Overconfidence and Trading Volume

(1)

Overconfidence and Trading Volume

An analysis of the relationship between investor overconfidence and trading volume

Delano Brons Thesis Msc. Finance University of Groningen

‘Overconfidence is a very serious problem. If you don’t think it affects you, that’s probably

because you’re overconfident’ - Carl Richards.

Keywords: Investor Overconfidence, Trading Volume, Turnover, Return, Rationality, Behavioural Finance.

JEL - Classifications: G02; G11

Groningen, 22 January 2016

(2)

Abstract

(3)

Table of Contents Introduction ... 3 1. Literature review ... 5 1.1 Theory on overconfidence ... 5 1.2 Market return ... 6 1.3 Trading volume ... 7 1.4 Other explanations ... 8

1.5 Empirical work on overconfidence and trading volume ... 9

Hypothesis (H1) ... 10

2. Data & Methodology 2.1 Data ... 11

2.2 Empirical methodology (VAR Model) ... 11

2.3 Variables & Descriptive data ... 14

3. Empirical results 3.1 VAR Model results ... 21

3.2 Impulse response function ... 25

3.3 Detrended VAR model ... 28

3.4 Without crisis periods ... 30

(4)

High trading volume is often detected in financial global markets. It is the view of many studies such as Heath and Tversky (1991), Coval and Moskowitz (1999), Lui, Strong and Xu (1999), and Tourani-Rad and Kirkby (2005) that rational market theories are not able to explain all trading volume in full. Therefore, a number of authors present behavioural biases to do so (Odean, 1998; Barber & Odean, 2001; Grinblatt & Keloharju, 2009). These studies have tended to focus on irrational investor behaviour rather than on the basic thought of rationality. The basic thought of rationality, which assumes perfect markets and perfect people is highly debatable. Besides it is becoming generally accepted that investors can behave irrational. Throughout the so-called behavioural finance literature, one behavioural bias, namely overconfidence is prominently present. Overconfidence is widely used to explain irrational behaviour (Odean, 1998; Barber & Odean, 2001; Glaser & Weber, 2009). The fact that many considerations have to be taken into account for a trading decision makes trading a difficult task. Especially in these difficult tasks, human emotions affect decision-making (Odean, 1998a). As Barber and Odean (2001) argue that overconfidence is most present in low predictable and difficult tasks, it is likely that overconfidence is present in trading decisions. Consequently, investor overconfidence is a popular phenomenon to study.

The aim of this study is to investigate if investor overconfidence is present on financial markets, more specifically the Dutch stock market. Next, I will investigate if overconfidence leads to high volume trading. The goal of this study is to examine and provide comprehensive and empirical evidence of the relation of overconfidence with trading volume. A considerable amount of research about the relationship between overconfidence and trading volume is performed in many financial markets in the world. However, the existence of overconfidence is mostly examined in the United States (US) (Statman, Thorley & Vorking, 2006). The outcomes of studies done in the US might differ from a comparable study done in the Netherlands. Because of a different market and time period, which includes the recent financial crisis in 2008, this paper contributes to the existing literature.

(5)

volume. For this research, monthly time-series data is analysed from September 2000 to September 2015. The shares are from the most traded companies in the Netherlands on the Dutch stock market; they are listed in the Amsterdam Exchange Index (AEX). By having a sample period of 2000 to 2015 a comparison can be made between the post-Internet and pre- and post-global financial crisis periods. In short, I attempt to confirm that overconfidence leads to excessive trading and will use the relation between return and turnover to confirm the existence of overconfidence.

Statman et al. (2006) also state that overconfidence is about investors’ belief about trading in general. On the other hand, the disposition effect is more about specific stocks in their portfolio. Aspara and Hoffman (2013) define the disposition effect as the willingness to sell winning stocks too early and keep losing stocks for too long. Both seem to increase trading volume. However, the focus of this study is about investors’ belief on trading in general. Either way, there are multiple explanations for trading activity. For this reason, this paper makes a clear distinction between overconfidence and the disposition effect. Hence, investor overconfidence is used to try explaining high trading volume.

The results show that market return affects market turnover. The relationship between market return and market turnover is definitely present. However, in contrast to previous literature, the relationship is found to be both negative and positive. To confirm the existence of overconfidence, it should be an exclusive positive relationship. Hence, the presence of overconfidence on the Dutch stock market cannot be confirmed. Second, due to the Internet crisis and the financial crisis, which both affected the global financial markets, the crisis averages of return, turnover and volatility are extraordinary. Despite of the extraordinary crisis averages, the empirical tests show that both crisis periods do not significantly affect the results. Furthermore, I find that previous turnover continues to affect turnover for multiple following months and is thus autocorrelated.

The remainder of this paper is organized as follows. Section I describes overconfidence, trading volume, and past theoretical and empirical work on behavioural finance pertinent to the present study. Section II contains the data and empirical methodology. It also provide the main descriptive information. Section III presents the results. Finally, Section IV concludes the paper.

