You are on page 1of 18

Feedback trading behavior in Dhaka Stock Exchange

(DSE), Bangladesh

Abu Taher Mollik
Accounting, Banking and Finance
Faculty of Business and Government
University of Canberra
Building 6, Room: 6C26;
Phone: (+61) 0422 239 066

M Khokan Bepari
School of Commerce and Marketing
Faculty of Arts, Business, Informatics and Education
Central Queensland University, Australia

DSE 20 return series is expected to show a high feedback trading led autocorrelation. O53 Keywords: Dhaka Stock Exchange. sell when they fall” and negative feedback trading means “sell when prices rise. It has important implications for other inefficient and emerging markets in that inefficiency in the form of return predictability in those markets may not imply abnormal profit opportunities. G14. Feedback trading. However. However. Abstract This paper empirically tests the autocorrelation structure in the return series of Dhaka Stock Exchange (DSE) in the framework of feedback trading behavior of noise traders. for this over-reaction in the following days and prices tend to move in the opposite direction inducing negative autocorrelation. buy when they fall”. as is argued by Dean and Faff (2008). Thus linking the observed autocorrelation structure in return series to feedback trading behavior of irrational investor. Positive feedback trading. The daily return series of DSE General index and DSE 20 index have been used to capture the effect of non- synchronous trading and the full effect of feedback trading. the study concludes that return autocorrelation and inefficiency in DSE does not imply abnormal profit opportunities. Asymmetric GARCH model. Introduction This paper investigates the autocorrelation structures of DSE return series in the framework of feedback trading hypothesis. Feedback trading means trading based on historical data. the results of the study found support for the existence of positive feedback trading in DSE. Cutler et al. G15. Feedback trading results in excess volatility which destabilizes stock prices (DeLong et al. Return autocorrelation. N25. . Indeed. Positive feedback traders reinforce price movements such that prices will continually overshoot the levels suggested by fundamental company information. 1. while positive feedback trading implies “buy when prices rise. While DSE General return series is expected to have high non-synchronous trading induced autocorrelation. Thus presence of a sufficiently large number of feedback traders in the stock market is reflected in the autocorrelation of returns. suggesting s that bad news and good news do not have differential effect on the conditional variance of DSE return series. 1990) and induces return predictability (Bohl and Reitz (2004). this feedback trading induced return predictability does not imply abnormal profit opportunities because higher volatility associated with feedback trading ‘makes it harder for rational risk –adverse investors to exploit the predictable pattern of stock prices (Koutmos. G12. JEL Classification: G11. (1991)) via higher level of autocorrelation. feedback trading does not have asymmetric effect on DSE return series in up and down market.. Market corrects. 1997)’.

. Thus the present paper contributes to the literature by providing a new explanation of the inefficiency in an emerging stock market. Berlung & Liljeblom. Once new information affects highly traded stock prices. 2004. the second section of the paper provides a review of literature. 1985. . a time lag occurs when the thinly traded stock prices are affected resulting in autocorrelation . The last section concludes. Rahman & Hossain. Boudoukh et al. 2000). and Pierdzioch & Schertler. However. inclusion of thinly traded stocks in the index cause positive autocorrelation in the index returns. and Cohen 1980) as reasons for return autocorrelation. Non-synchronous trading assumes that the new information is reflected in highly traded stock prices earlier than in thinly traded stock prices. Perry. 1993.1993) . An important property of time varying risk premia is that return and return autocorrelations should be consistent and visible in all assets . the study concludes that return autocorrelation and inefficiency in DSE does not imply abnormal profit opportunities. Mobarek & Keasey. a time lag occurs when the thinly traded stock prices are affected resulting in autocorrelation . (1994) summarizes these hypotheses into three basic schools of thought. 2006. The motivation of the study is to investigate the underlying causes of autocorrelation structure in DSE.Characterized by a smaller number of investors. and a high level of volatility. Boudoukh et al. various studies suggest that non-synchronous trading and transaction costs are not sufficient to fully explain the autocorrelation in stock return (see. In doing so. Loyalist. When index returns are analyzed in this context. assumes that the new information is reflected in highly traded stock prices earlier than in thinly traded stock prices. In periods of high trading volume investors’ estimates of the market risk aversion is more accurate and thus the autocrrelation in the stock retrun is reduced . It has important implications for other inefficient and emerging markets in that inefficiency in the form of retrun predictability in those markets may not imply abnormal profit opportunities. Literature Review Finance literature explains the autocorrelation structures in the stock return from various perspectives. Safvenblad (1997) observed that this time varying risk 1 Initially developed by Fisher (1966) and subsequently augmented by Lo and MacKinlay (1990). What are the plausible causes of observed autocorrelation structure in DSE return series? Is it induced by feedback trading behavior of irrational investors? The study addresses these issues. Lo and MacKinlay (1990))1 and transaction cost (Mech . Pierdzioch. 1994. for instance. for example. However. lower level liquidity.. Linking the observed autocorrelation structure in return series to feedback trading behavior of irrational investor. Revisionists also assumes efficient market and suggest time varying risk premia as the reasons for autocorrelation in index return. Dhaka Stock Exchange (DSE) is found to be inefficient in its weak form (see. Section three provides an introduction to the microstructure of DSE while section four explains the empirical models of feedback trading and volatility along with the methodology used and findings of the study. Once new information affects highly traded stock prices. Significant level of autocorrelation has been found in the DSE return series. 2005). assumes an efficient stock market and put forward the concept of non-synchronous trading ((Fisher (1966). 2. 1998. Time varying risk premia is based on the notion that aggregate level of risk is mean reverting but unobservable (Campbell et al.

