The effect of Gender, Age and Education on Transaction Frequency
Sample on individual investors from a Dutch discount broker
MSc Finance – Master Thesis
June 2017 Michael Steenkist 11418206 Supervisor dr. Liang Zou Abstract
This study investigated the effect of gender, age and education on transaction frequency of individual investors. From the existing literature it is known that men are more overconfident than women and that overconfidence increases with age and the level of education. Another finding from the existing literature is that more overconfident investors do more transactions. Based on these findings, three hypotheses were created to test whether the investor characteristics gender, age and education affect transaction frequency due to overconfidence. These relationships were tested for the transaction frequency in all asset categories and specifications for stocks and ETFs, options, and futures. The results show that men do more transactions than women, but only in stocks and ETFs and not for the other categories. Age is positively related to the transaction frequency. The level of education is negatively related to the amount of transactions in all asset classes and options, which is the opposite of what was expected.
Statement of originality
This document is written by Student Michael Steenkist who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements
First, I would like to thank the Dutch broker that allowed me to use their data for research purposes. The data contained confidential information on their clients and the broker itself and has always been treated with ultimate care and warranted anonymity of the clients. This unique dataset made it possible to do research on individual investors and to write this thesis. Second, I would like to thank my supervisor dr. Zou for the support and freedom he has given me. From the first moment he gave me the freedom to write a thesis on what I found interesting and supported that throughout the process.
Table of contents
I. Introduction 4
II. Literature Review 6
A. Individual investors 6
B. Overconfidence and competence 8
C. Brokers 10
D. Hypotheses 10
III. Data and methodology 11
A. Data 11
B. Methodology 17
IV. Results 19
A. All asset categories 19
B. Stocks and ETFs 22
C. Options 24 D. Futures 25 V. Conclusion 26 VI. Discussion 28 References 30 Appendices 32
I. Introduction
Previous literature has shown that investing and emotions are related to each other (Hirshleifer & Shumway, 2003; Kahneman & Tversky, 1979), indicating that people are less rational creatures than some economic theories assume, as for example the efficient market hypothesis (Fama, 1970). Barber and Odean (2001) show that male investors do significantly more trades than female investors, and that their returns are therefore reduced more due to the incurred transaction costs. They argue that overconfidence is the reason why men trade more than women. This thesis elaborates on how gender affects investor behaviour and specifically on the transaction frequency. Besides gender, this thesis will test how the characteristics age and education are related to investor behaviour. By using a unique dataset from a discount broker in the Netherlands on individual investors, it is possible to analyse these relationships. The aim of this study is to gain more insight in how investor characteristics affect investment decisions and thereby contribute to the existing literature in behavioural finance.
Behavioural finance can be considered as a mix between psychology and finance. Literature on this topic is relatively new compared to the original and well known literature in finance. Kahneman and Tversky (1979, 1992) are considered as two of the first to do research in the field of behavioural finance by developing the prospect theory. Kahneman and Tversky (1979) proposed different problem sets to students and faculty staff, who had to choose between two options. For instance, they could choose to win $4,000 with an 80% probability, but with a 20% probability of winning $0, or they could take $3,000 with a 100% certainty (Kahneman and Tversky, 1979, Problem 3). The expected value of the first option is 80% of $4,000, which is $3,200. The expected value of the second option is $3,000. However, 80% of the respondents chose for the second option, to take the $3,000 for sure. They argue that based on the expected utility theory people want to maximize their expected utility. It therefore suggests that people are risk averse or that their utility functions are concave, because the first option has a higher expected value than the second. In the opposite problem set, the respondents could choose between losing $4,000 with an 80% probability and a 20% probability of losing $0, or to lose $3,000 for sure (Kahneman and Tversky, 1979, Problem 3’). The expected value of the first option is negative $3,200, which is a $200 higher loss compared to the expected value of the second option. However, 92% of the respondents chose for option one, where they had a small probability of not losing any money. Based on multiple of these problem sets, they argue that the value function of people has a S-shape and is concave for gains, convex for losses and steeper for losses than for gains (Kahneman and Tversky, 1979, figure 3). This means that the displeasure that people experience when they lose money is larger than the pleasure of winning the same amount; this is called loss aversion. People are willing to take more risk in case they are in the loss region, as was shown in Problem 3’. The prospect theory undermines the expected utility theory by showing that people do not maximize their expected utility in every situation.
Since the paper by Kahneman and Tversky (1979), the amount of research on behavioural finance has increased. More existing research in psychology is linked to financial phenomena to
explain investor behaviour and the irrationality of financial decisions. An example is the paper by Hirshleifer and Shumway (2003), who look at the relationship between sunshine and market returns across 26 countries from 1982 to 1997. They argue that psychological research has predicted that sunny weather is related to positive moods. Surprisingly, they find that sunny weather is significantly correlated with positive stock market returns. The idea that the weather can have an impact on the stock market contradicts the idea that investors make rational decisions and always maximize their utility. The relevance of the content of this last article is rather low for this thesis. It does however indicate that the financial markets, and inherently the people who invest on the market, are not always rational. This shows the importance of research in the field of behavioural finance, which is the aim of this thesis.
By looking at data on customers from a Dutch discount broker it is possible to see how the different investor characteristics gender, age and education affect investor behaviour. This dataset contains information on all transactions these customers have done in 2016. This thesis differs from the paper by Barber and Odean (2001) in the sense that a lot has changed in twenty years (their data set is from 1991 – 1997) regarding investment products, transactions costs and the ease of doing a transaction. The Dutch discount broker offers an online investment account, which makes it possible to trade all over the world in multiple asset categories within a few mouse clicks. Also, more investor characteristics, such as age and education will be used to see how these affect investor behaviour. Besides that, Barber and Odean (2001) only look at stock transactions, whereas this thesis also looks a broader range of asset categories. Although they use a larger time period and more trading accounts, the dataset for this thesis contains more transactions, meaning more observations. The dataset is very unique in the sense that it is not publicly available and there are not many other datasets on individual investors containing such detailed information. This makes this thesis a contribution to the existing literature in behavioural finance and especially in the area of individual investors.
In section II, the existing literature on individual investors, overconfidence, competence, and discount brokers is discussed. Section III elaborates on the dataset and the methodology that is used to test the hypotheses. In section IV, the regression results are shown and discussed. Section V contains the conclusion and in section VI the limitations of this thesis and notices for further research are discussed.
