Herd behavior in advanced European countries

(1)

Faculty of Economics and Business

Herd behavior in advanced European countries

(2)

2

Abstract

The goal of this paper is to find if investors herd in advanced European countries during market stress periods. I use the dummy variable test of Christie and Huang (1995) and the non-linearity test of Chang et al. (2000) to measure for herding. The dataset consists of daily returns of the United Kingdom, Germany and Italy from 01/01/2003 to 12/31/2012. I test for herding in the full sample, the sample characterized by capitalization and for herding in industry sectors. I also test for herding during crisis periods. The results of the dummy variable test do not indicate there is herding in advanced European countries. However, the non-linearity test indicates there is herding in advanced European countries.

(3)

3

1. Introduction

People are influenced by others in their decision making in almost every activity; when deciding at which restaurant to eat or deciding which football club to support and cheer to victory. Highly attended

restaurants and football clubs tend to be more appealing to people and will attract even more customers or fans. People imitate the behavior of others, this group behavior is called herding.

When Twitter announced IPO plans on 4 October 2013, investors massively purchased stock of Tweeter Home Entertainment Group, a firm that filed for bankruptcy in 2007. At a certain point the stock Tweeter Home Entertainment Group had soared up to a staggering 500% on a single day. The trade of Tweeter shares was halted when mass confusion was realized, but at that time millions of transactions had already taken place. The reason for this misunderstanding was that the ticker symbol of Tweeter Home

Entertainment Group is TWTRQ and the sign of the ticker symbol of Twitter is TWTR. The funny thing is that the letter Q in the ticker symbol stands for a firm that is involved in bankruptcy proceedings.

Nofsinger and Sias (1999) define herding as a group of investors trading in the same direction over a period of time. According to this definition, the example of Twitter shows that herding is present in financial markets. I will use an adaptation of the definition proposed by Christie and Huang (1995). They define herding as investors who suppress their own beliefs and base their investment decisions solely on the collective actions of the market, even when they disagree with its prediction.

According to Bikhchanandani and Sharma (2000), herding can be divided in two different concepts; intentional herding and spurious herding. Intentional herding is the result of investors imitating each

other’s behavior. Spurious herding occurs when investors face the same problems, have similar information and make individually the same choices. Spurious herding is an efficient outcome whereas intentional herding does not need to be efficient. Despite this clear distinction in herding, it is nearly impossible to determine when to speak of intentional or spurious herding, because according to Bikhchanandani and Sharma (2000) many factors can influence an investment decision. This is in line with Hirshleifer and Teoh (2003), since they conclude that in most instances it is very likely that a combination of rational and

irrational behavior among investors causes herds. Therefore, this study does not focus on the explanation of herding, but answers the question whether there is herding present in financial markets.

(5)

5 negative market portfolio returns. Chiang and Zheng (2010) suggest that extreme portfolio returns

consistently occur during crisis periods. Therefore, a crisis period is treated as a time period where there is market stress. Therefore, I expect that during days of extreme portfolio returns and during a crisis period investors will herd.

Herding may lead to a situation where the market price fails to reflect all relevant information. Banerjee (1992) and Bikhchanandani et al. (1992) assume that this can lead the market towards inefficiency. However herding does affect investors in more ways. According to Chang et al. (2000), a greater number of securities is needed to achieve the same level of diversification when herding is present in a financial market. Investors who are aware of herding can benefit, since Tan (2008) states that herding can influence stock prices and drive them away from their fundamental value. According to Demirer and Kutan (2006)

herding behavior can also destabilize financial markets especially in crisis conditions and government agency and policy makers should be aware of this. Herding may, according to Tan (2008), be relevant for asset pricing models, because stock price movements may impact a stock’s risk and return characteristics. Therefore, it is important for investors, government agencies, policymakers and academics to know whether herding is present in financial markets.

Herding among market participants has been widely studied in recent years. However, there are only a few studies that focus exclusively on herding among market participants in European countries, like Caparrelli et al. (2004). This paper will test for herding effects in advanced European countries. I will test for herding effects in the stock markets of the United Kingdom (UK), Germany and Italy. These countries are selected, because the market capitalization of listed companies in these countries combined is approximately five trillion US dollar. This is almost halve of the market capitalization of all listed companies in Europe. This paper differs from other research done in Europe in that I will not only test for herding in the market as a whole. Besides that I test whether herd behavior is market capitalization specific which is comparable to the study of Caparrelli et al. (2004) or industry sector specific. Furthermore I also test if herding is present in advanced European countries during crisis which is comparable to the study of (Chiang and Zheng, 2010). However there has been no other research that investigates herding in advanced European countries so extensively, showing the importance of this study, which is answering the following research question:

(6)

6 All data employed in this paper is collected from Thomson Reuters Datastream. The data period of the sample used in this paper ranges from 1 January 2003 till 31 December 2012. I use daily stock returns from common shares in the UK, Germany and Italy. The dummy variable regression test of Christie and Huang (1995) and the non-linearity test of (Chang et al. 2000) are employed to measure for herding.

(7)

7

2. Literature

Many papers have studied the concept of herd behavior. This is not surprising, because herding can have a great impact on financial markets. Many papers have developed theoretical models of herd behavior (Bikhchandani et al. 1992; Devenow and Welch, 1996; Scharfstein and Stein, 1990). There are also

empirical studies that mainly focus on detecting the existence of herd behavior among mutual fund managers (Lakonishok et al. 1992; Wermers, 1999) or among financial analysts (Gleason and Lee, 2003; Welch, 2000). The disadvantage of these studies is that they can only state if specific groups herd and cannot measure whether herding is present in a market as whole.

