The existence of herding behaviour measured by
co-movement
Evidence from Germany and the UK
Nicole Elianne Timmerhuis University of Groningen
S1901656 Faculty of Economics and Business
Grote Kromme Elleboog 7 MSc. Finance
9712 BJ Groningen January 2015
The existence of herding behaviour measured by
co-movement
Evidence from Germany and the UK
ABSTRACT
This paper examines the existence of herding behaviour in Germany and the UK, based on the co-movement of stock returns as an indicator of herding behaviour, during January 1998-December 2008. In addition, it tests if herding behaviour is stronger present during periods of market stress and if there is correlation between Germany and the UK. Overall, the results show that there is no evidence for the existence of herding behaviour in Germany. For the UK, it reports evidence for the existence of herding behaviour, which is stronger present during periods of market stress. Lastly, the paper observes a strong correlation between Germany and the UK.
JEL classifications G01, G02, G12, G15
1.
INTRODUCTION
‘Men, it has been well said, think in herds; it will be seen that they go mad in herds, while they only recover their senses slowly, and one by one.’ Charles Mackay (1884)
Over the years, many academic researchers tried to understand the investment behaviour of market participants and its impact on the market. According to the efficient market hypothesis of Fama (1970), all investors are rational and receive the same market information. This implies that stocks would reflect all market information and trade at their fair value. However, practice shows that investors do not always respond rationally. Their irrational behaviour influences the markets and ensures that prices may be driven away from their equilibrium values. In the past decades, this caused a growing interest in behavioural finance. Behavioural finance studies the influences of psychological factors on the economic decisions of investors. Herding behaviour is one of the most interesting concepts of behavioural finance. It can be defined as a similar behaviour pattern among investors, mainly due to mutual imitation (Hirschleifer and Teoh, 2003).
In this paper, the existence of herding behaviour is investigated in Germany and the United Kingdom (UK). Besides the existence, it examines if herding behaviour is stronger present during periods of market stress, i.e. declining markets. Herding behaviour can be distinguished into a rational and irrational form. The rational view states that investors intentionally mimic the actions of other investors, convinced that these investors are better informed or possess better analytical skills than they do. The irrational view believes that investors base their decisions on the decisions of other investors rather than on their private information and analysis, without any fundamental reason (Scharfstein and Stein, 1990; Devenow and Welch, 1996). Both explanations of herding behaviour illustrate that investors follow the market consensus. Herding behaviour suggests that investors trade in the same direction, which would imply that stock returns behave more similar (Nofsinger and Sias, 1999). Based on this co-movement of stock returns, Christie and Huang (1995) develop a model to measure the existence of herding behaviour with the use of stock returns. The relationship between the cross-sectional standard deviation (CSSD) of stock returns, or the co-movement of stock returns, and the market return in the US equity market is an indicator of herding behaviour. Chang, Cheng and Khorana (2000) extend the CSSD model with a small adjustment and turn it into the cross-sectional absolute deviation (CSAD) to measure the co-movement of stock returns. By doing this, the non-linear relationship between CSAD and the market returns can be captured. This paper follows the approach of Christie and Huang (1995) and Chang, Cheng, and Khorana (2000) to find evidence of the existence of herding behaviour in Germany and the UK.
markets (low returns), high trading volume and high volatility. Herding behaviour is stronger present during periods of market stress. During an economic crisis, the market is in stress. For perfect example is the recent global crisis of 2007-2008. Therefore it is not surprising that herding behaviour became once again a very interesting concept. Economou, Kostakis, and Philippas (2011) wonder if herding behaviour has a big impact on the recent global crisis, especially in South-European countries that faced a hard time surviving the crisis (Greece, Italy, Portugal and Spain). They combine the models to measure herding behaviour of Christie and Huang (1995) and Chang, Cheng, and Khorana (2000) with the empirical research of Gleason, Mathur and Peterson (2004) and test the existence of herding behaviour in South-European countries. Thus, besides the existence of herding behaviour, they examine if herding behaviour is stronger present during periods of market stress.
So, the herding behaviour in South-European countries was already investigated, but what about the North-European countries? The economy of Europe can be seen as a whole, but does this imply that the investors in North-European and South-European countries are similar as well? Do North-European countries, which in general survived the global crisis of 2007-2008, exhibit herding behaviour? This research adds to the existing literature by examining the existence of herding behaviour in two North-European countries, Germany and the UK. In addition, I examine the presence of herding behaviour during periods of market stress. Herding behaviour during market stress might have a big impact on market equilibriums, therefore it is worthy to examine if it is stronger present during this periods. The article of Economou, Kostakis, and Philippas (2011) is the guideline of this empirical research. Following their research, I provide insight to the impact of herding behaviour during the recent global crisis of 2007-2008 in Germany and the UK. The two countries are qualified as developed countries by the MSCI1 and survived the recent global crisis. To examine the existence of herding behaviour I construct a dataset consisting of the stock returns from all listed companies in Germany and the UK during the sample period. The sample period is January 1998-December 2008, and I use daily returns and monthly returns. The monthly returns are used as a robustness check. I use the co-movement of stock returns as a measurement of herding behaviour. The co-movement of stock returns is calculated as the cross-sectional absolute deviation (CSAD), which is used by Chang, Cheng and Khorana (2000). I expect that in presence of herding behaviour, the stock returns will move similar with the market returns and therefore the CSAD will decrease (or increase at a decreasing rate) with an increase in the market return. In general, my results show that there is no evidence of herding behaviour in Germany and evidence of herding behaviour in the UK.
Gleason, Mathur, and Peterson (2004) discuss that during market stress, several characteristics have a stronger presence. This characteristics are low market returns, high trading volume and high volatility. I test if herding behaviour is stronger present during market stress with regard to market returns. I expect that herding behaviour would be stronger present during declining markets, i.e. with low returns. To make my research more robust, I examine if herding behaviour is stronger present
during periods with high trading volumes and high volatility as well. I can conclude that there is no evidence of a bigger magnitude of herding behaviour during market stress in Germany. The UK exhibits more herding behaviour during periods of market stress.
At last, I wonder if the movements of stock returns in Germany can be explained by the co-movement in the UK. Can the sectional dispersion of one country be explained by the cross-sectional dispersion of the other country? I discover a strong correlation between Germany and the UK, indicating that herding behaviour in one country has influence on the herding behaviour on the other country. According to this strong correlation, it would be convincing if the markets exhibit similar movements. This is a remarkable result taking into account that the previous regressions all show a difference between Germany and the UK.
2.
LITERATURE
In its most general form, herding behaviour can be defined as a similar behavioural pattern of individuals (Hirschleifer and Teoh, 2003). If investors behave similarly, they form a herd. A simple explanation for this behaviour could be that all investors receive market information at the same time, resulting in the purchase or sell of stocks at the same time. Besides the market information, is it possible that investors base their decisions by the observations of other investors (Devenow and Welch, 1996). In the financial literature, herding behaviour can be distinguished into two perspectives, the rational and irrational view.