II. Literature Review
A. Individual investors
As discussed in the introduction, a dataset on individual investors of a Dutch discount broker makes it possible to do this research. Brad Barber and Terrance Odean are one of the few people who also had access to data from a discount broker to do research on the behaviour of individual investors. They have used a dataset from a large discount broker in the United States for multiple of their researches that have been published in A-journals. Five of their articles are discussed here, because of the relevance for this thesis and to better understand individual investors.
Shefrin and Statman (1985) named the tendency of investors to sell winners too early and hold losers too long the disposition effect. In Odean (1998), the disposition effect is tested using the data from the US discount broker. He describes this effect as an implication of extending the prospect theory by Kahneman and Tversky (1979). The disposition effect is based on the aversion of investors to realize their losses (loss aversion) that was discussed earlier. Odean (1998) uses all transactions of 10,000 customers during January 1987 through December 1993, which accumulates to a total of 162,948 records. By looking at the proportion of gains and losses realized, he shows that the disposition effect holds and individual investors significantly prefer to sell winners and hold on to losers. There is however an exception for December, due to the tax-motivated selling of losing positions in this month.
Odean (1999) describes that in psychological research it is shown that people tend to be overconfident about their abilities and skills. He uses the same dataset as Odean (1998) to test the hypothesis that investors trade too much due to their overconfidence. He argues that overconfident investors tend to overestimate their expected returns and belief their profits will be high enough to cover the transaction costs. Therefore overconfident investors do transactions that actually have a negative expected return, because they overestimate their ability and skill to realize high-expected profits. First, he finds that the investors tend buy stocks that have shown higher or lower returns in the last six months than the stocks they sell. Second, the results show that the gain from buying securities does not outperform the securities they sell sufficiently to cover the transaction costs. Lastly, he shows that the stocks these investors buy subsequently underperform compared to the stocks they sell. This means the investors would be better off to do exactly the opposite transactions.
Barber and Odean (2000) also show that investors that trade actively underperform compared to the market and to those who trade less frequently. For their research they used data from the large US discount broker, but now with data from 1991 through 1996. They show that the investors that are in the highest quantile of active traders have a stock turnover of more than twice the size of their portfolio. This while the average investor has a stock turnover of 75% of the size of their portfolio. The most active investors underperform the market with 5.5% annually and the average investors with 1.1%. However, when they account for the fact that these investors trade small, low value stocks with
high risk, the underperformance increases to 10.3% and 3.7% respectively. Their main conclusion is that doing transactions is indeed hazardous for your wealth. The argument they provide for why investors decide to trade more actively is similar to Odean (1999), which is that investors trade too much due to overconfidence.
Individual investors differ in their investment behaviour from institutional investors. Barber and Odean (2008) look at the effect of attention and news on the behaviour of individual and institutional investors. They argue that individuals have less time and possibilities to incorporate all available information in making investment decisions, also referred to as inattention. Therefore individual investors tend to buy stocks that are in the news, have a very high trading volume or have shown excessive positive or negative returns. These stocks are also referred to as attention-grabbing stocks. In their paper, the first hypothesis they test is that the buying behaviour of individual investors is affected more by attention grabbing stocks than their selling behaviour. This is because individual investors tend to sell stocks they already own, and are reluctant to short-sell. Their results show that individual investors also buy twice as much as they sell of a specific stock that experienced excessive trading volume. In case a stock experiences an extremely low return on the day before, individual investors tend to buy almost twice as much of these stocks as when they sell. The second hypothesis is that these attention-grabbing stocks less affect the investment decisions of institutional investors. Barber and Odean (2008) show that this is indeed the case and explain that this is probably caused by the fact that professional investors have more time and resources to incorporate all available information.
The most cited paper is Barber and Odean (2001) and is very relevant for this thesis. In this paper they do research on the difference between the investment behaviour and returns of men and women. Based on psychological research they argue that men are more overconfident than women. Due to this surplus in overconfidence, men are inclined to trade more excessively than women. As discussed in Barber and Odean (2000), investors that trade more actively underperform compared to investors that trade less. Therefore they tested the hypotheses that men do more stock trades than women and that the returns of men are affected more due to this excessive trading. Their results show that men did 45% more stock trades, which resulted in a reduction of the returns because of the transaction costs. For men the returns were 2.65% reduced per year, while this was 1.72% for women. These results are the strongest when they look at the difference between single men and single women. The finding that men trade more than women is also tested in this thesis. However, Barber and Odean (2001) only test this for stocks, while this thesis also focuses on a broader range of asset categories. Furthermore, this thesis tests the relationship of age and level of education with transaction frequency.
B. Overconfidence and competence
Overconfidence is a returning topic in the behavioural finance literature and is an important factor in this thesis. Therefore more of the existing literature on overconfidence and the link with investor characteristics are discussed in this section. However, the definition of overconfidence is not homogeneous throughout the existing literature. Menkhoff, Schmeling and Schmidt (2013) discuss three different definitions, based on prior literature. First, overconfidence can be related to the term ‘confidence interval’ in experimental research. Research participants usually have to provide an upper and lower bound of a certain variable. When this confidence interval is too tight, this can be seen as an underestimation of the variance of the variable. Therefore overconfidence can be seen as the underestimation of the variance of a specific variable. Second, they state that overconfidence can be unrealistic positive self-evaluation. Third, overconfidence can be seen as illusion of control and optimism, which leads to overestimation of the mean of a certain variable, such as the probability of a successful outcome. In this thesis, existing literature on the relationship between overconfidence and gender, age and education will be used to form hypotheses for the effect of these characteristics on transaction frequency. The three definitions of overconfidence by Menkhoff et al., (2013) are used in discussing the effect of overconfidence on investor behaviour and how it is related to the characteristics. The findings of existing literature on these relationships are discussed here. Besides overconfidence, perceived competence is also discussed in the behavioural finance literature. Competence can be interpreted as the skill, knowledge and understanding a person has in a certain situation. Overconfidence and perceived competence are not the same, but can be related to each other. To further elaborate on these topics, articles on overconfidence and competence are discussed subsequently.
As discussed earlier, Odean (1999), and Barber and Odean (2000, 2001) argue that overconfidence affects the investment behaviour of individual investors. These papers show that overconfident investors do more transactions and that this results in lower returns. Furthermore, it is shown that men are more overconfident than women and therefore trade more, which has a negative effect on their returns (Barber & Odean, 2001).
