Testing the weak efficient market hypothesis using Bangladeshi panel data

Transcription

1 Chu V. Nguyen (USA), Muhammad Mahboob Al (Bangladesh) Testng the weak effcent market hypothess usng Bangladesh panel data Abstract Ths emprcal study nvestgates whether the Dhaka Stock Exchange market n Bangladesh s weakly effcent by modfyng and estmatng Dockery and Kavussanos multvarate model usng a set of panel data. Consstent wth a pror expectaton, and prevous research results, the fndngs suggest that the Dhaka Stock Market s not nformatonally effcent. The results also ndcate that when the number of stocks ncluded n the sample exceeds 3, the null hypothess of the effcent market hypothess s rejected throughout. A well developed market economy s a necessary condton for an effcent equty market. Thus, to buld an effcent stock market, Bangladesh authorty should frst concentrate ts effort on developng ts market nfrastructure. Keywords: effcent market hypothess, panel data, Dhaka Stock Exchange, regresson model. JEL Classfcaton: C01, E32, E44, E51. Introducton The growth of equty markets and the globalzaton of fnancal markets are often the subject of major research studes n developng countres. Addtonally, testng dfferent forms of the effcent market hypothess seems to be the most popular theme of these emprcal nvestgatons. Insttutonal and large nvestors strve for portfolos wth nternatonal stocks because of the potental hgher returns. Even wth the 5 percent and 10 percent stpulatons n the U.S. mutual funds, transactons by these nvestors are nowhere near odd lot orders. These, n turn, prevent most developng economes from beneftng from recent ncreases n the nternatonalzaton of portfolo nvestments. Ths phenomenon pushes most developng economes nto a downward spral or a vcous crcle: because ther equty markets are not deep and broad enough. In addton, they lack the necessary nfrastructures to attract nternatonal portfolo nvestments; but because they cannot attract the nternatonal portfolo nvestments, ther markets cannot be deep and broad and they cannot mprove ther nfrastructures. Informaton on the characterstcs of markets and the lsted companes n the equty markets cannot be dssmlated quckly due to nsuffcent electronc and other nfrastructure ssues that renders equty markets n developng countres not even weakly effcent. Earler research fndngs, ncludng that of Hoque, Kabr, and Rahman (2005) reported that the Bangladesh and the Indan equty markets show sgnfcant mean-reverson and predctable behavors n ther daly return seres. However, the authors found that the U.S. and Japanese stock markets exhbt some mean-reverson, but largely follow unpredctable patterns over the perod. Ths study proposes to assess whether the Dhaka Stock Exchange exhbts a weak form of the effcent market hypothess. Chu V. Nguyen, Muhammad Mahboob Al, Tha paper s organzed as follow. Secton 1 presents a lterature revew. Secton 2 descrbes Bangladesh equty market. Secton 3 presents methodology. Secton 4 show data. Secton 5 reports and dscusses the emprcal results. The fnal secton provdes the concludng remarks and some polcy mplcatons. 1. Lterature revew The effcent market hypothess dates back to when the busness cycle theorsts tred to analyze the behavor of stock markets over tme n search of ndcators of the busness cycles. Kendall (1953) examned the stock market behavor, and to hs surprse, he could not dentfy any predctable patterns n the stock prces. Over tme, as equtes became ncreasngly mportant sources of corporate fnance, theorsts such as Fama (1970) and numerous others began to hypothesze and to emprcally test stock prce behavors leadng to the formulaton of effcent market hypothess. The two most common procedures for testng the effcent market hypothess are the runs test DeFusco, McLeavey, Pnto and Runkle (2004) and seral correlaton analyss (Conrad and Kaul, 1988; Lo and McKnlay, 1988). The correlaton analyss tests for the sgnfcance of postve or negatve correlatons n stock returns over tme. Specfcally, the correlaton tests for ndependence,.e., random walk, by statstcally determnng whether the rate of return on day t correlates to the return on day t - 1, t - 2, or t - 3. The runs test examnes a seres of stock prce changes and these changes are desgnated a plus (+) f t s an ncrease n prce or a mnus ( ) f t s a decrease n stock prce. The result s a set of pluses and mnuses. A run occurs when two consecutve changes are the same: two or more consecutve postve or negatve changes are defned as one run. When the prce changes n a dfferent drecton, for example, when a negatve prce change s followed by a postve prce change, the run ends and a new run may begn. To test for random walk, the number of runs 11

