Does herding behaviour vary in bull and bear markets: Perspectives from Egypt

Transcription

1 From the SelectedWorks of Pandre Samson Spring March 5, 06 Does herding behaviour vary in bull and bear markets: Perspectives from Egypt Ayman H. Metwally, Arab Academy for Science & Technology Tarek Eldomiaty, International University, Cairo Lina Ahmed Abdel-Wahab, Arab Academy for Science & Technology This work is licensed under a Creative Commons CC_BY International License. Available at:

2 6 Int. J. Behavioural Accounting and Finance, Vol. 6, No., 06 Does herding behaviour vary in bull and bear markets? Perspectives from Egypt Ayman H. Metwally Arab Academy for Science, Technology & Maritime Transport, Graduate School of Business, P.O. Box 033, Elhorria, Alexandria, Miami, Egypt Fax: (+0) Tarek Eldomiaty* Misr International University, Faculty of Business Administration and International Trade, P.O. Box Postal Code 34, Cairo, Egypt Fax: +0 () *Corresponding author Lina Ahmed Abdel-Wahab Arab Academy for Science, Technology & Maritime Transport, Graduate School of Business, P.O. Box 033, Elhorria, Alexandria, Miami, Egypt Fax: (+0) Abstract: This paper aims at examining the existence of herding behaviour in two opposite market condition: bullish and bearish. The paper extends Christie and Huang (995) and Chang et al. (000) methods. The sample includes daily and monthly data of Egyptian listed companies from January 007- December 0. The results show that: a) firms exhibit herding behaviour when the market is down; b) the Egyptian market is inefficient during bull and bear markets; c) there is asymmetric effects of trading volume and volatility on herding behaviour. The paper contributes to the literature as it shows the determinants of herding behaviour in a small market where dissemination of financial information is relatively slow and uncertainty is relatively high, thus herding differs during bull and bear movements. The paper carries practical implications in terms of detecting mispricing of stocks and market destabilisation. Keywords: herding behaviour; asymmetric herding; bull; bear; emerging markets; Egypt. Reference to this paper should be made as follows: Metwally, A.H., Eldomiaty, T. and Abdel-Wahab, L.A. (06) Does herding behaviour vary in bull and bear markets? Perspectives from Egypt, Int. J. Behavioural Accounting and Finance, Vol. 6, No., pp Copyright 06 Inderscience Enterprises Ltd.

3 Does herding behaviour vary in bull and bear markets? 7 Biographical notes: Ayman H. Metwally is an Associate Professor at the Arab Academy for Science and Technology and Maritime Transport. Tarek Eldomiaty is a Professor of Finance and Dean of Faculty of Business at Misr International University (MIU). His research interests and publications cover the areas of corporate finance and investments. He published many papers in scholarly journals such as Review of Quantitative Finance & Accounting, International Journal of Emerging Markets, International Journal of Commerce and Management, Asia-Pacific Journal of Business and Economics, International Journal of Revenue Management, International Business and Economics Research Journal, Journal of Economic and Administrative Sciences, Advances in Financial Planning & Forecasting, Macroeconomics and Finance in Emerging Market Economies, Journal of Emerging Markets Finance, and Quarterly Review of Economics & Finance. Lina Ahmed Abdel-Wahab is an Assistant Lecturer at the Arab Academy for Science and Technology and Maritime Transport. Introduction In recent years, there has been much interest in incorporating ideas from the social sciences in the economics and finance research. Behavioural finance is the application of psychology to financial behaviour (Economou et al., 00; Shefrin, 000). Herding behaviour is one of the behavioural biases that have attracted increasing attention over the past decade (Balcilar et al., 0). The literature defines herding as an obvious intent by investors to ignore their beliefs (or information) and copy the behaviour of other investors (Balcilar et al., 0; Bikhchandani and Sharma, 00). In finance, it is widely known that investors buy undervalued stocks and sell overvalued stocks. However, if investors exhibit herding behaviour, then they would buy what other investors are buying and sell whichever they are selling regardless of whether such a decision follows the wide known rule of trading securities (Caparrelli et al., 004; Lakonishok et al., 99). Herding in financial markets is of significant interest for both economists and practitioners. Economists are interested in the behavioural effect on stock prices. The potential effect on their return and risk characteristics have consequences for asset pricing models. Practitioners instead are interested in herding among investors since it might create profitable trading opportunities. The influence of investor herds has the power to drive prices away from their fundamental values (Tan et al., 008). Inspired by the imperative critical ramifications for both economists and practitioners, the mixed results of previous studies and the absence of exploration in the Egyptian market, this paper aims to examine herd behaviour in extreme market conditions using monthly data from the Egyptian stock market for the year 007 0, the presence of herding is tested as suggested by Chang et al. (000) and Christie and Huang (995). Moreover, the existence of asymmetric herding associated with market returns, trading volume and return volatility will be also examined. The current paper extends other studies of herding behaviour to an emerging market, not yet being thoroughly investigated especially under conditions of bearish and bullish market phases. The paper