The rational perspective, discusses that herding behaviour arises from reputational reasons. It centres on externalities, optimal decision making being influenced by information difficulties or incentive issues (Devenow and Welch, 1996). Investors intentionally mimic the actions of other investors, convinced that these investors are better informed or have better analytical skills than they possess (Devenow and Welch, 1996). Investors do not expect to receive excessive returns by their herding behaviour, but, at the minimum, the average market return (Gleason, Mathur, and Peterson, 2004). They protect their own interests and reputation (Demirer and Kutan, 2006). According to Scharfstein and Stein (1990), the fear of being poorly assessed or judged by others if they make the wrong decision triggers herding behaviour.
likely to be stronger present during extreme market conditions. Periods of market stress are defined as periods with low returns, high trading volume and high volatility (Gleason, Mathur, and Peterson, 2004). The uncertainty during market stress influences the investors and creates anxiety of making incorrect decisions and incurring a loss. It negatively influences the ability to analyse rationally and investors will search for confirmation and tend to follow market consensus.
Both explanations of herding behaviour illustrate that investors do not base their decisions on their own analysis and information, but follow the market consensus. Herding behaviour suggests that investors trade in the same direction, which would imply that stock returns behave more similar (Nofsinger and Sias, 1999). Regardless of what the motive for co-movement of stock returns is, the tendency to show similarity in their behaviour, and thus act like a herd, can be an important threat for the stability and the efficiency of financial markets (Bikhchandani and Sharma, 2001; Kremer and Nautz, 2013). The efficient market hypothesis believes that all investors are rational and receive the same market information. Therefore, stocks would reflect all market information and trade at their fair value (Fama, 1970). The existence of herding behaviour violates this assumption. Herding behaviour suggests that investors are not necessarily rational and do not always derive the share price by rational analysis of firms, but by observing and following investors actions (Lao and Singh, 2011). Herding behaviour may destabilize the market by driving prices away from their fundamental values. Stocks do not only reflect investors’ rational expectations, but also the irrational decisions of the investors in the market (Demirer and Kutan, 2006). Therefore, the efficient market hypothesis does not account for the existence of herding behaviour in the stock market. In literature, the market efficiencies that cannot be explained by rational asset pricing models, are linked to herding behaviour (Christie and Huang, 1995; Tan, Chiang, Mason, and Nelling, 2008). Herding behaviour has a substantial effect on stocks and is therefore important to study. When investors suppress their own private information and analysis and base their decisions on collective information, the fundamental value of the stock might be lower than its market value (Tan, Chiang, Mason and Nelling, 2008).
Kraus and Stoll (1972) discuss that herding behaviour destabilizes stock prices, which in their perspective means that prices move away from their fundamental values in the same direction. To measure herding behaviour, they use the magnitude of parallel trading. Herding behaviour towards the market consensus shows movements independently from the given market conditions. This can be measured by examining the relationship between stock returns and the market returns (Hwang and Salmon, 2004). The degree of co-movement of stock returns is an indicator of herding behaviour. It is measured by the cross-sectional variation in stock returns, or the extent to which stock returns diverge form the market returns (McEnally and Todd, 1992). Even though herding behaviour is often irrational, knowledge of the magnitude of herding behaviour gives insight in possible changes in stocks (Dhaene, Linders, Schoutens, and Vyncke, 2012).
According to the efficient market theory, returns of individual stocks are expected to deviate from the market returns. If stocks show co-movement in their returns, it may be the result from the release of information or from friction and noise in the trading process (Amihud and Mendelson, 1989). The co-movement of stock returns provides meaningful information about market situation. Even though the stock returns may perform differently, they may move in the same direction. Therefore, co-movement in stock returns is a measurement of herding behaviour (Guo and Shih, 2008). After Kraus and Stoll (1972), Christie and Huang (1995) use co-movement of stocks to identify the presence of herding behaviour. They examine the relationship between the CSSD and the market returns to measure the co-movement of stocks. Thus, according to the literature, herding behaviour is a signal of market inefficiency and violates the efficient market hypothesis and rational asset pricing models. But does herding behaviour really exists?
The literature on the existence of herding behaviour shows some mixed results. Scharfstein and Stein (1990) discover the existence of rational herding behaviour under certain circumstances. They argue that managers mimic the actions of other managers to cover bad performance. Since they are concerned about their reputation, they ignore their own private information and follow the other investors. Lakonishok, Schleifer, and Vishny (1992) proposed a statistical model, the LSV-measure. They examine whether money managers buy or sell simultaneously the same stocks as other managers buy or sell. This definition of herding is tested on their sample of 796 equity funds, but their results suggest that the managers do not show evidence of herding. Using the LSV-measure, Grinblatt, Titman, and Wermers (1995) find little evidence of herding behaviour in their sample on portfolio changes of 275 mutual funds between 1974 and 1984. They find that 77% are momentum investors, who buy stocks that were past winners. The tendency to do this is strongly correlated with the tendency of a fund to herd. Subsequently, Wermers (1999) proceeds their research with data on quarterly equity holdings of virtually all mutual funds that were in existence between 1975 and 1994. Using the LSV-measure Wermers finds that for the average stock there is some evidence of herding by mutual funds. He observes much higher levels in trades of small stocks and in trading by growth-oriented funds.
Khorana (2000), but do not find any evidence of herding during periods of extreme market movements in Exchange Traded Funds (ETFs) at the American stock exchange. An interesting finding is the different reactions of investors to information during periods of extreme market movements. In declining markets investors tend to respond quickly with a higher incentive to mimic the market, compared to rising markets where investors respond more slowly. Demirer and Kutan (2006) examine the presence of herd formation in Chinese markets. They use both individual firm- and sector-level data, and separate the Shanghai and Shenzhen stock exchanges at the sector level. Their analysis focusses on the return dispersions during periods of unusually large upward and downward changes in the market index. The results show no evidence of the existence of herding behaviour. Consistent with this finding, Gleason, Mathur, and Peterson (2004) observe that the dispersion in stock returns during extreme downside movement is much lower than it is during upside movements, indicating that stock returns behave more similar during declining markets. Tan, Chaing, Mason, and Nelling (2008) examine whether herding behaviour exists in Chinese A-share and B-share stocks, both on the Shanghai and the Shenzen stock exchanges. For each of these four markets they test whether herding behaviour exists and whether this behaviour is stronger present during periods of market stress. The characteristics of market stress are low returns (declining markets), high trading volume and high volatility. Their results show that herding behaviour is present in all four markets in both rising and declining markets. For the behaviour of A-share investors they find that herding behaviour is stronger present during periods of high returns (rising markets), high trading volume and high volatility, while they find no evidence for a stronger presence of herding behaviour of B-share investors during periods of market stress compared to normal periods.