2 for a gven seres s compared to a table of expected values for the number of runs that should occur n a random seres. Emprcal studes, usng these testng procedures, have yelded mxed results. Hstorcally, earler evdence was qute favorable to the hypothess, but n recent years, deeper analyses of evdence cast doubt on ther valdty as a whole. The followng are taken as evdence n favor of the effcent market hypothess. The nvestments n mutual funds cannot be expected to earn an abnormally hgh return. Good performance n the past does not guarantee that the nvestment advsor or a mutual fund wll perform well n the future. Favorable earnngs announcements or announcements of stock splts do not, on average, cause stock prce to ncrease. Future changes n stock prces, for all purposes, are unpredctable. On average, techncal analysts do not fare better than other fnancal analysts nor do they outperform the market. However, small-frm effect, January effect, market overreacton, excessve volatlty, mean reverson, new nformaton not beng mmedately ncorporated n stock prces are taken as evdence aganst the effcent market hypothess. Dockery and Kavussanos (1996) appled a regresson model to the panel data to test the weak form of the effcent market hypothess on the Athens Stock Market. Dockery and Kavussanos began by consderng a smple regresson model, where the current stock prce s regressed on an ntercept constant, the prce n the last perod, and a dsturbance. The authors argued that the necessary condton for the Athens Stock Market to be weakly effcent s that the estmated constant and the estmated coeffcent of the past perod prce cannot be statstcally dfferent from 0 and 1, respectvely. However, they posted that snce the stock prces are not statonary, they have to model ther frst dfferences. To ths end, perodc changes n stock prces are regressed on an ntercept constant and a dsturbance. Dockery and Kavussanos artculated that n the new expresson, the necessary condton for the market to be weakly effcent s that the estmated ntercept constant could not be statstcally dfferent from zero. 2. Bangladesh equty market Formal tradng on the Bangladesh Stock Exchange, known as the Dhaka Stock Exchange (DSE), began n 1956, two years after the establshment of the East Pakstan Stock Exchange Ltd. on Aprl 28, The tradng was suspended for fve years startng from the ndependence war of the country whch held n The Chttagong Stock Exchange (CSE) was establshed n December The Bangladesh Securty and Exchange Commsson (SEC) was establshed on June 8, 1993 by the SEC Act of 1993 to regulate the tradng of stocks, to protect securty nvestors, to ensure proper nsurance and complances, and to promote far, transparent and effcent securty markets (Source: a15.htm). The common characterstcs of equty markets n poor, developng countres are that the markets are narrow and shallow as a consequence of slow progress. These phenomena provde opportuntes for unethcal and even llegal manpulatons often causng stock market crashes. These crashes usually cause severe fnancal damages to nvestors. Ths panful stuaton occurred wth both the Bangladesh stock exchanges n the summer and fall of Durng ths epsode, the DSE ndex ncreased from 832 n January 1966 to a peak of 3,567 on November 14, 1996 and dropped back down to n November To control the fnancal damages and decrease n nvestors confdence caused by the early epsodes leadng to the November 1996 crash and wth support from the Asa Development Bank, the Bangladesh government ntroduced the Captal Market Development Program on November 20, The purpose of the program was to (1) strengthen market regulaton and supervson; (2) develop the stock market nfrastructure; (3) modernze stock market support facltes; (4) ncrease the lmted supply of securtes n the market; (5) develop nsttutonal sources of demand for securtes n the market; and (6) mprove polcy coordnaton (Source: Wse and Al (2007) argued that a strong and vbrant securtes market s an mportant source of captal fundng: nsttutonal and prvate nvestors nvarably become a major source of nvestment captal for prvate frms. Due to the relatvely underdeveloped state of the