non-fundamental trading strategies seem to play a more important role in emerging markets relative to mature stock markets. Their results suggest that positive and negative feedback trading strategies exist in both types of markets but are more pronounced in emerging stock markets than in mature markets. the UK and the US). Bohl and Siklos (2005). Some investors seek trends in past stock prices and base their portfolio decisions on the expectation that such trends will persist. and finds evidence of positive feedback trading which is more pronounced in down market than in up market. A distinction is. Poland & Russia) and in mature markets (Germany. Hong Kong. Belgium. on the contrary. Germany. investigates the trading behavior during stock market downturn and find that there is evidence of positive feedback trading . Altay (2006). however. Heretics. Negative feedback trading. Japan. Koutmos and Saidi (2001) investigate six emerging Asian stock markets viz. Philippines. Taiwan. Hungary. If large number of traders implements the feedback trading strategy. based on daily data on the Dow Jones Industrial Average index since 1915 to 2004. Positive feedback trading should result in negative autocorrelation of returns because it gives rise to a short-run overreaction of stock market prices to new information. Koutmos et al. Italy. in contrast. with little evidence that market declines are followed with higher volatility than market advances. 1992). In terms of behavioral finance. this type of investors is usually called a noise trader or a feedback trader. made between positive and negative feedback trading. they assume the psychological factors to be responsible for autocorrelation in stock returns. that has been observed in almost all developed stock markets. There are a number of studies related to the investigation of positive feedback trading and stock return autocorrelation in the stock markets of different countries. current stock prices become related to the previous prices resulting in autocorrelation in the index returns. herding. Sentana and Wadhwani (1992) find existence of positive feedback trading in the US stock market. Hence. is against the validity of efficient market hypothesis. studies the autocorrelation structure in Istanbul Stock Exchange. concludes that the positive feedback trading causes negative autocorrelation in stock returns. Singapore. They also find that feedback trading effect on return autocorrelation is more pronounced in down market than in up market. Bohl and Siklos (2004) study the feedback trading behavior in emerging markets (Czech Republic. and Thailand . UK). the Cyprus Stock Exchange. In terms of behavioral finance. Feedback trading includes several trading strategies like profit taking. should result in positive autocorrelation of returns (Sentana and Wadhwani . On the contrary. They find that the negative innovations influence on volatility more than positive ones (leverage effect). Koutmos (1997).premia is more visible in long term expected retrun than short term one. the so-called 'leverage effect'. Bohl and Reitz (2002) show strong evidence of leverage effect in German market. (2006) document positive feedback trading in an emerging market. investigating stock market indices of six developed countries (Australia. contrarian investment and dynamic portfolio allocation. Malaysia.

Brokerage fees are 1 percent on trade (0. in public. it has now 280 listed companies as of March 2009. G. They attribute the stock market crash to feedback trading. while the evidence on leverage effect of feedback trading is inconclusive in the emerging markets.The market capitalization of DSE was USD 10822 million in December 2007 which was 16% of the GDP of Bangladesh. Nofsinger and Sias (1999) study feedback trading behavior of institutional investors versus individual investors and find that either institutional investors positive-feedback trade more than individual investors or institutional herding impacts prices more than herding by individual investors. (3) Block Market: A place where bulk quantities of shares are traded through pick and fill basis.5 percent buying and 0.72 percent while the banking sector alone constitutes 32. In spot market. fuel and power. insurance and investment) is 74. (4) Odd Lot Market: Odd lot scripts are traded here based on pick and fill basis. Therefore. odd and block markets are cleared and settled on T+3 & T+ 7 basis while that of spot market are cleared and settled on T+0 & T+1 basis. (2) Spot Market: Spot transactions are done here through automatic matching which must be settled within 24 hours. odd and block market. N & Z are five groups of instruments.during episodes of stock market crashes. A.20 market index. pharmaceuticals. All the listed companies can be divided into 17 sectors. The micro-structure of DSE is also characterized by limited provision of information of firm’s performance to market participants as most of the firms fail to hold regular . T+0 and T+ 1 is the clearance and settlement basis. Kallinterakis and Ferreira (2006) . .5 percent selling). Starting with 9 companies. are cleared and settled through the DSE Clearing House on T+1 and T+3 basis . 3. using data from the Portuguese PSI.85 percent. calculated from date of trading. Market Micro Structure of DSE The DSE is a small capital market. empirical literature. The four markets in the system are: (1) Public Market: Only trading of market lot share is done here through automatic matching. The present study investigates the issues of feedback trading and leverage effect in DSE in an attempt to provide further evidence on emerging market. Trading is done through automated on-line system every day except Friday and other government holidays. For clearing and settlement purposes all the listed instruments are divided into 5 groups and the market is divided into 4 types. Transactions of all groups of shares except group Z. to date. Instruments of group Z traded in public . In the total market capitalization share of largest 5 sectors (banks. B. test for herding and positive feedback trading and find the presence of statistically significant herding and positive feedback trading. after netting. They also conclude that herding tends to rise when the market exhibits a definitive direction and tends to decline when the market experiences fluctuations. provides evidence that feedback trading is more pronounced in emerging markets than in developed markets.

