You are on page 1of 12

Introduction: Speed of adjustment and quality of adjustments are the two basic issues with which stock market efficiency is related. One can term a stock market an efficient one if it incorporates all the news on a very rapid and unbiased manner. Academicians generally segregate stock market efficiency into three layers (namely weak form of efficiency, semi-strong form of efficiency and strong form of efficiency). Each of the aforementioned layers becomes even comprehensive in terms of the adjustment procedure. A stock market is efficient in weak form if the stock prices of all the listed firms reflect all the past information. In a stock market which is efficient in the weak form there will be no existence of any price patterns with prophetic significance, since the current price movement will not depend on the previous price movements. So, there is no real opportunity for the traders to earn abnormal return by analyzing the past price trends and patterns. So, the existence of weak form of efficiency rules out any real possibility of profitable trading rules. Stock prices in a weak form efficient market will respond only to new information and new economic events.

It had been an empirically proven fact that stock markets in the developing countries and least developing countries are not efficient in the semi strong and strong form. The researcher is trying to pinpoint the level of stock market efficiency (in weak form) for an emerging and developing stock market Dhaka Stock Exchange (DSE). DSE is the prime stock market of Bangladesh providing liquidity to millions of shareholders. By the end of 2010, total market capitalization of this stock market was around $50.28 billion with a listing of around 750 companies.

Background of the study: Louis Bachelier, a France based mathematician, first postulated the concept of stock market efficiency in his dissertation named "The Theory of Speculation” published in 1900. Even though during the 1930‟s independent and scattered works revolved around the theory of speculation upto the 1950‟s Bachelier works had been quite unrecognized. Only a few academicians believed that U.S.A. stock price and other related financial series followed random walk model. Alfred Cowles concluded that professional traders had failed to outperform the market on a long-run basis. Professor Eugene Fama while doing his Ph.D. at the University Of Chicago Booth School Of Business developed efficient market hypothesis as a part of his program during the 1960‟s. Efficient market hypothesis as a mechanism of stock return analysis gained enormous popularity during the 1990‟s when behavioral finance economists became the mainstream in the financial community.

Amount of academic research has been quite minimal centering on different layers of efficiency revolving stock markets of emerging markets. Since DSE can hardly be compared with the stock markets of Indonesia, India and Malaysia in terms of market attractiveness and market size, only a few academicians have felt the urge to conduct research to explore the different layers of efficiency in case of DSE. The researcher had tried to check out the extent of weak form of efficiency in case of Dhaka Stock Exchange across time slabs, across share categories, across sectors. One of the prime focuses of the study was to delineate any sign of improvement in terms of the weak form of market efficiency in case of DSE over time.

Development of research ideas, questions and objectives:

Research Idea: Weak form of market efficiency evidence from DSE

Research Questions:

The research will be conducted based on the following research questions and the researcher will try to find out the solutions for the following questions:

  • Q. 1

Whether DSE is efficient in terms of the weak form of market efficiency or not?

  • Q. 2

Is there any evidence of significant improvement in terms of the weak form of market efficiency

in DSE?

  • Q. 3

Does the level of weak form of market efficiency in DSE vary significantly among share

categorizes and industries?

Research objectives:

The following research objectives are going to be fulfilled:

  • 1. To test whether DSE is efficient in terms of the weak form of market efficiency or not

  • 2. To find out whether there is any evidence of significant improvement in weak form efficiency in DSE ( over time)

  • 3. To check to what extent weak form of market efficiency in DSE vary significantly among share categorizes and industries

Scopes and limitations of the proposed study:

1. The very basic construction mechanism of DSE market index is erroneous since it is not based on free float. So the conclusions that will be made about the weak form of market efficiency (based on the results to be generated by using market index data) could have some erroneous portion.

  • 2. Both run test and serial correlation test the level of independence for the whole set of data. It

may be possible that price changes are dependent only in special conditions.

  • 3. There is no way to refute the possibility of sustainable abnormal profit by using anomaly

based trading strategy in case of DSE, since the researcher will not check the effectiveness of all

the market anomaly based strategies for this stock market. Moreover, the unrealistic assumption (non existence of market imperfections) that will be made while checking the effectiveness of momentum driven trading strategy in case of DSE could even refute the drawn conclusion.