securtes markets n Bangladesh, long-term ndustral fnancng s supported largely by the bankng sector. As to the Bangladesh equty market, Alom, Khan, Morshed (2010) suggested that the weak-form neffcency of the stock market s most lkely caused by a combnaton of the lack of ts development and polcy choces resultng n napproprate regulatons. Uddn (2009) observed that market neffcency n Bangladesh s more lkely the consequences of an neffcency regulatory system and nformaton transparency. Moreover, as recently as of November 2007, Bepar and Mollk (2008) argue that the DSE s stll n the prmtve stages of development wth only 273 lsted companes whose total market captalzaton to GDP rato equals 16 percent. Out of the lsted companes, as measured by the percentage of the total market captalzaton, the market share of the bankng sector, share of the largest fve sectors 12

3 n the economy, and share of the fve largest corporatons are respectvely 58.8 percent, percent, and percent. These statstcs clearly suggest market concentraton and the domnance of the bankng ndustry. Addtonally, the fscal polcy of the country has been very weak and, hence, not conducve to promotng nvestments n the captal markets, especally n the secondary markets. In lght of the ntroducton of the Captal Market Development Program a decade or so ago and other mprovements n the recent past; the Bangladesh stock market s of partcular nterest for further emprcal research. Ths study dffers from prevous endeavors n whch t modfes Dockery and Kavussanos multvarate model to allow the use of a set of panel data as a set of pooled data from the Dhaka Stock Exchange market to test the weak-form of the effcent market hypothess. Ths nvestgaton also examnes the senstvty of the numbers of stocks ncluded n the sample to the results. 4. Methodology In testng the weak form of the effcent market hypothess usng panel data from Athens Stock Market, Dockery and Kavussanos defned Pt and P, t 1 as the prces of stock at tme t and t - 1 and postulated the followng regresson model: P t P, (1),t 1 t where = 1, 2,.N and t = 1, 2,.I, N s number of ndvdual stocks and I s number of observatons for stocks. and are respectvely ntercept and slope constants. t s a Gaussan error term whch may exhbts contemporaneous correlaton between stocks. The weak effcent market hypothess may be stated as: =1 and =0, H a : 1 and 0. Dockery and Kavussanos (1996, p. 122) argued that t s usually the case that stock prces are nonstatonary, n whch case n order to avod nference problems stock returns are consdered. The authors defned the stock returns as the dfferences between the logarthmc values of the stock prces prevalng n the current perod and the last perod, and rewrote ther regresson equaton as: rt, (2) t where = 1, 2,.N and t = 1, 2, T. They posted that the weak effcent market hypothess becomes: =0, H a : 0. Equaton (2) s a system of N equatons wth cross equaton correlatons allowed n the resduals. Dockery and Kavussanos (1996) artculated that the seemngly unrelated estmaton procedure, proposed by Zeller (1962), would yeld more effcent parameter estmates than the ordnary least squares method. A Wald test statstcs, usng only the unrestrcted model (2), can be used to test the set of null and alternate hypotheses of the effcent market hypothess. Let h ( a ) 0 be the set of restrctons, gven a vector of estmates a, the assocated covarance estmate V (a), and the covarance matrx of the restrcted V[ h( a)] ( h / a)' V ( a)( h / a), the Wald test statstcs, evaluated at the unrestrcted estmate a, s: 1 W h( a )V [ h( a ] h( a )'. (3) Ths test statstcs has an asymptotcal Ch-squared dstrbuton wth the degrees of freedom equal to the number of restrctons (Dockery and Kavussanos, 1996, p. 122). In formulatng ther model, specfed by equaton (2), Dockery and Kavussanos (1996) defned r t as the dfferences between the logarthmc values of the stock prces n the current perod and the prevous perod. Therefore, the value of r t for any stock s nevtably affected by the prce level of that stock, as measured by the unt of currency. Because of the dfferences n ther prces, the same percentage change n the prces of two stocks traded n the prevous day would result n two dfferent quantty changes. Therefore, to account for ths dfference, Dockery and Kavussanos (1996) used the multvarate model n whch each stock ncluded n the sample s modeled and estmated by one of the equatons n the system. However, most