kurtosis. due to persistence of a large number of non- active instruments.537*** ***Significant at 1 % level Rejection of normality can be partially attributed to temporal dependencies in the moments of the series especially second-moment temporal dependencies. this index should not have significant non-synchronous trading effect.665*** Ljung-box Q statistics (15) 50. we use DSE 20 return series. is therefore measured as Rt = ln( Pt) –ln (Pt-1 ) Descriptive statistics for the DSE Gen and DSE 20 daily returns are depicted in Table 1. 4.. Given that DSE 20 index consists of the most 20 blue chips companies of DSE which are frequently traded.410*** 33. The Ljung–Box (LB) test statistics applied to returns .07470 Std.0000 0. Both the series have negative skewness. Because of these microstructure biases.07360 -0.010492 0. 1994).676*** 2406.011070 Skewness -.06410 Minimum -0. Data and Research Findings The DSE General daily price index (consisting of all shares in DSE) and DSE 20 price index (blue chips companies) for the period of January 1. and kurtosis greater than three suggesting a long left tailed and leptokurtic distribution.00082 0.05760 0. measures for skewness.628*** 2 Ljung-box Q statistics (15) 78. The Jarque-Bera tests also reject the null hypothesis of normality of the return series.092 Kurtosis 5.0000 Ljung-box Q statistics (1) 27. Table 1: Descriptive statistics Statistics DSE Gen DSE 20 Mean 0.00034 Maximum 0. 2001 to December 31. Moreover. The statistics reported are the mean. To solve this problem. The processing of new information in DSE is rather weak. 0. The natural log of the relative price is computed for the daily intervals to produce a time series of continuously compounded returns.929*** Probability 0. which is the one period return in period t. the standard deviation. there is a lack of professional financial community who can analyze stock market data for the investors.634 Jarque-Bera 1699. 2007 are analyzed. and Jarque-Bera statistics.164 -. DSE return series may have non-synchronous trading induced autocorrelation which may limit our investigation of feedback trading induced autocorrelation.064*** 65.118 5. The distributions of the retrun series do not follow normal distribution. Rt . limited role of mutual funds and lack of professionally managed pension fund and limited number of investment and broker houses.annual general meetings and to provide audited financial statements on time to its shareholders. It is common to test for such dependencies using the Ljung–Box (LB) test statistics (Bollerslev et al. Dev.039*** 89.

GARCH (1. One indication at this point might be that the autocorrelation of DSE 20 return series is greater than that of DSE General return series. constant autocorrelation and autocorrelation depending on the volatility of returns may provide information about the reasons for the autocorrelation. As is explained in Cambell. Dividing the total autocorrelation into two parts. non-synchronous trading would imply positive autocorrelation in index returns. due to. The hypothesis that all autocorrelations up to 15th lags are jointly zero is rejected for both the returns and the squared returns in both the return series. One day lagged autocorrelation parameter (ρ) for both the indices are positive and statistically significant at 1 percent level.(testing for first moment dependencies) and squared returns (testing for second moment dependencies) are provided. the evidence is not clear as to whether non- synchronous trading or feedback trading is the cause of autocorrelation in return series. at this point. It simply provides an indication that first- moment and second moment dependencies are present and justifies the use of ARCH-type specification for the variance. However. . It also fails to explain whether volatility and autocorrelation are linked because of the presence of positive feedback trading. Lo and MacKinlay (1997). the LB-statistic is not capable of detecting any sign reversals in the autocorrelations due to positive feedback trading. These findings can be considered as evidence in favour of non-synchronous trading and/or feedback trading hypothesis. and we observe positive autocorrelation in Dhaka stock exchange. 1) model parameter estimations (table 2) of daily returns present statistically significant one-day lagged autocorrelations for both the DSE Gen and DSE 20 index return series. However. Thus there might have other causes behind the observed autocorrelation. Given that DSE 20 index is supposed to be less influenced by non-synchronous trading it should have lower magnitude of autocorrelation than that of DSE General Index. This provides evidence of temporal dependencies in the first moment of the distribution of returns. perhaps non-synchronous trading or market inefficiencies. Autocorrelation of the squared returns provides evidence of time- varying second moments.