  • 4. Sampling (cluster sampling) will be used to check the evidence of weak form of market

efficiency across segments and categories (using the return series of individual firms). Even though, the researcher feel that the number of firms chosen from each industry and share categories is adequate to represent the basic nature of that industry but still the conclusions could be refuted upto a significant level due to the sampling error.

Literature review: The researcher now wants to focus on the summarized philosophy of the master academicians revolving on the research area of concern. L. Bachelier (1900) successfully postulated a complete theory of random walk behavior in case of stock price in his path breaking journal “Théorie de la Spéculation”. The Bachelier- Osborne model assumes that price changes at each transaction are independent and price changes from transaction to transaction form identically distributed random variable. This model further assumes that the distribution of price changes from transactions to transactions have finite variance. Under such conditions, using the theorem of central limit, Osborne concluded that we could expect daily, weekly and monthly price changes to have normal or Gaussian distribution. According to M. Osborne (1959) the steady state distribution function of the natural log of stock market return resembles the probability distribution for a particle in Brownian motion. In his study he used real life assumptions like tick size, discrete time frame, integral number for transaction etc.

Fischer Black (1986) had delineated the impact of noise on the functionality of a financial market. According to him, noise at times makes the market to act in an inefficient manner but at the same time noise prevents the information traders to extract continuous benefits from the inefficiencies. On the inefficient pricing that exists in the stock market he commented - “What‟s needed for a liquid market causes prices to be less efficient.” Kyle (1984) formed a model where he had shown that more noise traders can make markets more efficient.

Eugene F. Fama (1970) pinpointed three information subsets by which stock prices get adjusted. The weak form of efficiency related tests concern only with the historical prices. He conducted study on the behavior of U.S.A. stock exchange market prices taking the daily price of individual stocks that constituted DJIA (from January 1956 September 1962). Using the natural log of the stock‟s price return he tested whether the successive price changes were independent or not. According to him, since a time series like the stock market return should not be expected to be characterized by perfect independence, a minimum acceptable level of dependence needed to be tested. He tested independence both from the perspective of a statistician and a trader. Dependence was tested by the usual serial correlation model, run test mechanism and Alexander‟s Filter technique. Any distribution was consistent with random walk behavior as long as it correctly characterized the process of generating the price changes. Of course the theory did imply that the parameters of distribution should be stationary or fixed. According to Fama, as long as the independence assumptions hold, the stationary could be explained loosely. He concluded that there was no evidence of large degree of dependence in daily, four day, nine day and sixteen day price changes.

The early studies on testing weak form of market efficiency generally supported the weak- form efficiency of the markets considering a low degree of serial correlation but such findings were rationale only for developed stock market. Kendal (1953) observed that weekly price changes of British firms were approximately normally distributed. Moore (1962) confirmed that the randomly selected eight stocks of NYSE followed the normal probability pattern even though he had found too many large price changes in case of using the log of weekly return series. Lo and Mackinlay (1988) focused on 1216 weeks return series data (from September 1962 to December 1985) to check out the random walk behavior in case of U.S.A. market. The random walk model was strongly rejected for the entire sample period and for all the sub periods for a variety of aggregate return index and size sorted portfolios. Fama and French (1988) reported

negative correlation in daily returns in case of U.S.A. market but suggested that it was small in absolute magnitude. Hudson, Dempsey and Keasey (1994) found that the technical trading rules have moderate level of predictive power but not sufficient to enable excess return in U.K market.