of the fnancal reports on changes n stock prces are n percentage changes from one perod to another. If the return s expressed n terms of percentage change, then t can be nterpreted as the return on one unt of currency nvested n a stock, magnfed by one hundred tmes; therefore, the mpacts of the prce level s removed from t. As a result, dfferent magntudes of daly changes reflect only the mpact of the good or bad news about the stock, excludng the mpact of prce level prevalng at the end of the last perod. Let s t be the percentage change from P, t 1 to P t, where Pt and P, t 1 are the prces of stock at tme t and t 1, then equaton (2) can be rewrtten as: 13

4 s t, (4) t where = 1, 2, N and t = 1, 2, T;, t are respectvely an ntercept constant and a Gaussan error term whch may exhbts contemporaneous correlaton between the stocks. Equaton (4) s of a unvarate nature whch can be estmated usng all data ponts n the panel data set as the set of pooled data. The hypothess for the weak effcent market hypothess may be stated as: = 0, H a : 0. and a Wald test statstcs, usng only the unrestrcted model (4), can be used to test the set of null and alternate hypotheses of the effcent market hypothess wth only one restrcton. Thus, the asymptotcal Ch-squared has one degree of freedom. 4. Data Ths emprcal nvestgaton utlzes daly data from the Dhaka Stock Exchange, for the months of March, Aprl, December 2006 and January 2007, whch are randomly selected and they provde a total of 75 tradng days (19 days n March, 18 days n Aprl, 18 days n December 2006 and 20 days n January 2007). Among the group A, of almost one hundred lsted stocks, only 30 of them were traded n every day of the 75 tradng days. Ths yelds 2,250 panel data ponts (30 stocks x 75 days) that can be pooled. The reported percentage change on returns on the stocks was used to estmate the model specfed by equaton (4). 5. Emprcal results Ths secton examnes the emprcal fndngs on whether or not the Bangladesh stock market s weakly effcent. Specfcally, equaton (4) s estmated usng the percentage changes n the prces of 30 sample stocks for the months of March, Aprl, December 2006 and January The calculated Ch-squared test statstc and the p-value are reported n Table 1. Based on the calculated Chsquared statstc wth one degree of freedom of , the null hypothess of the weak form of the effcent market hypothess should be rejected at any conventonal level of sgnfcance. Ths rejecton strongly suggests that the Bangladesh stock market has not been nformatonally effcent despte the ntroducton of the Captal Market Development Program a decade or so ago. Due to nsuffcent technologcal nfrastructure, nformaton about lsted companes cannot be dssmlated through the fnancal communty quckly. The data shows the daly tradng volumes are low and much more unsteady. Therefore, the Dhaka Exchange traded stocks are lkely less lqud as compared to the well developed markets. Consequently, the market, collectvely, would be very slow to respond to relevant nformaton, whch allows nvestors the ablty to make forecasts based on condtonal means of stock prces. Ths s not a characterstc of the perfect captal market structure. Table 1. Wald test statstc and p-value 2 (1) value p-value One of the ssues n appled statstcs s the queston of how representatve the samples are. The ssue s more relevant when researchers do not have much flexblty n selectng ther samples. The sample used n ths emprcal nvestgaton s an example of ths. Out of almost one hundred stocks n group A the best stocks traded on the Dhaka Stock Exchange only thrty of the stocks traded every day n the seventy fve tradng days n March, Aprl, December 2006 and January If a longer perod were selected, the number of stocks, whch are traded every day, would be even lower. For example, there were forty eght stocks traded every day n the thrty seven tradng days n March and Aprl, 2006; but only thrty stocks traded n all seventy fve days n the sample perod selected by ths study. Therefore, t s of some nterest to nvestgate whether the result reported n Table 1 s senstve to the number of the stocks selected and, hence, how representatve they are of the whole Dhaka Stock Exchange. To ths end, ths study performs the cumulatve Wald test by ncludng 1, 2, 3, 4, 5, 10, 15, 20, 25, and the whole sample of 30 stocks. The results are summarzed n Table 2. Table 2. Wald statstcs for varous numbers of stocks Number of stocks n the sample 2 (1) statstc p-value Analyss of the fndngs, reported n Table 2, reveals that when the number of stocks n the sample exceeds 3, the weak form of the effcent market s rejected at any conventonal sgnfcant level. 14