0000 1 0. 2.0000 Durbin-Watson stat.5326 -6. As the volatility rises.2221*** p-value 0. the autocorrelation is captured by the coefficient of y1 . Their findings are similar to those of previous findings using other models. Koutmos (1997) introduced the term with SW (1992) model to capture the asymmetric impact of feedback trading in up and down market.1729 5.4601 Schwarz criterion -6. 4 In the model. 1) model Parameters DSE Gen DSE 20 Rt = α + ρRt – 1 + єt α 0. ***Significant at 1 % level. When the volatility is zero.0000 0. 4 Sentana and Wadhwani (1992) model does not have the term . 3 See Sentana and Wadhwani (1992) for the derivation of the model.0000 0.0001 0.19E – 06*** p-value 0.0002 p-value 0. . Shiller-Sentana-Wadhwani (SSW) model is widely applied in finance literature2 to divide the total autocorrelation into two parts.492 Akaike info criterion -6.02E-06*** 9.1864*** 0.2053*** p-value 0.091 5905.1734*** 0.We use this term for the same purpose. Sentana and Wadhwani.5574 -6.3569 ρ 0.0000  2t =  +  0 + 1 + t 2 2 t 1 t 1  1. negative feedback trading dominates at low volatility levels and positive feedback trading dominates at high levels of volatility.4404 Log likelihood 5209. impact of feedback trader in the autocorrelation structure of return series are given by the sign of  1 and  2 .0840 0.4214 * Significant at 10% level.0003* -0. 1992) theoretical model that provides a testable implication for feedback trading .1207 F statistics 5. 1984.8362*** 0. The model has two inbuilt autocorrelation parameters that help to capture effect of feedback trading .7146*** p-value 0. The reversal in the sign of return autocorrelation is consistent with the presence of positive feedback traders in the stock market.0000 0 0. constant autocorrelation and autocorrelation depending on the volatility of returns. 3 We use this specific characteristic to provide empirical evidence on the presence of positive feedback traders in DSE.0862 2. Asymmetric GARCH models are combined with SSW (Shiller. The model and the empirical results are provided in table 3. Table 2: Autocorrelation estimation of GARCH (1. the 2 One important exception in recent years is the work by Dean and Faff (2008) who uses Markov switching regime model to document feedback trading induced sign reversal in autocorrelation structure in Australain equity and bond index.0000 0. Following Sentana and Wadhwani (1992). The model has a testable implication that during periods of low volatility returns are positively autocorrelated and during periods of high volatility return autocorrelation turns negative.

coefficient  0 . the statistically significant negative  2 parameters for both the return series singles out feedback trading as the cause of return autocorrelation and at the same time rejects the effect of non-synchronous trading and time varying risk premia on the auto-correlation coefficient. As non-synchronous trading should not affect the autocorrelation structure of the highly liquid DSE 20 return series. Period of high volatility are also period of high trading volume. In the mean equation. For both DSE Gen and DSE20 return series. While the  3 parameter estimates for the return series are positive. Any increase in autocorrelation (in absolute terms) or its sign reversal form positive to negative one should be attributed to feedback trading. Negative and statistically significant  2 parameter in the daily return series for both the indices is an important evidence of positive feedback trading in DSE.autocorrelation is captured by the coefficient of y1 + y 2 . The differential effect of feedback trading in two return series can be rationalized based on the fact that DSE 20 index is highly liquid and more frequently traded.  . Negative autocorrelation in returns series suggests positive feedback trading. Negative feedback trading results in positively auto correlated stock returns and positive feedback trading results in negatively autocorrelated stock returns. is not statistically significant for both the return series. This is further confirmed by the fact that DSE 20 return series shows similar autocorrelation structure which is unexpected in case of a highly liquid. 1992). Thus in the period of high volatility the effect of non-synchronous trading should be small. which increases with the increase in volatility. A statistically significant positive  3 coefficient implies a stronger feedback effect in down market while a negative  3 coefficient implies a stronger feedback effect in the up market. The sign of the autocorrelation will be determined by the type of the feedback trading prevalent among the feedback traders. Given that DSE 20 index constitutes of securities with high level of market capitalization and high level of liquidity. While the statistically significant autocorrelation coefficient  1 (constant part of the autocorrelation coefficient ) may provide evidence of feedback trading and /or non-synchronous trading. The sign of  3 parameter indicates the asymmetry of feedback trading in up and down markets. It implies that the conditional variances (σ2t ) . These findings suggest that feedback trading has symmetric effect on DSE General return series in both up and down market and asymmetric effect on DSE 20 return series with a greater effect in the down market relative to the up market. ‘Models of non-synchronous trading do not usually predict negative autocorrelation in index returns’ (Sentana and Wadhwani. it is significant only in case of DSE 20.  0 and 1 parameters in the conditional variance equation are statistically significant at 1 percent level which means the conditional variance is affected by the last information arrival into the market as well as one day lagged conditional variance. a measure of impact of the rational investors on returns. Because positive feedback trading is associated with high volatility. a similar autocorrelation structure in both the indices reinforces the existence of positive feedback trading in DSE . risk averse investors are highly exposed to risk in down market and they off load these securities quickly in herding.

