Internatonal Journal of Busness and Economcs, 00, Vol., No., 9-5 Maret Openng and Stoc Maret Behavor: Tawan s Experence Q L * Department of Economcs, Texas A&M Unversty, U.S.A. and Department of Economcs, Tsnghua Unversty, P.R.C. Astract Ths paper studes the mpact of maret openng on stoc maret ehavor. Tawan opened ts stoc maret n January, 99. Usng stoc maret data from Tawan for oth efore and after the openng event, we found that whle there s no sgnfcant changes n the stoc mean returns, volatlty s sgnfcantly reduced three months after Tawan opened ts stoc maret. As a result, the maret effcency, as measured y the Sharpe rato, sgnfcantly ncreases three months after the openng event. Key words: maret openng; volatlty; maret effcency JEL classfcaton: G4. Introducton Captal molty has een a sgnfcant feature n the gloal fnancal marets. Many countres opened up ther stoc marets to foregn nvestors n the past twenty to thrty years. It s mportant for governments and nvestors to evaluate the mpact of the maret openng on the maret ehavor. For example, does the openng ncrease speculaton n the maret and thus mae the maret more volatle and more vulnerale to foregn shocs? Kawaatsu and Morey (999a, 999) examne whether emergng maret equty prces have ecome more effcent after fnancal leralzaton. They fnd that leralzaton does not mprove the effcency of emergng marets. In fact, they fnd Receved June 5, 00, accepted Octoer 5, 00. *Correspondence to: Department of Economcs, Texas A&M Unversty, College Staton, TX 77843, U.S.A. Emal: q@econ.tamu.edu. I would le to than two anonymous referees and Thomas C. Chang for helpful comments that lead to an mproved verson of the paper. Clff J. Huang should e credted for encouragng me to mae ths sumsson. I would also le to than John Grffn and Dong L for ther helpful comments on an earler verson of ths paper. Ths research s supported y the Socal Scences and Humantes Research Councl of Canada, the Natural Scences Engneerng Research Councl of Canada, the Ontaro Premer s Research Excellence Award, the Bush program n Economcs of Pulc Polcy, and the Prvate Enterprse Research Center, Texas A&M Unversty.
0 Internatonal Journal of Busness and Economcs that the marets were already effcent pror to the actual leralzaton. Henry (000) fnds that emergng maret openngs are assocated wth a 3.3% ncrease per month n maret returns durng an eght-month wndow leadng up to the mplementaton of ts ntal stoc maret leralzaton. Henry argues that ths s due to the fact that stoc maret leralzaton may reduce the leralzng country s cost of equty captal y allowng for rs sharng etween domestc and foregn agents. In ths paper we nvestgate the mpact of maret openng on the maret ehavor n the Tawan Stoc Exchange. We wll examne the change n maret return, volatlty, and maret effcency. In the next secton we wll ntroduce the methodology. Secton 3 presents our results and s followed y some concludng remars n Secton 4.. Methodology We use the Tawan Stoc Exchange (TSE) ndex data to emprcally compare the maret ehavor efore and after the event day (openng day). Specfcally we examne the change of maret return, volatlty, and maret effcency. Tawan opened ts stoc maret on January, 99. We use data from January, 990 to January, 99. Our data covers a two years perod wth one year of data efore the openng event and one year of data after t. The data are retreved from Datastream. We use contnuously-compounded daly returns n ths study: r t = ln( pt / pt ), where p s the TSE ndex level on day t.. Volatlty t We examne the change of volatlty y comparng the varance of the returns n efore and after the event day. We wll use { r t} t= to denote the n days of r t n mmedately efore the openng event, and { r t} t= to denote the n days of r t mmedately after the openng day. These two groups data are ndexed y =,. Let σ denote the varance of r t, =,. Then the null hypothess of no volatlty change s H a 0 a : σ = σ aganst the alternatve hypothess H : σ σ. Note that r t s the t th oservaton n the th group ( =, ; t =,..., n ), and defne xt = rt r., where r = n. n t = rt s the sample mean of n { rt } t= ( =, ). The test statstc [Levene statstc, see Levene (960)] we use s the rato of etween-group to wthn-group varance for x : n ( n = ( x. x.. ) x = =.) t t x = L =, () ( n ) t