5 Concludng remarks and polcy mplcatons In lght of the ntroducton of the Captal Market Development Program a decade or so ago and other recent mprovements n captal markets, the Bangladesh stock market s of partcular nterest for emprcal work. To ths end, ths emprcal nvestgaton modfes Dockery and Kavussanos multvarate model and uses a set of panel data from the Dhaka Stock Exchange market to test the weakform of the effcent market hypothess. Consstent wth a pror expectaton, the known characterstcs of the Dhaka Stock Exchange, and earler emprcal research results, the emprcal fndngs strongly suggest that Dhaka Exchange stock market s not nformatonally effcent. The fndngs further ndcate that when the number of stocks ncluded n the sample exceeds 3, the null hypothess of the effcent market hypothess s rejected throughout. Due to the common characterstcs assocated wth poor developng countres, ther equty markets are References Banks and Bank Systems, Volume 6, Issue 1, 2011 not even weakly effcent. Ths s not vewed postvely gven that most well developed countres have effcent equty markets. Ths observaton coupled wth the status of a poor developng country and ther attendant problems suggest that to develop an effcent equty market, Bangladesh should frst concentrate ts effort on developng better market nfrastructures for a more effectve market economy. Measures should be taken to stop corrupton n the equty market so that dstorton can not take place. Cartel among ssuers, traders, bourses, regulators, large nvestors and audtors may be prevented n the equty market. Practce of corporate governance should be properly done. Legal nfrastructure ought to be developed. Fscal polcy should be properly re-desgned. In ths envronment, a strong poltcal wll to reform the system and a strong commtment to mplement the reforms are needed to establsh a more compettve and effcent overall market economy that would be conducve for buldng an effcent stock market. 1. Alom, K., Khan, A.I., and M., Mokdum (2010). Are the Emergng Captal Markets Weak form Effcent? Evdence from the Model of the Dhaka Stock Exchange, Conference proceedngs of paper presented n the Internatonal Conference on Knowledge Globalzaton 2010 on Empowerment through Knowledge Sharng and Progress through Knowledge Transfer jontly organzed by North South Unversty, Bangladesh and Suffolk Unversty, USA held on May 8 to May 10, pp Bepar, M.K., and Mollk, A. (2008). Bangladesh Stock Market Growng? Key ndcators based assessment, Journal of Busness Admnstraton Onlne, Vol. 7, No. 2, avalable at Fall Conrad, J. and Kaul, G. (1988). Tme Varaton n Expected Returns, Journal of Busness, Vol. 61, pp DeBondt, W.F.M., and Thaler, R. (1985). Does the Stock Market Overreact?, Journal of Fnance, Vol. 40, No. 3, pp DeFusco, R.A., McLeavey, D.W., Pnto, J.E. and Runkle, D.E. (2004). Quanttatve Methods for Investment Analyss, 2nd Edton, Charlottesvlle, VA: CFA Insttute. 6. Dockery, E., and Kavussanos, M. G. (1996). Testng the Effcent Market Hypothess Usng Panel Data, wth Applcaton to the Athens Stock Market, Appled Economc Letters, Vol. 3, pp Fama, E.F. (1970). Effcent captal markets: a revew of theory and emprcal work, Journal of Fnance, Vol. 25, No. 2, pp Hoque, H. Al A.B., Kabr, M.M., and Rahman, M. (2005). Predctablty n stock returns evdence from conventonal and new varance rato tests, Journal of Fnance and Bankng, Vol.7, No. 1,2, June-December, pp Kendal, M. (1953). The Analyss of Economc Tme Seres, Part A: Prces, Journal of Royal Statstcal Socety, Seres A, Vol. 96, pp Lo, A.W. and MacKnlay, A.C. (1988). Stock Market Prces Do Not Follow Random Walks: Evdence from a Sample Specfcaton Test, Revew of Fnancal Studes, Vol. 1, pp Uddn, M.B. (2009). Determnants of market prce of stock: a study on bank leasng and nsurance companes of Bangladesh, Journal of Modern Accountng and Audtng, July, Vol.5, No. 7 (Seral No.50), pp Zellner, A. (1962). An Effcent Method of Estmatng Seemngly Unrelated Regressons and Tests of Aggregaton Bas, Journal of Amercan Statstcal Assocaton, Vol. 57, pp Wse, V. and Al, M.M. (2007). Equty Fundng Through Intal Publc Offerngs: the Case of Bangladesh, Global Busness & Economcs Anthology, USA, Volume 2, December, pp