Caporale, Economou, and Philippas (2008) follow Christie and Huang (1995) and Chang, Cheng, and Khorana (2000), and are pioneers in studying the countries in Europe. They examine the presence of herding behaviour in extreme market conditions using data from the Athens stock exchange in Greece. Using daily, weekly and monthly returns, they observe herding behaviour between 1998 and 2007. Besides this, they observe the presence of herding behaviour during the stock market crisis in 1999. Chaing and Zheng (2010) contribute to the existing literature of herding behaviour by extending the sample markets with 18 countries. They separate this global data into three categories, advanced markets, Latin American markets and Asian markets. By apply daily data to this 18 countries during the period 1988-2009, they find significant evidence for herding behaviour in the advanced (with the exception of no herding behaviour in the US) and Asian markets. In the Latin American markets they do not find any evidence of herding behaviour either. In 2010, Economou, Kostakis, and Philippas extend the research of Caporale, Economou, and Philippas (2008) by including the stock markets of Portugal, Italy and Spain between 1998 and 2008. Besides the existence of herding behaviour in these markets, they test for herding behaviour during market stress (with respect to market returns, trading volume and return volatility) and the existence of herding behaviour during the global financial crisis of 2007-2008. Their results report evidence of herding behaviour in Italian and Greek stock markets and nu herding behaviour in the Spanish stock markets. For Portugal, they only find herding behaviour during declining markets. Remarkably, during the global financial crisis of 1007-2008, they find herding behaviour in Portugal and Greece and no herding behaviour in Italy and Spain. In 2011, the authors extend their research by examining the co-movements between the cross-sectional dispersion of the four stock markets. They find evidence for strong a correlation between the countries, indicating that movement of stock returns in one market can be explained by the movements of the stock returns in the other market. Besides expending globally, Economou, Kostakis, and Philippas (2011) find evidence suggesting that stock return dispersion in the US play a significant role in explaining the non-US markets herding activity. This leads to the last question: is it possible to explain the co-movement of stock returns in one country with the co-movement of stock returns in the other country?
§ Herding behaviour does not exist in the North-European countries Germany and the UK, during the period January 1998-December 2008.
§ In Germany and the UK herding behaviour is stronger present during market stress, i.e. declining markets.
§ Germany and the UK are closely related and therefore there is a strong correlation in the co-movement of stock returns between the two countries.
3.
METHODOLOGY
In 1992, Lakonishok, Schleifer, and Vishny stated that the herding behaviour of investors may cause stock returns to move in the same direction. Two studies that have proposed methods of detecting herding behaviour by co- movement using stock return data are Christie and Huang (1995) and Chang, Cheng, and Khorana (2000). Christie and Huang (1995) suggest that under the traditional definition of herding behaviour, an intuitive measure of its market impact is the dispersion in the cross-section of stock returns, defined as the CSSD of returns. The rationale is that during these periods of market pressure, movements of stock returns have the tendency to be more clustered. This co-movement of stock returns is independent of their fundamental characteristic. As individual returns begin to vary from the market return, the level of dispersion increases. In other words, if herd behaviour is present the individual returns would move together with the market return. The co-movement of stock returns reveals if investors behave the same and follow a herd. To measure this dispersion with respect to market return, Christie and Huang (1995) propose the CSSD method, which is expressed as:
𝐶𝑆𝑆𝐷! = (𝑅!,!− 𝑅!,!)! ! !!! 𝑁 − 1 (1)
Where N is the number of firms in the market portfolio, Ri,t is the observed return of the stock of a firm
i at time t, Rm,t is the cross-sectional average stock returns of N firms in the market portfolio at time t.
Ri,t, the observed return of the stock of a firm i on time t is calculated by: !!,!!!!!,!!!
!,!!! . I calculate Rm,t, the
average market portfolio return, by using the equally weighted average of stock returns on this particular time t as a proxy of Rm,t.
market returns increases, the CSAD of the individual return should increase at a decreasing rate, or decrease because investors are conforming to the market consensus. The cross-sectional absolute deviation is given by:
𝐶𝑆𝐴𝐷! = 𝑅!,!− 𝑅!,! ! !!! 𝑁 (2)
Where N is the number of firms in the market portfolio, Ri,t is the observed return of the stock of a firm
i at time t, Rm,t is the cross-sectional average stock returns of N firms in the market portfolio at time t.
To find an answer on the main question if herding behaviour exists, the co-movement of the stock returns and the market returns is examined. This is done by a regression model in which the CSAD is regressed with respect to the market return:
𝐶𝑆𝐴𝐷!,! = 𝛼 + 𝛾! 𝑅!,! + 𝛾! 𝑅!,! !+ 𝜀! (3)
Both 𝑅!,! and 𝑅!,! ! terms appear in the right-hand-side of equation (3). Chang, Cheng, and Khorana (2000) note that rational asset pricing models assume a linear relation between the dispersion in individual asset returns and the return on the market portfolio. As the absolute value of the market return increases, so should the dispersion in individual asset returns. One would expect a positive estimate of γ1. If the investors exhibit the tendency to herd, the stock returns would behave in
the same direction. The positive and linear relationship between individual return and market return will be violated. During relatively large price swings, the existence of herding behaviour is captured with a negative and nonlinear relationship between CSAD and market return. The co-movement of stock returns is indicated by a negative and statistically significant γ2 coefficient (Chang, Cheng, and
Khorana, 2000). The cross-sectional dispersion of stock returns will increase at a decreasing rate or decrease in the presence of herding behaviour. In the absence of herding behaviour I expect γ1 > 0 and
γ2 = 0 in equation (3). If investors exhibit herding behaviour, I expect γ2 < 0.
To examine whether the co-movement of stock returns is stronger present during periods of market stress, I follow the approach of Economou, Kostakis, and Philippas (2011) and include some dummy variables in the benchmark model, equation (3). With the following regression, I examine whether there exists herding behaviour in rising or declining markets and if this behaviour is stronger present during rising or declining markets.
𝐶𝑆𝐴𝐷!,! = 𝛼 + 𝛾!𝐷!" 𝑅!,! + 𝛾! 1 − 𝐷!" 𝑅!,! + 𝛾!𝐷!" 𝑅!,! !
+ 𝛾! 1 − 𝐷!" 𝑅!,! !+ 𝜀!
Where Dup is a dummy variable taking the value one on days with positive market returns and the value
zero on time t with negative market returns. The regression of equation (4) tests whether stock returns behave more similar during rising or declining markets. I expect that in the absence of herding behaviour γ1 > 0 and γ2 > 0 in equation (4). If herding behaviour exists, I expect γ3 < 0 and γ4 < 0, with
γ4 < γ3 if this behaviour is stronger present during times with negative market returns.
If herding behaviour is stronger present during market stress, and thus during negative returns, it must reveal during periods with other characteristics as well. Besides negative returns, high trading volume and high volatility are some other indicators of market stress. Consequently, if herding behaviour is prevailing during periods with negative stock returns, it should be prevailing in periods during high trading volume and high volatility as well. Therefore, these two characteristics are suitable to use as a robustness test. Following Economou, Kostakis, and Philippas (2011), I perform a regression with respect to trading volume and a regression with respect to volatility. In both regressions I include a dummy variable to separate periods with high and low trading volume and high and low volatility.
I follow the approach of Tan, Chaing, Mason, and Nelling (2008) to define trading volume and volatility. Trading volume for each market is the aggregate trading volume for all active shares on a given trading day. Volatility is calculated as the square of the market portfolio return.