Now the researcher want to focus on the various researches conducted on weak form of market efficiency in case of Dhaka Stock Exchange and the researcher will try to summarize the results in this part of the literature review. Mobarek and Keasey (2000) looked out for evidence of weak form of market efficiency in case of Dhaka Stock exchange but based on the study using the smoothed daily market return series, they had to conclude that the market was not efficient. This result was consistent in case of sub samples and for individual firms. The result of both parametric tests and non parametric tests confirmed the conclusion. ARIMA (2,0,1) was the best fit for the return series and they concluded that this order of (p,d,q) should be used in building up any predictive model. A study by Ahmed (2002) revealed that the Dhaka stock exchange return did not follow the presumed the random walk theory rather he had found serial correlation. Ljung-Box statistic (LBQ) was used for the period from January 1990 through April 2001 and for two sub-periods to find out the generalized autocorrelation status at different time lags. Even though, for the first time slab he had found positive autocorrelation, for the overall time frame the negative correlations were evident. When he tried to study the level of semi strong form of efficiency he found that it might take around one month for any publicly available information to get reflected in the stock price in case of DSE. Chowdhury, Sadique and Rahman (2001) investigated market anomaly based strategies‟ effectiveness (especially calendar effect based strategy) in case of Dhaka Stock Exchange. While they had used regression, the result made them infer that there was seasonality in the market return. When they checked existence of weekend effect they had found these results - average return on the last trading day of the week produced significant negative return (at 10% level), no significant abnormal returns on the last day of the week (5% level), no significant average return on the opening day of the week and Mondays produced persistently significant average negative returns. Hossain (2004), using a different methodology (to track sector based momentum effect), found that when portfolios of bank stocks were constructed on day one of the week with the opening prices and sold on Monday or the last day of the week, there was a considerable chance that the portfolio would provide significantly higher mean daily return (MDR). Even when the portfolios were produced randomly, the effectiveness of such momentum based strategy remained intact. According to Hassan, Islam and Basher (2000) DSE market returns displayed significant serial correlation of -0.07, implying stock market inefficiency. The results also showed a significant relationship between conditional volatility and the stock returns, but the risk- return parameter was negative (-0.1072) and statistically significant. Kader and Rahman (2004), by using k% filter rule, found strong evidence of profitable trading rule at a sustainable basis, which violates the random walk hypothesis.

Methodology: By conducting this research the researcher will try to track the level of efficiency in terms of weak form in Dhaka stock exchange. There will be three broad arenas of research test of weak form of efficiency for the overall market (with a special focus on the possible evidence of improvement in terms of the weak form of market efficiency), test of weak form of efficiency across different share category prescribed by the stock market regulators (for example A, B, and Z) and finally test of weak form of efficiency across various sectors of the DSE (like

bank, NBFI, ceramics, food etc). Both parametric tests like (Auto correlation, and ARIMA model) and non parametric test (run test) will be performed to check whether any particular return series (it could be the return series for the index representing the overall market or the return series relevant for any firm) follows the random walk model or not - randomness of the return is the true reflection of the weak form of efficiency and irrelevance of any chartist model. The null hypothesis will be that the overall market or the stock price of any enlisted and regularly traded firm is efficient in terms the weak form of efficiency. Since stock market indices are theoretically the closest proxy of the overall market the researcher have decided to go for DSE all share index to track the extent of weak form of efficiency in case of Dhaka stock exchange. In the calculation of the index, market value is the weighting device so the data will be free from any sort of abnormal market price attached to any particular stock. The researcher have decided to go for the month end index value and since he has a longer range time frame in his mind (from 1987 to 2010), the researcher does not think thin or infrequent trading can pose any significant bias to the research result. To check whether Dhaka stock exchange is heading towards the attainment of weak from of efficiency gradually, the researcher has decided to divide the long time period into two segments to test the significance of the improvement associated with the regulatory reforms, huge jump in the market capitalization, and inclusion of quality stocks in the attire. The two chosen segments are: 1. from the year 1987 to 1996 (up to the month of June) and 2. from year 1997 (excluding January to June) to year 2010 (upto the month of July). The highly volatile period (from July 1996 to June 1997) will be kept out of the account because of the very awkward result that it can produce. Later on, the time slab of 13 years (from 1997 to 2010) will be broken into two segments to check out the possibility of any improvements in terms of the weak form of market efficiency. The two chosen segments are: 1. from the year 1997‟s July to 2003 (December) and 2. from year 2004 (from January) to year 2010 (upto the month of July). In the later parts of the study the researcher will also use daily return series. If the result shows that the overall market is not efficient in terms of the weak form efficiency, then the researcher will try to check whether the formation of any trading rule can give any investor abnormal return or not. To be more precise, the researcher will try to track the effectiveness of momentum based strategy (holding the wining portfolio and selling short the losing portfolio), in case of D.S.E. based on a sample size of 65 regularly traded firms in Dhaka Stock exchange. Moreover to achieve the third research objective separate test for randomness will be conducted for various classes (like A, B, and Z category) of stocks not for the overall market. Based on some randomly chosen firms, the researcher will also test, whether there is any significant variation in terms of weak form of market efficiency across various industries.