The dummy variable regression test of Christie and Huang (1995) can measure whether herding is present in a market as whole. They develop a herding measure based on the cross-sectional standard deviation (CSSD) of single stock returns with respect to market returns. Christie and Huang (1995) hypothesize that under conditions of market stress individual investors tend to suppress their own beliefs and follow the market consensus. This will result in a decrease in dispersions and is therefore an indicator of herding. This is consistent with experimental evidence from social psychology. Asch (1956) finds that individuals abide the group decision and neglect their own private information, even when they perceive the group to be wrong. Chang et al. (2000) developed a less stringent method to detect herding in the market. They presume that rational asset pricing models suggest an increase in dispersions 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 implies that the dispersions are an increasing function of the market return. As a measurement of dispersion the authors use the cross-sectional standard absolute deviation (CSAD). Hence, the presence of herd behavior 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 the methodology of Christie and Huang (1995), the method of Chang et al. (2000) is able to measure herding during normal and market stress periods.

(8)

8 The advantage of both the methodologies of Christie and Huang (1995) and Chang et al. (2000) is that they are able to measure whether herding is present in the market as a whole. The methodologies of Christie and Huang (1995) and Chang et al. (2000) are widely accepted as a measurement for herding. Numerous studies worldwide have applied their methods to measure herding effects (Lindhe 2012).

In Asia Tan et al. (2008) examine herding behavior in the Chinese stock market. Their dataset consist of daily stock returns of dual-listed Chinese A-share and B-share stocks from 1996 to 2003. They did find evidence of herding within the Shanghai and Shenzhen A-share markets. On the Shanghai and Shenzhen A-share markets domestic individual investors play an important role, they are the dominant group of investors on these markets. Furthermore they find evidence within both the B-share markets, which are dominated by foreign institutional investors.They did not find that the Asia crisis influences herd behavior on these markets.They did find asymmetry of herd behavior in the A-share stocks but not in the in the B share stocks. Demirer and Kutan (2006) also examine the Chinese stocks from January 1999 to December 2002 on the Shanghai and Shenzen stock exchange. They use the Christie and Huang (1995) method, and did not find any evidence of herding on the Chinese stock exchange.

In the US, Gleason et al. (2004) investigate whether investors herd at the American stock exchange for the period January 1999 to September 2002. They examine whether investors herd during periods of market stress. Gleason et al. (2004) use intraday data of nine sector exchange traded funds traded at the American stock exchange. They use the methodology of Christie and Huang (1995) and Chang et al. (2000). They did not find evidence in favor of herding during periods of market stress.

In Australia Henker et al. (2006) test whether herding occurs in the Australian equity market as a whole and in industry sectors. To measure for herding they use the Christie and Huang (1995) and the Chang et al. (2000) models. They use a sample of the 160 most actively traded stocks on the Australian stock exchange for the period 2001 to 2002, with a market capitalization of more than half a billion U.S. dollar. They did not find evidence in the Australian equity market as whole and they also did not find evidence in industry sectors.

(9)

(10)

10 2.1 Hypotheses

Herding in financial markets occurs, following Henker et al. (2006) when investors ignore or at least underweight the fundamentals of stocks and follow the performance of the market. According to Christie and Huang (1999), this type of herd behavior is more prone to emerge during periods of market stress, because investors are willing to disregard their own beliefs and conform to the group decision if there is more uncertainty in the market. This is consistent with experimental evidence from social psychology. Asch (1956) finds that individuals abide the group decision and neglect their own private information, even when they perceive the group to be wrong. Therefore I expect that investors herd in advanced European countries during periods of market stress.

H1 : Investors herd on the stocks markets of advanced European countries during periods of market stress.

According to Gleason (2003), theoretical and empirical evidence on herding suggests that in equity markets herding may occur when there is lack of information regarding assets. Because it is more difficult to obtain information about small capitalization firms I expect herding is more prevalent with small capitalization firms. Wermers (1999) find evidence of mutual fund herding. Institutional investors tend to move into or out of small capitalization stocks based on analysts’ predictions. A possible reason could be that there is a relative small amount of information available on small capitalization stocks. Wermers (1999) finds little herding taking place in an average stock and greater herding in small stocks.

H2 : In small capitalization firms greater herding effects are expected to be found than in large capitalization firms.

According to Bikhchanandani and Sharma (2001) it is unlikely that investors observe each other’s holdings of an individual stock soon enough to change their own portfolios. The possibility of herding at the

individual stock level is therefore less likely. Therefore it is more likely to find herding at the level of investments in a group of stocks (stocks of firms in an industry sector) after the impact of fundamentals has been factored out. Also different groups of stocks have different risk characteristics and in some industry sectors there is more uncertainty than in others. Moskowitz and Grinblatt (1999) provide evidence of a strong industry momentum effect in the industry components of stock returns, supporting the idea of herding at an industry level. Therefore I expect that herding is sector specific.

(11)

11 Chiang and Zheng (2010) examine herding behavior in 18 global markets and they find supportive

evidence for herding in markets during crisis periods. Chiang and Zheng (2010) suggest that extreme portfolio returns consistently occur during crisis periods. Therefore a crisis period is treated as a time period where there is market stress. Therefore, I expect that during crisis periods investors will herd.

(12)

12

3. Methodology & Data

3.1 Methodology

I use the methodologies of Christie and Huang (1995) and Chang et al. (2000) to detect for herding. Other methods mainly focus on detecting the existence of herd behavior among mutual fund managers like Lakonishok et al. (1992) and Wermers (1999) or among financial analysts Gleason and Lee (2003) and Welch (2000). The disadvantage of these methods is that they use data which is not publicly available and only examine specific investor groups. The advantage of the methodologies developed by Christie and Huang (1995) and Chang et al. (2000) is that they only need data which is publicly available and measure herding around the market as a whole.