The next equations are robustness checks if stock returns move more similar during periods that exhibit characteristics of market stress. To test for co-movement during periods of high or low trading volume, I include a dummy variable in my benchmark model, equation (3). The dummy variable, DHVolume, takes the value one on days with high trading volume. High trading volume is defined as high
if the volume on time t is greater than the average volume of the last 30 days. On the other hand, it will take the value zero if the volume on time t is lower than the average volume of the last 30 time units (days or months). The model to examine the asymmetry with respect to trading volume is represented by:
𝐶𝑆𝐴𝐷!,! = 𝛼 + 𝛾!𝐷!"#$%&' 𝑅
!,! + 𝛾! 1 − 𝐷!"#$%&' 𝑅!,! + 𝛾!𝐷!"#$%&' 𝑅!,! !
+ 𝛾! 1 − 𝐷!"#$%&' 𝑅!,! !+ 𝜀!
(5)
As with the regression for rising and declining markets, I expect that in absence of herding behaviour γ1 > 0 and γ2 > 0 in equation (5). If stocks returns behave more similar during high or low trading
volume, I expect γ3 < 0 and γ4 < 0, with γ3 < γ4.
The last examination for prevailing herding behaviour during market stress is with respect to market volatility. Once again a dummy variable, DHVolatility, takes the value one on time t with high
regarded to be high if the volatility on time t is greater than the volatility average of the previous 30 returns. I estimate the following regression model:
𝐶𝑆𝐴𝐷!,! = 𝛼 + 𝛾!𝐷!"#$%&'$'&( 𝑅!,! + 𝛾! 1 − 𝐷!"#$%&'$'&( 𝑅!,!
+ 𝛾!𝐷!"#$%&'$'&( 𝑅!,! !+ 𝛾! 1 − 𝐷!"#$%&'$'&( 𝑅!,! !+ 𝜀!
(6)
I expect that in absence of herding behaviour γ1 > 0 and γ2 > 0 in equation (6). If stocks returns behave
more similar during high or low trading volume, I expect γ3 < 0 and γ4 < 0, with γ3 < γ4.
As mentioned above, I expect herding behaviour to be prevalent during periods of market stress, such as the global crisis of 2007-2008. If herding behaviour exists, it must exist during this recent crisis. To make the previous regressions more robust, I perform a regression including a dummy variable for the crisis period, DCRISIS in the benchmark model.
𝐶𝑆𝐴𝐷!,! = 𝛼 + 𝛾! 𝑅!,! + 𝛾! 𝑅!,! !+ 𝛾!𝐷!"#$#$ 𝑅
!,! !+ 𝜀! (7)
This dummy takes the value one during the crisis period and the value zero otherwise. Following Economou, Kostakis, and Philippas (2011), I examine this regression for two different crisis periods. The first crisis period is the broad period and refers to the entire period August 2007-December 2008. The second period is the narrow one and refers to the period around the collapse of Lehman Brothers. This period contains the months September 2008-October 2008. Rationale behind this, is that the collapse of Lehman Brothers had a big impact on the exchanges during a period in which the exchanges already were in market stress. For this regression I expect γ1 > 0 and γ2 = 0 in equation (7).
If herding behaviour reveals during crisis periods, I expect γ3 < 0.
The last regression I perform is to test whether the co-movement in stock returns of country A influences the co-movement in stock returns of country B. This cross-country effect is estimated by the following model:
𝐶𝑆𝐴𝐷!,! = ∝ +𝛾! 𝑅!,! + 𝛾! 𝑅!,! !+ 𝛾!𝐶𝑆𝐴𝐷!,!+ 𝜀! (8)
4.
DATA
The data necessary for this research is retrieved from two different sources; Thomson Reuters Datastream and Orbis. Thomson Reuters Datastream contains worldwide information on stocks, equities, indices, macro-economic data and financial data on companies. Orbis is a database on over 80 million companies worldwide. An overview of all listed companies during the sample period (January 1998-December 2008) is retrieved. The needed data consists of all listed stocks in the testable markets, Germany and the UK. Germany and the UK are chosen since they are countries in the North of Europe, whereas previous research focussed on South-European countries (Economou, Kostakis, and Philippas, 2011). Both countries are qualified by the MSCI2 as developed countries in Europe. However, the
market in these two countries is different. The social market economy, or Rhine capitalism, is the leading form of market capitalism in Germany. This form of capitalism is mainly based on the prominent role of banks. Banks and the government are not passive, but take regulatory measures if this is necessary. For example, investors are protected by this regulations against destructive competition. Contradicting, the leading market capitalism in the UK is the Anglo-Saxon model. This model is based on the belief that self-interest and decentralized market can function in a self-regulating, balanced manner. The regulations and taxes are low, providing an overall ease of doing business. However, investors are less protected by the government against the risks of the market (Cernat, 2004). My dataset is constructed for the period from January 1998 till December 2008. The reason to choose this time period it to make is possible to compare my results with the results of Economou, Kostakis, and Philippas (2011). I include both inactive and active stocks, so my dataset is free of survivorship bias. To discover which stocks were listed during this period, I collect this information from Orbis. In Germany the number of listed stock firms varies between 310 and 822, whereas for the UK the number of listed companies varies between 384 and 837. After establishing lists of all the companies listed on the stock exchange of Germany and the UK during my sample period, I can retrieve the data from Thomson Reuters Datastream. Following Economou, Kostakis, and Philippas (2011), I collect daily and monthly stock returns, trading volumes and market values. Some observations are excluded from my sample, so that the observations used are observed on daily/months when both markets were open for trading. I use daily returns to test my hypothesis, and use monthly returns as a robustness test. The reason to use daily returns is because herding behaviour exhibits in short time horizons (Tan, Chiang, Mason, and Nelling, 2008). The trading volumes and market values are used to calculate the regressions for the other robustness tests.
Table 1 reports the descriptive statistics for de CSAD measure and the average daily market return (Rm,t). This is calculated with value weights and equal weights for both Germany and the UK.
There are almost 2.900 daily observations for this measure. As can be seen in the table, the UK has a
broader range in which the CSAD varies compared to Germany. CSAD of Germany varies between 0.00% and 7.52% while the UK varies between -1.31% and 33.06% for the value weighted daily returns. For the equal weighted returns, Germany has a broader range in which the CSAD varies (0.00%-21.66%). Compared to the equally weighted return measure, the value weighted return measure reports more similar descriptive statistics between Germany and the UK.
The correlation matrix for the calculated CSAD measures and the market returns for Germany and the UK is presented in Table 2. To establish this correlation matrix I use the daily return observations, it shows that the correlation of the measures between both countries is higher than the correlation of the market return between the two countries. The descriptive statistics and correlation matrix of the monthly returns are reported in Table A.1 and A.2 in the Appendix.
Table 1: Descriptive statistics
Germany UK
𝐶𝑆𝐴𝐷!,! 𝑅!,! 𝐶𝑆𝐴𝐷!,! 𝑅!,!