Heath and Tversky (1991) show that people prefer to bet on their own judgments and knowledge in case they feel skilful or knowledgeable regarding a certain topic. Their research consisted of multiple experiments with questions in diverse areas, such as history, geography or sports. First, people were asked to state their amount of confidence in providing the correct answer. Second, they could choose to submit their own answer or to submit a randomized answer with a probability to be the correct answer that was equal to the confidence they stated. The results show that people tend to choose their own answer in case they are confident about their knowledge of a certain question. However, when people do not feel knowledgeable about the topic, they prefer to choose the random answer with equal probability. Based on these results Heath and Tversky (1991) came up with the
competence effect. This is the effect that people prefer to rely on their own perceived knowledge and skill.
Graham, Harvey and Huang (2009) link the findings of Heath and Tversky (1991) to investor behaviour. They investigate whether the competence effect affects trading frequency and the tendency of investors to invest too much in their home market, also known as home bias. Besides that, they look at the relation between investor characteristics (age, gender, education, income and portfolio size) and the level of competence. With use of the UBS/Gallop investor survey, they are able to gather data on trading frequency, home bias, investor competence, overconfidence and optimism towards the U.S. stock market (Graham et al., 2009, Table 1). Firstly, their results show that investors that feel competent do more transactions and invest more in international markets. Second, they find that ‘male investors, and investors with larger portfolios or more education, are more likely to perceive themselves as competent than are female investors, and investors with smaller portfolios or less education’ (Graham et al., 2009; p. 1). Third, they do not find any significant relationship between age and perceived competence. In contrast to Barber and Odean (2001), they state that competence is not the same as overconfidence. However, they argue that overconfidence makes investors see themselves as more competent. Therefore investors that are overconfident about their abilities, knowledge and expected utility have higher perceived competence. Graham et al. (2009) and Barber and Odean (2001) both find that male investors trade more frequently than female investors, and explain that this is because men are more overconfident and perceive themselves as more competent. In addition, Graham et al. (2009) show that investors with larger portfolios or higher education also perceive themselves as more competent and therefore tend to trade more. The main difference between the research by Graham et al. (2009) and this thesis is the type of data that is used. They used survey data, while for this thesis data from a Dutch retail broker on real transactions is used. Therefore, it is interesting to see if the results are in line with each other.
Although Graham et al. (2009) do not find a significant relation between age and competence; there are psychological papers that did find a significant relation between age and overconfidence. Hansson, Rönnlund, Juslin and Nilsson (2008) examine the age difference in probability judgment and intuitive confidence intervals. They argue that prior research has found contradicting relationships between age and overconfidence. For instance Forbes (2005) and Pliske and Mutter (1996) have found a negative relationship, while Crawford and Stankov (1996) found a positive relationship. Hansson et al. (2008) use the ‘naïve sampling model’ and find a positive relationship between age and overconfidence. Because of these contradicting findings it is not clear what the relationship between age and overconfidence is. As Hansson et al. (2008) used multiple of the articles on age and overconfidence and found a positive relationship; it is tested in this thesis if age has a positive relation with transaction frequency. Odean (1999), Barber and Odean (2000, 2001), and Graham et al. (2009) have showed that overconfidence is related to excessive trading. A positive relationship between age and transaction frequency can therefore suggest that older people are more overconfident.
Menkhoff et al. (2013) do research on overconfidence of institutional investors, investment advisors and individual investors. They also argue that older investors are more miscalibrated, indicating they are more overconfident. However, they also state that experience is an important factor and that age is not a good measure for experience. Experience even seems to have the opposite relationship with overconfidence, meaning that more experienced investors are less overconfident. Customers of the Dutch broker have to state their trading experience when they open a trading account. This makes it possible to incorporate trading experience in this thesis to account for the findings by Menkhoff et al. (2013).
C. Brokers
Fong, Gallagher and Lee (2014) argue that it is important to consider which type of broker is used for research on individual investors. They look at the difference in informativeness of trades of investors at discount brokers and service brokers. First, the results show that trades from full-service brokers are more informative compared to those of retail brokers. Second, they have found that past returns, news and volatility have a positive relation with the volume of discount retail brokers. This last finding is in line with the results of Barber and Odean (2008), which is that individual investors tend to be net-buyers of attention-grabbing stocks. However, they state that these factors are not related to the net volume of full-service retail brokers. Therefore Fong et al. (2014) argue that it is important to consider which type of broker is used. They recommend being cautious with generalizing the results of a specific study for all individual traders, because the broker type can be of influence. Based on their broker classification criteria, the broker that is used for this thesis can be classified as a discount retail broker.
D. Hypotheses
Based on the literature review, it is possible to state the hypotheses that are going to be tested in this thesis. The hypotheses will be tested for all asset classes, stocks and ETFs, options and futures. These specifications are made based on the statistics on the dataset from the Dutch broker and will be explained in section III of this thesis. First, we know that men are more overconfident than women; that this is also true for investors and that overconfidence leads to excessive trading (Odean, 1999; Barber and Odean, 2000, 2001). Therefore the first hypothesis is that male investors have a higher transaction frequency than female investors.
H1: Male individual investors have a higher transaction frequency than female investors.
Second, we know that age has a positive relation with overconfidence (Hansson et al., 2008; Menkhoff et al., 2013). Therefore the second hypothesis is that older investors have a higher
transaction frequency because they are more overconfident. However, there is also prior literature that argues that age has no significant (Graham et al., 2009) or even a negative relationship (Forbes, 2005; Pliske and Mutter, 1996) with overconfidence. Using the unique data from the Dutch discount broker, it is possible to test whether this relationship of age and transaction frequency is indeed significantly positive. This would suggest that older investors are indeed more overconfident. Trading experience is not the same as age and even has a negative relationship with overconfidence (Menkhoff et al., 2013). Therefore trading experience is included in the models to test the hypotheses and control for this effect.
H2: Age is positively related to the transaction frequency.
Lastly, it is known that people with more education are more overconfident or perceive themselves as more competent (Graham et al., 2009). Therefore the third hypothesis is that people with a higher level of education trade more frequently due to this overconfidence. Most of the investors from the dataset have Dutch education. Appendix 1 gives an overview of the levels of education that are used in this thesis.
H3: The level of education is positively related to the transaction frequency.