Since in all the analysis the input variable is going to be return series (either the market return or the stock return), at first the researcher must explain the definition of return. First of all since the data represents monthly return only month beginning index or stock price and month ending index or stock price will matter and for this study the interim price or index value will be irreverent. For conducting research with the stock market index, return is the historical monthly appreciation or depreciation of the index value (the difference between current index value and the base index value where base value means the index of the just previous month) divided by the base index value. In case of calculating monthly return for listed stocks the formula will be very straight cut; it will be monthly capital gain or loss divided by the base price ( price that existed at the last trading day of the previous month) .Cash dividend adjustment, stock dividend adjustments will be incorporated in the calculation of return. For simplicity, the right share issue

related affairs will not be embedded in the calculation of the return series; although to get the true monthly stock return such adjustments are essential. Luckily very firm of the randomly chosen firm will turn out to be a right issuer. At some case daily return calculation will be needed and at that time return will simply refer to the capital gain or capital loss. At times natural log of the market return and stock based return will be used as the input variable since logarithmic returns are more prone to be distributed normally which is a precondition of almost any statistical technique.

Natural log of market return = Ln (I t

/ I t-1 ), where Ln represents natural log, I t

is the index

value at t (t - at times will be day or it can be month and at times even a year), I t-1 is the index

value at period t-1(t-1 this can at time represent the previous day or previous month or even a previous year).

Natural log of the individual share return = Ln [(P t * S.D.A. + C) / P t-1 )] where Ln represents natural log, P t is the stock price at t (t - at times will be day or it can be month and at times even a year), S.D.A is the adjustments that should be made for the stock dividend adjustment, C is the amount of cash dividend and finally P t-1 is the stock price at period t-1(t-1 this can be at time represent the previous day or previous month or even a previous year).

Since by using the run test result, tracking the autocorrelation coefficients at different lags and fitting ARIMA model of a specific order (0,1,0) the researcher is trying to track the randomness in either stock price or in the index value and eventually trying to make inference about the random walk behavior ( so impliedly weak form of market efficiency) either for the firms or the overall market, the researcher should now try to give the readers some of the conceptual ideas behind the mechanisms of run test, autocorrelation and ARIMA.

The run test is a widely used approach to test and detect statistical independencies (randomness) ignoring any assumption regarding the properties of distribution. A run can be defined as a succession of identical symbols which are followed or preceded by different symbols or no symbol at all. The number of runs is computed as a sequence of the return changes of the same sign (such as; ++++−−−+++−−++++++−−−−). The null hypothesis of the test is that the observed series will be a random series. When the expected number of run is significantly different from the observed number of runs, the test reject the null hypothesis that the monthly returns are random. A lower than expected number of runs indicates market‟s overreaction to information while higher number of runs reflects a lagged response to information. Either situation would suggest an opportunity to make excess returns. The cutting point which is used to dichotomize the data set (return series in our case) could be specified as a particular number, or the value of a statistic (like mean, median or mode). In this analysis the researcher will use mean as the cutoff point. Cases with values that will be less than the cut point (which is mean in this case) will be assigned to one group, and cases with values that will be greater than or equal to the cut point (which is mean in this case) will be assigned to another group. For each of the data points (monthly or daily - market or stock return), the difference D i = X i cut point will be calculated. If D i 0, the difference will be considered positive, otherwise negative. The number of times the sign changes, that is D i 0 and D i+1 <0 or D i < 0 and D i+1 0 as well as the number of positive (n p ) and negative (n a ) signs, will be determined. The number of runs (R) will be the number of sign changes plus one. The run test converts the total number of

runs into a Z statistic. For large samples the Z statistics gives the probability of difference between the actual and expected number of runs. If the Z value is greater than or equal to1.96, then we should reject the null hypothesis at 5% level of significance and we should also conclude that the market or the stock price of any firm is not efficient in terms of the weak form of efficiency. But if the Z value is lower than 1.96, then we should accept the null hypothesis at 5% level of significance and we should also conclude that the market or the stock price of any firm is efficient in terms of the weak form of efficiency. The Z value is calculated as the difference between the actual and expected number of runs divided by the standard deviation, so

Z = (R- μ r )/ σ r , here μ r goes for the expected number of runs and σ r is the standard deviation.