Panel A: Value weighted market returns
Mean 0.0139 0.0002 0.0272 0.0004 Median 0.0124 0.0004 0.0255 0.0000 Maximum 0.0752 0.1452 0.3306 0.2413 Minimum 0.0000 -0.1232 -0.0131 -0.1375 Std. dev. 0.0073 0.0193 0.0156 0.0187 Observations 2869 2869
Panel B: Equally weighted market returns
Mean 0.0219 -0.0001 0.0133 0.0131 Median 0.0209 0.0004 0.0123 0.0118 Maximum 0.2166 0.0834 0.0549 0.0571 Minimum 0.0000 -0.0588 0.0000 0.0000 Std. dev. 0.0081 0.0095 0.0055 0.0068 Observations 2869 2869
Notes: This table represents the descriptive statistics for the cross-sectional absolute deviation (CSADi,t) of individual stock
returns and the return of the market portfolio (Rm,t) of both Germany and the UK. The sample period is January
1998-December 2008 using daily observations. The CSADi,t is calculated by: 𝐶𝑆𝐴𝐷!= !!,!!!!,!
! !!!
! , where Rm,t is the return of the
market portfolio, Ri,t is the return for stock I, and N is the number of stocks in the market portfolio.
Table 2: Pairwise cross-market correlations
Germany CSAD Germany Rm,t
Value weighted Equally weighted Value weighted Equally weighted
UK CSAD 0.7215 0.6238 UK Rm,t 0.3588 0.0075
Notes: This table represents the pairwise correlation coefficients of the cross-sectional absolute deviation (CSAD) and the return of the market portfolio Rm,t for Germany and the UK. The sample period is January 1998-December 2008 using daily
5.
RESULTS
To find an answer on the first hypothesis if herding behaviour does exist in Germany and the UK, I run the regression from equation (3). The results from equation (3) can be found in Table 3, estimated for both Germany and the United Kingdom for the sample period January 1998-December 2008. The table is divided into two parts, panel A and panel B. Panel A represents the results using value weighted market returns. The analysis is mainly based on the results using value weighted market returns. The rationale behind this, is that the value weighted return method gives the closest reflection of the real world. To strengthen the analysis, I perform several robustness tests. The first robustness test is rerunning the regression using equally weighted returns. These results are shown in panel B of Table 3. The second robustness test I perform is using monthly returns instead of daily returns, for both the value weighted returns and equally weighted returns. The results of the regressions using monthly returns can be found in Table A.3 in the Appendix.
One can observe a positive and significant relation between market return (Rm,t) and the
CSAD for both countries (Germany γ1=0.2458 and the UK γ1=0.3118, using value weighted
returns). This relationship is shown by the coefficient γ1 and independent of the use of the value
weighted return method or the equally weighted return method. A positive relationship between CSAD and market returns is consistent with the theory behind standard asset pricing models. As a second robustness test, I perform the regression with monthly returns. This regression shows a positive coefficient γ1 as well, and is therefore robust with the regression using daily returns.
The squared market return indicates whether stocks behave more similar and reveal herding behaviour. A negative sign of coefficient γ2 indicates that the cross-sectional return dispersion increases
at a decreasing rate or decreases with respect to the market return during extreme market movements. Starting with the regression using value weighted returns, I find a positive and statistically significant number for γ2 indicating no herding behaviour in Germany (γ2=0.7070). The results of the UK show the
opposite, a negative and statistically significant coefficient (γ2=-0.8582), indicating the existence of
herding behaviour. As a robustness check, I perform the same regression using equally weighted returns. As can be seen in Table 3, for the UK a positive and statistically significant coefficient is estimated. This indicates no herding behaviour in the UK, which is not consistent with the evidence found in the method using value weighted market returns.
The second robustness check I perform, is by using monthly returns instead of daily returns. Using the value weighted and monthly returns I find no herding behaviour for Germany and herding behaviour for the UK. This is robust with the findings of my main analysis using value weighted and daily returns.
hypothesis which postulates that herding behaviour does not exist in the North-European countries Germany and the UK, during the sample period. A possible explanation could be the difference in capitalism in the countries. The underlying principle on which the social market capitalism of Germany is based, is the prominent role of banks and on extensive cross-ownership links in corporate finance and control. Not only banks but individual investors as well have an influence on corporate managements. This would imply individuals have good knowledge of the situation and therefore depend less on the behaviour of others.
The Anglo-Saxon model is based on the belief that markets should be self-regulating. There are only a few large shareholders in the UK. A change in behaviour of one of these large investors has immediately an impact on the market, reflected in the majority of the stock returns. Instead of the millions of small individual investors in the German system, a few large shareholders dominate the market. Furthermore, the investors in the UK are more exposed to the risks of the market, due to less regulation. Therefore, it is possible, that the tendency of investors to exhibit herding behaviour is stronger. Co-movement of stock returns can appear and the existence of herding behaviour could be explained by this.
Table 3: Estimates of herding behaviour in the full sample period
Constant 𝑅!,! 𝑅!,! ! 𝑅!𝑎𝑑𝑗.
Panel A: Value weighted market returns
Germany 0.0079 0.2458 0.7070 29.79%
(77.85)*** (18.37)*** (2.25)**
UK 0.0073 0.3118 -0.8582 22.82%
(68.66)*** (29.16)*** (-8.68)***
Panel B: Equally weighted market returns
Germany 0.0171 0.5946 -0.2629 34.00%
(89.60)*** (13.72)*** (-1.51)
UK 0.0114 0.0017 0.5200 1.68%
(114.13)*** (0.11)* (4.51)***
Notes: This table reports the estimated coefficients for the benchmark model: CSADi,t=α+γ1Rm,t+γ2 Rm,t 2
+εt, where
CSADi,t represents the cross-sectional absolute deviation of stock returns with respect to the market portfolio return Rm,t for
Germany and the UK. The sample period is January 1998-December 2008 using daily observations. ***, ** and * represent the statistical significance of the coefficients at a 1%, 5% and 10% level. The values in the parentheses are the z-statistics, these are calculated using GARCH(1,1) model considering heteroscedasticity and autocorrelation.
that the herding coefficients are equal on days with rising and declining market prices. The second robustness test I perform is by using monthly returns instead of daily returns. Equal to the regression with the daily returns, I perform the regression with value weighted monthly returns and equally weighted monthly returns. The results of the regressions using monthly returns can be found in Table A.4 in the Appendix.
During market stress, herding behaviour is expected to be stronger present. Thus, during declining markets, I expect that herding behaviour will be stronger present compared to rising markets. Besides declining markets, there are several other characteristics of market stress. During market stress, trading volume and volatility is expected to be high. Therefore, examining the herding behaviour during periods with these characteristics is a good robustness test. First, I run the regression of equation (5), including a dummy variable for trading volume. Again this is done for daily (Table 5) and monthly returns (Table A.5 in the Appendix). After this I run the regression of equation (6), in this regression there is a dummy variable for volatility. Results of this regression can be found in table 6 for the daily regressions and Table A.6 in the Appendix for the regression using monthly returns. The last robustness test I perform is by including a dummy variable for the crisis periods. These results are shown in Table 7 for the regression with daily returns and Table A.7 in the Appendix for the regression with monthly returns.
Analysing the results of the regression, a negative and statistically significant value for coefficient γ3 indicates herding behaviour during rising markets and a negative and statistically
significant value for coefficient γ4 indicates herding behaviour during declining markets. Using daily
value weighted market returns, I find a positive and statistically significant coefficient during rising markets for Germany (γ3=0.0085). This positive coefficient indicates that stocks do not move more
similar during rising markets. During declining markets, the coefficient (γ4=0.5172) is positive as well.