475 Akaike info criterion -6.2836 12.0000 Durbin-Watson stat.0212 0.0001 0 3.0538 0. Terraza .7669 1 0.5594 -6.5590 0.1344 Log likelihood 5217.0000 1 0. C.0000 0. Statistically significant positive  0 for both the returen series in our analysis would.0000 0 0.0001 -0.3981 1.03E-06*** 8.GARCH (1.8368**** 0.2846*** p-value 0.3803 p-value 0. In this case c=2 and τ=1 are selected. suggest the lack of rational traders’ influence on prices during volatility changes.439 5923. and M. 1. The main characteristic of the non-linear trading strategy in the mean equation of the above model is that it can take into account dynamics produced by both positive and . and in such a case  0 would be negative.1448** -389.3764 0.1) model for testing feedback trading hypothesis Parameters DSE Gen DSE 20 Rt = α+  0 (σ t) .7195*** p-value 0.2221*** p-value 0. Following Kyrtsou (2005) and Kyrtsou and Serletis (2006) we employ a Mackey- Glass-GARCH (MG-GARCH) process model to reconfirm the existence of feedback trading. The optimal c and τ are chosen on the basis of Log Likelihood and Schwarz criteria. Taking the complex behavior in stock markets into account. When the expected volatility rises. The model has either negligible or zero autocorrelations in the conditional mean.1538*** p-value 0.0011*** p-value 0.0000 0. **Significant at 5% level.1825 0.2425*** 0.6968*** p-value 0.9672 1. 2003 ).do not have statistically significant effect on index returns and thus investors are not compensated.0002 3 0.0007   +  0 + 1 + t 2 2 2 t = t 1 t 1  1.0000 2 -365. Our results reveal just the opposite.5332 -6.0000 0. with higher return for taking higher risk in DSE.1726*** 0.9789 F statistics 7. and a rich structure in the conditional variance. rational investors influence prices negatively. it is more robust than the traditional stochastic approach to model the observed data by a nonlinear chaotic model disturbed by dynamic noise (Kyrtsou. Table 3: Parameter estimations of Koutmos version of SSW.(  1 +  2 2 2 σ t) Rt -1 +  3 Rt 1 +єt α -0.4536 ***Significant at 1 % level. This model also permits to capture volatility-clustering phenomena which is treated as an endogenous process . in the CAPM framework.0000 0. therefore.4702 Schwarz criterion -6.66E-06*** p-value 0.

0000 0 0.0000 0.0000 1 0.61E-07*** 8.9954 Log likelihood 5219. All the parameters in the variance equations are statistically significant and the sum of the ARCH term and GARCH coefficient is close to one for both the return series. If  0 +  1 > 0. The coefficients  0 and  1 vary over time. . The estimated results are reported in Table 5 . 1.0037 0. We employ EGARCH model to test the robustness of leveraged effect captured in Koutmos verson of SSW- GARCH model.0037 0.4642 Schwarz criterion -6.0000 0.8427*** 0. 1 Rt 1 + єt ( c=2 and τ=1) 1  R c t  0 169.0000 Akaike info criterion -6.09E-06*** p-value 0.0000 1 -169. Table 4: Parameter estimation of MG-GARCH (1. The results of the estimate are reported in table 4.  0 +  1 >0 for both the index.0000 0. These findings support the existence of positive feedback trading in DSE. 1) model Parameters DSE Gen DSE 20 Rt  Rt =  0 .2164*** p-value 0.4491 Durbin-Watson stat.0000   +  0 + 1 + t 2 2 2 t = t 1 t 1  9.727821*** p-value 0.659917 -6.1660*** 0.1155*** p-value 0.8385*** p-value 0.9282 1. while  0 +  1 < 0 reveals negative feedback (Kyrtsou and Labys.514 ***Significant at 1 % level Asymmetric response to shocks is made explicit in Nelson’s (1991) Exponential GARCH (EGARCH) model.9216*** 146.negative feedback traders. 2007). Besides.779 5916.6712*** -145. Its advantage over simple GARCH and AR- GARCH alternatives has been shown in Kyrtsou and Terraza (2003) and Kyrtsou and Karanasos (2006). implying volatility clustering and persistence. we observe positive feedback behavior. For both the index  0 > 0 and  1 <0. It allows volatility to respond more rapidly to falls in the market than to corresponding rises which is an important stylized fact for financial assets and is known as leverage effect.677054 -6. Both the parameters are statistically significant.