Q L where x. = n n t= x t n = x = t= t and x... () n + n a Under H 0 and the assumpton that r t s ndependently and normally dstruted, the test statstc L has an F (, n + n) dstruton. However, as ponted out y Brown and Forsythe (974), the Levene statstc L s senstve to non-normalty n the dstruton of r t. It s well nown that stoc returns, especally daly returns, have fat-taled dstrutons and hence are non-normal. To deal wth non-normalty of stoc return data, we wll use a ootstrap method to otan the crtcal values of the test statstc L. The ootstrap steps are as follows: Collect data for days efore and after the openng event, the daly returns are used to compute the Levene statstc. The ootstrap sample s randomly drawn wth replacement and susequently dvded nto two groups, wth the frst assgned to group and the remanng to group. The Levene statstc s computed usng the ootstrap group and group data. Ths procedure s repeated,000 tmes to gve,000 ootstrap Levene statstcs whch are used to yeld ootstrap sgnfcance levels for the orgnal Levene statstc. See Efron and Tshran (993) for general dscussons on usng the ootstrap method to approxmate fnte sample null dstrutons of test statstcs. Smlar approaches are also used y L, Ln, and L (997) and L and Ln (998) to examne the volatlty changes n Chna s stoc maret.. Mean Return To assess the mpact of maret openngs on mean return, we compute the mean returns efore and after the event. Denotng these y µ = E( r t ) ( =, ), we test the null hypothess of equal mean return: H 0 : µ = µ aganst the alternatve hypothess H : µ µ. For tradng days efore and days after the event, we denote the return y r, r,..., r ) and r, r,..., r ), respectvely. The test statstc s ( ( r r t r ( ) =, ˆ (3) σ +σˆ where r and σ ˆ are the sample mean and sample standard devaton of { rt} t=, respectvely ( =, ). To address the non-normal dstruton of stoc returns, we test H 0 usng ootstrap sgnfcance levels. The ootstrap procedure s the same as dscussed n the prevous secton except that now we use the ootstrap sample to
Internatonal Journal of Busness and Economcs compute the t (r) test defned n (3). The ootstrap procedure s repeated,000 tmes to gve,000 ootstrap t (r) statstcs whch are used to yeld ootstrap sgnfcance levels for the orgnal t(r) statstc..3 Sharpe Rato To examne the reward-to-rs rato of Tawan s stoc maret after the openng event, we compare the Sharpe rato change efore and after the event day. A hgher return or a smaller volatlty leads to a hgher Sharpe rato. Recall that r = n t= r t s the sample mean and σ s the sample standard devaton of { rt} t=. Also let r f, t denote the contnuously-compounded daly rs-free return on day t of the th group,.e., r f, t = ln( mt mt ), where m t s the rs-free ndex on day t for the th group. Let r f, = E[ rf, t ] denote the mean value of the rs-free return for group. Group s Sharpe rato s defned as the rato of excess return to return s standard devaton µ = E( r )] [ t SR µ rf, =, σ where µ = E( r t ). The null hypothess s H c : SR = SR aganst the alternatve hypothess H c : SR SR. c Our test statstc for testng H 0 s gven y where 0 SR = ( SR SR ), D,, SR ) σˆ,, = ( r rf, wth r = n t= r t and r = f, n t= r f, t. It can e shown that SR D has an asymptotc (as ) standard normal c dstruton under H 0. To etter approxmate the fnte sample null dstruton of the test statstc, we use the ootstrap method to otan the sgnfcant level of the test statstc. The ootstrap procedure s smlar to that dscussed earler. 3. Emprcal Results We emprcally examne the stoc maret ehavor change n Tawan s stoc maret after the openng event.
Q L 3 3. Maret Volatlty Frst we report the result of testng whether the volatlty changes after the openng event. In Tale we report the sample standard devaton of stoc returns for days efore and after the openng event, where = 5, 30, 50, 75, 00, 50, 00. As can e seen from Tale, except for the = ± 5 case, sample standard devatons after the event day are all smaller than the correspondng counterparts efore the event day. The Levene statstc and ts sgnfcance levels (usng the ootstrap method) are also reported n Tale. We reject the null of σ = σ n favor of σ < σ for 75 at the % level. Ths outcome suggests that y openng the maret, the volatlty of Tawan s stoc maret dd not change for the mmedate three months after the event. However, etween three months to one year after the openng event, Tawan s stoc volatlty s sgnfcantly reduced. Tale. Testng Volatlty Change ± 5 ± 30 ± 50 ± 75 ± 00 ± 50 ± 00 Std Dev (Before) 0.0445 0.0495 0.0493 0.0488 0.049 0.0486 0.044 Std Dev (After) 0.056 0.0447 0.040 0.0354 0.0330 0.098 0.087 Levene Statstc 0.540 0.3840.8840 9.800 6.0700 6.900 48.9300 Sgnfcance Level 0.5930 0.5490 0.850 0.000 0.0000 0.0000 0.0000 3. Mean Return The computed mean returns r, the test statstc t (r) for testng H 0, and