators provideboth avaluable serviceas arbitraguers,whoenablehedgers totransferrisk,andgreatermarket depth. T he bene¯cial e®ect on volatility may go either way. O n the one hand, the additional liquidity provided by speculators may also lead to a decrease involatility(H odgsonandN icholls (1 991 )).D anthine(1 978),G rossman(1 988) andG rossmanandM iller(1 988)provideargumentsforadecreaseincashmarket volatility because ofan increasein marketliquidity and adecreasein the cost toinformedtraders ofrespondingtomispricing.T his wouldbeofbene¯tiftoo 1 little information leads to toomuch volatility as argued byA ttanasio(1 990 ). O n theotherhand,ifspotprices dobecomemorevolatilethis is good because itre° ects alltheinformation available(R oss (1 989)). Finally,totheextentthatthemarketisincomplete,thefuturesmarkethelps tocompletethemarketbyallowingpreviouslyunavailablecombinations ofrisk and return tobereplicated. O pponents offutures markets assertthatthey are destabilizingbecause of speculationandarbitragestrategies suchas computertrading.W hereas proponents see bene¯ts from the availability oflow-costtrading,Stein (1 987)shows thattradingin futures by uninformed speculators destabilizes the market.In this case, prices become less informative. T he resulting loss in con¯dence in ¯nancialmarkets may lead to a reduction in liquidity thatcan have negative welfaree®ects byincreasingtransactions costs andtherebyraisingtherequired 1 T hisissim il ar to Singleton's(1987) point regard ingexc hange ratesw here the m ainthrust isthat m ore inf orm ationc anac tual ly l ead to a red uc tioninvolatility 3 realreturn on an asset.M alkiel(1 979)and P indyck (1 984)alsosuggestthat an increaseinvolatilityincreases therequiredriskpremium and this leads toa highercostofcapitaland lowerstockprices. M any ofthese implications are very di±culttotest.Forexample,changes in spotmarketdepth maybeduetosecularchanges thatare unrelated tothe introduction offutures markets. Similarly, an increase in volatility following the introduction ofa futures marketmay be due to the greaterrevelation of information (R oss (1 989))ordue to the destabilizing in° uence ofspeculators (Stein(1 987)).O nealternativeistousethemethodologyofG allowayandM iller (1 997)whouseamatchingsampletoeliminate any secularchanges.U nfortunately,theJSE is highly concentrated and thinlytraded (B eelders (20 0 0 a))so this method is infeasible. T he advantage ofusing the B &S92 framework is thatwe can testforthe sensitivityofspotvolatilitytotradingactivityonthefutures marketafterpartialling outthe e®ectofspottradingactivity.W ithin this framework we can determineiffutures tradingactivityhas apositivee®ectonspotmarketvolatility. 3 Spot Vol atil ity and Spot T rad ingAc tivity T hepositivecorrelationbetweenvolumeandabsolutepricechanges arederived from the interplay oftwoprinciples: the asymmetry ofinformation generates tradingandtheextentofthedisagreementdetermines thesizeofthetrades.In this section we brie° y reviewthe theories thatprovide testable hypotheses of therelationship betweenspotvolatilityand spottradingactivity. First,(Clark(1 973),Epps and Epps (1 976),T auchen and P itts (1 983)and H arris (1 986))proposed a mixture ofdistributions hypotheses where the sequentialarrival of new information generates both trading volume and price movements with both increasingduringperiods characterized bynumerous informationshocks.A nymeasureofactivitysuchas volume,openinterestorthe numberoftransactions canactas themixingvariable. Second,Copeland(1 976,1 977),M orse(1 981 ),Jennings,Starks andFellingham (1 981 )and Jennings and B arry(1 983)proposemodels wherein newinformationis disseminatedsequentiallytotraders andimperfectlyinformedtraders cannotinferthepresenceofinformedtrading.T hesequentialarrivalofinformationgenerates bothpricechanges andvolumewherebotharepositivelyrelated tothenumberofinformation shocks arrivingperunittime. T hird, within the class ofstrategic models, A dmati and P ° eiderer(1 988) show that traders who have discretion in choosing when to trade, choose to trade when volume is large.Consequently, transactions and price movements are bunched in time and the e®ectofvolume on price movements depends on recentvolumelevels.Kyle's (1 985)modelimplies thatlargervolumesupports moreinformedtradersandthemarketdepthvarieswithnon-informationaltradingactivity.Kyle(1 985)de¯nesmarketdepthastheorder° owrequiredtomove prices by one unit.B essembinderand Seguin (1 993)argue thatmarketdepth 4 varies withrecenttradingactivityanditcanbeproxiedbyendogenouslydetermined open interestand open interestis morelikelytore° ecthedgingactivity than speculation.T hehypothesis is thatvolatilityis lowerwhen open interest is largeafterconditioningon contemporaneous volume. Finally,withintheclassofcompetitivemodelstradesizeis positivelyrelated to the quality orprecision of the information possessed by informed traders (P ° eiderer(1 984),Kim andV errechia(1 991 )).A n adverseselection problem is introducedbecauseinformedtraders wanttotradelargeamounts atthecurrent price. It is shown that absolute price changes and volume - the aggregate demand ofallinvestors -arepositivelyrelated within this class ofmodels. A s wementionedintheintroduction,Karpo®(1 987)documents manystudies thatsupportthe positive relationship between absolute price changes and volume.R ecently, Jones, Kauland L ipson (1 994)suggestthatthe numberof transactions is morerelevantthanthesizeofthetransaction(volume).Intheir study of835 securities on theN A SD A Q -N M S master¯le oftheCenterforR esearchin SecurityP rices (CR SP )they¯nd thatvolumeorthesizeofthetrade has noexplanatorypowerbeyondthatcontainedinthenumberoftransactions. B ased on the theories and empiricalwork todate, we obtain fourtestable hypotheses: (i)spotvolatility and spot volume are positively correlated;(ii) spotvolatilityandopeninterestarenegativelyrelated.W ederivethetwomore hypothesesfrom Jones,KaulandL ipson(1 994)whosuggestthatthenumberof transactions canbesubstitutedforvolume:(iii)spotvolatilityandthenumber oftransactionsarepositivelyrelated;(iv)thesizeofthetradehasnoexplanatory powerbeyond thatcontained in thenumberoftransactions.U nfortunatelywe donothavethenumberoftransactions forthespotmarket,butwedohavethe numberoftransactions forthe futures market, sowe analyze the relationship between spotvolatilityand futures transactions. 