1.417 5926.0016 0 5.4721 criterion Schwarz criterion -6.1266 0. .0000 1 0.2780*** p-value 0.0017 0.8672 Log likelihood 5217.9131*** p-value 0. In the model It-1 is the news entering in the market in time t-1 and the effect of good news (є t –1  0) on the conditional variance is  0 . and the effect of bad news (є t -1<0 ) is (  0 + 1 ).769 ***Significant at 1 % level.9624 F statistics 6.8770 0. along with the SSW model .6428 -6. We employ TGARCH model to test the robustness of leveraged effect captured in Koutmos verson of SSW-GARCH model.0005 1 0.0017 0.6216*** p-value 0.0000 0.3169*** 0. The estimated results are reported in Table 6.6702 -6.1) Parameters DSE Gen DSE 20 α -3.9675 10. which differentiates the effect of good news and bad news on conditional variance and provides the opportunity to test the feedback trading behavior after the good news and bad news enter the market.2506*** -0.0000 0.4480 Durbin-Watson stat.0000 2 -388.1) model.4977 2 0.8260 10.0000 0.28 1. Bohl and Siklos (2004).5376*** -1.9588 1.2386*** p-value 0.0000 0 0.8786 0.12E-05 -0. constructed a TGARCH (1.3698*** p-value 0.(  0 +  1 σ2t) Rt -1 + єt  t 1  ln(  2t ) =  +  0 + 1 ( t 1 ) +  2 ln  t21  t 1  t 1 SSW-EGARCH(1.0000 0. This means that a statistically significant 1 coefficient is evidence in favor of asymmetric effect of good news and bad news on conditional variances.1) model Rt = α+ μ (σ2t) . Table 5: Parameter estimation of SSW –EGARCH(1.0274 p-value 0.28 Akaike info -6.0798*** p-value 0.9688*** 0.0000 0 + 2 1.4171*** 379.0057  -0.0010*** p-value 0.

0692 11.9392 1. Hence.598 ***Significant at 1 % level. if є t –1  0 SSW-TGARCH(1.5894 -6.0086 1 -0.2572 0.9232*** p-value 0.9622 F statistics 7.0253 -0. When the expected volatility rises.5684* -393. The  0 parameter which implies the effect of conditional volatility on index returns is statistically insignificant for DSE Gen under both the methods.0000 0.1980*** 0.99 0. 1) model.0426 p-value 0.0005*** -0.7325*** p-value 0.0000 2 266.1984*** 0.0677 0.(  0 +  1 σ2t) Rt -1 + єt  2t =  +  0 t21 + 1 It-1  t21 +  2  t21 + t . if є t -1<0 0.003 5918. ** Significant at 5% level.2326*** p-value 0. rational investors influence prices negatively.GARCH (1.1) Parameters DSE Gen DSE 20 α 0. therefore.0000 0.80E-07 8. Statistically significant positive  0 for both the return series in our analysis would.7930*** 0.0129 0 3.Our results reveal just the opposite.5632 -6. the findings of our analysis may be interpreted to mean that the rational traders don’t influence prices when .0099*** p-value 0.2400 9. * Significant at 10% level The findings also support the result of the Koutmos version of SSW.0001  7. Table 6 : Parameter estimation of SSW-TARCH (1.1392 2 0.4805 Schwarz criterion -6. It-1 =  1 .5753 0.20E-06*** p-value 0.0031 0. its sign is positive.2807*** p-value 0. Although. 1.0000 1 0.0000 0 0.0559 Log likelihood 5197.97 Akaike info criterion -6. suggest the lack of rational traders’ influence on prices during volatility changes.  0 is expected to be negative .1) model Rt = α+ μ (σ2t) .0000 0.1) and SSW – EGARCH(1.2593 0.0000 (  0 + 1 ) / 0 1.4437 Durbin-Watson stat.0008** p-value 0.13 0. the  0 parameter is statistically significant for DSE 20 return series under both the model.82 0 + 2 0.