This implies that during declining markets there is no herding behaviour in de Germany. However, this result is not statistically significant and therefore I can only conclude that there is no herding behaviour during rising markets in Germany. For the UK, using value weighted returns, I find statistically significant and negative coefficients indicating the presence of herding behaviour in the UK (γ3
=-0.8144 and γ4=0.-1.6142). The magnitude for herding in declining markets (γ4) is bigger than the
magnitude for herding in rising markets (γ3). Controlling this discovery with a Wald test, I find that
herding behaviour is statistically significant stronger present in declining markets in the UK.
As robustness check, I perform the regression using equally weighted returns. Again for Germany a positive and statistically significant coefficient in rising markets is found, indicating no herding behaviour during rising markets in Germany. Controlling my value weighted results with the equally weighted results for the UK, I discover some remarkable results. Using equally weighted returns, I find a negative relationship between CSAD and market return for the UK (γ1=-0.0009 and
γ2=-0.3019). This would indicate that the cross-sectional dispersion decreases during all times if the
for the coefficients γ3 and γ4, I find contradicting results compared to the regression using value
weighted daily returns. The coefficients for herding behaviour are positive and statistically significant (γ3=0.5204and γ4=53.3611), indicating no herding behaviour in the UK. Even though the coefficients
for herding (γ3 and γ4) are positive and statistically significant, due to the negative relationship between
CSAD and the market returns, they are not reliable.
Table 4: Estimates of herding behaviour in rising and declining markets Constant 𝐷!"𝑅
!,! 1 − 𝐷!"𝑅!,! 𝐷!" 𝑅!,! ! 1 − 𝐷!" 𝑅!,! ! 𝑅!𝑎𝑑𝑗.
Panel A: Value weighted market returns
Germany 0.0079 0.2384 0.2548 0.0085 0.5172 29.89%
(78.33)*** (14.95)*** (16.45)*** (1.96)** (1.39)
UK 0.0073 0.3108 0.3581 -0.8144 -1.6142 22.84%
(68.13)*** (2.92)*** (1.11)*** (-9.46)*** (-6.98)***
Panel B: Equally weighted market returns
Germany 0.0172 0.5379 0.5108 7.2723 -1.0877 45.03%
(91.76)*** (9.84)*** (12.08)*** (2.46)** (-0.66)
UK 0.0114 -0.0009 -0.3019 0.5204 53.3611 1.64%
(111.53)*** (-0.06) (-2.19)* (4.48)*** (3.84)***
Panel C: Wald test for equality of herding coefficients
Value weights Equally weights
Germany UK Germany UK γ1 – γ2 -0.0164 -0.0472 0.0271 0.3010 Chi-square (0.89) (2.19) (0.27)*** (4.70)** H0: γ1 = γ2 γ3 – γ4 0.3338 0.799868 8.3600 -52.8408 Chi-square (0.38) (11.21)*** (6.56)** ( 14.39)*** H0: γ3 = γ4
Notes: This table reports the estimated coefficients for the following regression model: 𝐶𝑆𝐴𝐷!,!= 𝛼 + 𝛾!𝐷!"𝑅!,! +
𝛾!1 − 𝐷!" 𝑅!,! + 𝛾!𝐷!" 𝑅!,! !
+ 𝛾! 1 − 𝐷!" 𝑅!,! !
+ 𝜀!, where CSADi,t is the cross-sectional absolute deviation for
market i during time t, Rm,t is the return of the market portfolio of market i during time t and Dup is a dummy variable taking
the value 1 on days with positive market returns and the value zero on time t with negative market returns. The sample period is January 1998-December 2008 using daily observations. ***, ** and * represent the statistical significance of the coefficients at a 1%, 5% and 10% level. The z-statistics are presented in the parentheses and estimated with GARCH(1,1) model considering heteroscedasticity and autocorrelation.
As a second robustness check I rerun the regressions with monthly returns. Inadequately, most of these coefficients are not statistically significant. The only statistically significant result is the positive coefficient during rising markets for Germany (γ3=5.6850), indicating no herding behaviour in
Germany. This is consistent with the analysis using value weighted daily returns. Therefore, this robustness test strengthens my finding of no herding behaviour in Germany. The other results are somewhat mixed, but as stated above not statistically significant and therefore irrelevant.
analysis, I partly reject my second hypothesis, postulating that the presence of herding behaviour is stronger during periods of market stress.
The tendency to follow the actions of others is stronger during market stress, i.e. declining markets. Besides low returns, are periods of market stress also characterised by high trading volume and high volatility. If herding behaviour is stronger present during market stress, this would indicate that herding behaviour is stronger present during periods with low returns (declining markets), high trading volume and high volatility.
In the previous regression, I examined if herding behaviour exists during market stress with respect to high and low returns. To control my findings, I expand my robustness tests by running two new regressions including different dummy variables. Instead of rising or declining markets I add a dummy for high and low trading volume in the third robustness regression of equation (5) and a dummy for high and low volatility in the fourth robustness regression of equation (6).
The third robustness test is the controlling regression with respect to trading volume. The regression model examines whether the CSAD reveals if herding behaviour is present during periods of high or low trading volume with respect market to returns. The results are reported in Table 5. I compare results using value weighted market returns, reported in Table 5 panel A, and equally weighted market returns, reported in Table 5 panel B. Panel C represents the Wald test, testing the null hypothesis that the herding coefficients are equal on days with high and low trading volume.
Using value weighted daily returns I find statistically significant evidence for no herding behaviour in Germany (γ4=1.3097) in markets during periods with low trading volume and statistically
significant evidence of herding behaviour in the UK (γ4=-0.7751) during periods of low trading
volume. This is consistent with the result of periods with low returns from the rising and declining regression. However, during market stress the trading volume is expected to be high, which implies that herding behaviour would be stronger present during a period with high trading volume compared to low trading volume. Unfortunately, using value weighted daily returns, I do not find any statistically significant coefficients.
Controlling this with the equally weighted results, I find statistically significant and positive coefficient indicating no herding behaviour during periods of high or low trading volume (γ3=0.3950
and γ4=0.6259) for the UK. This is inconsistent with the main analysis. However, the relationship
between CSAD and market return is negative (γ1 and γ2), which is in contradiction with the CAPM.
Table 5: Estimates of herding behaviour in days of high and low trading volume Constant 𝐷!"#$%&'𝑅 !,! 1 − 𝐷!"#$%&' 𝑅!,! 𝐷!"#$%&' 𝑅 !,! ! 1 − 𝐷!"#$%&' 𝑅!,! ! 𝑅 !𝑎𝑑𝑗.