3.1.1 Dummy variable tests

Christie and Huang (1995) propose the CSSD as a measure to detect herd behavior. The CSSD is calculated as follows,

𝐶𝑆𝑆𝐷_𝑡 = √∑𝑛_𝑖=1(𝑅𝑖𝑡−𝑅𝑚𝑡) 2

𝑁−1 , (1)

where Rit is the return of stock i at time t and Rmt is the cross-sectional average of N returns in the aggregate market portfolio at time tand N is the number of stocks in the aggregate market portfolio. The CSSD is a natural measure to capture the influence of herd behavior, however results based on this measure can, according to Christie and Huang (1995), be sensitive to outliers. To test for robustness of the results the regression is repeated by using the CSAD. The CSAD is calculated as follows,

𝐶𝑆𝐴𝐷_𝑡 =∑𝑛𝑖=1│𝑅𝑖𝑡− 𝑅𝑚𝑡 │

𝑁 , (2)

where Rit is the return of stock i at time t and Rmt is the cross-sectional average of N returns in the aggregate market portfolio at time tand N is the number of stocks in the aggregate market portfolio.

Christie and Huang (1995) hypothesize that investors are drawn to the consensus of the market when there is market stress. I proxy for periods of market stress by examining the daily return distribution of a

(13)

13 the daily return distribution as periods of market stress. Christie and Huang (1995) develop the following dummy variable regression models, which I also use to detect herding;

CSSDt = α + βLDtL + βUDtU + εt (3) CSADt = α + βLDtL + βUDtU + εt (4)

where α is a constant, βL and βU are coefficients,DtL and DtU are dummy variables and εt is the error term. DtL = 1 if the market return on day t lies in the extreme lower tail of the distribution and DtL = 0 otherwise. DtU = 1 if the market return on day t lies in the extreme upper tail of the distribution and DtU = 0 otherwise. There is evidence of herding when βL and βU are significant and negative.

3.1.2 Non-linearity tests

Chang et al. (2000) develop a non-linearity test, which is a less stringent method than the dummy variable regression to detect for herding in the market. Chang et al. (2000) presume that rational asset pricing models predict that equity return dispersions will increase with the market return. Furthermore they assume that the relationship between equity return dispersions and the market return is linear. If investors ignore their private information and instead follow the performance of the market, then the linear and increasing relation between equity return dispersions and the market return is violated. The relationship can become non-linear increasing or even decreasing, this is evidence of herding. As a measurement of equity return dispersion Chang et al. (2000) use the CSAD. In contrast to the methodology developed by Christie and Huang (1995) the method of Chang et al. (2000) can measure herding during normal periods and of market stress periods. Tan et al. (2008) Chiang and Zheng (2010) find evidence in favor of asymmetry of herding in up and down markets. To allow that herding behavior may be different for up markets than for down markets, two separate regressions are run;

CSADtdown = α + γ1down │Rmtdown│ + γ2down (Rmtdown)2 + εt (5) CSADtup = α + γ1up │Rmtup│ + γ2up (Rmtup)2 + εt (6)

(14)

14 returns. Absolute values are used for │Rmtdown│ and │Rmtup│, because the sign does not matter, only the value.

If investors herd around the market during periods of relative large market movements this would result in a non-linear relationship between CSAD and the average market return. This non-linearity would be captured by a statistically significant value of the γ2 coefficient, where a negative and statistically significant γ2 coefficient is evidence of herding.

3.1.3 Sample specifications

I also test if herding is more prevalent in small capitalization stocks than in large capitalization stocks. Therefore, companies are allocated to either a group small or large capitalization stocks. I use the same definition as Caparrelli et al. (2004) to determine whether a company is a small or large capitalization company. Companies with a market value below (above) 1 billion euros are considered small (large) capitalization companies. For each year the market capitalization of a company is obtained from Thomson Reuters Datastream. The market capitalization determines whether a company is classified either as a small or as a large capitalization company in that specific year.

Furthermore, I test whether herding is industry sector specific. Therefore, I allocate companies to an industry sector based on their Standard Industry Classification code (SIC code). Table 1 shows the SIC codes and the corresponding industry sector. Consistent with the methodology used by Henker et al. (2006), each industry sector portfolio must contain at least 20 companies at any time, which contributes to reliable results. The data set does not contain any companies in SIC code group 9000 to 9999 and therefore no further descriptive statistics and results are presented about this industry sector.

Table 1. SIC code groups

SIC code Industry sector

SIC 0000 – 1999 Agriculture, forestry, mining, construction

SIC 2000 – 2999 Manufacturing of food, lumber, paper, tobacco

SIC 3000 – 3999 Manufacturing of metal, electronics, machinery, transportation equipment

SIC 4000 – 4999 Transportation and public utilities

SIC 5000 – 5999 Wholesale and retail trade

SIC 6000 – 6999 Finance insurance and real estate

SIC 7000 – 8999 Services

SIC 9000 – 9999 Public Administration

(15)

15 data sample consist of 254 observations for the UK and 253 observations for Germany and Italy, and is called crisis period sample. Because of this small amount of observations, the regression models in equations (3) and (4) are ill suited to measure herding, since in almost all industry sectors there was only one observation in the upper (1%) or lower (1%) tail of the distribution. This is not sufficient to test for herding, and therefore I only test for herding by the non-linearity test of Chang et al. (2000). The dummy variable regression test of Christie and Huang (1995) is not necessary to identify periods of market stress during the crisis period, since Chiang and Zheng (2010) suggest that extreme portfolio returns consistently occur during crisis periods.