4 8 A.H. Metwally et al. focuses on the existence of asymmetric herding associated to trading volume and market volatility. The remaining part of this paper is organised as follows: Section presents the literature review. Section 3 discusses the data and variables employed in the tests as well as the methodology used in this paper. Section 4 presents the empirical results. Finally, conclusion and recommendations for future research will be provided in Section 5. Literature review There are two polar views of the investment behaviour of market participants in financial markets, loosely speaking the traditional and the behavioural finance views (Lindhe, 0). The traditional framework for finance is largely built on the efficient market hypothesis (EMH) and its applications. Fama (970) defined a market as efficient if prices always fully reflect all available information (Lindhe, 0). It is based on assumptions about investor rationality and arbitrage (Lindhe, 0). The contrasting view, i.e., behavioural finance, is mainly focused on investor psychology and limits to arbitrage (Lindhe, 0; Barberis and Thaler, 003). The field of behavioural finance is said to have developed in response to a host of anomalies that cannot be explained by traditional financial models. In the aftermath of several stock crashes investor psychology has been recognised as an important influence in financial markets. Recent studies have primarily focused on the fact that individuals tend to imitate the actions of others (De Bondt et al., 008). In fact, economists as well as practitioners believe that there is extensive herding among investors in financial markets (Lindhe, 0; Devenow and Welch, 996). Herding is a behavioural anomaly which defies the EMH. According to EMH investors make informed decisions and determine their expected returns based on equilibrium models like capital asset pricing model (CAPM), therefore all of the securities are fairly priced (Prosad et al., 0).. Herding behaviour The literature on herding behaviour has offered several definitions of the concept. Banerjee (99) defines herding as everyone doing what everyone else is doing, even when their private information suggests doing something quite different. This is a general form of herding and can be applied to various situations in everyday life. In the field of behavioural finance, herd behaviour is frequently used to describe correlations in the trades as a result of interactions between market participants (Chiang and Zheng, 00). Bikhchandani and Sunil (00) argue that investors can be part of a herd if they are conscious of, and influenced by, the actions of others. The authors stated that an investor is herding if the information of other agents investing in a product causes the investor to change his or her decision from not investing to investing in the product. The opposite scenario when the investor already made an investment without knowledge of the other investors decisions and changes his or her mind when learning that others have decided not to make that investment is also a case of herding. Hwang and Salmon (004) pinpointed that Herding arises when investors decide to imitate the observed decisions of others or movements in the market rather than follow their own beliefs and information (Lindhe, 0). The most common causes of herding are information asymmetry and uncertainty. Avery and Zemsky (998) and Park and

5 Does herding behaviour vary in bull and bear markets? 9 Sabourian (0) offer algorithmic analysis that links herding behaviour with price volatility. The latter is caused by the existence of information asymmetry between parties of a transaction. In this sense, these studies suggest that market transparency (in terms of wide monitoring of price movements) helps converting investors behaviour from irrational to rational herding. The uncertainty plays an extended crucial role in determining the trend of herding. Connolly et al. (005) and Chiang et al. (03) report that as uncertainty increases, investors who lack clear market signals and fundamentals are prone to act independently. This result offers a support to the above-mentioned conclusion that market transparency encourages investors collective behaviour rationally.. The influence of herding behaviour Christie and Huang (995) conclude that the influence of investor herds in the financial market has been frequently used explanation. In the aftermath of several financial crises there has been an increased interest in the existence of herd behaviour. It is frequently argued that financial crises are a result of widespread herding among market participants (Chari and Kehole, 004). In fact, both economists and practitioners believe that extensive herding among investors in financial markets takes place (Lindhe, 0; Devenow and Welch, 996). Researchers sometimes separate the underlying motives of herd behaviour into a rational and irrational behaviour, occasionally called spurious respectively intentional herd behaviour. Kremer and Nautz (03) recognise three types of herding; unintentional, intentional and spurious herding. They define unintentional herding as Unintentional herding is mainly fundamentally driven and arises because institutions may examine the same factors and receive correlated private information, leading them to arrive at similar conclusions regarding individual stocks. They define intentional herding as intentional herding is more sentiment-driven and involves the imitation of other market participants, resulting in simultaneous buying or selling of the same stocks regardless of prior beliefs or information sets. This type of herding can lead to asset prices failing to reflect fundamental information, exacerbation of volatility, and destabilisation of markets, thus having the potential to create, or at least contribute, to bubbles and crashes in financial markets (Sulasalmi, 04)..3 Approaches of herding behaviour Herd behaviour is generally studied through one of the two different approaches. The first, which this paper follows, is often called market wide herd behaviour (Caporale et al., 008; Ohlson, 00) since it focuses on cross-sectional correlations of the entire stock market. It can also include studies that adjust for thinly traded stocks or focus on the entire distribution of a larger subsample, such as a sector. Henker et al. (006) points out that in this sense it does not detect single investor behaviour, instead it concentrates on the tendencies of the entire market or distribution of the measured population. If herd behaviour is present on the market wide level, returns of individual stocks will be more than usually clustered around the market return. This implies that investors neglect their private opinions and received information in favour of the market consensus (Ohlson, 00).