Once again, μ r = [2n p n a / (n p

+ n a )] +1 and σ r is

n p )} / {(n p + n a ) 2 * (n p + n a -1)}]

calculated as follows: σ r = √ [{2n p n a * (2n p n a - n a -

Auto-correlation test is a reliable measure for testing of either dependence or independence of

random variables. Informally, it is the similarity between observations as a function of the time separation between them. In case of the analysis, autocorrelation is the tendency of market returns to depend on the values of the lag period values. For the random nature (impliedly the

achievement of weak form of efficiency) of the overall stock market or any firm‟s stock, the

autocorrelation coefficient of respective return series needs to be zero or closer to zero. Hypothesis test needs to be developed in order to determine whether a particular autocorrelation coefficient is significantly different from zero. Once again the null hypothesis is that a particular autocorrelation coefficient is equal to zero or the overall market or any stock price follows the random walk behavior. The analysis covers monthly index based return (both smoothed and non- smoothed) for DSE all share index and the smoothed and unsmoothed monthly return series of randomly chosen firm. If the series is of nonstationary nature then the return series will appear to grow or decline over time and the autocorrelation coefficients of different lags will eventually fail to die out rapidly. If the return series exhibit seasonality then the autocorrelation coefficients at the seasonal lag (4 th lag in case of quarterly data, 12 th lag in case of monthly data) or multiple of seasonal lags will demonstrate statistically significant coefficients (in other words significantly different from zero). If the return series is of a stationary nature then the basic statistical properties such as the mean and variance remain will constant over time and the autocorrelation coefficients will decline to zero quite rapidly, generally after the second or third time lag. The autocorrelation coefficient for the k th lag will be calculated by the following formula:

r

n

(

Y

t

t

k

  1

Y

)(

Y

t

K

Y

)

n

t 1

(

Y

t

Y

)

2

Here r k = autocorrelation coefficient for the k th lag

Y = mean of the data set Y t = Observation in time period t Y t-k = Observation at time period t-k

To conclude whether any autocorrelation coefficient is statistically different from zero or not the researcher will compare the t-statistics of respective lags with the upper and lower limit value. If the t-statistics fall within the region then the null hypothesis is accepted that is the

autocorrelation coefficients at different time lags are not significantly different from zero and vice versa.

t = ( r k - ρ k ) / SE (r k ) Here r k = autocorrelation coefficient for the k th lag ρ k = Population autocorrelation coefficient (assumed to be zero ) SE (r k ) = Standard error of autocorrelation function. It will be calculated as (1+2r 2 k )/n. Upper limit = + t- table value (1.96) * SE (r k ) (level of significance = 5%) Lower limit = - t- table value (-1.96) * SE (r k ) (level of significance = 5%) Since number of observations will always be higher than 29 at a 5% confidence level t table value is +/- 1.96

Instead of testing the autocorrelation coefficients separately, the researcher will alternatively use the very common portmanteau test known as Box-Ljung test. It tests whether any of a group of autocorrelations of a time series (return series in our case) is different from zero. The Q statistic will be calculated by the following formula:

autocorrelation coefficients at different time lags are not significantly different from zero and vice versa. tautocorrelations of a time series (return series in our case) is different from zero. The Q statistic will be calculated by the following formula: Here, r = autocorrelation coefficient for the k lag T = Number of observations s = Number of time lags to be tested k = the time lag If Q statistics exceeds the critical value of a chi-square distribution ( ) with s degrees of freedom, then at least one value of r is statistically different from zero at the specified significance level ( in our case the confidence level will be set at 5%). At 12 and 24 degrees of freedom, the table value for a chi-square distribution at 5% level of significance is 21.02 and 36.415. The Null Hypothesis will be that none of the autocorrelation coefficients up to lag s are significantly different from zero. ARIMA (a dynamic time series model) will be used to check whether the return series depends on the past values of the return series and past disturbance elements. The model is generally referred to as an ARIMA (p, d, q) model where p, d, and q are integers greater than or equal to zero and refer to the order of the autoregressive, integrated, and moving average parts of the model respectively. Autoregressive orders (p) specify which previous values from the series are used to predict current values. Difference (d) specifies the order of differencing applied to the series before estimating models. Moving average orders specify how deviations from the series mean for previous values are used to predict current values. ARIMA models form an important part of the Box-Jenkins approach to time-series modeling. If the market is efficient in the weak form then the coefficients will not differ significantly from zero and the best fit will be found in ARIMA (0, 1, 0) model. In conducting the ARIMA for testing the random walk behavior of the return series the researcher will follow the following gradual and generic steps: Step 1: Model identification : At first by observing autocorrelation coefficient scenario carefully, the researcher will decide whether the time series is of stationary nature or not. Nonstationary series have an ACF that remains significant for half a dozen or more lags, rather " id="pdf-obj-7-41" src="pdf-obj-7-41.jpg">