3.2 Data

All data employed in this paper is collected from Thomson Reuters Datastream. The data period of the sample used in this paper ranges from 1 January 2003 till 31 December 2012, and is called complete

sample. I use daily returns of common shares for the UK, Germany and Italy. The data sample incorporates newly issued stock during the investigation period. Investment companies and mutual funds are excluded from the data sample. Daily returns are only present in the data set when the stock is traded on that specific day. By performing this action all dead stocks are removed from the sample.

The combined market capitalization of listed companies in the UK, Germany and Italy is approximately 5 trillion US dollar, which is almost half of the market capitalization of all listed companies in Europe1. The UK has a market capitalization of approximately 3 trillion US dollar, Germany has a market capitalization of approximately 1,5 trillion US dollar and Italy has a market capitalization of 0,5 trillion US dollar. According to de Jong and Semenov (2002), there exists a strong relationship between a country’s cultural characteristics and its origin legal system. There are cultural differences between countries based on English legal origin (Australia, Canada, New Zealand, UK, US), German legal origin (Austria, Germany, Switzerland) and French legal origin (Belgium, France, Greece, Italy, Portugal, Spain). Countries based on the same legal origin have similar scores on some of Hofstede’s (1980) dimensions, scores that differ from countries based on other legal origins. The UK, Germany and Italy are selected in this research, because with their relative large market capitalization and different legal origin they serve as a good proxy for advanced European countries.

Table 2 presents descriptive statistics of CSSD and CSAD for the UK, Germany and Italy. Panel A reports the statistics for all listed companies in a country and is called full sample. Panel B reports the statistics for the sample categorized by market capitalization and panel C contains statistics for the sample categorized

(16)

16 by industry sector. The UK data set contains 2527 observations and the data sets of Germany and Italy consists of 2542 observations. The mean CSSD in the full sample is the highest in Germany with a value of 4,52 and the lowest in Italy with a value of 2,28. The CSAD is consistently lower than the CSSD, which is consistent with Christie and Huang (1995) who state that the CSSD is much more sensitive for outliers. The reported CSSD and the CSAD in panel B are higher for small capitalization than for large

capitalization companies. In panel C, the industry sector agriculture, forestry, mining, construction consists of a minimum of 5 companies for Germany and 6 companies for Italy. The industry sector wholesale and retail trade consist of 9 companies for Germany and 5 for Italy. These industry sectors in Germany and Italy are excluded from examination, since they do not meet the minimum of 20 observations criterion. Table 3 presents the descriptive statistics for the crisis period sample. Panel A reports the statistics for all listed companies in a country and is called full sample. Panel B reports the statistics for the sample

categorized by market capitalization and panel C contains statistics for the sample categorized by industry sector. The crisis period sample consists of 254 observations for the UK and of 253 observations for Germany as well as for Italy.

(17)

17 Table 2

Descriptive statistics of CSSD and CSAD complete sample

Panel A – Full sample Full sample Panel B – Market capitalization Small Large

UK GR IT UK GR IT UK GR IT

CSSD Minimum 1,72 2,25 0,97 CSSD Minimum 1,85 2,43 0,92 0,63 1,02 0,67

CSSD Maximum 9,86 13,7 12,18 CSSD Maximum 11,08 14,34 14,34 9,06 20,01 14,64

CSSD Mean 3,64 4,52 2,28 CSSD Mean 4,02 4,43 2,52 1,87 3,77 1,67

CSSD Std. dev. 1,17 1,44 0,85 CSSD Std. dev. 1,31 1,33 0,93 0,80 2,27 0,72

CSAD Minimum 0,82 1,53 0,68 CSAD Minimum 0,86 1,61 0,64 0,49 0,75 0,52

CSAD Maximum 5,29 9,17 4,35 CSAD Maximum 5,26 9,39 4,75 4,90 7,68 4,32

CSAD Mean 1,85 2,53 1,48 CSAD Mean 2,00 2,67 1,61 1,29 1,94 1,21

CSAD Std. dev. 0,60 0,61 0,47 CSAD Std. dev. 0,63 0,61 0,48 0,52 0,74 0,45

N Minimum 417 230 139 N Minimum 315 167 101 90 53 42

N Maximum 1030 593 237 N Maximum 826 502 205 202 110 72

Number of observations 2527 2542 2542 Number of observations 2527 2542 2542 2527 2542 2542

(18)

18

Note: This table lists descriptive statistics of daily equally weighted cross-sectional standard deviations (CSSD) 𝐶𝑆𝑆𝐷𝑡= √

∑𝑛𝑖=1(𝑅𝑖𝑡−𝑅𝑚𝑡) 2

𝑁−1 and daily equally weighted cross-sectional absolute deviations (CSAD)

𝐶𝑆𝐴𝐷𝑡=∑ │𝑅𝑖𝑡

𝑛

𝑖=1 − 𝑅𝑚𝑡 │

𝑁 for the United Kingdom, (UK) Germany (GR) and Italy (IT). The data range is from 1/1/2003 to 12/31/2012. Small (large) capitalization stocks have a market capitalization below (above) 1billion. Table 2 (continued)

Panel C – Industry sector

Agriculture, forestry, mining, construction Manufacturing of food, lumber, paper, tobacco Manufacturing of metal, electronics, machinery Transportation and public utilities

Wholesale and retail trade

Finance, insurance and real

(19)