6 30 A.H. Metwally et al..4 Previous empirical studies A series of studies have attempted to detect herding using measures of dispersion around the market return during periods of significant changes in stock prices. According to Christie and Huang (995), these periods are particularly informative because a herd is more likely to form under conditions of market stress, when individual investors tend to suppress their own beliefs and follow the market consensus. They established an intuitive herding measure based on the cross-sectional standard deviation (CSSD) of single stock returns with respect to market returns. They examined the US stock market from July 96 to December 988 and their results were consistent with the predictions of rational asset pricing. Chang et al. (000) analysed the stock markets in the USA, Hong Kong, South Korea, Taiwan and Japan for the period Their results indicated no evidence of herding in the USA and Hong Kong, partial evidence of herding in Japan and significant evidence of herding in emerging markets notably South Korea and Taiwan. Moreover, Caparrelli et al. (004) examined herding in the Italian stock market for the period September 988 to January 00. Their results using Christie and Huang (995) measure did not indicate the presence of herding. However, according to the results of the non-linearity test of Chang et al. (000), herding was present during extreme market conditions. Gleason et al. (004) also examined herding using intraday data to examine whether traders herd during periods of extreme market movements in nine sector exchange traded funds (ETFs) traded on the American Stock Exchange for the period April, 999 to 30 September, 00 using the measures of Christie and Huang (995) and Chang et al. (000). According to their results investors do not herd during periods of extreme market movements using ETFs. Demirer and Kutan (006) applied the Christie and Huang (995) method to examine herding in Chinese stock market. They analyse individual firm-level returns as well as sector returns using daily stock return data from 999 to 00 and find no evidence of herding. According to their results, the Asian crisis period had no significant impact on cross-sectional standard deviations. Henker et al. (006) tested whether market wide and industry sector herding occurs intraday in the Australian equities market, using the Christie and Huang (995) and the Chang et al. (000) models. They used a sample of the 60 most actively traded stocks on the Australian Stock Exchange for the period According to their results neither market wide nor industry sector herding occurs intraday in the Australian Stock Exchange. Farber et al. (006) used the Christie and Huang (995) methodology to test for herding in Vietnam and its equity market Ho Chi Minh city securities trading centre (HSTC) for the period Their results confirm herding in extreme market conditions as expressed by Christie and Huang (995) and Economou et al. (00). Tan et al. (008) examined herding behaviour in dual-listed Chinese A-share and B-share stocks from 996 to 003. According to their results, there is evidence of herding within both the Shanghai and Shenzhen A-share markets that are dominated by domestic individual investors, and also within both B-share markets, that are dominated by foreign institutional investors. Evidence of herd behaviour over daily time intervals is much stronger, revealing the short-term nature of the phenomenon. The Asian crisis did not influence the markets under examination. Herding behaviour by A-share investors in the Shanghai market is more intense during periods of rising stock markets, high trading volume and high volatility. However, there is no asymmetry in the B-share market. Caporale et al. (008) examined herd behaviour in extreme market conditions using data from the Athens Stock Exchange. They tested for the presence of herding as described by

7 Does herding behaviour vary in bull and bear markets? 3 Christie and Huang (995) and Chang et al. (000). Their results indicated the existence of herd behaviour for the period When the testing period is broken into semi-annual sub-periods, there is evidence of herding during the stock market bubble of 999. In addition to that, Kallinterakis and Lodetti (009) examined the impact of thin trading over herding in the New Securities Stock Exchange of Montenegro using the non-linear model of Chang et al. (000). Their study covers the period March 003 May 008 and there was no evidence of herding even when correcting for thin trading. Recently, Chiang and Zheng (00) examined herding behaviour in 8 global markets (Australia, France, Germany, Hong Kong, Japan, UK, USA, Argentina, Brazil, Chile, Mexico, China, Indonesia, Malaysia, Singapore, South Korea, Taiwan and Thailand) from 5 May, 988, through 4 April, 009, using Chang et al. (000) model. They found evidence of herding in advanced stock markets (except the USA) and in Asian markets. Moreover, stock return dispersions in the US played a significant role in explaining the non-us market s herding activity. Finally, there was supportive evidence for herding in the USA and Latin American markets during crisis periods. Moreover, Economou et al. (00) examined herd behaviour in extreme market conditions using daily data from the Greek, Italian, Portuguese and Spanish stock markets for the years , i.e., the existence of asymmetric herding behaviour associated with market returns, trading volume and return volatility. Along with this, they also investigated the presence of herd behaviour during the global financial crisis of 008. The results of the study showed that herding is found to be stronger during periods of rising markets in these stock markets. Herding is present in the Portuguese stock market during periods of down returns and there is no evidence of herding in the Spanish stock market. Finally, it was claimed that there is an evidence of herding during the global financial crisis of 008 only for the Portuguese stock market and evidence of anti-herding for the Spanish and the Italian stock markets. Investor behaviour seems to have been rational for the Greek stock market during the global financial crisis (Subash, 0). Ohlson (00) examined the stockholm stock exchange for the existence of herd behaviour with a market wide approach during using three models developed by Christie and Huang (995) and Chang et al. (000), herd behaviour was found in upgrowing market days and also in bullish markets of 005 and 007. In days with the most extreme market movements herd behaviour was found in large cap stocks. Moradi and Abbasi (0) examined the Tehran exchange, indicating that there is no herding in the Tehran stock market during Lindhe (0) studied four Nordic countries (Denmark, Finland, Norway and Sweden), with regard to their propensity to exhibit herd behaviour. They applied the approach of Chiang and Zheng (00) to detect market-wide herding during Significant evidence of local market-wide herding was found in Finland but no evidence of herding was found in Denmark, Norway and Sweden. Moreover, Sulasalmi (04) examined the presence of herding in the Finnish stock market during by applying the market-wide approach. The results indicate the absence of herding behaviour in the Finnish market. As for Middle Eastern countries, Balcilar et al. (0) examined herding behaviour in the Gulf Arab stock markets (Abu Dhabi, Dubai, Kuwait, Qatar and Saudi Arabia) taking into account herding under different market regimes (low, high and extreme or crash volatility). Their results supported the presence of herding behaviour under the crash regime in all of the markets except Qatar which herds under the low and high volatility regimes. Ramadan (05) examined the presence of herding in the Amman stock