4 History ofthe South Af ric anFuturesM arket In A pril1 987 an informalfutures marketwas started in South A frica by the R and M erchantB ank(R M B ):itwas theexchange,theclearinghouseand the only marketmaker. A lthough the futures marketsurvived the O ctober1 987 marketcrash,volumes remained relatively thin.In September1 988 itwas decidedtobroadenthemarketandensurewidespreadtrustintheindependenceof theexchangeandclearinghousebyformingtheSouthA fricanFuturesExchange (Safex)and theSafexClearingCompany(P ty)L imited (Safcom).T wenty-one majordomestic¯nancialinstitutions,includingtheSouthA fricanR eserveB ank and the JohannesburgStock Exchange (JSE),subscribed to80 \ seats"in the ventureatapriceofR 25,0 0 0 ,therebyprovidingthestart-up capitalofR 2 mil2 lion. W hen the infrastructure,rules and procedures were completed in 1 989, 2 T he South Af ric anc urrency is c alled the \rand " and is m ad e up of10 0 c en ts. T he ab reviationf or the rand isan\R " b ef ore the am ount. In1980 , the exc hange rate w ith the U S$ w asapproxim ately one-f or-one;inJ une 2 0 0 0 ,it isapproxim ately R 6. 90 per U S$ . 5 afurther39 seats were issued ataprice ofR 35,0 0 0 each,givingan additional R1. 365 million in capital. Safcom took overthe managementofthe informalfutures marketin A pril 1 990 withthefollowingobjective:\ Safexseekstoprovidethesecureande±cient marketfortradingderivativesinSouthA frica. " T heSafexwaso±ciallylicensed on A ugust1 0 ,1 990 afterthepassingoftheFinancialM arkets ControlA ct.A t this point, the only listed products were futures contracts on the majorJSE (A llshare(A L SI),G old(G L D I)andIndustrial(IN D I))Indexes,themostactive long-datedbond (theE1 68 issued byEscom (ElectricitySupplyCommission)), thethreemonth liquid bankers acceptancerateand on theU S$ priceofgold. In O ctober1 990 theSA R B granted non-residents permission toparticipate ontheSafexviatheFinancialR andsystem.G rowthinthemarketwas initially very slow with volumes of 2,0 0 0 -5,0 0 0 contracts per day, but they began to soarafterthe Safexintroduced options-on-futures togetherwith aworld-class, portfolio-scanning-type margining system in O ctober 1 992. W ithin approximately1 2 months,volumes hadgrownby70 0 % andinD ecember1 993 volumes exceeded1 million contracts permonth forthe¯rsttimeandopeninterestwas greaterthan 50 0 ,0 0 0 contracts. O n June1 5,1 995 the underlyingindexes ofthe A L SI,IN D I and G L D I futures contracts were restructured and based on the largestand mostactively traded stocks in theirsector. T he newunderlying indexes were the A L SI40 , IN D I25 andG L D I1 0 indexes wherethenumberattheendoftheindexdenotes thenumberofstocks usedinits construction.A futures contractontheFinancialand Industrial(FIN D I30 )Index contractwas introduced slightly lateron O ctober6,1 995.O wingtothe size ofthe South A frican markets,many large companies appearin morethan twoindexes. In M ay 1 996 a fully-automated tradingsystem was introduced.T his coincided with thephasingin oftheJohannesburgEquities T rading(JET )system on the JSE from M arch 1 996 toJune1 996.In January 1 997open interestexceeded 1 million contracts forthe¯rsttime.T hemostsuccessfulcontracts are thosebasedontheA L SI40 andIN D I25 indexes.T heM ining1 5 (M IN I 1 5)and Financial1 5 (FIN I 1 5)contracts wereintroduced on M arch 3,1 998 toreplace the G L D I 1 0 and FIN D I 30 contracts,respectively, thatwere discontinued on M arch1 9,1 998.O nM arch1 8,1 999 theR esources 20 (R ESI 20 )contractbegan trading and replaced the M IN I1 5 contractwhich ceased trading the previous day. V olumes continue to be dominated by equity index products with the concentration being in the A L SI40 and IN D I25 indexes. O ptions accountfor approximately 50 % ofvolumes and 80 % ofopen interest,and tradeis increasinglydominatedbyinternationalmarketplayerswhoaccountforapproximately 40 % ofvolumes traded. Inkeepingwithits statedobjectiveofprovidinga\ secureande±cientmarketfortrading derivatives in South A frica" an A griculturalM arkets D ivision was introducedin1 995 andoptions onagriculturalproducts wereintroducedin M arch 1 998.In February 1 999 Sun° owerSeeds and CapeW heatfutures were introduced intheA griculturalM arket. 6 5 E m piric alM ethod T he empiricalmethod is thatofB essembinderand Seguin (1 992)whoiterate between thetwoequations, R t = ±+ Xn j=1 ¾ bt = ® + X4 ° jR t¡j + ´idi + i=1 X4 ½idi + i=1 Xn j=1 Xn j=1 ¯ j¾ bt¡j + Xn j=1 ¼ j¾ bt¡j + U t !jU t¡j + "t (1 ) (2) where R t is the return on day t and is calculated as R t = 1 0 0 ¢ln(P t=P t¡1 ); P t is the index atdp ate t, di denotes day-of-the-week dummies forM onday to T hursday,b ¾ t ´jU tj ¼=2 istheestimatedvolatilityandU t and"t aretheresiduals.T oobtain consistentestimates ofthe parameters we adoptthe following iterativemethodofestimation:¯rst,estimate(1 )withoutthelaggedvolatility, savetheresiduals as Ubt andcomputeb ¾ t.Second,estimate(2)andsavethepredicted volatility.T hird, estimate (1 )and include the predicted volatility from thesecondstep.Savetheresiduals as Ubt andcomputeb ¾ t.Finally,estimate(2) and interprettheparameterestimates. T oanalyzetherelationship between volatilityand theactivityvariables we includetheactivityvariables in(2)toobtain, ¾ bt = ® + X4 i=1 ´idi + Xn j=1 ¯ j¾ bt¡j + Xn j=1 !jU t¡j + m X ¹ kA k+ "t; (3) k=1 wheretheactivityvariablesareindexvolume,futuresvolume,openinterestand