19 Table 3

Descriptive statistics of CSSD and CSAD crisis period sample

Panel A – Full sample Full sample Panel B – Market capitalization Small Large

UK GR IT UK GR IT UK GR IT

CSSD Minimum 2,65 2,73 1,56 CSSD Minimum 2,80 2,80 1,55 1,36 1,70 1,14

CSSD Maximum 9,86 12,25 8,75 CSSD Maximum 11,08 12,45 9,92 6,56 17,83 4,77

CSSD Mean 4,50 4,85 2,75 CSSD Mean 4,84 4,71 2,92 2,79 4,36 2,15

CSSD Std. dev. 1,36 1,64 0,77 CSSD Std. dev. 1,48 1,52 0,81 1,07 2,48 0,76

CSAD Minimum 1,50 1,81 1,15 CSAD Minimum 1,58 1,95 1,18 1,02 1,28 0,84

CSAD Maximum 5,30 9,17 4,35 CSAD Maximum 5,26 9,39 4,37 4,90 7,68 3,71

CSAD Mean 2,58 3,00 1,91 CSAD Mean 2,66 3,07 2,01 2,02 2,50 1,62

CSAD Std. dev. 0,85 0,97 0,49 CSAD Std. dev. 0,86 0,96 0,47 0,77 1,01 0,59

N Minimum 562 323 196 N Minimum 428 256 156 123 62 47

N Maximum 905 593 217 N Maximum 772 502 181 126 84 49

Number of observations 254 253 253 Number of observations 254 253 253 254 253 253

(20)

20

Note: This table lists descriptive statistics of daily equally weighted cross-sectional standard deviations (CSSD) 𝐶𝑆𝑆𝐷𝑡= √

∑𝑛𝑖=1(𝑅𝑖𝑡−𝑅𝑚𝑡) 2

𝑁−1 and daily equally weighted cross-sectional absolute deviations (CSAD)

𝐶𝑆𝐴𝐷_𝑡=∑𝑛𝑖=1│𝑅𝑖𝑡− 𝑅𝑚𝑡 │

𝑁 for the United Kingdom, (UK) Germany (GR) and Italy (IT). The data range is from 1/1/2008 to 12/31/2008. Small (large) capitalization stocks have a market capitalization below (above)1 billion. Table 3 (continued)

Agriculture, forestry, mining, construction Manufacturing of food, lumber, paper, tobacco Manufacturing of metal, electronics, machinery Transportation and public utilities

Wholesale and retail trade

Finance, insurance and real

(21)

21

4. Results

Table 4 shows the results of the dummy variable regression tests and the non-linearity test of the complete sample. Panel A presents the results for the full sample, Panel B presents the results for the sample

categorized by market capitalization and Panel C shows the results for the sample categorized by industry sector.

4.1.1 Dummy variable regression tests

I first discuss the results of the dummy variable regression tests of Christie and Huang (1995). Equations (3) and (4) are estimated for the 1% criterion as well as the 5% criterion. The 1% criterion restricts DtL (DtU) to 1% of the lower (upper) tail of the market portfolio return distribution and the 5% criterion restricts DtL (DtU) to 5% of the lower (upper) tail of the market portfolio return distribution.

I expect that during periods of market stress cross-sectional deviations will decrease and therefore expect significant and negative values of βL_{and β}U_. _{However, the results indicate otherwise; all estimates of}_βL_and

βU

are significant and positive across all samples. Therefore, I find no evidence of herding in the full

sample, the sample categorized by capitalization or the sample categorized by industry sector. These results are consistent for all countries. Positive and significant values of βL

and βUindicate that cross-sectional deviations increase in periods of market stress.The estimates of βL

and βUare consistently higher in the 1% criterion test than in the 5% criterion test.This indicates that equity return dispersions actually tend to increase during more extreme periods of market stress. These results are consistent with the predictions of rational asset pricing, but contradict the prediction of herd behavior during periods of market stress. These results are in line with the results of Christie and Huang (1995) and Henker et al. (2006). Christie and Huang (1995) estimate equations (3) and (4) for the US stock market and test for herding in 12

industry sectors from July 1962 to December 1988. Henker et al. (2006) estimate equations (3) and (4) for the Australian stock market from 2001 to 2002 in 5 industry sectors. They both find that equity return dispersions actually tend to increase during periods of market stress in all industry sectors and in the full sample. They both show that equations (3) and (4) find similar results.

4.1.2 Non-linearity tests

Table 4 also shows the results of the non-linearity test of Chang et al. (2000). I perform the non-linearity test for the a down market by equation (5) and for the up market by equation (6). A negative and

(22)

cross-22 sectional return dispersion increases at a decreasing rate. Therefore a negative and statistically significant

γ2 coefficient is evidence of herding.

I find that all coefficients of γ1are significant and positive. This is consistent with the significantly and positive coefficients of βL_{and β}U _{in the dummy variable regression test. A}_{significant and positive}

coefficients of γ1 implies that CSADincreases with absolute market returns. In the up market in the full sample, γ1 is the highest for the UK (0,685) and the lowest for Italy (0,566). Furthermore, γ1 tends to be higher in up markets than in down markets. This suggests that dispersions in stock returns are on average wider on an up market than on a down market day. This is consistent with the findings of Chang et al. (2000).

In the down market I find mixed results for the UK. I find a significant and positive coefficient of γ2 for the sample large capitalization. A significant and positive coefficient of γ2 indicates that cross-sectional return dispersions will increase at an increasing rate. For the industry sectors manufacturing of food, lumber, paper, tobacco and wholesale and retail trade I find significant and negative coefficients of γ2. This indicates that as the average market return is high in absolute terms, the cross-sectional return dispersion increases at a decreasing rate. This is evidence of herding. I find no significant coefficients of γ2 in Germany and therefore no evidence of herding is found in Germany during down markets. In Italy I find evidence of herding in the industry sector services. However, the results of Italy should be treated with care, because the adjusted R2_{is low compared to other findings in other sectors. Also the coefficient of γ}

2 is only significant at the 5% level.