Here, r k = autocorrelation coefficient for the k th lag T = Number of observations s = Number of time lags to be tested k = the time lag

autocorrelation coefficients at different time lags are not significantly different from zero and vice versa. tautocorrelations of a time series (return series in our case) is different from zero. The Q statistic will be calculated by the following formula: Here, r = autocorrelation coefficient for the k lag T = Number of observations s = Number of time lags to be tested k = the time lag If Q statistics exceeds the critical value of a chi-square distribution ( ) with s degrees of freedom, then at least one value of r is statistically different from zero at the specified significance level ( in our case the confidence level will be set at 5%). At 12 and 24 degrees of freedom, the table value for a chi-square distribution at 5% level of significance is 21.02 and 36.415. The Null Hypothesis will be that none of the autocorrelation coefficients up to lag s are significantly different from zero. ARIMA (a dynamic time series model) will be used to check whether the return series depends on the past values of the return series and past disturbance elements. The model is generally referred to as an ARIMA (p, d, q) model where p, d, and q are integers greater than or equal to zero and refer to the order of the autoregressive, integrated, and moving average parts of the model respectively. Autoregressive orders (p) specify which previous values from the series are used to predict current values. Difference (d) specifies the order of differencing applied to the series before estimating models. Moving average orders specify how deviations from the series mean for previous values are used to predict current values. ARIMA models form an important part of the Box-Jenkins approach to time-series modeling. If the market is efficient in the weak form then the coefficients will not differ significantly from zero and the best fit will be found in ARIMA (0, 1, 0) model. In conducting the ARIMA for testing the random walk behavior of the return series the researcher will follow the following gradual and generic steps: Step 1: Model identification : At first by observing autocorrelation coefficient scenario carefully, the researcher will decide whether the time series is of stationary nature or not. Nonstationary series have an ACF that remains significant for half a dozen or more lags, rather " id="pdf-obj-7-49" src="pdf-obj-7-49.jpg">

If Q statistics exceeds the critical value of a chi-square distribution ( ) with s degrees of freedom, then at least one value of r is statistically different from zero at the specified significance level ( in our case the confidence level will be set at 5%). At 12 and 24 degrees of freedom, the table value for a chi-square distribution at 5% level of significance is 21.02 and 36.415. The Null Hypothesis will be that none of the autocorrelation coefficients up to lag s are significantly different from zero.

ARIMA (a dynamic time series model) will be used to check whether the return series depends on the past values of the return series and past disturbance elements. The model is generally referred to as an ARIMA (p, d, q) model where p, d, and q are integers greater than or equal to zero and refer to the order of the autoregressive, integrated, and moving average parts of the model respectively. Autoregressive orders (p) specify which previous values from the series are used to predict current values. Difference (d) specifies the order of differencing applied to the series before estimating models. Moving average orders specify how deviations from the series mean for previous values are used to predict current values. ARIMA models form an important part of the Box-Jenkins approach to time-series modeling. If the market is efficient in the weak form then the coefficients will not differ significantly from zero and the best fit will be found in ARIMA (0, 1, 0) model. In conducting the ARIMA for testing the random walk behavior of the return series the researcher will follow the following gradual and generic steps:

Step 1: Model identification: At first by observing autocorrelation coefficient scenario carefully, the researcher will decide whether the time series is of stationary nature or not. Nonstationary series have an ACF that remains significant for half a dozen or more lags, rather

than quickly declining to 0. Differencing is necessary when trends are present (series with trends are typically nonstationary and ARIMA modeling assumes stationary) and is used to remove their effect. The order of differencing corresponds to the degree of series trend-first-order differencing accounts for linear trends, second-order differencing accounts for quadratic trends, and so on. Every case the researcher have decided to go for the first order of differencing which is ∆Y t = Y t Y t-1. Once a stationary series is obtained the researcher will identify the appropriate form of the model that will be used. Here the researcher will try to match various theoretical set of autocorrelation and partial autocorrelation associated with the respective ARIMA model. For example, if autocorrelation die out exponentially to zero and partial autocorrelation cut off, the model will require autoregressive terms. The number of spikes will indicate the order of the auto regression. If the autocorrelations cut off and partial autocorrelation die out the model will require moving average terms. The number of spikes indicates the order of the moving average if both the autocorrelations and partial autocorrelation die out, both the autoregressive and moving average terms need to be indicated.