8 3 A.H. Metwally et al. exchange (ASE) during using Chang et al. (000) model. According to the results, herd behaviour was found in the Amman stock exchange during the period under investigation. Heshmat and Mahran (04) examined the Egyptian stock exchange during using all the stocks listed in EGX00. The results found no presence of herding at the market level. On the other side, at the sector level herd behaviour was present in only 3 sectors out of 5 (basic resources, construction and materials and real state). 3 Estimation methods, data and variables This section presents the approach of this paper as well as two pioneering models to detect herding behaviour in the equity market. The pioneering methods to detect herding behaviour was presented by Christie and Huang (995) and then this model was further developed by Chang et al. (000), henceforth referred to as CH and CCK. Using these models, a theoretical framework was developed to examine market-wide herding behaviour and asymmetric herding during market stress. The CH model suggests that a suitable measure of the market impact of investors herds is dispersion. As it measures the average proximity of individual returns to the market return. Dispersions are bounded from below zero. When the individual returns differ from the market return the level of dispersions increase. Thus, market-wide herding would indicate a decrease in dispersions. The cross-sectional standard deviation is used as a measurement of dispersion (CSSD). In addition, the authors suggest that individuals are more likely to follow the performance of the market during periods of large market movements. This means that investors will base their investment decisions only on the performance of the market. As a result, individual returns will not differ significantly from the market return. This means that the level of dispersions, i.e., CSSD will be lower than during normal market conditions. This is in contrast to rational asset pricing models were dispersions are assumed to increase during periods of large market movements. The authors also present a measurement for the cross-sectional absolute deviation (CSAD). In their paper, CCK extends the work of CH and presents a modified and less stringent method to detect market-wide herding. They assume, as CH, that rational asset pricing models suggest an increase in the dispersion during periods of market stress. In addition, they argue that rational asset pricing models would predict the relation between dispersions in individual assets and the market return to be linear. This means that the dispersions are an increasing function of the market return. As a measurement of dispersion the authors use CSAD which they based on the conditional version of the CAPM. Hence, the presence of herd behaviour in the market would not only imply a decrease in dispersions, but also a non-linear relation between the dispersions and the market return. This means that the dispersions will decrease or at least increase at a less-than-proportional rate with the market return. In contrast to CH, the method of CCK is able to detect herding during more normal conditions in addition to periods of market stress. Christie and Huang (995) and Chang et al. (000) supported that herding can be expressed using cross-sectional analysis of asset returns, since the smaller cross-sectional dispersion of returns indicates parallel movement with the cross-sectional mean return, that is to say movement to some type of market consensus.

9 Does herding behaviour vary in bull and bear markets? 33 Christie and Huang (995) estimated the cross-sectional standard deviation (CSSD) of single stock returns with respect to market returns, which is expressed as: CSSD N ( R ) i, t Rm, t i= t =, () N where R i,t is the observed stock return of firm i at time t, R m,t is the cross-sectional average return of the N returns in the market portfolio at time t, and N is the number of stocks in the market portfolio. The CSSD of returns was then regressed against a constant and two dummies in order to identify the extreme market phases, with DL = if the market return on day t lies in the extreme % and 5% lower tail of the distribution of market returns (and zero otherwise), and DU = if it lies in the extreme % and 5% upper tail of the same distribution (and zero otherwise): CSSD = α + bd + bd + е, () L U t t t t where the α coefficient denotes the average dispersion of the sample excluding, the regions corresponding to the two dummy variables. According to this model, statistically significant negative values for b and b indicate the presence of herd behaviour. When the individual returns herd around the market consensus, dispersions are predicted to be relatively low. By contrast, rational asset pricing models predict an increase in the dispersion because individual assets differ in their sensitivity to the market return. Although the cross-sectional standard deviation of returns is an intuitive measure for capturing herding, it can be considerably affected by the existence of outliers. That is why Christie and Huang (995) as well as Chang et al. (000) proposed the use of the CSAD as a better measure of dispersion: N R R = (3) N i= it, mt, C SADt, where R i,t is the observed stock return of firm i at time t, R m,t is the cross-sectional average return of N stocks in the portfolio at time t, and N is the number of stocks in the portfolio. The equation for the CSAD corresponding to equation () is the following: CSAD = α + bd + bd + е (4) L U t t t t Chang et al. (000) proposed the first non-linear model framework for testing herding. Their empirical model is built on the intuition that under CAPM assumptions, rational asset pricing models predict that the equity return dispersions are not only an increasing function of the market return but also that the relation is linear. In the presence of herding, the relation can become non-linearly increasing or even decreasing. Chang et al. (000) proposed an alternative approach to the one suggested by Christie and Huang (995), using the entire distribution of market returns, as in the following equation: CSAD = α +ϒ R + ϒ R + ε. (5) t m, t m, t t While this method is similar in spirit to Christie and Huang s (995), they may provide conflicting results with regard to the presence of herding. That is because the Christie and Huang (995) approach is a more stringent test, which requires a far greater magnitude of non-linearity in order to find evidence of herding (Tan et al., 008).