(23)

23 Table 4

Estimates of herding behavior complete sample

Panel A - Full sample

CSSD CSAD Down Up

5% Criterion 1% Criterion 5% Criterion 1% Criterion

α βL _βU _α _βL _βU _α _βL _βU _α _βL _βU _α _γ 1 γ2 Adj R2 _α _γ 1 γ2 Adj R2 Full sample UK 3,566***_0,892***_2,653*** _3,605***_1,192***_3,790*** _1,778***_1,109***_1,945*** _1,816***_1,447***_2,654*** _0,015***_0,607***_-1,166 _0,443 _0,014***_0,685***_12,618***_0,523 (157,2) (7,6) (15,5) (158,6) (6,5) (12,1) (171,5) (20,6) (24,9) (166,7) (16,6) (17,6) (62,1) (12,4) (-0,8) (71,6) (13,1) (5,4) GR 4,443***_0,618***_2,694*** _4,488***_1,084***_3,610*** _2,468***_0,865***_1,781*** _2,505***_1,328***_2,665*** _0,022***_0,422***_-0,054 _0,369 _0,021***_0,678***_-0,022 _0,488 (158,0) (4,0) (14,1) (159,4) (4,1) (9,2) (224,4) (14,4) (23,8) (220,7) (12,5) (16,9) (77,9) (9,3) (-0,1) (116,7) (27,5) (-0,0) IT 2,203***_1,002***_1,735*** _2,256***_1,092***_2,440*** _1,419***_0,880***_1,184*** _1,458***_1,067***_1,933*** _0,012***_0,365***_-0,021 _0,410 _0,011***_0,566***_-1,249* _0,461 (135,6) (12,5) (16,2) (135,8) (7,6) (9,7) (169,5) (21,3) (21,4) (165,2) (14,0) (14,5) (66,6) (12,1) (-0,0) (74,2) (21,2) (-1,8)

Panel B – Market capitalization

CSSD CSAD Down Up

(24)

24 Table 4 (continued)

CSSD CSAD Down Up

(25)

25

This table reports the estimated coefficients of the following regression models: CSSDt = α + βLDtL + βUDtU+εt, and CSADt = α + βLDtL + βUDtU+εt where DtL (DtU) equals 1 if the market return on day t lies in the extreme lower (upper) tail of

the distribution, otherwise DtL (DtU) equals 0. Estimates of herd behavior are shown for the United Kingdom (UK) Germany (GR) and Italy (IT). The 1% and 5% criterion refers to the percentage of observations in the upper and lower tail of

the market portfolio return distribution used to determine extreme price movement days. This table also reports the estimated coefficients of the following regression model for the Up market: CSADtup= α + γ1up │Rmtup│ + γ2up (Rmtup)2 + εt

and the following regression model for the down market: CSADtdown= α + γ1down │Rmtdown│ + γ2down (Rmtdown)2 + εt Adj R2 is the adjusted R2.The data range isfrom01/01/2003 to 12/31/2012. The coefficients of the t-test are indicated in

parentheses.

*Statistical significance at the 10% level. ** Statistical significance at the 5% level *** Statistical significance at the 1% level

Table 4 (continued)

CSSD CSAD Down Up

(26)

26 Chiang and Zheng (2010) also test for herding in the UK and Germany and use non-linearity tests to

measure for herding. Chiang and Zheng (2010) use data that range from 4/25/1989 to 4/24/2009 and consists of 140 stocks for the UK and 125 stocks for Germany. They find that investors herd on up and down market days in the UK and Germany. They do not categorize their data by capitalization or industry sector. I do not find any evidence of herding in the UK and Germany in the full sample. Nonetheless, I do find herding in a couple of industry sectors. A possible explanation for the different results is that the data set that Chiang and Zheng (2010) use, consists of less stocks. Also their data set has more observations and may therefore explain the different results. Caparrelli et al. (2004) test for herding in Italy and use non-linearity tests to measure herding. Their data range from 09/01/1988 to 01/08/2001 and consist of 151 stocks. They also categorize data by capitalization. Caparrelli et al. (2004) find evidence of herding in the full, small

capitalization and large capitalization sample during up market periods. During down market periods they find evidence of herding in the full sample and the large capitalization sample. These results are consistent with the results I find in the full sample and large capitalization sample during the up market period.

However, during the down market periods, I did not find evidence of herding. A possible explanation is that the data range is longer in the research from Caparrelli et al. (2004). Also the results that Caparrelli et al. (2004) find during the up market periods are stronger and more significant than the results that they find during the down market periods. The results they find at the down market are statistical significant at the 5% level.

Table 5 shows the results of the non-linearity test of Chang et al. (2000) during crisis periods. Panel A presents the results for the full sample, Panel B presents the results for the sample categorized by market capitalization and Panel C shows the results for the sample categorized by industry sector. During the up market periods, I find significantly and negative coefficients of γ2 for the full and small capitalization sample in Germany. The results of the full sample are significant at the 5% level and the results of the small sample test are significant at the 1% level. This suggests that investors herd in small capitalization stocks during the crisis in Germany and behave differently during a crisis than in a normal market. During the up market periods, I find significant and positive coefficients of γ2 for the UK. This is consistent with the results I find

(27)