Step 2: Model estimation: Then the researcher will try to estimate the parameters of the model. Parameters that are statistically different from zero needs to be retained in the model; on the other hand parameters that are not statistically different from zero needs to be dropped from the model. The confidence level will be set at 95%.

Step 3: Model checking: An overall model adequacy is checked is provided by a Chi-square test based on the Box-Ljung statistics. The Q statistic will be calculated by the following formula:

m

Q = n (n+2) Σ r 2 k / (n-k)

K=1

Here, r k = residual autocorrelation coefficient for the k th lag n = Number of residuals m = Number of time lags to be tested k = the time lag

If the p value associated with the statistics is small(less than 5%), the model is assumed to be considered inadequate and vice versa. By using the expert modular on SPSS, the researcher will try to find out the best possible time series model for respective return series. The criterions to judge for the best model are relatively small of BIC, relatively small of SEE; relatively high adjust R 2 .The SPSS expert modular allows any time series to fit with several well known models like simple exponential smoothing, Holt's exponential smoothing, Brown's linear trend etc. and obviously the Box-Jenkins‟s based ARIMA method.

Now the researcher wants to discuss briefly the mechanisms that he will follow to test the effectiveness of momentum effect driven strategy in DSE. The methodology in this research is almost in line with Jegadeesh and Titman (1993) original approach to study the momentum effect. Assumptions and procedures are as follows;

1. Two hypothetical scenarios will be tested for observing short term effectiveness of momentum effect driven strategy - 30 month return series ( monthly return) of 65 sample firms will be tested and for the long term effectiveness checking 10 year return series (

yearly return) of 65 sample firms will be tested. All the data, the researcher will use is going to be real life realized return series.

  • 2. There will be a one month lag and one year between the ranking period and holding period for the above mentioned two scenarios.

  • 3. After observing the ranking, for the next holding period the winner portfolio will be hold and the losing portfolio will be short sold ( in case of checking the effectiveness of the momentum effect driven strategy). The top capital gain based return generator 10 firms out of the sample of 65 will constitute the winner portfolio and the top 10 capital loss based losing shares will constitute the losing portfolio. In case of the winner and loser portfolio, stocks will have equal weights. After that holding period return from the winning portfolio and losing portfolio will be computed. In short, return from the momentum strategy is the return of the winning portfolio less the return of the losing portfolio. If the return pattern of the ranking period repeats then the winning portfolio shares will have more price hikes and the losing portfolio shares could have been returned by an even less funding (it is actually the very basic motive behind momentum strategy). In this case, the return from momentum effect driven strategy (the short sale of the losing portfolio and purchase of the winners at the beginning of the holding period) can be expected to be substantially higher than the broad based index DSE all share in our research, since the index is constituted of both the winners and losers. Moreover, since we have shorted losing portfolio, the negative returns on that portfolio are actually the short seller‟s gain. After each holding period there will be a new ranking and eventually new winning and losing portfolio. The researcher will check whether on a short term basis ( 30 month) this market anomaly based strategy could had outperformed the broad based market index ( the performance of an investor who has hold a market value weighted numbers for all traded, listed securities of DSE).

  • 4. The researcher will assume non existence of market imperfections in case of DSE, for this study. The winning and losing portfolios will be equally weighted and the net investment will be assumed to be zero. Further the researcher will assume no reinvestment of profits meaning that all over the term, the chartist‟s net investment will be zero.

The methodology for tracking the evidence of any weekend effect in DSE is very simple. The researcher will calculate the daily capital gain and loss based return of the DSE all share index for all the trading days in DSE Sunday, Monday, Tuesday, Wednesday and Friday. Because of the extra time and extra risk associated with the opening week‟s price, academicians expect that Monday‟s return will be substantial ( in our case Sunday) but actually there is an almost

universal tendency of stocks to exhibit relatively large returns on Friday ( in our case Thursdays) compared to those on Mondays ( in our case it will be Sunday). Based on the comparison of average daily return of the trading days over a 10 year time frame, the researcher will try to check the evidence of such market anomaly in case DSE. Since for comparison purpose the researcher has to select any week where trading had occurred in all the five days of the week, number of weeks that will be used for the analysis is expected to be on a lower note.