III. Data and methodology
A. Data
The data used for this thesis is from a discount broker in the Netherlands and consists of two datasets from 2016. The first dataset contains information on the transactions of all customers and the second includes information on investor characteristics. These two datasets have been merged based on account number, to create the main dataset. This main dataset contains information on 8,249 unique trading accounts and a total of 1,270,783 transactions, distributed over multiple asset classes. However, there have been multiple adjustments to this data set, to be able to test the hypotheses. These adjustments are discussed first.
First, the main dataset includes transactions and information on individual and business accounts. The business accounts have been omitted from the dataset, because this thesis does research on individual investors only. This drop accounted for 37,927 transactions and 303 unique trading accounts. Second, observations with a trading experience above 100 years have been dropped, because this is not very realistic. The employees had to fill in the customer information by hand until 2011, which could have led to mistakes. This drop led to a decrease of 43 unique accounts and 6,615 transactions. Third, there were some accounts with a negative net asset value (NAV). Based on the information from the broker, it is known that this are often ‘dead’ accounts of people that are closing
their account. Due to a small interest payment it can happen that the NAV becomes negative. Therefore these accounts have been omitted from the dataset, which accounted for 13 unique accounts and 536 transactions. Fourth, observations have been dropped if the age was lower than 18 or higher than 100. By law it is not allowed for the broker to open a trading account for a minor (under 18 years old). Besides that, the previous Customer Relationship Management (CRM) system of the broker did not correctly register all birthdates. Therefore some clients are registered at a default birthdate, which is the 1st of January 1900, which makes them 117 years old. Therefore all observations with an age below 18 or above 100 have been dropped, leading to a decline of 431,456 transactions and 808 unique trading accounts in the dataset. Fifth, based on the same reason as the age registration, not all genders have been registered correctly. Therefore, the observations with an unknown gender have been dropped. This has resulted in a decrease of 3,219 transactions and 33 unique trading accounts. The dataset that will be used to test the hypotheses includes a total of 791,030 transactions distributed over thirteen asset classes and 7,049 unique trading accounts that did at least one transaction in 2016.
Table 1
Investor Characteristics
This table contains statistics on characteristics of the individual investors that have an account with the Dutch broker and did at least one transaction in the year 2016. In Panel 1a, the statistics based on the unique trading accounts are shown. In Panel 1b, the statistics are based on all transactions that were made in the year 2016. ‘Gender’ can only be defined as male or female. ‘Age’ has been split up in five ranges, where the minimum age is 18 years old and maximum age 100 year old. ‘Education’ has been split up into two categories: university or lower than university. ‘Net Asset Value’ shows the total value of the investor portfolio in Euro, including positions in all asset classes and cash, at the beginning of 2016. ‘Trading Experience’ shows the amount of trading experience the investor had before opening an account at the broker and is measured in years. ‘N’ shows the total number of observations.
a. Unique trading accounts b. All transactions
Count (%) Mean Median Count (%) Mean Median
Gender 0.92 1.00 0.93 1.00 Male 6460 (91.64) 734176 (92.81) Female 589 (8.36) 56854 (7.19) Age 49.30 49.02 52.22 52.34 < 30 670 (9.50) 30376 (3.84) 30 – 40 1329 (18.85) 137123 (17.33) 40 – 50 1703 (24.16) 177029 (22.38) 50 – 60 1643 (23.31) 215479 (27.24) 60 > 1704 (24.17) 231023 (29.21) Education University 4654 (66.02) 466154 (58.93) < University 2395 (33.98) 324876 (41.07)
Net Asset Value (in EUR) 39836.88 6627.08 107892.04 22486.25
Trading Experience (in years) 10.01 7.00 11.76 10.00
A summary of the investor characteristics is shown in table 1. In panel 1a the statistics of the unique trading accounts are presented. First of all, men are in the majority with more than 91.64% of the trading accounts. This means that men hold 6,460 of the unique trading accounts. Women are in the minority with 8.36% of the trading accounts and hold 589 of the trading accounts. In panel 1b the statistics for all the transactions are shown. Men did 92.81% of the transactions, which is more than the percentage of unique trading accounts that are held by men. Vice versa, women account for 7.19% of the transactions, which is lower than the percentage of unique trading accounts that are held by women. This already implies that men do more transactions than women, which supports H1. However, it is not yet possibly to conclude that this difference is significant.
Second, the average (median) age of an investor is 49.30 (49.02) years old, based on the unique trading accounts. This is lower than the average age based on all the transactions of 52.22 (52.34) years old. The youngest group of investors (age between 18 and 30) accounts for 9.50% of the trading accounts, while they only account for 3.84% of all transactions. The second youngest group (age between 30 and 40) accounts for 18.86% of the unique trading accounts and 17.33% of the transactions. The third group (age between 40 and 50) also has a higher percentage of unique trading accounts than transactions, respectively 24.16% and 22.38%. The fourth group (age between 50 and 60) accounts for 23.30% of the trading accounts, but does 27.26% of all the transactions. Also the oldest group (age between 60 and 100) has a lower percentage of unique trading accounts than transactions, respectively 24.18% and 29.20%. These statistics support H2; that older investors do more transactions compared to younger investors. Again it is not possible to conclude that this difference is significant and has to be tested.
Thirdly, the statistics on education show that the majority of investors have university as the highest level of education; 66.02% of the unique trading accounts. Based on the statistics on all transactions, 58.93% of the transactions were made by investors that went to university. The investors with a lower level of education account for 33.98% of the unique trading accounts and 41.07% of the transactions. This suggests that higher educated investors do less transaction than lower educated investors, which is the opposite of what is expected. These statistics are not in support of H3; that higher educated people are more overconfident and therefore do more transactions.
Lastly, the control variables NAV and trading experience are shown. The average (median) NAV is 39,837 (6,627) Euro based on the unique trading accounts. The average NAV based on all transactions is 107,892 (22,486) Euro. This indicates that investors with a higher NAV do more transactions. The average (median) trading experience is 10.01 (7.00) years for the unique trading accounts. For all transactions the average is 11.76 (10.00) years. This suggests that investors with more trading experience, do more transactions.
The distribution of all transactions among asset categories and currency are shown in Appendix 2. The largest asset class is options with 39.88% of the transactions. The second and third
largest are stocks and ETFs with 26.67% and futures with 20.86%. These three asset classes account for 87.41% of all the transactions. To test the relation between the investor characteristics and transaction frequency, all asset categories will be used and specifications are made for these three largest categories. The testing of the hypotheses for the specified asset classes also serves as a robustness check. Statistics on the currencies in which the transactions are made are specified in panel 2b of Appendix 2. As this is a Dutch broker, the majority of the transactions are made in Euro; which is 56.96%. The second largest currency is the US Dollar with 39.90% of all transactions. Combined, these two currencies account for 96.86% of the transactions.