27 Table 5

Estimates of herding behavior crisis period sample Panel A - Full sample

Down Up α γ1 γ2 Adj R2 α γ1 γ2 Adj R2 Full sample UK 0,020*** _0,638*** _-2,787 _0,496 _0,021*** _0,313 _20,115*** _0,617 (22,8) (5,1) (-0,9) (18,5) (1,4) (2,7) GR 0,023*** _0,504*** _-1,251 _0,512 _0,023*** _0,807*** _-1,695* _0,674 (24,1) (4,4) (-0,6) (26,6) (9,5) (-1,8) IT 0,016*** _0,182*** _1,764 _0,445 _0,016*** _0,297*** _1,228 _0,549 (29,8) (3,0) (1,5) (26,9) (4,5) (1,1)

Panel B – Market capitalization

Down Up α γ1 γ2 Adj R2 α γ1 γ2 Adj R2 Small capitalization UK 0,020*** _0,803*** _-5,206 _0,569 _0,021*** _0,480* _28,874** _0,587 (23,6) (5,9) (-1,5) (18,5) (1,7) (2,4) GR 0,025*** _0,472*** _-1,030 _0,485 _0,023*** _0,894*** _-2,243** _0,704 (26,6) (4,0) (-0,5) (27,8) (10,4) (-2,3) IT 0,017*** _0,256*** _0,406 _0,508 _0,017*** _0,250*** _2,658** _0,574 (30,8) (4,3) (0,4) (33,3) (3,8) (2,1) Large capitalization UK 0,016*** _0,172* _1,970 _0,361 _0,018*** _0,040 _5,355*** _0,412 (15,5) (1,7) (1,2) (15,7) (0,3) (2,9) GR 0,019*** _0,211*** _3,069 _0,421 _0,018*** _0,582*** _-0,900 _0,625 (15,0) (1,7) (1,6) (17,7) (7,6) (-1,2) IT 0,012*** _0,216*** _0,683 _0,278 _0,013*** _0,318*** _-0,335 _0,402 (16,1) (2,7) (0,4) (15,5) (4,2) (-0,3)

(28)

28

Table 5 (continued)

Down Up α γ1 γ2 Adj R2 α γ1 γ2 Adj R2 Transportation and public utilities UK 0,015*** _0,682*** _-4,346 _0,377 _0,018*** _0,146 _24,363*** _0,384 (13,8) (4,5) (-1,2) (15,0) (0,6) (2,9) GR 0,024*** _0,322*** _0,455 _0,409 _0,025*** _0,458*** _0,051 _0,570 (24,6) (3,4) (0,3) (26,5) (6,1) (0,1) IT 0,014*** _0,274*** _0,653 _0,371 _0,014*** _0,526*** _-0,346 _0,395 (18,3) (3,2) (0,4) (12,8) (4,1) (-0,2) Wholesale and retail trade UK 0,017*** _0,628*** _-4,287 _0,332 _0,021*** _0,137 _9,416** _0,414 (13,0) (4,4) (-1,5) (14,5) (0,7) (2,1) Finance, insurance and real estate UK 0,019*** _0,400** _1,884 _0,350 _0,018*** _0,833*** _-2,644 _0,461 (15,9) (2,6) (0,6) (11,4) (4,2) (-0,6) GR 0,023*** _0,565*** _-2,536 _0,513 _0,024*** _0,509*** _1,488 _0,579 (25,9) (5,5) (-1,4) (23,0) (5,1) (1,1) IT 0,014*** _0,253*** _1,576 _0,427 _0,014*** _0,350*** _1,067 _0,511 (19,0) (3,2) (1,1) (16,7) (3,9) (0,8) Services UK 0,018*** _0,569*** _-0,692 _0,538 _0,018*** _0,628*** _11,824 _0,499 (24,5) (4,6) (-0,2) (16,6) (2,7) (1,2) GR 0,024*** _0,547*** _-2,230 _0,453 _0,024*** _0,777*** _-0,016 _0,577 (26,0) (4,2) (-0,8) (23,0) (6,8) (0,0) IT 0,015*** _0,312*** _-1,715 _0,233 _0,018*** _0,222** _0,981 _0,309 (18,4) (3,8) (-1,2) (20,0) (2,6) (0,8)

This table reports the estimated coefficients of the following regression model for the Up market: CSADtup= α + γ1up │Rmtup│ + γ2up

(Rmtup)2 + εt and the following regression model for the down market: CSADtdown= α + γ1down │Rmtdown│ + γ2down (Rmtdown)2 + εt Adj R2 is

the adjusted R2_{. Estimates of herd behavior are shown for the United Kingdom (UK) Germany (GR) and Italy (IT).The data range is}

from01/01/2008 to 12/31/2008. The coefficients of the t-test are indicated in parentheses. * Statistical significance at the 10% level.

(29)

29

5. Conclusion

In this paper herd behavior in advanced European countries is studied. This paper focuses on herding around the market, this occurs when investors ignore the individual fundamentals of a stock and instead follow the performance of the market. I use daily data from 01/01/2003 to 12/31/2012 and the dummy variable

regression test of Christie and Huang (1995) and the non-linearity test of Chang et al. (2000) to determine whether herding is present.

I did not find evidence of herding with the dummy variable regression test in all samples. However, I find that equity return dispersions actually tend to increase during periods of market stress. These results are consistent with the predictions of rational asset pricing and with the results that Christie and Huang (1995) and Henker et al. (2006) both find. They find that equity return dispersions actually tend to increase during periods of market stress in all industry sectors and in the full sample.

I did find evidence of herding with the non-linearity test. The different results can be explained because the non-linearity test is a less stringent method than the dummy variable regression test to detect herding in the market. According to Chang (2000), the dummy variable regression test requires a far greater magnitude of non-linearity in the return dispersion and mean return relationship for evidence of herding.