Panel A: Value weighted market returns
Germany 0.0079 0.2482 0.2297 0.5599 1.3097 29.96%
(78.34)*** (17.05)*** (13.64)*** (1.59) (2.97)***
UK 0.0074 0.2878 0.2914 1.1889 -0.7751 23.59%
(64.39)*** (10.37)*** (23.75)*** (1.49) (-7.49)***
Panel B: Equally weighted market returns
Germany 0.0172 0.5923 0.4930 -2.4445 2.9133 34.55%
(86.23)*** (13.45)*** (8.26)*** (-1.45) (0.95)
UK 0.0113 -0.0067 0.0059 0.3950 0.6259 2.99%
(129.50)*** (-0.23) (0.34) (2.77)*** (4.74)***
Panel C: Wald test for equality of herding coefficients
Value weights Equal weights
Germany UK Germany UK γ1 – γ2 0.0185 -0.0036 0.0992 -0.0127 Chi-square (1.11) (0.02) (4.21)** (0.14) H0: γ1 = γ2 γ3 – γ4 -0.7499 1.9640 -5.3578 -0.2309 Chi-square (1.77) (6.28)* (3.09)* (1.41) H0: γ3 = γ4
Notes: This table reports the estimated coefficients for the following regression model: 𝐶𝑆𝐴𝐷!,!= 𝛼 + 𝛾!𝐷!"#$%&'𝑅!,! +
𝛾!1 − 𝐷!"#$%&' 𝑅!,! + 𝛾!𝐷!"#$%&' 𝑅!,! !
+ 𝛾!1 − 𝐷!"#$%&' 𝑅!,! !
+ 𝜀!, where CSADi,t is the cross-sectional
absolute deviation for market i during time t, Rm,t is the return of the market portfolio of market i during time t and DHVolume is
a dummy variable, which takes the value 1 if the trading volume is higher than the average of the last 30 days and takes the value zero if the trading volume is lower than the average of the last 30 days. The sample period is January 1998-December 2008 using daily observations. ***, ** and * represent the statistical significance of the coefficients at a 1%, 5% and 10% level. The z-statistics are presented in the parentheses and estimated with GARCH(1,1) model considering heteroscedasticity and autocorrelation.
The fourth robustness test regarding herding behaviour during market stress is conducted with respect to volatility of market returns. The regression model examines whether the CSAD reveals if herding behaviour is present during periods of high or low volatility with respect to market returns. The results are reported in Table 6. The results using value weighted returns, reported in Table 6 panel A, are compared with the results using equally weighted results, reported in Table 6 panel B. Panel C represents the Wald test, testing the null hypothesis that the herding coefficients are equal on days with high and low volatility.
Using daily value weighted returns, I find significant and positive evidence of no herding behaviour during markets with high volatility for Germany (γ3=0.9385). This is consistent with earlier
findings of no herding behaviour in Germany. In the UK, using value weighted daily returns, negative and statistically significant coefficients are found. Indicating that herding behaviour exhibits in markets with high (γ3=-0.9490) and low volatility (γ4=-0.8834). Conducting a Wald test reveals that herding
For Germany, the two robustness tests, with regard to trading volume and volatility, provide results that strengthen the rejection of the second hypothesis, which postulates that herding behaviour is stronger present during periods of market stress. However, the same robustness tests show mixed results for the UK. The analysis with regard to the trading volume report that herding behaviour does not exist during periods of market stress and the analysis with regard to volatility, provides evidence of a stronger presence of herding behaviour during market stress, which is in line with the second hypothesis. Thus, after the robustness tests, I can still partly reject the second hypothesis.
Remarkable for the UK using equally weighted daily returns, I find statistically significant and positive coefficient indicating no herding behaviour during periods of high or low volatility (γ3=0.5398
and γ4=1.5138). This is inconsistent with the main analysis using value weighted daily returns. The
relationship between CSAD and market returns is negative (γ2), which is in contrast with the CAPM.
Remarkable, this negative relationship is revealed for the UK in every regression of equation (3), (4), (5) and (6) using the equally weighted daily returns. Because the regression with equally weighted returns is a robustness check it weakens my results that prove herding behaviour in the UK.
Table 6: Estimates of herding behaviour in days of high and low volatility Constant 𝐷!"#$%&'$'&( 𝑅!,! 1 − 𝐷!"#$%&'$'&( 𝑅!,! 𝐷!"#$%&'$'&( 𝑅!,! ! 1 − 𝐷 !"#$%&'$'&( 𝑅!,! ! 𝑅!𝑎𝑑𝑗.
Panel A: Value weighted market returns
Germany 0.0078 0.2302 0.2829 0.9385 0.7318 30.89%
(64.76)*** (15.89)*** (10.46)*** (2.77)*** (0.71)
UK 0.0073 0.3358 0.2975 -0.9490 -0.8834 23.44%
(70.38)*** (24.94)*** (25.35)*** (-9.37)*** (-4.29)***
Panel B: Equally weighted market returns
Germany 0.0169 0.6056 0.7327 -2.6621 -12.9299 34.13%
(67.87)*** (12.90)*** (7.38)*** (-1.49) (-1.82)*
UK 0.0110 0.0239 -0.0073 0.5398 1.5138 3.91%
(72.51)*** (1.00) (-0.04) (4.63)*** (4.67)***
Panel C: Wald test for equality of herding coefficients
Value weights Equal weights
Germany UK Germany UK γ1 – γ2 -0.0527 0.0384 -0.1271 0.0312 Chi-square (4.62)* (8.35)*** (3.08)* (1.10) H0: γ1 = γ2 γ3 – γ4 0.2068 -0.0656 10.2678 -0.9739 Chi-square (0.04) (0.09) (2.47) (15.59)*** H0: γ3 = γ4
Notes: This table reports the estimated coefficients for the following regression model: 𝐶𝑆𝐴𝐷!,!=
𝛼 + 𝛾!𝐷!"#$%&'$'&(𝑅!,! + 𝛾! 1 − 𝐷!"#$%&'$'&( 𝑅!,! + 𝛾!𝐷!"#$%&'$'&( 𝑅!,! !+ 𝛾! 1 − 𝐷!"#$%&'$'&( 𝑅!,! !+ 𝜀!, where
CSADi,t is the cross-sectional absolute deviation for market i during time t, Rm,t is the return of the market portfolio of market
i during time t and DHVolatility is a dummy variable, which takes the value 1 if the volatility is higher than the average of the
After examining if herding behaviour is stronger present during periods of market stress, I control my results by examining the presence herding behaviour in periods of the recent global crisis of 2007-2008. I use two alternative definitions for the crisis period. The broad one refers to the entire period August 2007-December 2008, while the narrow one refers to the period around the collapse of Lehman Brothers, i.e. September 2008-October 2008 (Economou, Kostakis, and Philippas, 2011). The estimated coefficients are shown in Table 7. I rerun the regression with the use of monthly observations as a robustness, the results of this regression are obtained in Table A.7 in the Appendix. Both tables are divided into two panels, panel A reporting the results using value weighted returns and panel B reporting the results using equally weighted returns.
For both Germany and the UK, in crisis period 1 and crisis period 2, using value weighted and equally weighted returns, I observe a positive and statistically significantly coefficient for the crisis. This indicates evidence for the absence herding behaviour during crisis periods. These findings are not consistent with my earlier findings, thus it weakens my results. The robustness test using monthly returns shows some mixed results. For Germany and the UK during crisis period 2, the results of the regression with regard to crisis, equation (7), using value weighted daily returns strengthen my findings and confirm the absence of herding behaviour during crisis periods.