The main dependent variable for this thesis is the transaction frequency. Transaction frequency is seen as the amount of transactions an investor does within the sample period and is not the same as a trade. A trade consists of at least two transactions, an opening and a closing transaction. However, an investor could decide to increase or decrease an existing position. Therefore a trade can consist of more than two transactions. For the purpose of this thesis, the transaction frequency for an individual investor is used. The statistics of the unwinsorized transaction frequency for the different investor characteristics are shown in Appendix 3. All of the 7,049 unique trading accounts and thirteen asset classes are used for these statistics. The mean is much higher compared to the median for all statistics, which suggests that there are some outliers that increase the averages. In Appendix 4, two histograms on the transaction frequency are shown. Histogram 4a visualizes the distribution of the transaction frequency and shows the outliers in the dataset. In histogram 4b, the transaction frequency is winsorized and trimmed at the 99th percentile. By comparing the range of the x-axis, it is clear that the outliers have been removed from the dataset. The winsorized data is used to test the hypotheses, because using the original dataset will lead to biased results due to the outliers. The statistics of the winsorized transaction frequency for the different investor characteristics are shown in table 2 and are now discussed.
First of all, the average amount of transactions an individual investor has done in 2016 is 80.81, while the median is only 14. This implies that there are still investors with a very high transaction frequency, which the maximum of 1,565 transactions per year also indicates. However, when this is compared to the statistics from Appendix 3, it is clear that winsorizing already accounted for the extreme outliers. For the unwinsorized dataset, the median was also 14, but the mean was 112.22 and the maximum 17,614. Second, the winsorized statistics show that men have done 81.57 transactions on average, with a median of 14. Both the average and median are higher compared to the average amount of transaction of women. Women show an average amount of 72.47 transactions, with a median of 10. This finding is in support of H1, that men do more transactions than women; which was also shown in table 1. Thirdly, the average as well as the median amount of transactions increases over age. This is in support of H2, that older investors do more transactions compared to younger investors. Lastly, an investor with university as highest level of education does 72.94 transactions on average and the median is 13. The investors with a lower education did 96.20 transactions on average
and the median is 16. This suggests that lower educated investors do more transactions than higher educated investors, which is in line with the statistics of table 1 and is the opposite of what H3 argues.
Table 2
Transaction frequency
This table contains statistics on the transaction frequency of the individual investors for all the asset classes. One transaction is equal to a frequency of one. ‘All investors’ includes all the individual investors independently of the characteristics gender, age and education. The definition of the characteristics are similar to table 1. The transaction frequency is winsorized and trimmed at the 99th percentile to account for outliers.
Count Minimum
Lower
Quartile Median
Upper
Quartile Maximum Mean
Standard Deviation All investors 6979 1 4 14 63 1565 80.81 179.75 Gender Male 6394 1 4 14 64 1565 81.57 180.72 Female 585 1 3 10 45 1210 72.47 168.66 Age < 30 668 1 3 8 29 1161 44.12 120.25 30 – 40 1318 1 3 10.5 38 1537 57.28 149.05 40 – 50 1685 1 4 15 68 1424 84.05 181.74 50 – 60 1625 1 4 17 74 1565 92.15 197.86 60 > 1683 1 4 20 94 1565 99.59 196.42 Education University 4618 1 4 13 56 1490 72.94 165.51 < University 2361 1 4 16 82 1565 96.20 203.92
As discussed in the literature review, the amount of transactions an investor does has a negative impact on the returns. To give an insight in the return distribution for the transactions, table 3 contains the statistics on relative returns for stock and ETF transactions in the year 2016. It is not possible to calculate the return for all asset classes due to the information that is available in the dataset. This is for instance the case for options and futures. When options expire in-the-money, they get exercised. This does not have a profit or loss as result, but a position in the underlying value in case of a physical settlement or cash in case of a cash settlement. The broker system only registers a return on an option position when it is closed with an offsetting option transaction and not when the option expires. Because of this reason, the calculation of the relative returns for transactions in options does not give a fair view on the performance of the investors. For futures, the relative returns are calculated based on the initial margin requirements of the future contract. In the main dataset the margin requirements are not specified, meaning it is not possible to calculate the relative returns. If the relative returns where calculated based on the transaction value, these will be too low compared to the real relative return. Last note on the calculation of the relative returns, is that for stocks they are based on a first-in first-out principle and are not time-weighted. Therefore the relative returns only show the return from opening a position to closing a position. These are only transactions in Euro and US Dollar, as these currencies account for 96.86 percent of all the transactions. As the main dataset does not include all opening and closing transactions, the returns have been calculated by using the
transaction value of the closing order. For the calculation of returns for long positions that were closed with a sell transaction, the absolute returns had to be subtracted from the transaction value (1). For short positions that where closed with a buy transaction, the absolute return had to be added to the transaction value (2). The absolute return is after transaction costs, fees and taxes. The relative returns as calculated here, provide information on the return investors made from opening and closing a stock or ETF position independently from any dividends or holding period, because this information is not available.
(1) 𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑅𝑒𝑡𝑢𝑟𝑛!"#$= (!"#$%#&'()$ !"#$% !" !!! !"#$%&' !"#$% ! !"#$%&)!"#$%&'( !"#$%& × 100%
(2) 𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑅𝑒𝑡𝑢𝑟𝑛!!!"#= (!"#$%#&'()$ !"#$% !" !!! !"#$%&' !"#$% ! !"#$%&)!"#$%&'( !"#$%& × 100%
Firstly, table 3 shows that the average relative return on stocks is 3.74%. The median relative stock return for all the investors is 0.21%. Second, men have a lower average and median compared to women. Men have an average (median) relative return of 3.63% (0.21%), while females have an average of 6.03% (0.38%) percent. However, based on the statistics of Table 2, men do more transactions. Therefore the compounding effect, which is interest on interest or in this case return on return, could still cause men to have an higher total relative return than women. To be able to test this, more information is needed on the change of each individual investor his portfolio value over time. Third, the average relative return per transaction decreases with age. The median however, does not show a similar pattern as the averages. Again, this does not mean that older investors have a lower total return, but that the average relative return per transaction is lower. Based on the information from table 2, older investors seem to do more transactions. This could again cause a compounding effect resulting in higher total returns. Lastly, the statistics for education level show that the investors with a university degree have an average return per transaction of 3.52% and a median of 0.16%. Investors with lower education have an average return of 4.09% per transaction and a median of 0.33%. This is interesting, because the statistics in table 2 showed that the transaction frequency was also higher for lower educated investors. This suggests that people with a lower level of education, do more transactions with higher relative returns per transaction. Appendix 5 and 6 show the absolute returns for the three largest asset classes: stocks and ETFs, options and futures. These statistics give an overview of how the performance per transaction is for the different investor characteristics and how this differs for the three asset classes. These statistics are intended to give more insight into the dataset, but are not used in testing the hypotheses.