1992)’ because ‘higher volatility (risk) makes it harder for rational risk-averse investors to exploit the predictable pattern of stock prices (Koutmos. Hence. A minimum requirement is that the standardized residuals are zero-mean and unit-variance i. the leverage effect to be positive 1 should be negative and statistically significant.i. Statistically significant (at 1 percent significant level)  1 parameters and their positive sign for both the return series indicate positive autocorrelation arising from non-synchronous trading. All the parameters in variance equation are positive (for TGARCH model). positive feedback trading is an important reason of short-term movements in DSE. A statistically significant and positive 1 parameter in case of TGARCH model implies leverage effect in the market. as a result. processes. bad news has greater influence on the conditional variance than good news.d. it does not imply abnormal profit opportunity because ‘higher predictability arising from higher serial correlation is still compatible with equilibrium (Sentana and Wadhwani. Correct specification of the models are ensured by diagnostic checking of conditional variance equations in order to test the hypothesis that the normalized estimated residuals are i. higher predictability of returns which may imply weak form of efficiency in DSE. This argument may be supported further by our findings that rational traders don’t influence prices when the volatility rises in DSE (implied by insignificant γ0 parameter). thereby. However. Thus volatility is persistent.i. Lagrange multiplier (LM) test for the presence . The parameter which is of greater importance at this point is 1 . thus non-negativity restriction of conditional variance is fulfilled. GARCH models are estimated using robust standard errors under the assumption of normality. In the conditional variance equation. 1 parameter is not statistically significant and does not have the expected sign for both DSE Gen and DSE 20 return series under both the models. For both the index  0 > 1 . 1997)’. Whereas the statistically significant  2 parameters (at 10 percent significant level for DSE Gen. bad news and good news has the same magnitude of impact. These findings may be interpreted to mean that there is no leverage effect in DSE return series. Positive feedback traders cause negative autocorrelation in returns and. The LB-Q statistics are not significant for up to 15 lags of squared standard error for all the specifications. In case of EGARCH model. which shows the asymmetric effect of information on conditional variance.the volatility changes in DSE. parameters  0 +  2 is close to one in case of TGARCH while in case of EGARCH the sum is greater than one. In our analysis. (  0 + 1 ) /  0 is positive but not significantly high. and at 1 percent significant level for DSE 20) and their negative sign for both the return series imply positive feedback trading in DSE. thus the residuals of the estimated models are white noise with no serial dependence. In other words. Statistically significant and positive  2 parameter for both the index under EGARCH model indicates a high level of volatility clustering implying that the positive stock price changes are associated with further positive changes and vice versa. The validity of these empirical findings depends on the correct specification of the model.d.

All the models are in broad agreement with the findings that bad news and good news does not have asymmetric impact on the conditional variance and thereby the feedback trading has the same effect on DSE return series in both up and down market. the Cyprus Stock Exchange. All the models are correctly specified. These findings are consistent with the findings of Koutmos et al. and thereby. Non- synchronous trading hypothesis does not lend support to the observed autocorrelation structure in DSE retrun. Future studies should investigate other market microstructure related explanation of autocorrelation pattern in DSE. The Influence of Positive Feedback Trading on Return Autocorrelation: Evidence for the German Stock Market. (2006).. T. Autocorrelation in Capital Markets: Feedback Trading in Istanbul Stock Exchange. Why the feedback trading is not asymmetric in DSE in up and down market also merit attention for investigation. 5.19. E. Erdinc . Bohl. The findings imply significant autocorrelation in the DSE return series supportive of positive feedback trading. S. Thus the residuals do not exhibit conditional autoregressive heteroskedasticity. Reitz.of ARCH effect in the residuals reveals no such effect for the models used. Working Paper . ( 1998). Market Serial Correlation on a Small Security Market: A Note. 2 :10-21. The analysis of the asymmetric character of feedback trading in up and down market and asymmetric effect of good news and bad news on conditional volatility reveal that bad news and good news has the same magnitude of effect on the conditional variance of the DSE return series. It has important implications for other inefficient and emerging markets in that inefficiency in the form of return predictability in those markets may not imply abnormal profit opportunities. M. Linking the observed autocorrelation structure in return series to feedback trading behavior of irrational investor. the study concludes that return autocorrelation and inefficiency in DSE does not imply return predictability and abnormal profit opportunities. 43. References Altay. the feedback trading is not asymmetric in DSE in up and down market.GARCH models together this study concludes that positive feedback trading exists in DSE. 5: 1265-1274. Journal of Finance. Thus considering SSW-GARCH models and MG. Journal of Financial Management and Analysis. (2002). Berlung. Conclusions This paper examines the auto correlation structure of DSE return series in the framework of behavioral finance to test the feedback trading hypothesis. (2006) wherein positive feedback trading is documented in an emerging market. with little evidence of asymmetric feedback trading in up and down market. and Liljeblom.

(1988). L. Investment analysis and the adjustment of stock prices to common information. 4. Trinity College Dublin. A. T. T. P. Trading volume and cross- autocorrelations in stock returns. Vol IV (Elsevier Science B. .F. Richardson. Handbook of Econometrics. Time-variation in expected returns. (1994). T. Sydney..zbw. 409-425. Conrad. 39.kiel. Swaminathan. Journal of Bollerslev. Trading Behavior during Stock Market Downturns: The Dow.. M. (1990). and B. L. (2005). Dean. (2008). Nelson.30. (2004).) Ch 49.F Engle and D. Swaminathan. M. (1993). Journal of Finance 55:913–35. T. J. Quantitative Finance Research Centre . L. Journal of Finance 45. (1994). ARCH Models. (1966). J. Whitelaw. and Siklos . Brennan. Journal of Business. in R. Jegadeesh. H.V. A Tale of Three Schools: Insights on Autocorrelations of Short-Horizon Stock Returns. 7. McFadden. Engle and D. Chordia. R. Boudoukh. Some New Stock Market Indexes. Kaul. (2000).P. University of Technology. 1915 – 2004. Research paper series . Evidence of feedback trading with Markov switching regimes. Waldmann . Review of Quantitative Finance and Accounting.. and Siklos. Summers and R. Series of Postgraduate Research Programme “Capital Markets and Finance in the Enlarged Europe”. L. Warren G. Review of Financial Studies. and Robert W .191- 225.. and G. Fisher. Bohl. paper presented at the workshop of Applied Economic Research of the Deutsche Bundesbank and the Institute for International Integration Studies. Positive Feedback Investment Strategies and Destabilizing Rational Speculation. M. 61. and B. Shleifer. P. M. Available on line at opus.. B. J..P. Faff . L. 1/2002 Bohl. Emperical Evidence on Feedback Trading in Mature and Emerging Stock Markets. 539-73. M. DeLong. J. 133-151. 379 -395. eds. N. Review of Financial Studies 6:799–824. J.