(6)

1. Literature review

This section defines what overconfidence is, how it originates and summarizes the most relevant studies in this field. It reports on the impact of overconfidence on investors’ trading behaviour and the financial markets. I will divide this section into a theoretical and empirical part. Finally, I will summarize the most important findings.

1.1 Theory on overconfidence

(7)

information and their ability to pick stocks. Barber & Odean (2001) further argue that overconfident investors think their belief is better than others. On the whole, one could argue that overconfidence is often present in daily emotions, decisions and actions. As a result, a large number of researchers investigate the phenomenon overconfidence in financial markets (Odean, 1998a; Gervais & Odean, 2001; Statman et al., 2006; Glaser & Weber, 2009). Different approaches could be applied when researching overconfidence, but the academic literature mainly takes shape in the theoretical models of Odean (1998b), and Gervais and Odean (2001). These theoretical models mostly focus on financial market participants. In 1985, Kyle developed the term ‘noise traders’. It represents all investors with irrational behaviour. This could be due to a behavioural bias, or other irregularities. Most likely, irregularities are a result of overconfidence, as Odean (1998) argues this as the most primary behavioural bias. Kahneman and Tversky (1996, p. 515) state that ”confidence is the subjective probability or degree of belief associated with what we think will happen”. To some extent, this definition is directly applicable to the three types of overconfidence of De Bondt and Thaler (1994). They describe overplacement, overestimation and overprecision. They state that investors overestimate their own trading skills and accuracy of their knowledge, and therefore, think their investment decisions are correct too often. In the view of Kahneman and Tversky (1996), investors believe that they know what will happen. Additionally, Brown and Sarma (2007) argue that overestimation of one’s own trading skills is the most common type. This relates to ‘miscalibration’, which regards the accuracy of one’s own beliefs. For instance, a realized high return is perceived as if it is a result from one’s own trading abilities. Furthermore, overconfidence could lead to poor stock and market evaluations. Also, the belief of trading in general is too optimistic, which leads to a better-than-average effect. Owing to the better-better-than-average effect, investors believe they outperform others. Moreover, it makes the assumption that overconfidence is most likely present in daily trading decisions more reasonable.

1.2 Market return

(8)

factors. Daniel et al. (1998) argue that the level of overconfidence is based on all earlier experiences. Initially, an investor starts with a normal amount of believe in himself. The belief of investors grows or diminishes with his experiences on the financial market. If investors realize positive returns, the belief will grow and vice versa. The theoretical models of Gervais and Odean (2001) develop through this line of argument. Hence, it seems to be the case that high market returns make investors more confident about their trading abilities. In other words, the confidence of an investor is determined through past outcomes. Even if positive returns are the direct result of a bull market, overconfident investors perceive it as if it is due to their own trading ability. On the other hand, low market returns make investors less confident about their trading abilities. Though, investors often seem to weigh losses more heavily than gains.

1.3 Trading volume

(9)

Granting the former argument is true, then more trading opportunities are available, which eventually lead to more trading activities. Odean (1998) argues that the most inconsistent market participants are individual investors. They often do not have the expertise and knowledge that professional investors have. So for them it is harder to remain an optimal trading activity and find an optimal trading option. As mentioned earlier, it makes trading a difficult task and it is in these difficult tasks, where overconfidence is most present.

1.4 Other explanations for trading volume

(10)

more they think their own opinion is correct. The confidence strengthens one’s own opinion and corresponding actions.

1.5 Empirical work on overconfidence and trading volume

(11)

the picture when trying to explain trading activity at investors. High trading volume can be seen as an irrationality of the financial market participants.

In summary, I can conclude that there are various explanations for trading volume and multiple approaches in researching overconfidence. In this thesis I examine the relationship between overconfidence and trading volume. Though, volatility and cross-sectional volatility are included as control variables for other explanations. Despite the findings in previous studies, the fact that overconfidence is a behavioural bias makes overconfidence hard to measure. Even so the lead-lag relationship between market return and market turnover seems to be strong enough to be a good indication for overconfidence. For the most part the studies use trading volume to investigate the presence of overconfidence on financial markets. Following Statman et al. (2006) trading volume is calculated by using turnover rate and since the relation between market return and market turnover seems to be strong enough to confirm the existence of overconfidence, this gives the following hypothesis:

(12)