Table 3
Relative Stock and ETF Returns per transaction (in %)
This table contains statistics on the relative returns for closing transactions in USD and EUR traded stocks and ETFs. A closing transaction is considered as a transaction that reduces the quantity of an existing position partly or completely. ‘Total’ includes all the individual investors independently of the characteristics gender, age and education. The definitions of the characteristics are similar to table 1. The relative stock returns are winsorized and trimmed at the 0.1th percentile and the 99.9th percentile to account for outliers. All values are
stated in percentage points (%), excluding ‘Count’.
Count Minimum
Lower
Quartile Median
Upper
Quartile Maximum Mean
Standard Deviation Total 90900 -99.96 -3.17 0.21 3.67 4499.33 3.74 90.06 Gender Male 86598 -99.96 -3.17 0.21 3.64 4499.33 3.63 87.06 Female 4302 -99.95 -3.14 0.38 4.28 4229.83 6.03 137.11 Age < 30 4694 -99.88 -2.79 0.11 3.53 4229.83 5.77 119.01 30 – 40 12682 -99.96 -3.11 0.20 4.71 3910.00 5.35 85.54 40 – 50 24133 -99.95 -2.21 0.43 3.85 4499.33 4.78 106.53 50 – 60 25230 -99.96 -3.03 0.11 2.77 4062.67 2.69 85.30 60 > 24161 -99.96 -4.41 0.19 4.13 3585.00 2.57 70.60 Education University 55680 -99.96 -3.28 0.16 3.51 4499.33 3.52 89.45 < University 35220 -99.96 -2.99 0.33 3.94 4406.00 4.09 91.00 B. Methodology
The statistics on the transaction frequency and investor characteristics indicate that male, older and lower educated investors do more transactions. The ordinary least squares (OLS) regression method is used to test if the investor characteristics gender, age and education indeed have an impact on the trading behaviour of the investors. To test if these characteristics have an impact on the trading frequency of investors, the number of transactions per unique trading account is used as the dependent variable. An overview of the statistics of the transaction frequency is given in table 2 and discussed in the data section. The independent variables are gender, age and education. For gender, a ‘dummy’ variable is used, that equals one if the investor is a male and zero if the investor is a female. Age is simply measured in years. For the level of education a dummy is used. This dummy equals one if the investor has university as the highest level of education and zero if the level of education is lower than university. In the main dataset, the education levels are qualified according to the Dutch education system. For an overview of the different levels of education and the working of the dummy variable, see Appendix 1. As control variables, the trading experience and NAV of an investor at the beginning of 2016 are used. As discussed in the literature review, Menkhoff et al. (2013) argue that trading experience is an important factor for overconfidence and that the relationship is the opposite from age. Therefore trading experience is used as a control variable in the regressions. Also, Graham et al. (2009) show that investors with larger portfolios perceive themselves as more competent, which
indicates that they are more overconfident. The NAV measures the size of the portfolio including cash positions and is used as a control variable in the regressions. Because of the difference in size compared to the transaction frequency, the log value of the NAV is used.
To test the relationship between investor characteristics and transaction frequency, four different dependent variables with eight regressions each are used. The four dependent variables are the transaction frequency for all asset classes and then specifications are made for stocks and ETFs, options, and futures. As Appendix 2 shows, these three asset classes account for the highest amount of transactions. By specifying the asset categories it is possible to check if the relationships are the same for different asset categories, which can be used as a robustness check. Barber an Odean (2001) only looked at transactions in stocks, which makes the specification for the different asset classes in this thesis unique. For the dependent variables, the currency and size of the transaction is not taken into account. This means that each transaction is considered equally. For each dependent variable there are eight different regressions. This is to test the effect of each single independent variable, with and without the control variables and makes it possible to check the robustness of the coefficient for a single independent variable. An overview of the setup of these eight regression models is shown in Appendix 7.
The main regression model (3) contains all the independent and control variables that are used. Here TFij is the transaction frequency for investor i in asset class j. 𝐺𝑒𝑛𝑑𝑒𝑟! is the dummy variable
that equals one if investor i is a male. 𝐴𝑔𝑒! is the age of investor i measured in years. The education dummy Universityi equals one if investor i has university as the highest level of education. Experiencei is the trading experience measured in years and log(NAVi) the natural logarithm of the net asset value of investor i. Lastly, 𝜀! represents the error term.
(3) 𝑇𝐹!" = 𝛽! 𝐺𝑒𝑛𝑑𝑒𝑟!+ 𝛽! 𝐴𝑔𝑒!+ 𝛽!𝑈𝑛𝑖𝑣𝑒𝑟𝑠𝑖𝑡𝑦!+ 𝛽! 𝐸𝑥𝑝𝑒𝑟𝑖𝑒𝑛𝑐𝑒!+ 𝛽! log 𝑁𝐴𝑉! + 𝜀!
By using this regression formula it is possible to test the relationship between the investor characteristics gender, age, and education and the transaction frequency. If the coefficient for the characteristic is significant at 5% or lower, it is possible to argue that the coefficient is different from zero. This implies that there is a significant relationship between an independent variable and the dependent variable. The level of significance for a single coefficient is calculated using a simple t-test. Based on the existing literature, it is expected that men are more overconfident than women and that overconfidence increases with age. As overconfidence has a positive relationship with transaction frequency, it is expected that gender and age both have positive coefficients. This suggests that men and older investors do more transactions compared to women and younger investors. For education, it is expected that the investors with a higher level of education do more transactions, because they are more overconfident. This means that the coefficient for the university dummy is expected to be positive. Although the control variables are not the focus of this thesis, it is expected that trading
experience has a negative relationship with overconfidence. Therefore investors with more trading experience are expected to do less transactions and the coefficient for trading experience should be negative. For the NAV it is expected that investors with a higher portfolio value are more overconfident, which implies they trade more. Therefore the coefficient for the NAV is expected to be positive. Besides using a simple t-test to measure the significance of one coefficient, the F-statistic is used as an approximate test to examine the significance of the overall regression model. It is possible that a single independent variable has an insignificant coefficient, but that the joined independent variables are actually relevant factors to explain the dependent variable. The t-statistic and F-statistic are calculated by using software program STATA.