In conducting the research, sampling will be an issue since the researcher wants to track the level of weak form of market efficiency across industry and across category. For the first case the researcher will choose 65 firms and for the second case he will choose 30 firms. At first the

whole population of the listed firms will be subdivided into different stratum like category A, B, Z etc or in case of industry wise analysis stratum like banks, cement, ceramics etc. From the strata to choose the appropriate firm simple random sampling will be the preferred sampling mechanism since all the firms will get almost the same probability of getting selected for the sample. The data that will be used for the study is going to be of secondary nature. DSE library will provide the researcher all these published information. Since the research wing of DSE is efficient the timing, quality, relevance and reliability of the data is beyond any level of suspicion. The researcher is not going to use any questionnaire for data collection. SPSS and Microsoft Excel (two prime software) will be used to analyze all the data set.

Conclusion: The expected result from this study is inefficiency in weak form. Someone may question the researcher about the real benefit of having an inefficient market (inefficient in the lowest grade!), since in this dynamic world every particles are trying to be as efficient as it could be. But since it is better to have an inefficient stock market rather than not having any, we should concentrate on the ways by which DSE could one day become efficient in the weak form. There is a positive wind over the market blowing for the last two or three years, we should be efficient enough to cash on it.

This market is noise and rumor driven and there is little scope for information trader since they are blocked by an array of noise traders who every now and then create panic. Still the market is small enough for the international speculator and gambler. The regulators must make sure that the playing field is level for everybody, instead of resisting the correction process. Public information has to be disseminated instantly and in an unbiased manner.

If we can make sure that only news are affecting the stock price and nothing else, once the market becomes corrected it will start to behave randomly (obviously the whole process may take quite a long time), since by the law of the universe, news are like heads and tails of a coin (on an average there is equal possibility that the next news will be a good one or a bad one.)

References

  • 1. Bachelier, L. (1900) Théorie de la Spéculation.

  • 2. Osborne, M. (1959) „Brownian Motion in the Stock market‟ Operational Research. Vol. 7. No. (2), 145 73.

  • 3. Black, F. (1986) „Noise‟ Journal of Finance. Vol. XLI. No. (3), 529-43.

  • 4. Kyle, A. (1984) „Market Structure, Information, Futures Market and Price Formation‟ International Agricultural Trade. 45-63.

  • 5. Fama, E. (1965) The Behavior of Stock Market PricesJournal of Business. Vol. 38. 34-105.

  • 6. Fama, E. (1970) „Efficient Capital Markets: A Review of Theory and Empirical Work‟ Journal of Finance. Vol. XXV, No. 2. 383 417.

  • 7. Kendal, M. (1953) „The Analysis of Economic Time Series‟ Journal o f the Royal Statistical Society. Vol. 96. 11-25.

9. Lo, A. & Mackinlay, A. (1988) Stock Market Prices Do Not Follow Random Walks: Evidence from a Simple Specification TestReview of Financial Studies. Vol. 1, No. 1. 41-66.

  • 10. Fama, E. & French, K. (1988) Permanent and Temporary Components of Stock Prices‟ Journal of Political Economy. Vol. 96. 246 273.

  • 11. Hudson, R., Dempsey, M., & Ke a s e y, K e v i n (1994) „A n o t e o n t h e we ak - form efficiency of capital markets: The application of simple technical trading rules to UK Stock prices-1935 to1994Journal of Banking & Finance. Vol. 20. 1121-1132

  • 12. Mobarek, A., & Ke a se y, K. (2000) „Weak-form market efficiency of

an emerging

market: Evidence from Dhaka Stock Market of Bangladesh‟

  • 13. Ahmed, F. (2002) „Market Efficiency in Emerging Stock Markets : The Case of Dhaka Stock Exchange‟

  • 14. Chowdhury, S., Sadique, M. & Rahman, M. (2001) „Capital Market Seasonality: The Case of Dhaka Stock Exchange (DSE) Returns‟ South Asian Journal of Management. Vol. 8. No. 3 & 4. 1-7.

  • 15. Hossain. F. (2004) „Days of the Week Effect in Dhaka Stock Exchange: Evidence from Small Portfolios of Banking Sector‟ The Jahangirnagar Review. Vol. XXVIII. 73-82.

  • 16. Hassan, M., Islam, A. & Basher, S. (2000) „Market Efficiency, Time-Varying Volatility and Equity Returns in Bangladesh Stock Market‟

  • 17. Kader, M. & Rahman, A. (2004) „Testing the Weak-Form Efficiency of an Emerging Market: Evidence from the Dhaka Stock Exchange of Bangladesh‟ AIUB Journal of Business and Economics. Vol. 4. No. 2. 109-132