10 34 A.H. Metwally et al. The relationship between CSAD t and R m,t is used to detect herd behaviour. A statistically significant negative coefficient γ implies the presence of herd behaviour. This is likely to increase the correlation among individual asset returns, and the dispersion between asset returns will either increase at a decreasing rate or decrease in the case of severe herding. If market participants are more likely to herd during periods of large price movements, then there should be a less than proportional increase (or decrease) in the CSAD measure. In the absence of herding, the relationship is linear and increasing, that is the dispersion increases proportionately with the increasing returns of the market. Herding behaviour during extreme market returns It has been suggested that the direction of the market movement has an effect on the relationship between CSAD and the market return, referring to herding being more likely to present in either bull or bear markets. To test for this asymmetry, the dataset is divided based on either two instances (increasing or decreasing market returns) and analysed separately as in Chiang et al. (007). In accordance with the approach to examine the asymmetric effect of market returns, the following models were developed. In a bull market: ( ) CSAD = α +ϒ R + ϒ R + ε, if R 0. (6) In a bear market up up UP up UP t t mt, mt, t mt, ( ) CSAD = α +ϒ R +ϒ R + ε, if R < 0, (7) DOWN DOWN DOWN DOWN DOWN t t m, t m, t t m, t where γ and γ would capture the linear and the non-linear relation, respectively, between CSAD and the market returns. There are many scenarios for γ and γ. If γ is negative then this is a concrete evidence of herding. However, if γ is positive, but γ is negative then this is an indicator that CSAD is increasing at a decreasing rate with the returns of the market which the CCK model would also interpret as herding. Yet, if γ is positive, then this is an indicator that the market is inefficient, but herding is not present (Chang et al., 000, pp ; Caparrelli et al., 004, p.6). The absolute values UP DOWN R and R are used because we are concerned with the size of the return, not mt, mt, UP DOWN with its sign. This also makes a comparison between γ and γ possible. Tan et al. (008) examined the existence of asymmetric herding behaviour associated with market returns, trading volume and return volatility. Trading volume asymmetric effects Tan et al. (008) used the following equations in order to examine the possible asymmetric effects of high and low trading volume. During high trading volume ( R ) CSAD = α +ϒ R + ϒ + ε, (8) V-HIGH V-HIGH V-HIGH V-HIGH V-HIGH t m, t m, t t During low trading volume ( R ) CSAD = α +ϒ R + ϒ + ε, (9) V-LOW V-LOW V-LOW V LOW V-LOW t m, t m, t t

11 Does herding behaviour vary in bull and bear markets? 35 where the superscripts V-HIGH and V-LOW refer to high and low trading volume. Trading volume V t is regarded as high if on day t it is greater than the previous 30 day moving average. Trading volume is regarded as low if it is less than the previous 30-day moving average. They also used 60-day, 90-day and 0-day moving averages and obtained similar results. In this paper, the 30-day moving average is used to detect the asymmetric effects of trading volume on the relation between CSAD and market returns. Asymmetric herding is present during high and low trading volume if the ϒ and ϒ are significant and negative. Return volatility asymmetric effects Moreover, they used the following equations in order to examine the possible asymmetric effects of high and low volatility. During high return volatility: CSAD = α +ϒ R + ϒ ( R ) + ε. (0) σ,high σ,high σ,high σ,high σ,high t t m, t m, t t During low return volatility: CSAD = α +ϒ R + ϒ ( R ) + ε () σ, LOW σ, LOW σ, LOW V-LOW V-LOW t t m, t m, t t where the superscripts (σ, HIGH) and (σ, LOW) refer to high and low return volatility and σ t is calculated as the square of the portfolio return in period t. The volatility is characterised as high if on day t it is greater than the previous 30-day moving average and low when it is below the 30-day moving average. In this paper, the presence of herding is tested as suggested by Cheng et al. (000). Moreover, we examine the existence of asymmetric herding behaviour associated with market returns, trading volume and return volatility as suggested by Tan et al. (008). To examine herd behaviour, data collected are daily data on stock prices and trading volume the actively traded stocks in the Egyptian stock markets available on Egypt for information Dissemination EGID on day t. Daily stock percentage log-differenced returns from January, 007 to December 0 was used. Variables In this section, the variables that will be used in CH and CCK models are discussed. In both models, there are eight regression equations used to detect herding in normal and extreme market conditions. The CH model measures the effect of herding behaviour on the market using CSSD of returns and this deviation determines the mean return closeness of every stock to the average. In equation (), the dependent variable is the monthly CSSD while, the independent variable is the market return. CCK model is used to measure asymmetric herding associated with market returns, trading volume and return volatility. Equations (6) and (7) are used to detect herding during bull and bear market so monthly CSAD is used as a dependent variable, while market return is the independent variable in both regression equations. Equations (8) and (9) are used to test for asymmetric herding behaviour associated with high and low trading volume. Monthly CSAD is used as dependent variable, while the market return during high or low trading volume is the independent variable.