numberoffuturestransactions.W eadoptthesamefourstepiterativeprocedure for(1 )and(3)as mentioned above. W euse thesamedecomposition as B &S92 toremovethe timetrend in the activity variables. First we take the natural log of the activity variable to remove heteroskedasticity.Second, we ¯ta 1 0 0 -day movingaverage tomodel the trend componentand ¯nally we use an A R IM A (0 ;1 ;1 0 )modeltoobtain a predicted and unpredicted component forthe detrended component ofthe activityvariable.In sum,thevariableis decomposed as follows: V t = trendt + trenddeviationt = M A V (1 0 0 )t + (P V t + U V t) (4) whereM A V (1 0 0 )t isthe1 0 0 -daymovingaverage,P V t isthepredicteddeviation from the trend and U V t is the unexpected deviation from the trend. B efore includingthesethreecomponent s ofvolumeas activityvariables in (3),wedeP meaneachseries sothat®=(1 ¡ j ¯ j)canbeinterpretedas theunconditional orlong-run variancewhentheactivityvariables areattheiraveragelevels. 7 6 Data P riortoJune 1 995, futures contracts on the South A frican Futures Exchange (SA FEX )were based on the broad, marketindexes thatrepresentthe largest percentage ofmarketcapitalization, i. e. the A llShare Index (A L SI), the IndustrialIndex(IN D I),theG old Index(G L D I)andtheFinancialIndex(FIN I). O n June 1 5, 1 995 the JSE introduced indexes based on the largerand more highly traded companies in each ofthese sectors forthe purpose ofbeingthe underlying index forfutures contracts on the SA FEX .T he A L SI40 , G L D I1 0 , IN D I25 andFIN D I30 mirrorthefourbroadindexes mentionedabovewherethe numberattheend ofeach indexrefers tothenumberofstocks thatis used in constructingthe indexand FIN D I denotes theFinancialand IndustrialIndex. M ichelle Francis ofthe JSE kindly provided the closing price and volume for theindexes fortheperiod 1 5 June 1 995 to31 D ecember1 999.T henumberof transactions, volume and open interestofthe futures contracts was obtained from theSA FEX website. W ehaveacompletedatasetfortheA L SI40 and IN D I25 futures contracts from 1 5 June 1 995 to 30 N ovember1 999. T he futures contractbased on the G L D I1 0 indexwas discontinuedin1 7September1 998 sothesampleperiodruns from September1 ,1 995 to1 7September1 998.A llotherfutures contracts are ignoredduetosmallsamplesizesorlackofdataduetothintrading.T heindexes 3 arevalue-weightedandareadjustedformergers,de-listings,butnotdividends. In conclusion, we focus on the futures contractbased on the A L SI40 , IN D I25 and G L D I1 0 contracts becausetheyhavesu±cientdatafortheanalysis. 7 R esul ts 7. 1 Desc riptive Statistic s W ereportthedescriptivestatistics ofthereturns ofthethreeindexes in T able 1 . W e include tests of skewness and kurtosis based on the G M M estimator becausethetraditionalestimatorsu®ers from thedrawbackthatits varianceis underestimatedwhenreturns arenon-normalandconditionallyheteroskedastic (P agan (1 996)).T he G L D I1 0 index is uniquely di®erenttothe othertwoindexes:thereturns fortheG L D I1 0 indexaretwotothreetimes as largeas those oftheIN D I25 and A L SI40 indexes and itdisplays statisticallysigni¯cantpositiveskewness.A lthoughthecoe±cients ofskewness oftheIN D I25 andA L SI40 indexes arenegative,theyarenotstatisticallysigni¯cant.W hereas theA L SI40 andIN D I25 indexes displaystatisticallysigni¯cantexcess kurtosis,theG L D I1 0 index has statistically signi¯cantly less kurtosis than the normaldistribution. Each ofthe indexes has a statistically signi¯cant¯rstorderserialcorrelation coe±cientandtheB oxL jungQ -statistics aresigni¯cantatthe1 % levelatboth 5 and 1 0 lags. T he signi¯cance ofthe Q -statistics of the squared returns is 3Detail ed inf orm ationab out the ind exc onstruc tionisavailab le at the w eb site ofthe J SE , http://w w w . jse. c o. za/. 8 consistentwith conditionalheteroskedasticityin thereturns. W edonotreportdescriptivestatisticsfortheactivityvariablesbecausesome ofthem were growingovertime.H owever,the generalfeatures ofthe activity variables can be seen in Figures 1 ,2 and 3.Forexample,the growth in index volume ofthe A L SI40 index is evidentin Figure1 C while the lackoftrend in ofthe numberoftransactions forthe A L SI40 and IN D I25 contracts is evident inFigures 1 D and3D .T hevolatilitythatfollowedtheA siancrisis is evidentin Figures 1 B and 3B forthe A L SI40 and IN D I25 indexes,respectively. Figures 1 F,2F and 3F display theuniquebehavioroftheopen interestthatdecreases atthe maturity date ofeach futures contract.W hereas the open interestfor theA L SI40 and IN D I25 contracttrend upwards,thereis notrend in theopen interestoftheG L D I1 0 index.T hefailureofthefutures contractbased on the G L D I1 0 indexis clearlyevidentinFigure2D wherethenumberoftransactions declines overtimedespitehigh volumes followingtheA sian crisis (Figure2C). 7. 2 Correl ationM atric es A s a precursor to the analysis, we brie° y highlight the ¯ve most important features ofthecorrelation matrix ofthereturns,absolutereturns and activity variables forthethreeindexes inT ables 2,4 and6.First,returns andabsolute returns are negatively correlated.T his is consistentwith negative skewness in thedistributions oftheA L SI40 andIN D I25 indexes,butnottheG L D I1 0 index. Second, the correlation between the absolute return and the activity variables arepositiveandaveragearound0 . 20 0 ;thisisconsistentwithapositiverelationship betweenvolatilityandtheactivityvariables(Karpo®(1 987)).T heabsolute return has the strongestrelationship with the unexpected futures volume and unexpectedtransactions volumes and1 0 0 -daymoving-averages,i. e.volatilityis positively related tothe overallleveloftradingand unexpected shocks tothe market. T he relatively large correlation between the absolute return and the 1 0 0 -daymoving-averages oftheindexvolume,futures volumeandopeninterest canbeexplainedbythehighleveloftradingactivityduringthehighlyvolatile period ofthe A sian crisis and the upward trend in activity overthe sample period. T hird,thethreemoving-averages arehighlycorrelatedwitheachotherwith manycorrelationcoe±cients inexcess of0 . 