2. Data and methodology 2.1 Data

For this study I choose to collect the data from Thomson Reuter DataStream. For the methodology I follow Statman et al. (2006), but with a different dataset. Monthly trading data is collected for return and turnover for the period September 2000 to September 2015 from the AEX. The AEX is a free-float adjusted market capitalization weighted index of the leading stocks traded on the Amsterdam Exchange. It consists of the top 25 companies in the Netherlands. One could argue that the presence of irregularities is best measured from stocks that are frequently traded. It seems to be more difficult to attract relevant data from stocks that are not traded often. The stocks of these companies are the most traded stocks in the Netherlands. Given these points, the AEX is a logical choice to investigate. Based on the analysis of the AEX dataset, I try to conclude for the Dutch financial market in full. From the base date (September 2000) the 25 stocks with the highest trading volume are listed on the AEX. However, during my sample period, the requirements to be listed on the AEX changed. Initially, the turnover in value and volume determined the stocks listed on the AEX. Subsequently, the market capitalization of the firms becomes leading. This change made the AEX internationally more accepted and simultaneously improved the AEX as an indicator for the whole Dutch stock market. A marginal note is that the change limits my data collection for some stocks. As a consequence of the change, six stocks are excluded from my research. For the remaining 19 stocks, presented in Figure A1 in Appendix A, I collect data for the whole sample period. Even though six stocks are excluded from the analysis, the remaining stocks are expected to obtain the same results. The trading file is conducted with data from constituent stocks. It includes the records of turnover by volume, outstanding number of shares, monthly returns, market capitalization of each stock and monthly volatility. The returns are accounted for dividends. More specifically, paying out dividends normally lowers the stock price of the company with the dividend amount, but in these returns the dividends are accounted for and do not lower the stock price. In short, based on the behaviour and movements of these stocks, I extend these findings to draw conclusions about the whole Dutch stock market.

2.2 Methodology

(13)

capable of telling how strong and for what period of time a variable will affect another variable. In this research, the effect of market return on market turnover is most important. I use an extended VAR model, the VARX model, to study the relation between market turnover and market return time-series. The VARX model includes control variables, which are: market volatility and dispersion. In specific, the VARX model of this study is:

mturn! mret! = α!"#$% α!"#$ + Ak mturn!!! mret!!! + Bl misg_!!! disp_!!! + ϵϵ!"#$%,!!"#$,! ! !!! ! !!! (1) Where !"#$%!

!"#$! is a n x 1 vector for n variables with t observations each, and Ak is a n x n

matrix measuring the coefficients of lagged terms of !"#$%!!!

!"#$!!! itself. With two control

variables, !"#$!!!

!"#$!!! is a 2 x 1 vector of exogenous variables, while Bl is a n x 2 matrix of

coefficients. N and M represent the optimal lag length for !"#$%!

!"#$! and

!"#$!!!

!"#$!!! ,

(14)

above the 5% criteria level and therefore the null-hypothesis of ‘residuals are multivariate normal’ cannot be rejected. It means that the residuals of the VAR model are normally distributed. Lastly, the VAR model is tested on heteroscedasticity. Both equations of the VAR model are estimated with White standard errors through an Ordinary Least Squares (OLS) method. Market turnover is the dependent variable in the first equation and market return is the dependent variable in the second equation. The heteroscedasticity test computes a p-value above the 5% criteria level, so the null-hypothesis of ‘homoscedasticity’ cannot be rejected. In other words, the VAR model is homoscedastic, which is desirable. These diagnostic tests are necessary to conduct, so that the results of the VAR model are reliable and can be interpreted.

Furthermore, the correlation between market return and market turnover can be tested with a Granger-causality test and a Wald-test. These tests could indicate correlation between the two endogenous variables. However, the correlation does not suggest anything about the duration of the impact and relationship between both endogenous variables. That being the case, impulse response functions are inserted. In general, the impulse response function is a shock to a VAR system. It identifies the responsiveness of the dependent variables in the VAR model, when a shock is put to a variable. The function is widely used to analyse and determine the effect on the endogenous variables. In this study, a shock to a change in one variable will bring a change in market return or market turnover. I will test the response of market turnover to an impulse of market return over time and vice versa. With this test the market turnover respond over time can be seen. It is a more specific manner to know how the endogenous variables are related to each other and to itself. I choose to use the Cholesky response of one standard deviation. It is worth mentioning that the Cholesky ordering of the impulse response functions is essential for the outcome of the test. For instance, if market turnover is ordered first and market return is ordered second, then the function rules out that changes in market turnover would contemporaneously be affected by shocks to changes in market return. The Cholesky decomposition then imposes that it starts at zero. Afterward it can become non-zero. Hence, in view of this research I choose to Cholesky order market return first and market turnover second. This way the immediate effect of market return on market turnover can be measured.

(15)

only excluding the financial crisis years. In this way I can obtain if these crisis years and periods have a significant effect on the results.

2.3 Variables

As stated in the methodology, the variables implemented in the VAR model are both endogenous and exogenous. The endogenous variables are market return and market turnover, the exogenous are monthly volatility and dispersion. The exogenous variables are better known as control variables. First of all, the turnover rate is used to indicate trading volume. Lo & Wang (2000) developed formulas for turnover rate and market turnover. The turnover rate indicates how often shares are traded, but simultaneously considers and eliminates growth. This makes turnover a proper rate to indicate trading volume. The following formula is used to compute turnover: 𝑇_! = 𝑋_! / 𝑁_!. First, 𝑋_! is monthly turnover by volume per stock. Second, 𝑁! stands for the number of outstanding shares per stock. Monthly turnover by volume dividing by the outstanding shares of the same period and 𝑇_! is computed, which is the monthly turnover rate per stock. With this formula, the monthly security turnover rate can be produced. Subsequently, for this VAR model the security turnover is aggregated into market turnover, which is done on a value-weighted base. These weights are based on the market capitalization of a stock relative to the whole market capitalization of the same period of the AEX. It represents the size of the company. The weights are accounted for the excluding of six stocks. In other words, the sum of the weights of the nineteen stocks included in the analysis is one. To be consistent, the weights are computed on a monthly basis. So, by implementing these weights to turnover rate of each stock, the monthly market turnover is produced. These calculations are conducted throughout the whole sample period September 2000 to September 2015. This makes the formula to compute market turnover as follows:

𝑇_!" = ! 𝑊𝑖 𝑇𝑖

!!! (2)

(16)

with trend and intercept has a probability of 0.00. However, the ADF tests on level with intercept or none, both have p-values above the 5% criteria level, so the null-hypothesis cannot be rejected. This is due to the fact that the long-term downwards trend is still present. Fortunately, the unit root test for the whole VAR model shows a stable model. In spite of some studies that might suggest that this long-term trend should also be removed to conduct reliable and valid analyses, I will not detrend logged market turnover. If market return is detrended, then possible long-term results will not be measured. Detrending logged market turnover will eventually lead to biased results, because past returns can affect trading volume for a long period of time (Gervais & Odean, 2001; Statman et al., 2006). Hence, I will present non-detrended log turnover as the base model and comment on how the results differ when conducting a VAR model with detrended logged market turnover. In the VAR model and throughout this paper I will label the non-detrended logged turnover series as mturn.

Figure 1: Monthly market turnover for AEX

This figure shows the raw monthly market turnover.

(17)

Table 1: Descriptive statistics monthly market turnover

This table shows the mean, standard deviation, minimum and maximum of the monthly market turnover in percentages. The different sub periods are based on the Internet –and financial crisis years. The Internet bubble, together with 9/11 terrorist attack, led to the crises years of 2000 and 2001. The peaks of financial crisis years are 2008 and 2009. Sub period B is the whole sample period, excluding both crisis periods.

Figure 1 shows very high turnover in the first years of the sample. Shortly after, turnover dropped and began decreasing from mid-2002. It decreases to just below 10% and after 2011 it stays below 10%. Both spikes, in 2001 and 2008 are likely results of the crises. The spikes in 2000 and 2001 are most likely due to the Internet crisis and 9/11. The insecurities following from these both events together led to excessive trading and thus extraordinary high monthly turnover. Table 1 shows descriptive statistics of different periods, inclusive and exclusive crisis periods. Sub period A in Table 1 shows the highest mean of almost 27%. It also shows an average turnover of almost 13% for the whole sample period and the standard deviation is more than 6%. This means that during the Internet crisis the average turnover is more than two standard deviations higher. For this reason, one could state that Internet crisis period is substantial different from the average of the whole sample period. Second, the spike in September 2008 is caused by the financial crisis. More specifically, the fall of Lehman Brothers led to unusual global financial market movements. In these following months many stocks were traded excessively. Also the impact of the financial crisis1_{is worth looking at.}

After the financial crisis, which has an average turnover rate of over 13%, turnover dropped to below 8%. Especially, compared to the pre-financial crisis years of 2002-2007, the turnover dropped substantial. In view of this research, one could argue that the financial crisis of 2008 has had its impact in the confidence of investors. More specifically, it is likely that the financial crisis of 2008 and 2009 decreased the confidence level of investors. The drop in turnover and thus trading volume might be explained by changes in overconfidence.

1_{2008 and 2009 were arguably the heaviest years of the financial crisis. In view of the whole}

sample period of 15 years, I choose 2008 and 2009 as the financial crisis years.

Full Sub. A Sub. B Sub. C Sub. D Sub. E

Period September 2000 - 2000 - 2001 2000 - 2015 2002 - 2007 2008 - 2009 2010 - 2015 September 2015 internetcrisis excl. crisis pre fin. cris. during crisis post fin. cris.

(18)

(19)

Figure 2: Monthly market return of AEX

This figure shows the monthly market return, mret.

Table 2: Descriptive statistics monthly market return

This table shows the mean, standard deviation, minimum and maximum of the monthly market return in percentages. Similar as table 1, the (sub) periods are based on both crises periods. Sub period B is the whole sample period, excluding both crisis periods.

Following Statman et al. (2006), I introduced two control variables to the VAR model. The first control variable is market volatility, labelled in the VAR model as misg. This is the monthly return volatility for the stocks of the AEX, measured in percentage points. The volatility of the AEX per month is attracted from Thomson Reuter Datastream. Statman et al. (2006) include it as a control variable in their VAR model based on the volume-volatility relationship, found by Karpoff (1987). The study of Pagano (1987) is also a motivation for including this control variable in the VAR model. Figure 3 shows multiple peaks in volatility, which is a result of high uncertainties on global financial markets. Table 3 presents descriptive statistics of different periods, inclusive and exclusive crisis periods.