In the down market I find mixed results for the UK. In the large capitalization sample the cross-sectional return dispersions will increase at an increasing rate. For the industry sectors manufacturing of food, lumber, paper, tobacco and wholesale and retail trade I find evidence of herding. There is no evidence of herding in Germany during down markets. In Italy there is evidence of herding in the industry sector services.

In the up market in the UK the cross-sectional return dispersions will increase at an increasing rate in all samples. Therefore, there is no evidence of herding in the UK in the up market. For Italy there is evidence of herding in the up market. There is evidence of herding in the full sample, large capitalization sample and in several industry sectors. This suggests that investors herd during periods when the Italian market is going up. However, the results should be treated with care because most results are only significant at the 5% level.

(30)

30 herding in the full sample and the large capitalization sample. These results are consistent with the results I find in the full sample and large capitalization sample during the up market period. However, during the down market periods, there is no evidence of herding. A possible explanation is that the data range is longer in the research from Caparrelli et al. (2004). Also the results that Caparrelli et al. (2004) find during the up market periods are stronger and more significant than the results they find during the down market periods. The results at the down market are only just statistical significant at the 5% level.

During the crisis period there is evidence of herding in small capitalization stocks in Germany. Investors in Germany behave differently during a crisis than in a normal market. I did not find evidence of herding in the UK during the crisis period. Chiang and Zheng (2010) also test for herding in the UK and Germany during crisis periods. Their results are consistent with the results I find for the UK, but not consistent with the results for Germany. A possible explanation for the different results is the different data range and the different number of stocks in the sample.

However, there are some limitations of the methodologies in this paper. One of the limitations of the

methodology of Christie and Huang (1995) is that it requires the definition of extreme returns. In this paper I define the upper (lower) 1% and 5% of the market return distribution as periods where market stress occurs. This definition of market stress is arbitrary. The greatest limitation is that both the models that I use only detect herding if herding manifests itself in stock returns. Therefore, when these models do not find any evidence of herding, it does not necessarily imply that herding is not present. Therefore, results must be treated with care, since this does not have to mean that other kinds of herding are not present.

(31)

31

Literature list

Asch, S., 1956, Studies of independence and conformity: I. A Minority of one Against a Unanimous Majority, Psychological Monographs, 70:9, 1-70.

Bikhchandani, S., Hirshleifer, D. and Welch, I., 1992, A Theory of Fads, Fashion, Custom, and Cultural Change as Informational Cascades, Journal of Political Economy, 100:5, 992-1026.

Bikhchandani, S., and Sharma, S., 2000, Herd Behavior in Financial Markets, IMF Staff Papers, 47:3, 279-310.

Caparrelli, F., D'Arcangelis, A. M. and Cassuto, A., 2004, Herding in the Italian Stock Market: A Case of Behavioral Finance, Journal of Behavioral Finance, 5:4, 222-230.

Chang, E., Cheng, J. and A. Khorana., 2000, An Examination of Herd Behavior in Equity Markets: An International Perspective, Journal of Banking and Finance, 24:10, 1651–1679.

Chiang, T.C., Li, J. and Tan, L. 2010, Empirical investigation of herding behavior in Chinese stock markets: Evidence from quantile regression analysis, Global Finance Journal, 21:1, 111-124.

Chiang, T. C. and Zheng, D., 2010, An empirical analysis of herd behavior in global stock markets, Journal

of Banking & Finance, 34:8, 1911-1921.

Christie, W.G., and Huang, R. D., 1995, Following the Pied Piper: Do Individual Returns Herd around the Market?, Financial Analysts Journal, 51:4, 31-37.

Demirer, R. and Kutan, A.M., 2006, Does herding behavior exist in Chinese stock markets?, Journal of

International Financial Markets, Institutions & Money, 16:2, 123-142.

Denevow, A. and Welch, I., 1996, Rational Herding in Financial Economics, European Economic Review, 40, 603-615.

Economoua, F., Kostakisb, A., and Philippas, N., 2010, An Examination of Herd behavior in Four Mediterranean Stock Markets, working paper.

(32)

32 Henker, J., Henker, T. and Mitsios, A., 2006, Do Investors Herd Intraday in Australian Equities?,

International Journal of Managerial Finance, 2:3, 196-219.

Hirshleifer, D., and Teoh, S., 2003, Herd Behaviour and Cascading in Capital Markets: a Review and Synthesis, European Financial Management, 9:1, 25-66.

Hofstede G., 1980, Culture’s Consequences, Beverly Hills, Sage.

Jong, E., and Semenov, R., 2002, Cross-Country Differences in Stock Market Development: A Cultural View retrieved from: http://som.eldoc.ub.rug.nl/FILES/reports/themeE/2002/02E40/02e40.pdf

Lindhe E., 2012, Herd Behavior in Stock Markets: A Nordic Study, master thesis, Lunds Universitet.

Moskowitz, T.J. and Grinblatt, M., 1999, Do industries explain momentum?, Journal of

Finance, Vol. 54:4, 1249-90.

Nofsinger, J.R. and Sias, R.W., 1999, Herding and feedback trading by institutional and individual investors, Journal of Finance, 54:6, 2263-95.

Scharfstein, D.S. and Stein, J. C., 1990, Herd Behavior and Investment, The American Economic Review, 80:3, 465-479.

Standard & Poor's, Global Stock Markets Factbook and supplemental S&P data retrieved from: http://www.indexmundi.com/facts/indicators/CM.MKT.LCAP.CD/rankingsAP.CD/rankings

Tan, L., Chiang, T.C., Mason, J. R. and Nelling, E., 2008 Herding behavior in Chinese stock markets: An examination of A and B shares, Pacific-Basin Finance Journal, 16:1, 61-77.

Herd behavior in advanced European countries