90 0 .Fourth,thepredictedindex,futures and transaction volumehavecorrelation coe±cients thataverage around 0. 40 0 ,buttheyhavealowercorrelationwiththepredictedopeninterest. Fifth, thecorrelation between the unexpected index volume,unexpected futures volumeandtransactionvolumeaveragearound0 . 250 .Incontrast,theunexpected futures volumeandunexpectedtransactions volumehaveacorrelationof0 . 71 1 . T osum up,theactivityvariables arepositivelyrelatedtovolatilityandweneed tobeconcernedaboutthemulticolinearitybetween themoving-averages. 9 7. 3 T he ALSI4 0 Ind ex T heestimated parameters ofequations (2)and(3)arecontainedinT able3 for theA L SI40 Index.M odel1 is thecombinationof(1 )and(2)wherenoactivity variables areincludedinthespeci¯cationoftheconditionalvariance.T heestimates havethreebasicfeatures:¯rst,volatility is positivelyseriallycorrelated - the¯rst¯velags ofvolatilityarestatisticallysigni¯cantand positive.T his is consistentwiththeexcess kurtosis inreturns andconditionalheteroskedasticity inassetreturns.Second,thecoe±cientonthe¯rsttwolaggedresiduals arestatistically signi¯cantand negative,i. e.the conditionalvariance increases when thereis anegativeshocktoreturns.T his is consistentwith asymmetrice®ects involatility(Christie(1 982))andwiththeresults ofB eelders (20 0 0 a)who¯nds T A R CH e®ects inthefutures indexes and broadermarketindexes.T hird,only the W ednesday dummy is statistically signi¯cantand positive.T his is consistentwith the results ofB eelders (20 0 0 b)who ¯nds a settlemente®ecton the JSE - transactions mustbesettled everyT uesdayorthe¯rstdaythereafter. Itis interestingtonotethatthecoe±cients on thelagged conditionalvarianceandlagged returns arequitesimilaracross thevarious models.Exceptfor the W ednesday e®ect, these three empiricalregularities are presentin each of thesixmodels.Each ofthesixmodels has astatisticallysigni¯cantF-statistic forgoodness-of-¯tatthe 1 % level. T he consistency in the magnitude ofthe parameters across models suggests thatnoneofthemodels is seriouslymisspeci¯ed. M odel2 is the combination of (1 )and (3)where the Index volume (X V ) is included as an activityvariable.T he1 0 0 -daymoving-average,predicted volumeandunexpectedvolumeeachhavepositivecoe±cients thatarestatistically signi¯cantatthe1 % level,i. e.theconditionalvarianceoftheindexis positively relatedtovolume(Karpo®(1 987)).T hesigni¯canceofthemovingaverageand the predicted volume may be due to the high levelofvolatility following the A sian crisis in 1 997 and 1 998, the introduction ofelectronic trading in 1 996 andtheresolutionofpoliticaluncertaintythathas leadtogreaterinternational interestin South A fricain thelatterpartofthe1 990 's. M odel3 isthecombinationof (1 )and(3)wherethefutures contractvolume (FV )is includedas theactivityvariable.T he1 0 0 -daymoving-averageand unexpected volumeeach havepositivecoe±cients thatarestatisticallysigni¯cant atthe1 % level.T hestatisticalsigni¯canceofthemovingaveragemaybedueto thesecularchanges thatwementionedabove.T hecoe±cientoftheunexpected futures volumeis 0 . 552 soan unexpected 1 0 % increasein volumeincreases the conditionalvarianceby6. 5% from 0 . 81 6 to0 . 871 assumingthattheconditional varianceequalled thelong-run varianceon theprevious ¯vedays. M odel4 is the combination of (1 )and (3)where the open interest (O I) is included as the activity variable. O nly the 1 0 0 -day moving-average has a positive and statistically signi¯cantcoe±cientatthe 1 % level.T he statistical signi¯canceofthemoving- averagemay beduetothesecularchanges thatwe mentioned above and the lack ofsigni¯cance ofthe othercomponents may be duetothelargechanges in open interestatthematurityofacontract. 10 M odel5 is thecombinationof (1 )and(3)wherethenumberoftransactions (T )is included as the activity variable.T he numberoftransactions does not haveanytrendsowedidnotcomputea1 0 0 daymovingaverage,butwedouse the A R IM A (0 ;1 ;1 0 )modeltodecompose itintoa predicted and unexpected component. T he coe±cient of the predicted component is insigni¯cant, but the coe±cientofthe unexpected componentis signi¯cantatthe 1 % leveland is statistically indistinguishable from 1 , i. e. a 1 0 % increase in the numberof transactions leads to1 0 % increasein thelong-run variance. Finally, model 6 refers to the combination of (1 )and (3)where all the activity variables are included as regressors. T he 1 0 0 -day moving average of the open interest is statistically signi¯cant at the 5% leveland captures the secularchangeswereferredtoearlier.T hepredictedlevelofopeninterestisalso statistically signi¯cant, butthe coe±cientis negative.T his is consistentwith theresults ofB &S92 whoconjecturethattheintroduction offutures contracts decreases spot volatility when the market is deep. Finally, the unexpected numberoffutures transactions is statistically signi¯cant with a coe±cientof 0. 981 , i. e. the unexpected number of transactions maintains its importance even afterpartiallingoutthee®ectofthe otheractivity variables.T his result is consistentwith Jones, Kauland L ipson (1 994)who ¯nd thatitis notthe volume,butthenumberoftransactions thatis thekeyvariablefordetermining volatility. W hen we compare the adjusted-R 2 ofeach modelwe note thatthe benchmark is given by model1 with an adjusted-R 2 of0 . 235.O utofmodels 2, 3, 4 and 5, model5 has the highestadjusted-R 2 of0 . 30 9. T he adjusted-R 2 of model5 is notmuchsmallerthanthatofmodel6 (0 . 31 8).T helargestincrease in the adjusted-R 2 occurs when weadd the numberoftransactions.B ased on explanatorypower,itappears thattheunexpectednumberoftransactions that is mostimportantadditionalvariable. In conclusion, spot volatility has a positive relationship with the secular trendinopeninterest,anegativerelationship withpredictedopeninterestanda positiverelationship withtheunexpectednumberoftransactions onthefutures market.In terms ofexplanatorypower,thenumberoftransactions is themost 4 importantvariable. 