G. . Terraza. (2006). (1997). (2007). M. and the profitability of momentum strategies. Journal of Finance 55:265–95. Is it possible to study chaotic and ARCH behaviour jointly? Application of a noisy Mackey-Glass equation with heteroskedastic errors to the Paris Stock Exchange returns series.(2005).porto. Andreas Pericli and Lenos Trigeorgis . (2006). C. A. 181-211. C. Kyrtsou.Evidence for neglected linearity in noisy chaotic models. Kallinterakis. pp. accepted for publication subject to revisions. Pinegar. Feedback Trading and the Autocorrelation Pattern of Stock Returns: Further Empirical Evidence. and M. M. Gregory. C. Delayed reaction to good news and the cross-autocorrelation of portfolio returns.. (2000). 45. 227-229 Lo. and S. Journal of Financial Economics.C.ucp. Koutmos. 12(3). Kyrtsou.. 3391-3394.. Kyrtsou. Available at http://dspace. Thornley (1996). Koutmos. Allaudeen and Yuanto Kusnadi. Journal of Finance. 15(10). A. 276 Kyrtsou. International Journal of Financial Markets and Institutions. (1993). 51:889–918. and MacKinlay. (2006). McQueen. T. European Journal of Finance. 3029-3056. Stock Return Cross- Autocorrelations and Market Conditions in Japan. 377(1). International Journal of Bifurcation and Chaos. Mech. G. C. Analyzing the link between stock volatility and volume by a Mackey-Glass GARCH-type model: the case of Korea.feg. Lim. 79. Computational Economics. 16. Journal of Business. and Karanasos. W. (6). Bad news travels slowly: Size. Journal of Econometrics. T. Hong. Detecting positive feedback in multivariate time series: the case of metal prices and US inflation. and W. Short-term Dynamics in the Cyprus Stock Exchange. Herding and Positive Feedback Trading in the Portuguese Stock Exchange: An Exploratory Investigation. Physica A. Labys . 21. Stein . 257. (2006). 34:307–44.. and J. (1990). H. Portfolio return autocorrelation. analyst coverage. C. S.Hamid . Vasileios and Mario Pedro Leite Ferreira . Journal of International Money and Finance. An Econometric Analysis of Nonsynchronous Trading.216. (2003).

Weak. Shiller. Shofiqur and Hossain Farhad. Dhaka Nelson . 20. R. 415–425. 2263-2295 Ogden. 8. Econmetrica. Kiel Institute for World Economics. J. Pierdzioch . Kiel Working Paper no. (2000). Kiel Working Paper no. 347-370. Whadwani. Brookings Papers on Economic Activity. (2004). Journal Financial and Quantitative Analysis.pdf Mozumder M. 2. Weak-Form Efficiency: Testimony of Dhaka Stock Exchange. D. (1999). Portfolio Serial Correlation and Non-synchronous Trading. Imtiaz. Introducing Financial Trading Rooms in Business Schhols. J. W. E. 1213. Asma and Keavin Keasey.Mobarek. June 12. Rahman. Journal of Business Research. (2006). Available at SSRN: http://ssrn. 457–498. Feedback Trading and Predictability of Stock Returns in Germany. Perry. Christian . (1985).J.. 1235. Economic 1-12. 203-17. Sentana. R. Conditional Heteroskedasticity in Stock Returns: A New Approach. (1997). Oslo. Herding and Feedback Trading by Institutional and Individual Investors. and R. 9. (2008).R. Feedback traders and stock return autocorrelations: evidence from a century of daily data. Available online at: http://www. 102. Journal of Finance 54.bath. 517-523. (1991). Review of Quantitative Finance and Accounting. Sources of Predictability of EuropeanStock Markets for High-Technology Firms. (2005). Christian and Andrea Schertler . The Fiancial Express. Presented at ENBS Conference.form Market Efficiency of an Emerging Market: Evidence from Dhaka Stock Market of Bangladesh. (1992). P. 59. . Empirical Analyses of Three Explanations for the Autocorrelation of Short Horizon Stock Index Returns.Sias. Kiel Institute for World Economics. (1984). and S. Pierdzioch . Stock prices and social dynamics.P.