12 36 A.H. Metwally et al. Equations (0) and () are used to test for asymmetric herding associated with high and low volatility. In this regression model, monthly CSAD is the dependent variable and the market return during high or low return volatility is the independent one. The theoretical framework provides the basis to develop set of testable hypotheses as shown in the following section. Hypotheses This section introduces the hypotheses that will be tested in light of the insights drawn from the literature review and the theoretical framework. Hypothesis : Herd behaviour is present in the Egyptian stock market. Hypothesis : There is a significant negative relationship between CSAD and market returns in both bull and bear markets. Hypothesis 3: There is a significant negative relationship between CSAD and market returns during high and low trading volume. Hypothesis 4: There is a significant negative relationship between CSAD and market returns during high and low volatility. Data The data used in this paper consist of daily closing prices, trading volume in shares, market returns and finally risk free rates for the time period January 007 December 0. All the data was collected from Egypt for information dissemination (EGID) which is a joint-venture company between the Egyptian Exchange EGX and NASDAQ (OMX). The data obtained is used to calculate the monthly returns for each stock (daily stock percentage log-differenced return was used), the following formula was applied: ( ( ) ( )) Rt = 00 log Pt log Pt, where R t is the monthly stock returns change between day t and day t and P t accounts for month ending price and P t is the month beginning price. The data are also used to calculate 30-day moving average trading volume as well as the return volatility (market return risk free rates). The calculated data are used in the CSSD and CSAD equations to detect the existence of herding and asymmetric herding behaviour. The timeframe covers a highly interesting period in the Egyptian market, during 5th of January revolution in 0 and the closing of the Egyptian stock market during February 0 accompanied by some other political events that significantly affected the investors trading decisions during this period. This period characterised by high volatility and extreme bear and bull markets that can be used to justify the presence of herd behaviour and give a solid explanation for the asset price bubbles (Rannou, 00) and the destabilisation of financial markets (Bikchandani and Sharma, 000). Monthly returns were preferred over daily or weekly returns because some stock prices do not change much on either daily or weekly basis (Omran, 005) this will make it harder to calculate the return dispersion and detect any decrease or increase. Another aspect about the data is that the available number of companies was limited compared to previous studies that used the same methodology to detect herding. This is mainly due to the limited number of listed companies in the Egyptian stock market.

13 Does herding behaviour vary in bull and bear markets? 37 The data used in the models were adjusted by excluding stocks characterised by thin trading or no trading, thus bias the results towards liquid stocks. Kallinterakis (007) defines thinly traded stocks a security that show an illiquid trading pattern over time. Therefore, the number of stocks varies every year, according to whether they are traded or not. 4 Empirical results 4. Descriptive analysis Regarding the current research, all data collected are quantitative data. Therefore, measures of central tendency and dispersion will be used to describe the dimensions in this paper. The Mean is a measure of central tendency; which can be defined as a single value that attempts to describe a set of data by identifying the central position within that set of data. The mean (often called the average) is the most likely used measure of central tendency (Weisberg, 99). As for measures of dispersion, they express quantitatively the degree of variation or dispersion of values in a population or in a sample. Along with measures of central tendency, measures of dispersion are widely used in practice as descriptive statistics. Some measures of dispersion are the variance and standard deviation. Through this section, the mean, variance, as well as the standard deviation are presented in Table. Table Descriptive statistics for market return, CSSD and CSAD Measures Market return CSSD CSAD Mean Median Mode a 0.050a Std. deviation Variance Range Minimum Maximum Number of stocks (N) Maximum 94 Minimum 96 This table reports descriptive statistics for the measure of monthly CSSD and CSAD of individual stock returns with respect to the market returns for the Egyptian market during the period January 007 December 0. Table shows the descriptive statistics of market return, CSSD and CSAD. It describes data that were used in tests focusing on the entire market. Using all the stocks traded in the Egyptian market after adjustments. It immediately becomes clear that the market return fluctuates substantially, covering a range of approximately 6%, this range is not exceptional when taking into consideration the effects of the economic crisis of 008 on the Egyptian stock market. This results in a significant decline in the market return reaching its minimum rate ( 33.%) in November 008. The Egyptian stock market