7. 4 T he G LDI10 Ind ex T heestimated parameters ofequations (2)and(3)arecontainedinT able5 for theG L D I1 0 Index.M odel1 is thecombinationof(1 )and(2)wherenoactivity variables areincludedinthespeci¯cationoftheconditionalvariance.T heestimates havethreebasicfeatures:¯rst,volatility is positivelyseriallycorrelated withthelargestcoe±cients atthesecondandfourthlag.Second,thecoe±cient on the ¯rstlagged residualis negative across allthe models butonly statistically signi¯cantin model4.T hird, none ofthe daily dummies are signi¯cant; 4 T he c aveat to thisc on c lusionisthat w e are not partialling out the e®ec t ofthe n umb er ofspot transac tions. 11 this is consistentwith B eelders (20 0 0 b)whofails to detectseasonality in the conditionalmean returns ofthe G L D I1 0 index.T he lowadjusted-R 2 of0 . 0 71 formodel1 is much lowerthan theequivalentmodelfortheA L SI40 indexand may be due to the greaterlevelofvolatility in the G L D I1 0 index. W ith the addition ofindexvolume(X V )in model2,theadjusted-R 2 increases to0 . 1 65. T he predicted and unexpected volume each have positive coe±cients thatare statisticallysigni¯cantatthe1 % level.T hecoe±cientofunexpectedvolumeis 1. 0 61 whichimplies thata1 0 % increaseinindexvolumeleads to5. 5% increase in thelong-runlevelofvolatility. T hecoe±cientofunexpected futures volumein model3 is 0 . 399 and is signi¯cant at the 1 % level. In comparison to model 2, the coe±cient ofindex volume is more than twice as large as the coe±cient on the futures volume. T urningtomodel4, none ofthe coe±cients ofthe components ofopen interest(O I)are statistically signi¯cant.In model5, the coe±cientofunexpected transactions is 0 . 979 whichimplies thata1 0 % increaseinthenumberoftransactions leads toa5. 8% increasein thelong-run variance.Finally, weconsider model6 thatincludes alltheactivityvariables as regressors.O nlythepredicted open interesthas nostatisticalsigni¯cance.T he coe±cients on predicted and unexpected index volume are both positive.T he coe±cients on predicted and unexpected futures volumeareboth negative.T his is consistentwith B essembinderand Seguin's conjecture thatfutures trading does notexacerbate spot volatility when the marketis deep.T he coe±cienton unexpected open interestis 1 . 298 and may re° ectincreased volatility atthe maturity ofthe futures contracts.Finally,the coe±cientofthe unexpected numberoftransactions is 0. 994,i. e.a1 0 % increasethenumberoftransactions leads toa5% increasein spotvolatility,ceteris paribus. W hen we compare the adjusted-R 2 ofeach modelwe note thatthe benchmarkis given bymodel1 with an adjusted-R 2 of0 . 0 71 .O utofmodels 2,3,4 and 5, model6 has the highestadjusted-R 2 of0 . 21 8.T he addition ofindex volume and transactions to the base models provides the biggestincreases in theadjusted R 2 . In conclusion,spotvolatilityhas apositiverelationship with indexvolume, anegativerelationship withfutures volumeandapositiverelationship withthe unexpectednumberoftransactions on thefutures marketand open interest. 7. 5 T he INDI2 5 Ind ex T he estimated parameters ofequations (2)and (3)are contained in T able 7 forthe IN D I25 Index. M odel1 is the combination of (1 )and (2)where no activity variables are included in the speci¯cation ofthe conditionalvariance. T he estimates have the same three basicfeatures as the A L SI40 and G L D I1 0 contract.Inmodels 2 and3,the1 0 0 -daymovingaverageandunexpectedindex volumeand futures volumeeach havepositivecoe±cients thatarestatistically signi¯cantatthe1 %.Inmodel3,the1 0 0 -daymoving-averageofopeninterestis statisticallysigni¯cantatthe1 % level.O nceagain,thestatisticalsigni¯canceof themovingaveragemaybeduetothesecularchangesthatwementionedabove. 12 Formodel4 we¯ndthatallthreecomponents ofthenumberoftransactions are statistically signi¯cant.T he 1 0 0 -day moving-average has acoe±cientof1 . 569 and the unexpected numberoftransactions has acoe±cientof1 . 0 28.Finally, in model6 only three coe±cients are signi¯cant: unexpected futures volume has a positive coe±cientof0 . 669, the 1 0 0 -day movingaverage oftransactions has a positive coe±cientof1 . 684 and predicted open interestonce again has a negative coe±cient. W hen comparing the adjusted-R 2 across the models, we ¯nd thatfutures volume and transactions provide the greatestincrease in explanatory powerovermodel1 .T hebasemodelhas an adjusted-R 2 of0 . 225 and this increases to0 . 273 and 0 . 274 when futures volume and the numberof transactions areadded,respectively. Inconclusion,spotvolatilityispositivelycorrelatedwiththe1 0 0 -daymoving averageoftransactions thatcaptures theseculartrendandpositivelyrelatedto theunexpected futures volume. 8 Concl usion Consistentwiththeexistingstudies,we¯ndthatspotvolatilityandtheoverall leveloftradingactivityon thespotand futures markets arepositivelyrelated. W henmarketdepthis measuredbypredictedopeninterestorpredictedfutures volume we ¯nd thatfutures market activity dampens spotmarket volatility. T his is consistentwith Kyle(1 985)whoconjectures thatgreatermarketdepth cansupportmoreinformed traders and theresults in B &S92. Finally, the unexpected number of transactions on the futures market is more strongly related to spotvolatility than unexpected futures volume and unexpected index volume forthe A L SI40 and G L D I1 0 contracts. In fact, for A L SI40 indexa1 0 % increaseintheunexpectednumberoftransactionsleadstoa 1 0 % increaseinvolatility.T heresponseofthevolatilityoftheG L D I1 0 indexto thesameincreaseintheunexpectednumberoftransactions is only5% although itis twiceas variableas theA L SI40 index.T heseresultsareconsistentwiththe results ofJones,Kauland L ipson (1 994)and may re° ectthefactthattraders who have access to new information are more likely to trade on the futures marketbecause itis less expensive than the spotmarket. T he one caveatto this conclusionis thatwedonotincludethenumberofspottransactions inour analysis;ahorseracebetweenthenumberofspottransactions andthenumber offutures transactions would clearlybeamorede¯nitivetest. R ef erences [1 ]A ttanasio,O . P.