the ootstrap sgnfcance level (usng days data) are reported n Tale. Tale. Testng Mean Return Change ± 5 ± 30 ± 50 ± 75 ± 00 ± 50 ± 00 Mean 0 (Before) -0.64 0.54.095 0.36-0.76-0.45-0.463 Mean 0 (After) -.5 0.96 0.099 0.33 0.3 0.078-0.07 Test Statstc 0.67 0.0.08 0.004-0.687 -.080 -.98 Sgnfcant Level 0.397 0.407 0.9 0.53 0.744 0.858 0.88 We oserve a systematc pattern that, for t 75, the mean returns after the openng are less than that efore the openng. However, for t 00, the mean returns after the openng ecome larger than those efore the openng. However, the sgnfcance levels (usng the ootstrap method) show that, for all the sample perod consdered, we cannot reject the null hypothess of µ = µ. Therefore, we conclude that there s no sgnfcant change n mean return one year efore and one year after the openng event. 3.3 The Sharpe Rato From the results of Tales and, we now that the varance (or standard
4 Internatonal Journal of Busness and Economcs devaton) of the return s sgnfcantly reduced after the openng event for 75, whle we cannot reject the hypothess that the mean returns reman unchanged efore and after the openng event for all the -values consdered. From these results one would expect that the Sharpe rato should e sgnfcantly ncreased after the openng day for 75. Tale 3 reports the result of testng the equalty of Sharpe rato usng days data. We oserve that whle we cannot reject the null hypothess of H c 0 for c 75, H 0 s rejected for 00 at the % level. Therefore, we conclude that the Sharpe rato s sgnfcantly ncreased three months after Tawan opened ts maret. Tale 3. Sharpe Rato Change ± 5 ± 30 ± 50 ± 75 ± 00 ± 50 ± 00 S-Rato (Before) -0.440 0.00 0.0 0.0648-0.0358-0.0875-0.050 S-Rato (After) -0.0 0.066 0.046 0.0884 0.070 0.06-0.0058 Sgnfcant Level 0.300 0.4690 0.5680 0.460 0.000 0.0000 0.000 The aove results show that, whle there s no sgnfcant change n mean return efore and after the openng day for Tawan s stoc exchange maret, volatlty sgnfcantly decreased three months after Tawan opened ts stoc maret. As a result, the Sharpe rato s sgnfcantly ncreased three months after the openng event. 4. Concludng Remars It should e mentoned that there mght e other factors affectng stoc return and volatlty. For example, the Persan Gulf War also occurred around the event date (I owe ths oservaton to a referee). Iraq occuped Kuwat on August, 990, and the Persan Gulf War too place from January 6, 99 to Feruary 8, 99. All these occurred durng our data perod. Of course t s dffcult, f not mpossle, to dsentangle the effect of the Persan Gulf War from the openng event. One way to mnmze the Persan Gulf War effect s to remove the data m days efore and after the event day. For example f we choose m = 00 or m = 50, we remove data three months or fve months efore and after January, 99. The Persan Gulf War should have much less effect on Tawan s stoc maret for the remanng perod. The estmaton results usng the remanng data stll show a sgnfcant reducton n volatlty, and a sgnfcant ncrease n Sharpe-Rato. Thus, our fndngs seem to e roust to the Persan Gulf War effect. In summary, the results ndcate that foregn nvestors have had a stalzng nfluence on the Tawan stoc maret. The leralzaton of the captal maret enales rs sharng etween domestc and foregn nvestors. Large nternatonal nvestors tend to study companes more thoroughly. The nvolvement of foregn nvestors means a etter dssemnaton of nformaton, whch leads to a more effcent maret.
Q L 5 References Brown, M. B. and A. B. Forsythe, (974), Roust Tests for the Equalty of Varance, Journal of Amercan Statstcal Assocaton, 69, 364-367. Efron, B. and R. J. Tshran, (993), An Introducton to Bootstrap. London: Chapman & Hall. Henry, P. B., (000), Stoc Maret Leralzaton, Economc Reform, and Emergng Maret Equty, Journal of Fnance,, 59-564. Levene, H., (960), Roust Test for Equalty of Varance, n Contrutons to Proalty and Statstcs, I. Oln, ed.,palo Alto, Calforna: Stanford Unversty Press, 78-9. L, D., S. K. Ln, and C. L, (997), The Impact of Settlement Tme on the Volatlty of Stoc Marets, Appled Fnance Economcs, 7, 689-694. L, D. and S. K. Ln, (998), Transacton Tax and Stoc Maret Behavor: Evdence from an Emergng Maret, unpulshed manuscrpt. Kawaatsu, H. and M. R. Morey, (999a), Fnancal Leralzaton and Stoc Maret Effcency: An Emprcal Examnaton of Nne Emergng Maret Countres, Journal of Multnatonal Fnancal Management, 9, 353-37. Kawaatsu, H. and M. R. Morey, (999), An Emprcal Examnaton of Fnancal Leralzaton and the Effcency of Emergng Maret Stoc Prces, Journal of Fnancal Research,, 385-4.