14 38 A.H. Metwally et al. started recovering and reached its maximum in January 0 (8.3%). It should be noted that these maximum and minimum represents extreme market returns and that their influence as outliers in the regressions are considered. The higher the mean, the more dispersion in absolute term has been prevalent during the sample period (Sulasalmi, 04). The CSSD mean is higher than the CSAD, but that is expected due to the nature of CSSD compared to CSAD. Additionally, all data was in some test separated according to up and down going market months (the so called bull and bear analysis). Finally, the maximum and the minimum number of stocks used in this paper is reported after adjusting for the inactive or thinly traded stocks as suggested by Kallinterakis (007) who stated that thinly traded stocks are an issue primarily for emerging markets. Many other authors supported this view including Hwang and Salmon (004), Henker et al. (006) and Ohlson (00). 4. Data testing In this section, the assumptions of normality will be verified, as one important assumption for parametric testing, such as regression analysis. Also, ordinary least squares method assumptions will be verified to be able to apply regression analysis and fit models according to CH, 995 and CCK, 000 models. Those assumptions are multi-collinearity, autocorrelation and heteroscedasticity. Regarding linearity, it had been assumed by CH and CCK models is working under the non-linearity assumption. 4.. Normality test This assumption is one of the most important assumptions of the ordinary least squares method used by regression analysis, as well as for all parametric tests in general. A dataset should be normal or well modelled by a normal distribution. A normality test is used to determine if a dataset is normal and to compute how likely it is for a random variable underlying the dataset to be normally distributed. An assessment of the normality of data is a prerequisite for many statistical tests because normal data is an underlying assumption in parametric testing. Van Der Waerden test could be used to converts the ranks from a standard Kruskal Wallis one-way analysis of variance to quantiles of the standard normal distribution. These are called normal scores and the test is computed from these normal scores. Applying test of Kolmogorov, Table was obtained. It shows the results of testing, where it could be claimed that the data are not all normally distributed, as P-value < 0.05, except for market return. In alignment with the above results, the rule of thumb will be applied, as a second stage to prove normality of the variables, using skewness and kurtosis. Table 3 shows the results obtained by applying rule of thumb. It shows that skewness and kurtosis of some variables such as CSSD is normally distributed as skewness and kurtosis are between 3 and 3, while other variables are shown to be not normally distributed. Despite the fact that few variables are not normally distributed, the authors claim that data are approximately normal using central limit theory as the sample size is large enough that amounts to 7 observations (Farrell, 006). At this point, it should be highlighted that this sample size is considered as an adequate sample for the required analysis. As mentioned by Siddiqui (03), an adequate

15 Does herding behaviour vary in bull and bear markets? 39 sample size is 0 observations per each predictor. In the current paper, two predictors are considered, which means the adequate sample starts by 40 observations ( 0). This means that a sample of 7 observations is safe for the required analysis. Table Normality testing by Kolmogorov Kolmogorov-Smirnov Shapiro-Wilk Statistic Df P-value Statistic Df P-value Market return * CSSD CSAD DL DU *Insignificant at %. Table 3 Normality testing by rule of thumb Observations Skewness Kurtosis Statistic Statistic Std. error Statistic Std. error Market return CSSD CSAD DL DU Testing autocorrelation Another important assumption of the OLS method used for regression analysis is that residuals should be independent. To check residuals independence, the Durbin Watson test was conducted to test for the models fit. Through the current paper, two main models will be fitted. Accordingly, autocorrelation problem will be checked using the Durbin Watson test for each model. The results are shown in Table 4, where it was found that the Durbin Watson computed values are all greater than.68, which is the Durbin Watson upper limit in the table. This means that the computed value is large enough to imply that residuals are independent from each other. Table 4 Durbin Watson testing for autocorrelation Models Description Durbin Watson test Model CSSD (Dependent variable) DU, DL.958* (Independent Variables) Model CSAD (Dependent variable) DU,DL (Independent variables) *Durbin Watson test is significant for the values greater than *

16 40 A.H. Metwally et al Testing multi-collinearity Multi-collinearity occurs when two or more predictors in a model are highly correlated so as they provide redundant information about the response. When checking multi-collinearity problem, it was found that none of the variables is suffering a problem of multi-collinearity. Table 5 shows that the VIF values are less than five, which means that the independent variables in the specified model are not inter-correlated among themselves implying that the problem of multi-collinearity does not exist (Studenmund, 0). Table 5 Multicollinearity testing using VIF Variables VIF* D U.005 D L.005 *VIF is acceptable for values less than Testing homoscedasticity The issue of homoscedasticity is addressed by examining the regression standardised residuals for both CSSD & CSAD, these are used to check the absence of heteroscedasticity, as residuals are scattered along the graph. The results show that data are close to being homoscedastic Testing linearity The issue of linearity vs. non-linearity is examined throughout the entire variables. The results of the partial plots show that the observed values fit nonlinear relationships and this is consistent with the assumptions of the models used. The Ramsey regression equation specification error test (RESET) test is a general specification test for the linear regression model this test could also be used to test whether the non-linear combinations of the fitted values help explain the response variable. The intuition behind the test is that if non-linear combinations of the explanatory variables have any power in explaining the response variable, the model is misspecified. 4.3 Christie and Huang (995) model results To test the hypothesis of this paper, CH model is applied to detect herding behaviour. CH model supported the idea that herding can be expressed using cross-sectional analysis of asset returns. The first hypothesis in this paper states that herding behaviour is present in the Egyptian stock market. To test this hypothesis CH model is implemented by calculating the monthly CSSD for each stock then regressed against two dummies (lying at the upper and lower % and 5% tails of the market return dispersion) in equations () and (), in order to identify the extreme market phases. DL is equal to if the market return on day t lies in the extreme % and 5% tail of the distribution of the market returns and zero otherwise and DU will be equal if it lies in the extreme upper tail of the same distribution and zero otherwise. Thus, with as a