(1 990 )\ A ssetP riceV olatilityandInformationStructures," Economics L etters,33,2,1 59-1 64. [2]A dmati,A .and P.P ° eiderer(1 988)\ A T heoryofIntradayP atterns:V olumeand P ricevariability,"T he R eviewofFinancialStudies,1 ,3-40 . 13 [3]B eelders,O wen(20 0 0 a)\ T heU nconditionalD istributionofSouthA frican StockR eturns,"workingpaper,D epartmentofEconomics,EmoryU niversity. [4]B eelders, O wen (20 0 0 b) \ Seasonality on the Johannesburg Stock Exchange,"workingpaper,D epartmentofEconomics,EmoryU niversity. [5]B essembinder,H endrikandP aulJ.Seguin(1 992)\ Futures-T radingA ctivityand StockP riceV olatility,"JournalofFinance,47,5,20 1 5-20 34. [6]B essembinder,H endrikand P aulJ.Seguin (1 992)\ P rice V olatility,T rading V olume and M arket D epth," Journalof Financial and Q uantitative A nalysis,28,1 ,21 -39. [7]Cagan,P.(1 981 )\ FinancialFuturesmarkets:IsM oreR egulationN eeded?" JournalofFutures M arkets,1 ,2,1 69-1 90 . [8]Campbell, R obertB .and Stephen J.T urnovsky (1 982)\ T he Stabilizing and W elfare P roperties ofFutures M arkets:A Simulation A pproach,"InternationalEconomicR eview,26,2,277-30 3. [9]Christie, A ndrewA .(1 982)\ T he Stochastic B ehaviorofCommon Stock V ariances:V alue,L everageandInterestR ateE®ects,"JournalofFinancial Economics,1 0 ,4,40 7-32. [1 0 ]Clark, P. K.(1 973)\ A Subordinate Stochastic P rocess M odelwith Finite V arianceforSpeculativeP rices,"Econometrica,41 ,1 35-1 55. [1 1 ]Copeland,T . E.(1 976)\ A M odelofA ssetT radingundertheA ssumptionof SequentialInformationA rrival,"JournalofFinance,31 ,September,1 1 491 1 68. [1 2]Copeland,T . E.(1 977)\ A P robabilityM odelofA ssetT rading,"Journalof Financialand Q uantitative A nalysis,1 2,N ovember,563-578. [1 3]Cox, J. C.(1 976)\ Futures T rading and M arketInformation," Journalof P oliticalEconomy,84,1 21 5-1 237. [1 4]D anthine,J.(1 978)\ Information,Futures P rices,andStabilizingSpeculation,"JournalofEconomicT heory,1 7,79-98. [1 5]Edwards, F. R .(1 988) \ D oes Futures T rading Increase Stock M arket V olatility,"FinancialA nalysts Journal,44,63-69. [1 6]Epps,T . W .andM . L .Epps(1 976)\ T heStochasticD ependenceofSecurity P riceChanges and T ransaction V olumes:Implications fortheM ixture-ofD istributions H ypothesis,"Econometrica,44,M arch,30 5-325. [1 7]G alloway, T ina M .and James M .M iller(1 997)\ Index Futures T rading and Stock R eturn V olatility:Evidence from the Introduction ofM Idcap 40 0 IndexFutures,"FinancialR eview,32,3,845-866. 14 [1 8]G rossman,S. J.(1 988)\ A n A nalysis oftheImplications forStockand Futures P riceV olatilityofP rgram T radingandD ynamicH edgingStratgies," JournalofB usiness,61 ,275-298. [1 9]G rossman,S. J.and J. E.Stiglitz (1 980 )\ O n theImpossibility ofInformationallyE±cientM arkets,"A merican EconomicR eview,70 ,3,393-40 8. [20 ]G rossman,S. J.andM . H .M iller(1 988)\ L iquidityandM arketStructure," JournalofFinance,43,61 7-633. [21 ]H arris, L .(1 986)\ Cross-Security T ests of the M ixture of D istributions H ypothesis," JournalofFinancialand Q uantitative A nalysis, 21 , M arch, 39-46. [22]H odgson, A llan and D es N icholls (1 991 )\ T he Impact of Index Futures M arkets on A ustralian Sharemarket V olatility," Journalof B usiness Finance andA ccounting,1 8,2,January,267-280 . [23]Jennings, R . H. ,L . T .Starks and J. C.Fellingham (1 981 )\ A n Equilibrium M odelofA ssetT radingwith SequentialInformation A rrival," Journalof Finance,36,M arch,1 43-1 61 . [24]Jennings,R . H .andC.B arry(1 983)\ InformationD isseminationandP ortfolio Choice," JournalofFinancialand Q uantitative A nalysis,1 8, M arch, 1 -1 9. [25]Jones,CharlesM . ,G autam KaulandM arcL .L ipson(1 994)\ T ransactions, V olume,and V olatility,"R eviewofFinancialStudies,7,4,631 -51 . [26]Ja®ee,D . M .(1 984)\ T heImpactofFinancialFuturesandO ptions onCapitalFormation,"JournalofFutures M arkets,4,3,41 7-447. [27]Karpo®, Jonathan M .(1 987)\ T he R elation between P rice Changes and T radingV olume:A Survey,"JournalofFinancialandQ uantitative A nalysis,22,1 ,1 0 9-1 26. [28]Kim,O .and R . E.V errechia(1 991 )\ M arketR eaction toA nticipated A nnouncements,"JournalofFinancialEconomics,30 ,273-30 9. [29]Kyle,A . S.(1 985)\ Continuous A uctions and InsiderT rading,"Econometrica 53,N ovember,1 31 5-1 335. [30 ]M alkiel,B .(1 979)\ T heCapitalFormationP roblem intheU nitedStates," JournalofFinance,34,391 -30 6. [31 ]M orse,D .(1 981 )\ P riceandT radingV olumeR eaction SurroundingEarnings A nnouncements:A CloserExamination," JournalofA ccountingR esearch,1 9,A utumn,374-383. [32]P agan, A drian (1 996)\ T he Econometrics ofFinancialM arkets," Journal ofEmpiricalFinance,3,1 5-1 0 2. 15 [33]P eck,A . E.(1 976)\ Futuresmarkets,SupplyR esponse,andP riceStability," Q uarterlyJournalofEconomics,90 ,40 7-433. [34]P indyck,R .(1 984)\ R isk,In° ationandtheStockM arket,"A merican EconomicR eview,74,335-351 . [35]P ° eiderer, P.(1 984)\ T he V olume ofT rade and V ariability ofP rices: A FrameworkforA nalysis inN oisyR ationalExpectations Equilibria,"workingpaper,Stanford U niversity. [36]R oss,S. A .(1 989)\ InformationandV olatility:T heno-arbitrageM artingale A pproach toT imingand R esolution Irrelevancy,"JournalofFinance,44, 1 -1 7. [37]Singleton, K.(1 987)\ Speculation and the V olatility ofForeign Currency Exchange R ates,"in Carnegie-R ochesterConference on P ublic P olicy 26, Edited byKarlB runnerandA lan H .M eltzer,N orth-H olland,9-56. [38]Stein,J. C.