-‐20% -‐15% -‐10% -‐5% 0% 5% 10% 15% se p-‐ 00 se p-‐ 01 se p-‐ 02 se p-‐ 03 se p-‐ 04 se p-‐ 05 se p-‐ 06 se p-‐ 07 se p-‐ 08 se p-‐ 09 se p-‐ 10 se p-‐ 11 se p-‐ 12 se p-‐ 13 se p-‐ 14 se p-‐ 15

Period September 2000 - 2000 - 2001 2000 - 2015 2002 - 2007 2008 - 2009 2010 - 2015 September 2015 internetcrisis excl. crisis pre fin. cris. during crisis post fin. cris.

(20)

Figure 3: Monthly volatility of AEX

This graph shows the monthly volatility of the AEX. There are clearly two peaks during the Internet crisis, and one peak during the financial crisis year 2008.

Table 3: Descriptive statistics of monthly volatility

This table shows the mean, standard deviation, minimum and maximum of the monthly volatility in percentages. Just as table 1, the (sub) periods are based on both crisis periods. Sub period B is the whole sample period, excluding both crisis periods.

It is worth mentioning that table 3 shows a substantial higher volatility than the sample years of Statman et al. (2006). Most likely, this is the result of the volatile years of the two crisis periods in the sample period. Moreover, both crisis periods could lead to more volatile years around those crisis years. For instance, after the financial crisis in 2008 and 2009, the value of the euro decreased and the interest European countries are obliged to pay for their loans increased. For countries with a high national debt, it became more difficult to loan money, so that these countries could hardly fulfil their financial commitments. Hence, multiple economists state that the financial crisis is not over yet in 2015. In short, these events probably affect the monthly market volatility of the whole sample period and not only the crisis years 2000, 2001, 2008 and 2009.

0% 10% 20% 30% 40% 50% 60% 70% 80% Se p-‐ 00 Se p-‐ 01 Se p-‐ 02 Se p-‐ 03 Se p-‐ 04 Se p-‐ 05 Se p-‐ 06 Se p-‐ 07 Se p-‐ 08 Se p-‐ 09 Se p-‐ 10 Se p-‐ 11 Se p-‐ 12 Se p-‐ 13 Se p-‐ 14 Se p-‐ 15

Period September 2000 - 2000 - 2001 2000 - 2015 2002 - 2007 2008 - 2009 2010 - 2015 September 2015 Internetcrisis excl. crises pre fin. cris. during crisis post fin. cris.

(21)

The second control variable is dispersion; this is the monthly cross-sectional standard deviation of returns of the AEX. Throughout this paper dispersion is labelled as disp. It is calculated by multiplying the corresponding market-capitalization weights to the squared deviation from the mean return for each stock (Campbell, 2007). Also this variable is calculated on a value-weighted base. Following Statman et al. (2006), the motivation to include this as a control variable is that stock variance is driven by multiple factors and the idiosyncratic risk of stocks. This means that cross-sectional volatility of the beta of stocks drives the significance of average stock variance in explaining market returns (Goyal & Santa-Clara, 2001). As earlier stated, dispersion considers firm- and industry specifics and is related to the idiosyncratic risk.

Dispersion is calculated with the following formula:

𝑑𝑖𝑠𝑝_!= ! 𝑊_! 𝑅_!− 𝑅

!!! ! (3)

Where 𝑊! stands for the weights based on market capitalization per stock, 𝑅! is the monthly return per stock and 𝑅 is the average market return per month of all companies. 𝑁 represents the number of periods. As stated before, dispersion is labelled as 𝑑𝑖𝑠𝑝.

(22)

Table 4: Descriptive statistics of dispersion

This table shows the mean, standard deviation, minimum and maximum of the monthly dispersion in percentages. The (sub) periods are based on both crisis periods. Sub period B is the whole sample period, excluding both crisis periods.

3. Empirical results 3.1 Results VAR model

With the VAR model, I test the relationship between the two endogenous variables; market turnover and market return, and two exogenous variables; market volatility and dispersion. This section is divided as follows. First, I will present the results of the VAR model with nondetrended log turnover, which is the base-model of this research. Afterwards, I will compare it with the possible different results of the VAR model with detrended log turnover. By means of impulse response function, I will give more insight in the relationship between the endogenous variables. Next, I will present the results of the VAR model of different sub periods, focusing on the impact of crisis periods.

First of all, I report on the results of the nondetrended VAR model. Table 5 presents the results. The columns of the table denote the dependent variables, the rows of the table are the lag terms and control variables. For each coefficient, I present the corresponding standard errors and p-values. The p-values are important, because they indicate if the variables are statistically different from zero. In general, p-values below 10% are significant; p-values below 5% and 1% are highly significant. If statistically significant, then the independent variable affects the dependent variable. At first, I report on the results of market turnover. Table 5 shows that the coefficient of the first market turnover lag is highly significant with market turnover. This means that market turnover of the current month remains influential for the coming month on market turnover. One could state that market turnover depends on its own behaviour. With a Wald-test it is possible to test for the jointly influence of all market turnover lags on itself. The null-hypothesis is that the lags jointly are not different from zero. As the p-value is 0.00, the null-hypothesis can be rejected. Given these results, it can be concluded that market turnover is autocorrelated. Second, Table 5 also shows that market

Period September 2000 - 2000 - 2001 2000 - 2015 2002 - 2007 2008 - 2009 2010 - 2015 September 2015 Internetcrisis excl. crises pre fin. cris. during crisis post fin. cris.