17 Does herding behaviour vary in bull and bear markets? 4 dummy variable the α parameter represents the rest of the population, the normal phase of the stock market, hence marked with the dummy variable zero in the test. The adjusted R value is used to explain the model s goodness of fit. That is, the percentage of variance in the independent variable that is explained collectively by the independent variables. It is useful as a quality indicator, especially in comparison to the other models since two of the models used the same dependent variable. Not surprisingly, the high value of the adjusted R is found with the 5% criterion model explains the dependent variable variance to a higher degree than tests with the % criterion. The results of the test of herding using CH model are given in Tables 6, 7. and 7.. It is obvious from the results that using both upper and lower % and 5% of the market returns, respectively, proxy for the periods of market stress. Tables 7. and 7. report a significant α, as P-value is equal to 0.000, which is less than Also, the coefficient of D L (b ) is negative significant as b is equal to and P-value is equal to 0.038, which is less than In addition, the coefficient of D U (b ) is insignificant because P-value is equal to which is greater than Table 6 Regression Model using CSSD as the dependent variable 5% Criterion % Criterion D L D U Table 7. CSSD regression model results Coefficient Std. error T-stat P-value Intercept (α) ** D L (b) * D U (b) Adjusted R = t-statistics based on Newey-west (987) consistent standard errors. *and ** represent statistical significance of 5% and % respectively. Table 7. CSSD regression model results Model Unstandardised coefficients Standardised coefficients T P-value B Std. error Beta (Constant) D L * D U According to the above-mentioned model, statistically significant negative values for b and b indicate the presence of herd behaviour. On the other hand, rational asset pricing models would expect the values of b and b to be significantly positive. In the current paper, b is shown to be negative significant indicating that during the time period under

18 4 A.H. Metwally et al. investigation investors in Egypt follow the performance of the market and ignores the individual characteristics of the stocks, which indicates the presence of herding when the market is going down. These results are consistent with the results of Chiang and Zheng (00), Hwang and Salman (004) and Khan et al. (00). On the other hand, b is shown to be insignificant, which indicates the absence of herding behaviour during periods of large market swings these findings are consistent with Lin and Swanson (003) who tested the Taiwan stock market using the same methodology. The insignificant value of b indicates that when the market is going up, there is an insignificant relation between equity dispersion and market returns. In both cases, it could be claimed that there is no evidence of rational asset pricing, as none of the coefficients is shown to be positively significant. The acquired results explain the presence of herding at monthly intervals of all investigated stocks in this research using the research method based on CSSD of stock returns during The results indicate that the stock return dispersion from the market return decreased in low market fluctuations, but in case of large market movements, the results were insignificant. So it could be stated that the presented results are inconsistent with the forecasting of the CAPM, which proposed that at the time of market stress, deviation level is increased since the sensitivity of every asset in relation to the market is different. These results are in accordance with the results of Sulasalmi (04), who found that the Finish stock market herds when the market is going down (low market fluctuations) using the CH model. On the contrary, CH found that extreme negative market returns days, do not disproportionally lower dispersion, but rather increase it in the US market. Meaning that no herding was found in the % and 5% tails. In the Egyptian market, the lower end of return dispersion is associated with decreased return dispersion. In comparison with previous studies, Demirer et al. (007) find no evidence of herding using the CH model in each of the regions of their sample; Africa, Asia, Developed countries, Eastern Europe, Latin America and the Middle East. 4.4 Chang et al. (000) model results Although the cross-sectional standard deviation of returns is an intuitive measure for capturing herding, it can be considerably affected by the existence of outliers. That is why Christie and Huang (995) as well as Chang et al. (000) proposed the use of the CSAD as a better measure of dispersion. The results of the CCK s linearity model is presented in Tables 9. and 9., it shows the results of the regression equation (5) the constant α is significant, as P-value is equal to 0.000, which is less than Also, the coefficient of D L is negative significant because b is equal to and its P-value is equal to 0.04 which is less than In addition, the coefficient of D U is insignificant as its P-value is equal to 0.66 which is greater than The following results correspond to the return dispersion model (5), estimated in the Egyptian stock market for the whole sample period January 007 December 0. The most important observation can be made from Tables 8, 9. and 9. that the CSAD has a negative significant relationship with low market returns (in case of the market is going down).