(1 987) \ InformationExternalities andW elfare-R educingSpeculation,"JournalofP oliticalEconomy,95,1 1 23-1 1 45. [39]T auchen,G . E.andM .P itts (1 983)\ T heP ricevariability-volumerelationship on Speculativemarkets,"Econometrica,51 ,485-50 5. [40 ]T urnovsky,S. J.(1 979)\ Futures M arkets,P rivate Storage,and P rice Stabilization,"JournalofP ublicEconomics,1 2,30 1 -327. [41 ]V eljanovski, C. G .(1 986)\ A n InstituionalA nalysis ofFutures Contracting,"in G oss, B . A . , Ed. , Futures M arkets:T heirEstablishmentand P erformance(Sydney:Croom H elm,1 986),1 3-40 . 16 T able1 :D escriptiveStatistics oftheJSE Futures Indexes A L SI40 SampleSize mean V ariance G L D I1 0 1 1 72 0. 015 0. 362 Skewness -1 . 20 0 iid t-statistic (-1 6. 754)c G M M t-statistic (-1 . 243) Kurtosis iid t-statistic G M M t-statistic L ags IN D I25 1 1 72 0. 005 0. 924 1 1 72 0. 014 0. 429 0. 597 -0 . 899 (8. 339)c (-1 2. 453)c (3. 746)c (-1 . 0 58) 1 4. 30 5 2. 61 9 (99. 759)c (1 8. 236)c (2. 1 37)b (6. 21 5)c 1 2. 427 (86. 656)c (2. 1 67)b A utocorrelations 1 2 3 4 5 0. 1 1 3c 0. 0 53c -0 . 0 22 -0 . 0 73b -0 . 015 0. 1 51 c 0. 016 0. 008 0. 012 -0 . 0 22 0. 0 67b 0. 011 -0 . 0 44 -0 . 0 39 0. 015 Q R (5) Q R (1 0 ) 25. 346c 27. 644c 27. 926c 34. 40 9c 9. 655a 1 3. 220 c Q R 2 (5) Q R 2 (1 0 ) 30 4. 75c 321 . 67c 275. 42c 397. 96c 284. 66c 31 9. 44c Superscripta,b andcdenotestatisticalsigni¯canceatthe1 0 %,5% and1 % levelofsigni¯cance,respectively. 17 T able2:T heCorrelation M atrixfortheA L SI40 Index R eturn A bs(R t) M A X V R eturn 1. 000 A bs(ret) -0 . 1 78c PX V UXV M A FV P FV U FV M A OI PO I 1. 000 M A XV PXV UXV 0. 011 -0 . 018 0. 0 33 0. 1 66c 0. 1 44c 0. 1 48 c M A FV P FV U FV -0 . 002 0. 006 0. 018 0. 1 61 c 0. 0 55a 0. 237c 0. 931 c -0 . 1 72c -0 . 006 -0 . 0 26 0. 40 3c -0 . 019 0. 015 0. 1 23c 0. 237c 1. 000 -0 . 1 72c -0 . 0 32 1. 000 -0 . 0 35 1. 000 M A OI PO I UOI 0. 008 -0 . 014 -0 . 015 0. 20 0 c -0 . 000 0. 0 33 0. 957c 0. 014 -0 . 006 0. 0 36 -0 . 0 34 0. 008 0. 0 29 0. 0 78 b -0 . 1 30 c 0. 929c 0. 0 32 -0 . 0 21 -0 . 1 29c 0. 20 1 c -0 . 0 93c -0 . 004 0. 1 0 4c 0. 0 57b 1. 000 0. 0 34 -0 . 0 22 1. 000 -0 . 1 0 6c PT UT -0 . 0 29 -0 . 0 25 0. 1 79c 0. 350 c 0. 011 -0 . 001 0. 41 3c 0. 016 0. 1 0 7c 0. 288 c 0. 0 60 b 0. 014 0. 762c 0. 0 65b 0. 0 26 0. 71 1 c -0 . 0 48 0. 017 0. 1 86c 0. 0 99c 1. 000 -0 . 0 30 1. 000 0. 0 0 7 -0 . 0 27 1. 000 T he pre¯xes M A , P and U denote moving average, predicted and unexpected,respectively,and thesu±xes X V ,FV ,O I and T denoteindex volume, futures volume,open interestand transactions,respectively.A superscripta,b and cdenotes signi¯cance atthe 1 0 %, 5% and 1 % levelofsigni¯cance forthe nullhypothesis thatthecorrelation equals zero. 18 T able3:M odelEstimates fortheA L SI40 Index M odel1 M odel2 M odel3 M odel4 M odel5 M odel6 Intercept 0. 325c 0. 41 3c 0. 41 2c 0. 40 4c 0. 425c 0. 535c ¾ bt¡1 0. 221 c 0. 1 85c 0. 1 97c 0. 20 8 c 0. 1 96c 0. 1 70 c ¾ bt¡2 0. 0 93c 0. 0 82a 0. 0 98 c 0. 0 89a 0. 1 0 4c 0. 0 87b ¾ bt¡3 0. 21 4c 0. 20 4c 0. 20 0 c 0. 1 96c 0. 1 96c 0. 1 83c ¾ bt¡4 0. 0 59a 0. 0 49 0. 0 48 0. 0 48 0. 0 52 0. 0 43 0. 0 55a 0. 0 41 0. 0 49 0. 0 43 0. 0 60 0. 0 46 U t¡1 -0 . 1 1 7c -0 . 1 32a -0 . 1 48 b -0 . 1 1 6a -0 . 1 47b -0 . 1 56b U t¡2 -0 . 0 649b -0 . 0 68 b -0 . 0 53a -0 . 0 66b -0 . 0 43 -0 . 0 48 U t¡3 -0 . 0 28 -0 . 0 41 -0 . 0 38 -0 . 0 33 -0 . 0 35 -0 . 0 41 U t¡4 -0 . 017 -0 . 019 -0 . 009 -0 . 016 -0 . 001 -0 . 005 U t¡5 0. 0 49b 0. 0 47 0. 0 52 0. 0 43 0. 0 58 0. 0 51 M onday 0. 0 65 0. 1 89 0. 1 69 0. 0 74 0. 103 0. 1 65 T uesday 0. 115 0. 118 0. 0 75 0. 112 -0 . 0 22 -0 . 013 0. 222b 0. 20 2a 0. 111 0. 20 1 a 0. 0 47 0. 017 0. 018 -0 . 019 -0 . 0 47 0. 017 -0 . 0 57 ¾ bt¡5 W ednesday T hursday 19 -0 . 0 878 T able3 (continued):M odelEstimates oftheA L SI40 M odel1 M odel2 M odel3 M odel4 M odel5 M odel6 M A XV 0. 1 33c -0 . 0 73 PXV 0. 51 3c 0. 51 4b UXV 0. 437c 0. 1 77 0. 1 78 c -0 . 1 84 P FV 0. 006 -0 . 1 24 U FV 0. 552c 0. 0 85 M A FV M A OI 0. 1 90 c 0. 424b PO I -0 . 0 96 -0 . 349b UOI 0. 678 0. 825 PT -0 . 0 90 -0 . 018 UT 0. 967c 0. 981 c 0. 228 0. 30 9 0. 31 8 23. 971 c 1 8. 50 8 c 31 . 689c 1 9. 784c 0. 81 6 0. 797 0. 843 0. 955 adj-R 2 F-statistic L R variance 0. 235 0. 246 25. 1 76c 20 . 239c 0. 90 8 0. 280 0. 781 T able3 contains theestimates ofmodels (1 ),(2)and (3).T heregressors ¾ bt¡j, j = 1 ;2:::;5 arethe¯velagged conditionalvariances and U t¡j,j = 1 ;2:::5 are the¯ve lagged residuals from (1 ).Fourday-of-the-weekdummies areincluded to mop up any seasonality in the conditionalvariance. T he pre¯xes M A , P and U denotemovingaverage,predicted and unexpected,respectively,and the su±xes X V ,FV ,O I and T denoteindexvolume,futures volume,open interest and T ransactions,respectively.A superscripta,band cdenotes signi¯canceat the1 0 %,5% and 1 % levelofsigni¯canceforthehypothesis thatthecoe±cient equals zero. 20 T able4:T heCorrelation M atrixfortheG L D I1 0 Index Rt Rt 1. 000 A bs(R t) -0 . 1 39c A bs(R t) PX V UXV P FV U FV PO I U PO I PT 1. 000 PXV UXV -0 . 011 0. 21 2c 0. 20 7c 0. 256c 1. 000 -0 . 0 77b 1. 000 P FV U FV -0 . 0 41 0. 1 0 8c -0 . 0 44 0. 258c 0. 0 84b 0. 003 0. 0 23 0. 31 3c 1. 000 -0 . 0 53 1. 000 PO I UOI -0 . 0 61 0. 1 33c -0 . 0 28 0. 006 0. 0 95b 0. 0 64 -0 . 0 22 -0 . 1 56c 0. 30 5c -0 . 0 66a 0. 0 92b 0. 0 23 1. 000 -0 . 1 84c 1. 000 PT UT 0. 0 46 0. 1 1 7c 0. 0 0 7 -0 . 0 72a c 0. 374 -0 . 009 0. 793c -0 . 1 59c 0. 0 43 0. 782c 0. 1 36c 0. 0 34 -0 . 0 20 -0 . 018 0. 017 0. 387c T hepre¯xes M A ,P and U denotemovingaverage,predicted and unexpected, respectively, and the su±xes X V ,FV ,O I and T denote index volume,futures volume,open interestand transactions, respectively.A superscripta, band c denotes signi¯cance atthe 1 0 %, 5% and 1 % levelofsigni¯cance forthe null hypothesis thatthe correlation equals zero. T he pre¯xes M A ,P and U denote moving average, predicted and unexpected, respectively, and the su±xes X V , FV ,O I and T denote index volume,futures volume,open interestand T ransactions,respectively.A superscripta,b and cdenotes signi¯canceatthe1 0 %, 5% and 1 % levelofsigni¯cance forthe hypothesis thatthe correlation equals zero. 21 1. 