(23)

(24)

endogenous variables, the impulse response function will be conducted and these results will be discussed later in this paper.

Table 5: Market VAR model

The remaining of Table 5 is on the next page.

(25)

Cont’d Table 5

***, **, * indicate significance at the 1, 5, and 10 percent levels, respectively (two-tailed).

Table 6: Granger-causality test

Table 5 also gives insights in the relations between the endogenous and exogenous variables. First of all, it shows that volatility does not significantly influence market turnover. Also the Wald-test shows an insignificant relationship between turnover and volatility. A possible explanation for this unexpected result could be that trading can only have a marginal effect on the stock price in high liquid stocks, because so many trades are made. Hence, some researchers might state that it is safer to trade in high volume stocks than in low volume stocks. On the other hand, with a p-value of 0.0499, dispersion does significantly positive influence market turnover.

Second, I report on market return. As none of the market return lags is significant with market return, I find that market return is not autocorrelated. Second, Table 6 also shows, with the null-hypothesis being “MTURN does not Granger Cause MRET”, a p-value of 0.1898. Hence, the null-hypothesis cannot be rejected. This indicates that past market

MRET (-5) coefficient 0.5604 -0.0636 standard error 0.3054 0.0830 p-value 0.0684* 0.4442 MRET (-6) coefficient 0.5248 -0.1672 standard error 0.3415 0.1012 p-value 0.1264 0.1005 MRET (-7) coefficient 0.1885 -0.1185 standard error 0.2488 0.0832 p-value 0.4499 0.1563 MRET (-8) coefficient -0.0008 -0.0886 standard error 0.2649 0.0790 p-value 0.9976 0.2638 C coefficient -0.3443 0.1073 standard error 0.1007 0.0422 p-value 0.0008*** 0.0120** MISG coefficient 0.1387 -0.3621 standard error 0.1852 0.0773 p-value 0.4551 0.0000*** DISP coefficient 0.7871 0.4028 standard error 0.9571 0.3224 p-value 0.0499** 0.2134

Null Hypothesis: Obs F-Statistic Prob.

(26)

turnover does not influence market return, which is in line with the EMH. On the other hand, two of the eight market turnover lags are significant with market return, at the 5% level. With a Wald-test the findings of the Granger-causality test checked once more. Again the Wald-test checks if the eight lags jointly affect market return and show with a p-value of 0.36 that market turnover is not different from zero. Hence, it does not affect market return. Third, volatility, with a coefficient of -0.3621, has a negative influence on market return. This relation is highly significant, at the 1% level. It is widely believed that high volatile financial markets do not always benefit the market return. That is to say, in volatile markets and more specifically when stocks are volatile, it is harder for investors to realize a positive market return. High volatility causes higher insecurities and therefore realizing positive returns becomes harder. Last, dispersion does not affect market return. The coefficient of dispersion is not significant. Considering the results of the control variables regarding both endogenous variables, it may be said that volatility and dispersion are both correctly inserted in the model as control variables.

In short, market turnover is autocorrelated and is affected by past market returns, both negative and positive. This is against the expectations, because it is not in line with the overconfidence hypothesis. Next, volatility does not affect market turnover. The VAR model and other supporting tests show that market turnover do not affect market return. To conclude, the VAR model results do not confirm the hypothesis of this research. Positive past market returns lead to an increase and a decrease of market turnover. That being the case, impulse response functions could give more insight in the relation between market return and market turnover. These functions are more specific about the influence of market return on market turnover. The results of the impulse response functions along with the VAR model could give a more elaborated conclusion.

3.3 Results impulse response function

(27)

(28)

month for market turnover. Besides the fact that it is statistically significant, the influence is also economically significant. As the influence is almost 20%, the turnover in the previous month is a good indication for the following level of turnover. Table 5 shows that this is found to be significant. Moreover, the influence of a shock remains of influence for multiple periods of time. The impact decreases, but always stays above 5%. The impact of market turnover on market turnover stays positive for a full year. The Wald-test result shows that this total influence is significant. It can be stated that market turnover is autocorrelated, and that past market turnover is very important in determining the level of market turnover in the following months.

(29)

-.01 .00 .01 .02 .03 .04 .05 .06 1 2 3 4 5 6 7 8 9 10 11 12

Response of MRET to MRET

-.01 .00 .01 .02 .03 .04 .05 .06 1 2 3 4 5 6 7 8 9 10 11 12

Response of MRET to LMTURN

-.10 -.05 .00 .05 .10 .15 .20 1 2 3 4 5 6 7 8 9 10 11 12

Response of LMTURN to MRET

-.10 -.05 .00 .05 .10 .15 .20 1 2 3 4 5 6 7 8 9 10 11 12

Response of LMTURN to LMTURN

Response to Cholesky One S.D. Innovations Figure 4: Impulse response function

The x-axis represents 12 periods. Each period is a month, so it represents a full year. The y-axis is the response of the dependent variable in percentages.