Remarkable, if I control this, running a regression using equally weighted returns, I observe a negative and statistically significant coefficient during both crisis periods for Germany (γ3=-2.6814 and
γ3=-2.5785). This finding weakens the earlier results of no herding behaviour during crisis periods for
Germany. Overall, I can conclude that the crisis periods reveal no evidence of the expectation that cross-sectional return dispersion would further decrease during the crisis period. The crisis does not influence herding behaviour. This is contradicting with the analysis that distinguishes rising and declining markets, where I find that herding behaviour was stronger present during market distress for Germany and the UK.
Table 7: Estimates of herding behaviour during financial crisis period
Constant 𝑅!,! 𝑅!,! ! 𝐷!"#$#$ 𝑅!,! ! 𝑅!𝑎𝑑𝑗.
Panel A: Value weighted market returns Crisis period: August 2007-December 2008
Germany 0.0079 0.2570 0.2744 1.5789 31.21%
(77.89)*** (19.37)*** (0.85) (3.92)***
UK 0.0073 0.3029 -0.8235 2.5386 24.58%
(68.02)*** (28.11)*** (-8.31)*** (5.18)***
Crisis period: September 2008-October 2008
Germany 0.0079 0.2551 0.4239 1.4521 30.77%
(75.82)*** (18.00)*** (1.22) (2.97)***
UK 0.0073 0.3110 -0.8570 1.9877 24.07%
(68.50)*** (28.92)*** (-8.58)*** (2.93)***
Panel B: Equally weighted market returns Crisis period: August 2007-December 2008
Germany 0.0172 0.5370 0.0024 4.0769 41.99%
(85.97)*** (11.57)*** (0.00) (2.15)**
UK 0.0113 -0.0017 0.4925 2.5908 24.96%
(127.40)*** (-0.11) (4.68)*** (4.27)***
Crisis period: September 2008-October 2008
Germany 0.0171 0.5893 -2.3137 10.7308 45.41%
(87.65)*** (12.78)*** (-1.22) (2.57)**
UK 0.0114 -0.0026 0.5153 3.4165 14.41%
(122.47)*** (-0.17) (4.61)*** (3.81)***
Notes: This table reports the estimated coefficients for the following regression model: 𝐶𝑆𝐴𝐷!,!= 𝛼 + 𝛾!𝑅!,! +
𝛾! 𝑅!,! !
+ 𝛾!𝐷!"#$#$ 𝑅!,! !
+ 𝜀!, where CSADi,t is the cross-sectional absolute deviation for market i during time t and
Rm,t is the return of the market portfolio of market i during time t. DCRISIS is the dummy included in the regression model,
taking the value 1 during days of the crisis period. Two alternative crisis periods are used. The sample period is January 1998-December 2008 using daily results. ***, ** and * represent the statistical significance of the coefficients at a 1%, 5% and 10% level. The z-statistics are presented in the parentheses and estimated with GARCH(1,1) model considering heteroscedasticity and autocorrelation.
Furthermore, to examine these correlation of cross-sectional dispersion in the two markets, I estimate equation (8) for both countries. The results in Table 8 using daily returns (Table A.8 in Appendix for monthly observations) show positive and statistically significant coefficients for both markets (γ3)
which indicates strong co-movement between the UK and Germany. Remarkable, I find positive and statistically significant coefficients for value and equally weighted market returns, using both daily and monthly returns. This is strong evidence in favour of the argument that the cross-sectional dispersion in the markets can be partly explained by the cross-sectional dispersion in the other market. Therefore, the results strongly confirm the last hypothesis, which postulates that Germany and the UK have a strong correlation in the co-movement of stock returns between the two countries. An explanation for this result could be the fact that both countries belong to Europe. Belonging to Europe provides a lot of privileges, especially economically. Trading across countries is easier and therefore it is possible that countries influence each other. The borders are not as strict as they were before, and all the countries in Europe can be seen as one. This indicates that Germany and the UK are one instead of two different countries, and therefore have the same co-movement of stock returns.
Table 8: Cross-country herding effects
Value weighted Equally weighted
𝐶𝑆𝐴𝐷!"#$%&' 𝐶𝑆𝐴𝐷!" 𝐶𝑆𝐴𝐷!"#$%&' 𝐶𝑆𝐴𝐷!" Constant 0.0031 0.0029 0.0129 0.0050 (18.39)*** (18.85)*** (16.98)*** (18.28)*** 𝑅!,! 0.1993 0.2081 0.3443 -0.0051 (15.12)*** (18.76)*** (4.19)*** (-0.40) 𝑅!,! ! 0.1590 -0.5241 3.9108 0.4081 (0.57) (-4.55)*** (1.10) (4.71)*** 𝐶𝑆𝐴𝐷!"#$%&' - 0.5021 - 0.3159 - (31.45)*** - (25.74)*** 𝐶𝑆𝐴𝐷!" 0.5528 - 0.4398 - (32.28)*** - (6.47)*** - 𝑅!𝑎𝑑𝑗. 64.00% 59.24% 52.73% 38.32%
Notes: This table reports the estimated coefficients for the following regression model: 𝐶𝑆𝐴𝐷!,! = ∝ +𝛾!𝑅!,! + 𝛾! 𝑅!,!
!
+ 𝛾!𝐶𝑆𝐴𝐷!,!+ 𝜀!, where CSADi,t is the cross-sectional absolute deviation for market i
during time t, Rm,t is the return of the market portfolio of market i during time t and 𝐶𝑆𝐴𝐷!,! is the cross-sectional absolute
deviation of the other market during time t. The sample period is January 1998-December 2008 using daily observations. ***, **, * represent the statistical significance of the statistics, the z-statistics are presented in the parentheses and estimated with GARCH(1,1) model considering heteroscedasticity and autocorrelation.
6.
CONCLUSION
This paper examines the existence of herding behaviour in the German and the UK stock market. The main finding is that herding behaviour in Germany does not exist, in contrast with the existence of herding behaviour in the UK. Furthermore, I do not find any evidence for the existence of herding behaviour during periods of market stress in Germany. In the UK besides the existence of herding behaviour, the behaviour is stronger present during periods of market stress.
To examine the existence of herding behaviour I constructed a dataset of stock returns for all listed stocks of Germany and the UK during the sample period January 1998-December 2008. The two countries are qualified as developed countries by the MSCI3 and survived the recent global crisis of 2007-2008. Besides this, they have a different capitalism form, which could be an explanation for the difference in results.
I construct the database twice, one time using daily returns and one time using monthly returns. The monthly returns are used as a robustness check. The CSAD is used to measure the co-movement, which indicates herding behaviour, with respect to market returns. After the regression of equation (3), I partly reject the first hypothesis, which postulates that herding behaviour does not exist in the North-European countries. For Germany the results are in line with the hypothesis. However, for the UK, I find evidence for the existence of herding behaviour.
Besides the existence of herding behaviour during the overall sample period, I examine the existence of herding behaviour during market stress. This is done by making a distinction between rising and declining markets and controlled by several robustness tests. The results of these tests to find an answer to the second hypotheses, which postulates that the existence of herding behaviour is