000 -0 . 1 23c T able5:M odelEstimates fortheG L D I1 0 Index M odel1 M odel2 M odel3 M odel4 M odel5 M odel6 Intercept 1. 0 61 c 1. 324c 1. 1 92c 1. 223c 1. 0 29c 1. 428 c ¾ bt¡1 0. 1 69c 0. 0 96b 0. 0 98 b 0. 1 37c 0. 0 78a 0. 0 26 ¾ bt¡2 0. 0 98 b -0 . 0 27a -0 . 1 0 6b 0. 0 91 b 0. 1 42c 0. 1 0 2b ¾ bt¡3 -0 . 006 -0 . 0 27 0. 016 -0 . 0 22 -0 . 003 -0 . 0 32 ¾ bt¡4 0. 1 20 c 0. 1 0 4c 0. 1 1 6c 0. 111b 0. 1 1 7c 0. 0 81 b ¾ bt¡5 0. 1 1 2c 0. 0 93b 0. 1 1 7c 0. 0 98b 0. 1 1 8c 0. 0 89b U t¡1 -0 . 017 -0 . 012 -0 . 0 37 -0 . 1 20 b -0 . 0 34 -0 . 0 40 U t¡2 -0 . 011 0. 006 -0 . 006 -0 . 0 23 -0 . 0 21 -0 . 015 U t¡3 0. 0 34 0. 0 25 0. 0 43 0. 007 0. 0 31 0. 0 29 U t¡4 0. 0 24 0. 0 31 0. 0 26 0. 018 0. 005 0. 010 U t¡5 -0 . 0 37 -0 . 0 22 -0 . 0 46 -0 . 0 48 -0 . 0 49 -0 . 0 46 M onday 0. 0 41 0. 356a 0. 1 71 -0 . 001 0. 353 0. 444 T uesday -0 . 0 62 -0 . 0 81 -0 . 1 36 -0 . 0 84 -0 . 006 -0 . 003 W ednesday -0 . 0 68 -0 . 0 80 -0 . 0 31 -0 . 1 79 0. 0 96 0. 0 64 T hursday -0 . 1 46 -0 . 1 83 -0 . 1 64 -0 . 1 24 0. 014 -0 . 019 22 T able5 (continued):M odelEstimates fortheG L FI1 0 Index M odel1 M odel2 M odel3 M odel4 M odel5 M odel6 PXV 0. 591 c 0. 71 6c UXV 1. 0 61 c 0. 674c P FV -0 . 1 97 -0 . 427b U FV 0. 399c -0 . 1 92c PO I -0 . 1 78 -0 . 1 87 UOI 0. 540 1. 298 b PT 0. 0 55 0. 377b UT 0. 979c 0. 994c adj-R 2 F-statistic L R variance 0. 0 71 0. 1 65 0. 109 0. 0 45 0. 1 69 0. 21 8 4. 71 9c 9. 371 c 5. 767c 2. 844c 7. 360 c 7. 669c 2. 118 1. 80 4 1. 538 2. 1 72 1. 754 1. 961 T able3 contains theestimates ofmodels (1 ),(2)and (3).T heregressors ¾ bt¡j, j = 1 ;2:::;5 arethe¯velagged conditionalvariances and U t¡j,j = 1 ;2:::5 are the¯ve lagged residuals from (1 ).Fourday-of-the-weekdummies areincluded to mop up any seasonality in the conditionalvariance. T he pre¯xes M A , P and U denotemovingaverage,predicted and unexpected,respectively,and the su±xes X V ,FV ,O I and T denoteindexvolume,futures volume,open interest and T ransactions,respectively.A superscripta,band cdenotes signi¯canceat the1 0 %,5% and 1 % levelofsigni¯canceforthehypothesis thatthecoe±cient equals zero. 23 T able6:T heCorrelation M atrixfortheIN D I25 Index Rt Rt 1. 000 A bs(R t) -0 . 1 46c A bs(R t) M A X V PX V UXV M A FV P FV U FV M A OI PO I 1. 000 M A XV PXV UXV 0. 002 -0 . 0 34 0. 0 20 0. 21 1 c 0. 0 40 0. 0 96c M A FV P FV U FV 0. 014 -0 . 0 42 -0 . 0 43 0. 256c 0. 1 34c 0. 240 c 0. 924c -0 . 0 46 0. 0 38 -0 . 1 1 9c 0. 373c -0 . 0 69b -0 . 0 40 0. 1 1 9c 0. 277c 1. 000 -0 . 0 82c -0 . 003 1. 000 -0 . 012 1. 000 M A OI PO I UOI 0. 015 -0 . 019 -0 . 0 29 0. 20 3c 0. 0 45 0. 0 21 0. 80 0 c 0. 003 0. 016 0. 016 -0 . 0 55a -0 . 0 26 0. 014 0. 014 -0 . 1 41 c 0. 821 c 0. 1 34c -0 . 011 -0 . 1 1 6c 0. 255c -0 . 0 96c 0. 008 0. 1 0 4c 0. 0 31 1. 000 -0 . 1 0 4c -0 . 0 36 1. 000 -0 . 1 33 M AT PT UT -0 . 017 -0 . 0 67 -0 . 0 93 0. 342c 0. 1 24c 0. 1 93c 0. 464c -0 . 0 78 b 0. 0 26 -0 . 0 89c 0. 349c -0 . 1 1 9c -0 . 0 30 0. 1 1 6c 0. 275c 0. 693c -0 . 1 27c -0 . 0 20 -0 . 0 49 0. 861 c -0 . 0 53a -0 . 0 46 0. 101c 0. 769c 0. 539c -0 . 1 40 c -0 . 002 0. 1 69 0. 281 0. 0 65 1. 000 -0 . 101c 1. 000 -0 . 0 27 -0 . 110c 1. 000 T hepre¯xes M A ,P and U denotemovingaverage,predicted and unexpected, respectively, and the su±xes X V ,FV ,O I and T denote index volume,futures volume,open interestand transactions, respectively.A superscripta, band c denotes signi¯cance atthe 1 0 %, 5% and 1 % levelofsigni¯cance forthe null hypothesis thatthecorrelation equals zero. 24 T able7:M odelEstimates fortheIN D I25 Index M odel1 M odel2 M odel3 M odel4 M odel5 M odel6 Intercept 0. 323c 0. 396c 0. 51 1 c 0. 40 9c 0. 621 c 0. 655c ¾ bt¡1 0. 248 c 0. 1 224c 0. 20 7c 0. 238 c 0. 1 75c 0. 1 72c 0. 0 36 0. 0 32 0. 017 0. 0 23 0. 017 -0 . 000 ¾ bt¡3 0. 229c 0. 222c 0. 224c 0. 222c 0. 1 98 c 0. 20 0 c ¾ bt¡4 0. 0 62 0. 0 60 0. 0 30 0. 0 53 0. 0 22 0. 014 ¾ bt¡5 0. 0 71 0. 0 50 0. 0 61 0. 0 56 0. 0 42 0. 0 36 U t¡1 -0 . 1 1 5b -0 . 1 29b -0 . 1 24b -0 . 1 20 b -0 . 1 24c -0 . 1 24c U t¡2 -0 . 018 -0 . 0 24 -0 . 016 -0 . 0 23 -0 . 0 21 -0 . 018 U t¡3 -0 . 006 -0 . 007 -0 . 005 0. 007 -0 . 001 -0 . 004 U t¡4 -0 . 008 -0 . 006 -0 . 002 0. 003 0. 001 -0 . 002 U t¡5 0. 0 26 0. 0 26 0. 0 23 0. 0 24 0. 0 28 0. 0 22 M onday 0. 1 28 0. 21 9 0. 222a 0. 1 33 0. 1 96 0. 234a T uesday 0. 1 29 0. 1 37 0. 0 41 0. 1 36 -0 . 0 36 0. 0 40 0. 232b 0. 220 a 0. 114 0. 21 0 a 0. 1 21 0. 103 0. 0 71 0. 0 46 -0 . 0 47 0. 0 56 -0 . 0 47 -0 . 0 75 ¾ bt¡2 W ednesday T hursday 25 T able7(continued):M odelEstimates fortheIN D I25 Index M odel1 M odel2 M odel3 M odel4 M odel5 M odel6 0. 262c -0 . 0 88 PXV 0. 21 0 0. 31 3 UXV 0. 30 6c 0. 0 68 M A XV 0. 482c 0. 1 75 P FV 0. 1 31 0. 230 U FV 0. 784c 0. 669c M A FV 0. 355c 0. 273 PO I 0. 1 32 -0 . 31 9b UOI 0. 288 0. 265 M A OI M AT 1. 569c 1. 684c PT 0. 272b -0 . 001 UT 1. 0 28 c 0. 325 0. 290 adj-R 2 F-statistic L R variance 0. 225 0. 232 0. 273 0. 222 0. 274 23. 741 c 1 8. 867c 23. 1 69c 1 7. 894c 23. 31 9c 1 6. 797c 0. 61 5 0. 71 6 0. 898 0. 757 0. 992 1. 0 43 T able3 contains theestimates ofmodels (1 ),(2)and (3).T heregressors ¾ bt¡j, j = 1 ;2:::;5 arethe¯velagged conditionalvariances and U t¡j,j = 1 ;2:::5 are the¯ve lagged residuals from (1 ).Fourday-of-the-weekdummies areincluded to mop up any seasonality in the conditionalvariance. T he pre¯xes M A , P and U denotemovingaverage,predicted and unexpected,respectively,and the su±xes X V ,FV ,O I and T denoteindexvolume,futures volume,open interest and T ransactions,respectively.A superscripta,band cdenotes signi¯canceat 26 the1 0 %,5% and 1 % levelofsigni¯canceforthehypothesis thatthecoe±cient equals zero. 27