3.2 Detrended VAR model and impulse response functions

As mentioned earlier, the long-term downwards trend in market turnover could bias the analysis, because the time-series is not stationary. With detrending log market turnover, the steering force that time may have on each time-series is removed. In other words, the long-term effects are ruled out. Simultaneously, detrending should create stationarity. This approach should result in slightly different coefficients, as the intercept and trend are simultaneously fitted together in a multivariate model. To check the validity of the first VAR model, I will describe the most important differences and similarities with this VAR model results. Again I conduct several diagnostic tests to determine the validity of this VAR model. First of all, the lag length criteria tests, together with serial correlation tests to verify this lag length, determine that this model has to have twelve lags. Both the LR and AIC suggest twelve lags for this model. The serial correlation test with two lags again shows serial correlation with highly significant p-values. The test with twelve lags shows only p-values

Part A Part B

(30)

(31)

-.01 .00 .01 .02 .03 .04 .05 1 2 3 4 5 6 7 8 9 10 11 12

Response of MRET to D(LMTURN)

-.10 -.05 .00 .05 .10 .15 .20 1 2 3 4 5 6 7 8 9 10 11 12

Response of D(LMTURN) to MRET

-.10 -.05 .00 .05 .10 .15 .20 1 2 3 4 5 6 7 8 9 10 11 12

Response of D(LMTURN) to D(LMTURN)

Response to Cholesky One S.D. Innovations

Figure 5: Impulse response function

Part A Part B

Part C Part D

Impulse response functions, where logged market turnover is detrended. The results are similar as the basic VAR model and impulse response function.

3.3 VAR models and impulse response functions without crisis periods

(32)

(33)

4. Conclusion

The purpose of this paper is to examine the presence of overconfidence in financial markets. According to previous literature, trading volume is a strong indicator for overconfidence. Following Statman et al. (2006), to confirm the presence of overconfidence, the lead-lag relation between market return and market turnover is researched with a VAR model. Market turnover represents trading volume.

The first key result shows that positive past market return indeed affects market turnover and thus trading volume in the following months. This relation is highly significant. The lead-lag relationship between return and turnover is strong and therefore could be used for further research. However, in contrast to previous research, my results show that positive past market returns lead to negative and positive market turnover. When in fact, to confirm overconfidence, it should only lead to positive market turnover. This finding is not consistent with the finding of Statman et al. (2006). They found that market return substantially increases trading volume in the U.S. market. As mentioned before, the results cannot confirm the presence of overconfidence on the Dutch stock market. Moreover, I found that the crisis periods, Internet crisis and financial crisis, do not significantly affect the results. Next, the second key finding is that market turnover is autocorrelated. This is consistent with the expectations and the findings of Statman et al. (2006).

(34)

(35)

5. References

Aspara, J., and Hoffmann, A. O., 2013. Selling losers and keeping winners: How (savings) goal dynamics predict a reversal of the disposition effect. Marketing Letters, 26(2), 201-211.

Barber, B. M., and Odean, T., 2001. Boys will be boys: Gender, overconfidence, and common stock investment. Quarterly Journal of Economics, 261-292.

Brown, R., and Sarma, N., 2007. CEO overconfidence, CEO dominance and corporate acquisitions. Journal of Economics and Business, 59(5), 358-379.

Campbell, J. L., 2007. Why would corporations behave in socially responsible ways? An institutional theory of corporate social responsibility. Academy of Management Review, 32(3), 946-967.

Coval, J. D., and Moskowitz, T. J., 1999. Home bias at home: Local equity preference in domestic portfolios. The Journal of Finance, 54(6), 2045-2073.

Daniel, K., Hirshleifer, D., and Subrahmanyam, A., 1998. Investor psychology and security market under-‐and overreactions. The Journal of Finance, 53(6), 1839-1885.

De Bondt, W. F. M, and Thaler, R. H., 1994. Financial decision-making in markets and firms: A behavioral perspective, in R. A. Jarrow, V. Maksimovic, and W. T. Ziemba: Finance, Handbooks in Operations Research and Management Science, 9(13), 385-410.

Fama, E. F., 1970. Efficient Capital Markets: A review of theory and empirical work. Journal of Finance, 25(2), 383-417.

Fan, J., and Wang, Y., 2007. Multi-scale jump and volatility analysis for high-frequency financial data. Journal of the American Statistical Association, 102(480), 1349-1362. Gervais, S., and Odean, T., 2001. Learning to be overconfident. Review of Financial studies,

14(1), 1-27.

Glaser, M., and Weber, M., 2009. Which past returns affect trading volume? Journal of Financial Markets, 12(1), 1-31.

Goyal, A., and Santa-Clara, P., 2003. Idiosyncratic risk matters! The Journal of Finance, 58(3), 975-1007.