As discussed in the data section, there are some large outliers in the main dataset that can bias the results. Therefore the dependent variable is winsorized and trimmed at the 99th percentile. The minimum transaction frequency is never smaller than one, because only the unique trading accounts that did at least one transaction in 2016 are used in the main dataset. Therefore it is not necessary to winsorize at the bottom. Furthermore, robust standard errors are used in all regressions to account for heteroskedasticity. Heteroskedasticity can result in invalid standard errors, leading to a wrong calculation of statistical significance.
IV. Results
A. All asset categories
First, the results with transaction frequency in all asset classes as the dependent variable are discussed. This is followed by the regression output for the specifications in the transaction frequency for stocks and ETFs, options, and futures. Besides testing if the investor characteristics have a different relationship with transaction frequency in other asset categories, these specifications serve as a robustness check. Table 4 contains the regression output for eight regressions with transaction frequency in all asset categories as the dependent variable. In the first regression, only gender is used as independent variable. Based on this output, gender has no significant effect on the transaction frequency in all asset categories. The second regression also includes the control variables trading experience and net asset value. The coefficient for gender remains positive but statistically insignificant. Both of the control variables have positive coefficients that are significant at a 1% level. An increase of one year of trading experience results in a 0.727 increase in the amount of transactions per year. A one-percent change in NAV increases the transaction frequency with 9.333 per year. By adding the control variables, the R-squared increased with at least 1.5 percentage points.
Table 4
Regression output (for j is all asset categories)
This table contains the regression output for eight regressions with transaction frequency in all asset categories per unique trading account as the dependent variable. The investor characteristics gender, age, education, trading experience and net asset value are similar to previous tables. Transaction frequency is winsorized and trimmed at the 99th percentile to account for outliers. ‘N’ is the number of observations per
regression. ‘R2’ is the r-squared and measures the explanatory power of each regression model. ‘F’ is the
F-statistic that is an approximation to test the joined significance of independent variables in the regression models. Standard errors are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.
(1) (2) (3) (4) (5) (6) (7) (8) Gender 9.097 2.875 9.769 4.740 (7.326) (8.911) (7.339) (8.957) Age 1.245*** 1.150*** 1.174*** 1.005*** (0.140) (0.196) (0.141) (0.199) University -23.268*** -28.956*** -19.158*** -24.984*** (4.852) (5.635) (4.873) (5.695)
Experience (in years) 0.727*** 0.024 0.834*** 0.186
(0.266) (0.302) (0.265) (0.305)
Log (Net Asset Value) 9.333*** 8.677*** 9.584*** 8.986***
(1.218) (1.210) (1.222) (1.216) Constant 72.472*** -9.897 19.477*** -51.126*** 96.203*** 7.517 26.712** -37.052** (6.968) (13.121) (6.716) (13.188) (4.196) (11.313) (10.943) (16.269) N 6979 5437 6979 5437 6979 5437 6979 5437 R2 0.000 0.015 0.009 0.020 0.004 0.020 0.012 0.024 F 1.542 25.034 78.875 36.082 22.997 33.163 31.565 25.373
The third regression has age as the only independent variable. The coefficient for age is positive and significant at a 1% level. This implies that older investors trade more than younger investors. A one-year increase in age results in a 1.245 increase in the transaction frequency. When the control variables are added to the model in the fourth regression, the coefficient for age remains positive and significant at a 1% level. However, the effect is smaller as a one-year increase in age now results in a 1.15 increase in transaction frequency. The coefficient for trading experience is negative, but statistically indifferent from zero. This implies that trading experience does not have an effect on the transaction frequency. It is possible that the age variable accounts for the effect of trading experience. However, this is not in line with Menkhoff et al. (2013), who argue that trading experience is an important factor for overconfidence and that the relationship is the opposite from age. The coefficient is negative, but because it is insignificant it is not possible to confirm the findings by Menkhoff et al. (2013). NAV still has a positive coefficient that is significant at 1%. An 1% increase in the NAV results in an increase of 8.677 transactions per year.
In the fifth regression output from table 4, only the university dummy is used as independent variable. University has a negative coefficient that is statically significant at a 1% level. This indicates that investors with a higher level of education do 23.268 fewer transactions per year compared to lower educated investors. In the sixth regression the control variables are added to the regression. The
coefficient for university got more negative and remains significant at 1%. Trading experience and NAV both have a positive and significant coefficient at 1%.
All of the independent variables for the investor characteristics are included in the seventh regression model. The coefficient for gender remains positive, but insignificant. Age still has a positive coefficient of 1.174 that is significant at 1%. This means that an investor that is one year older does 1.174 more transactions per year. The university dummy has a coefficient of -19.158 and is significant at 1%. This would be in line with the findings of regressions five and six; that investors with higher levels of education do fewer transactions. The control variables are added in regression model 8, which is the main regression model. The results are comparable to the previous regressions. The coefficient for gender is still positive and insignificant and the coefficient for age is positive and highly significant. A one-year-older investor does 1.005 more transaction per year based on this regression output. The coefficient for the university dummy is negative and significant at 1%, which is in line with previous results. Based on these coefficients, investors with a higher level of education do 24.984 fewer transactions per year. Experience in trading has a positive, but insignificant coefficient. The coefficient for NAV is positive and significant at 1%. Based on this output, an investor with an one percent higher NAV does 8.986 more transactions per year.
Based on the eight regressions with the transaction frequency in all asset classes as the dependent variable, there are already some interesting results. Firstly, gender does not seem to have a significant effect on the transaction frequency of investors. This is not in line with the existing literature on overconfidence, differences between gender, and investor behaviour. At this point there is not enough evidence in support of H1, that men do more transactions than women. Barber and Odean (2001) did find enough evidence that men trade more than women. However, they only used transactions in stocks for their research. For the regressions in table 4 the transactions in all of the thirteen asset classes are used, which could be one explanation for the different results. In the next section the regressions are specified in three different asset classes, to test whether the relation is different for other types of trading products. This makes it also possible to compare the results to Barber and Odean (2001). Second, the coefficient for age is positive and significant at 1% in all of the regressions. This is in support of H2, that older people do more transactions. Thirdly, the coefficient for university is negative and also significant at 1% in all the regressions. This indicates that investors with a higher level of education trade less, which contradicts the expectations based on the existing literature and is not in support of H3, but indicates the opposite. Lastly, the coefficient for trading experience is not significant in all regressions, also not in the main regression model. Therefore it is not possible to say what the effect of trading experience is on the transaction frequency and if this is in line with the findings of Menkhoff et al. (2013). The coefficient for the natural logarithm of the NAV is positive and significant at 1% in all of the regressions. This indicates that investors with larger portfolio values do more transactions. This is in line with the findings of Graham et al. (2009), that
investors with large portfolios are more overconfident. And that this overconfidence leads to higher transaction frequency as Odean (1999) and Barber and Odean (2000, 2001) argue.
The F-statistic for the first regression is very low. However, the other regression models have a F-statistic that is higher than the critical value. This makes the independent variables used in the models jointly significant at 1%. Even if the single independent variables do not have significant coefficients, they are jointly significant. This would mean that the independent variables used in the models have an effect on the transaction frequency. Therefore these regression models are considered to be a better fit than only using the intercept. To test if these results are consistent, the regression results for the specified asset categories are now discussed.
Table 5
Regression results (for j is stocks and ETFs)
This table contains the regression output for eight regressions with the transaction frequency in stocks and ETFs per unique trading account as the dependent variable. The investor characteristics gender, age, education, trading experience and net asset value are similar to previous tables. Transaction frequency is winsorized and trimmed at the 99th percentile to account for outliers. ‘N’ is the number of observations per regression. ‘R2’ is the r-squared and measures the explanatory power of each regression model. ‘F’ is the F-statistic that is an approximation to test the joined significance of independent variables in the regression models. Standard errors are in parentheses.
* p < 0.10, ** p < 0.05, *** p < 0.01. (1) (2) (3) (4) (5) (6) (7) (8) Gender 5.885** 6.016** 6.071*** 6.744** (2.343) (2.781) (2.340) (2.783) Age 0.301*** 0.352*** 0.297*** 0.351*** (0.049) (0.067) (0.050) (0.069) University -2.622* -2.981* -1.466 -1.518 (1.486) (1.664) (1.537) (1.719)
Experience (in years) -0.155* -0.364*** -0.131 -0.377***
(0.082) (0.098) (0.081) (0.099)
Log (Net Asset Value) 4.448*** 4.209*** 4.474*** 4.229***
(0.581) (0.571) (0.581) (0.572) Constant 20.032*** -19.766*** 10.652*** -27.311*** 27.182*** -12.831** 6.260* -32.505*** (2.228) (5.673) (2.279) (5.988) (1.228) (5.375) (3.680) (6.702) N 5590 4353 5590 4353 5590 4353 5590 4353 R2 0.001 0.027 0.007 0.032 0.001 0.026 0.008 0.034 F 6.309 21.791 38.343 25.450 3.113 20.707 16.973 17.571
B. Stocks and ETFs
Table 5 shows the output for the same regressions as in table 4, only now the dependent variable consists of only the transaction frequency in stocks and ETFs. In the first regression only the dummy variable for gender is used. The coefficient is positive and significant at 5%. This indicates that men do 5.885 transactions more in stocks and ETFs than women. The control variables are added in the second regression. The coefficient for gender increases to 6.016 and is still significant at 5%. For trading experience the coefficient is -0.155 and only significant at 10%. This is a weak indication
that investors with more experience do fewer transactions in stocks and ETFs. The natural log of the NAV has a positive coefficient of 4.448 and is significant at 1%. Based on this result, investors with a one percent higher NAV do 4.448 more transactions in stocks and ETFs.
The third regression in table 5 includes only age as independent variable. An one-year-older investor has a significant higher transaction frequency in stocks and ETFs of 0.301 transactions per year. The coefficient is significant at 1%. By adding the control variables, the coefficient for age remains positive and significant at this level. Based on this regression, an one-year older investor does 0.352 more transactions per year. Trading experience has a coefficient of -0.364 and is significant at 1%. An one-year increase in the trading experience results in 0.364 fewer transactions per year. NAV has a positive and highly significant coefficient of 4.209, which indicate that a 1% increase in NAV results in an increase of the transaction frequency with 4.209 per year.
In regression five, the university dummy is used as independent variable. The coefficient is negative, but only significant at 10%. This result suggests that investors with a higher level of education do fewer transactions in stocks and ETFs compared to investors with a lower education. However, as the coefficient has a low level of significance, the result is not very strong. In the sixth regression, the coefficient for university remains negative and only significant at 10%. Trading experience is insignificant, while NAV has a positive coefficient that is significant at 1%.
In the seventh and eighth regression, all of the independent variables are added. In both the models, the coefficients for gender and age remain positive and significant at respectively 5% and 1%. The university dummy is negative, but statistically insignificant. This indicates that education does not have an effect on the transaction frequency in stocks and ETFs. The coefficient for the trading experience is negative and positive for NAV; both are significant at 1%. Transaction frequency in stocks and ETFs therefore seems to have a negative relationship with the trading experience and a positive relationship with NAV.
Summarizing the results of table 5, it is clear that gender has a positive relationship with the transaction frequency in stocks and ETFs. This suggests that men do more transactions compared to women. This is in line with the research by Barber and Odean (2001) and is in support of H1. However, it does not support the results for gender from table 4. Age has a positive relationship with the transaction frequency in stocks and ETFs, which is in line with the earlier results for all asset categories. This is in support of H2, that older investors do more transactions compared to younger people. As Menkhoff et al. (2013) argue, trading experience has the opposite relationship of age. This also follows from the findings from table 5, as all coefficients for trading experience are negative in all the regressions. In the main regression model the coefficient for trading experience is significant at 1%. Also the relationship between the NAV and transaction frequency is in line with the expectations that where based on Graham et al. (2009). In all of the regressions, the coefficient for the natural log of the NAV is positive and highly significant. This indicates that investors with larger portfolio values do more transactions in stocks and ETFs.