Grinblatt, M., and Keloharju, M., 2009. Sensation seeking, overconfidence, and trading activity. The Journal of Finance, 64(2), 549-578.

(36)

Heath, C., and Tversky, A., 1991. Preference and belief: Ambiguity and competence in choice under uncertainty. Journal of Risk and Uncertainty, 4(1), 5-28.

Hodrick, R. J., and Prescott, E. C., 1997. Postwar US business cycles: an empirical investigation. Journal of Money, Credit, and Banking, 1-16.

Kahneman, D., Slovic, P., and Tversky, A., 1974. Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.

Kahneman, D., and Tversky, A., 1996. On the reality of cognitive illusions. Psychological Review, 103, 582-591.

Karpoff, J. M., 1987. The relation between price changes and trading volume: A survey. Journal of Financial and Quantitative Analysis, 22(01), 109-126.

Kaustia, M., 2010. Prospect theory and the disposition effect. Journal of Financial and Quantitative Analysis, 45(3), 791.

Keynes, J. M., 1930. A treatise on money. London: Macmillan.

Kyle, A. S., 1985. Continuous auctions and insider trading. Journal of the Econometric Society, 1315-1335.

Lo, A. W., and Wang, J., 2000. Trading volume: definitions, data analysis, and implications of portfolio theory. Review of Financial Studies, 13(2), 257-300.

Lui, W., Strong, N., and Xu, X., 1999. The profitability of momentum investing. Journal of Business Finance & Accounting, 26(9-‐10), 1043-1091.

Malkiel, B. G., 2003. The efficient market hypothesis and its critics. The Journal of Economic Perspectives, 17(1), 59-82.

Odean, T., 1998. Do investors trade too much? American Economic Review, 89, 1279-1298. Odean, T., 1998a. Are investors reluctant to realize their losses? The Journal of Finance,

53(5), 1775-1798.

Odean, T., 1998b. Volume, volatility, price, and profit when all traders are above average. The Journal of Finance, 53(6), 1887-1934.

Pagano, M., 1989. Trading volume and asset liquidity. The Quarterly Journal of Economics, 255-274.

Statman, M., Thorley, S., and Vorkink, K., 2006. Investor overconfidence and trading volume. Review of Financial Studies, 19(4), 1531-1565.

(37)

Tourani-‐Rad, A., and Kirkby, S., 2005. Investigation of investors' overconfidence, familiarity and socialization. Accounting & Finance, 45(2), 283-300.

Tversky, A., and Kahneman, D., 1973. Availability: A heuristic for judging frequency and probability. Cognitive Psychology, 5(2), 207-232.

(38)

Appendix A: AEX (Company/Stock) Figure A1: AEX

Companies excluded from analysis, because of changing AEX requirements: Altice, Delta Lloyd, Gemalto, NN Group, OCI and TNT Express.

Company Stock

Aalberts Industries AALB

Aegoon AGN

Ahold Koninklijke AH

Akzo Nobel AKZA

Arcelor Mittal MT ASML Holding ASML Boskalis Westminster BOKA DSM Koninklijke DL

Heineken HEIA

ING Groep INGA

KPN Koninklijke KPN Philips Koninklijke PHIA

Randstad RAND

RELX REN

(39)

-.10 -.05 .00 .05 .10 .15 .20 1 2 3 4 5 6 7 8 9 10 11 12

-.02 -.01 .00 .01 .02 .03 .04 .05 1 2 3 4 5 6 7 8 9 10 11 12

Response to Cholesky One S.D. Innovations Appendix B: Results

Figure B1: Market impulse response function (Reversed Cholesky ordering)

Part A Part B

Part C Part D

Part B shows the effect of Cholesky ordering. The response of market turnover is forced to start at zero, because the ordering (mturn – mret) rules out the effect of mret on mturn. Lmturn stands for log market turnover.

Table B1: Granger-causality test - detrended model

Dmturn stands for detrended log market turnover.

(40)

Table B2: VAR model (detrended log turnover)

The remaining of Table B2 is on the next page.

(41)

***,**,* indicate significance at the 1, 5, and 10 percent levels, respectively (two-tailed). Dmturn stands for detrended logged market turnover.

(42)

Appendix C: Results without crisis periods

Figure C1: Impulse response function without financial crisis

Figure C2: Impulse response function without both crisis periods

-.02 -.01 .00 .01 .02 .03 .04 1 2 3 4 5 6 7 8 9 10 11 12

-.05 .00 .05 .10 .15 .20 1 2 3 4 5 6 7 8 9 10 11 12

Response to Cholesky One S.D. Innovations

-.01 .00 .01 .02 .03 .04 1 2 3 4 5 6 7 8 9 10 11 12

-.04 .00 .04 .08 .12 .16 1 2 3 4 5 6 7 8 9 10 11 12

(43)

Table C1: Granger causality test - Without financial crisis

Table C2: Granger causality test - Without both crisis periods

MRET does not Granger Cause MTURN 145 3.1263 0.0006 MTURN does not Granger Cause MRET 0.6406 0.8038