Add Journal to My Library
Review of Finance
, Volume Advance Article – Mar 6, 2018

30 pages

/lp/ou_press/how-does-learning-and-education-help-to-overcome-the-disposition-WiR0tqWGw2

- Publisher
- Oxford University Press
- Copyright
- © The Author(s) 2018. Published by Oxford University Press on behalf of the European Finance Association. All rights reserved. For permissions, please email: journals.permissions@oup.com
- ISSN
- 1572-3097
- eISSN
- 1573-692X
- D.O.I.
- 10.1093/rof/rfy006
- Publisher site
- See Article on Publisher Site

Abstract The paper assesses how intelligence, education, and learning affect the disposition effect using our exhaustive NASDAQ OMX Tallinn dataset. We employ survival analysis to show that higher intelligence and stronger learning abilities as measured by education level and the type of education lessen the disposition effect. More highly educated and intelligent investors also learn faster by trading. We find that mathematical abilities are beneficial for overcoming the disposition effect and propose that learning ability is one of the most important components of intelligence in affecting the disposition effect. 1. Introduction The ability of investors to learn from their own mistakes is an important factor that can help prevent them taking short-sighted or biased investment decisions. There has been a lot of discussion about the importance of educating investors in how to avoid unnecessary volatility in the financial markets. One bias that is still prevalent in the financial markets is the overall tendency of investors to hold on to underwater positions and to give up winning positions too early. This well-known phenomenon is the disposition effect,1 and it has been scrutinized intensively since the 1990s. Most of the literature on the disposition effect uses the prospect theory developed by Kahneman and Tversky (1979) to explain the tendency to hold losers and sell winners. This view has been challenged since the end of 2000s though, and Barberis and Xiong (2009), Kaustia (2010), and Hens and Vlcek (2011) argue that the prospect theory cannot always explain the disposition effect.2 The results of recent theoretical studies imply that there are in all likelihood still unexplored reasons and explanations behind the disposition effect. A better understanding of the disposition effect can help to reduce the cost of the bias for investors.3 One of the reasons why the effect still needs investigation is that more detailed data become available only when more time has passed from the first identification of the bias. An important part of the recent disposition effect literature has focused on the learning process as one of the possible explanations. Feng and Seasholes (2005) were among the first to show that experience affects the disposition effect and later studies such as, for example, Seru, Shumway, and Stoffman (2010) distinguish between “learning from experience” and “learning about one’s ability” to trade, and they emphasize the importance of the latter. Even so, the characteristics of investors in the form of their abilities, knowledge, and skills can contribute positively or negatively to the disposition effect. The current literature has not studied in detail how education, intelligence, and innate cognitive abilities affect the disposition effect. Our contribution to the literature is to provide empirical insights into how education, intelligence, and certain mental abilities affect the disposition effect. We also show that learning abilities and the speed of learning can vary greatly and affect the disposition effect, even though we do not identify any strong effects from “learning about one’s abilities.” Our dataset allows us to measure intelligence and certain mental abilities in an educational setting before investors enter the stock market and to observe their subsequent behavior in the stock market. The study combines complete detailed transaction records from 2004 to 2012 from Nasdaq OMX Tallinn with data from the Estonian Ministry of Education and Science that gives information on individuals’ high school grades, examination results, university degrees, and subjects. The use of Estonian data provides a unique opportunity to work with a complete dataset which is not affected by subsample selection biases. We show how the performance of investors in terms of the disposition effect depends on their education level and standardized high school final examinations. We use the education level, education type, and examination results as proxies for intelligence. A higher level of education and certain types of education indicate higher intelligence and especially learning abilities as a component of intelligence. Although particular final examinations are a matter of choice, the choice of examinations is not arbitrary, because most university courses require certain basic examinations in the mother tongue, mathematics, and a foreign language, though there are exceptions, especially for humanities subjects. Combining the effect of education and intelligence with “learning by doing” allows us to ascertain how different abilities affect the disposition effect in a detail that has not been possible before. In addition, we can answer the question of whether the speed and importance of “learning by doing” depends on education and intelligence. We find support for the hypothesis that “baseline learning abilities”4 play an important role in affecting the disposition effect. One of our main findings is that investors with higher academic degrees or more challenging academic paths, which can serve as an indication of higher intelligence, are less influenced by the disposition effect and also learn faster by doing. We conclude that learning abilities are one of the most important components of intelligence in affecting the disposition effect. Such knowledge can help in designing more effective educational and training programs for investors. The remainder of the paper is organized as follows. Section 2 reviews the findings in the previous literature. We give an overview of our data in Section 3. Section 4 describes the methodology and Section 5 discusses the findings. The conclusion is presented in Section 6. 2. Related Literature The disposition effect has been one of the central topics in behavioral finance since the early days of the discipline, starting with the work of Shefrin and Statman (1985). Using prospect theory (Kahneman and Tversky, 1979), Shefrin and Statman (1985) argue that investors frame all choices in terms of potential gains and losses relative to a fixed reference point, and they employ an “S-shaped” valuation function. The effect is that the investor is risk averse in the gain region and risk seeking in the loss region. Parts of the recent literature reason that the explanation based on prospect theory can only partly explain the disposition effect. Studies by Barberis and Xiong (2009), Kaustia (2010), and Hens and Vlcek (2011), for example, argue that explanations based on prospect theory can only apply to a small number of cases and can yield both the disposition effect and the reverse of the disposition effect. Even though there are different approaches to explaining5 the disposition effect, empirical literature6 does not question the existence of the phenomenon and recent literature7 shows that some investors can be affected by the reverse disposition effect. The presence of the disposition effect is very strongly documented by many prominent authors. Using a stock market dataset from the USA, Odean (1998) shows that individual investors experience the disposition effect at the aggregate level. These findings are confirmed by numerous authors for many markets. Shapira and Venezia (2001) use Israeli data to show that both individual and professional investors exhibit the disposition effect, and the view is shared by Grinblatt and Keloharju (2001), who find that investors tend to be reluctant to realize their losses except in the tax-selling month of December. These findings are generally consistent with the study by Dhar and Zhu (2006). Dhar and Zhu (2006) argue, however, that there are differences at the individual level and almost 20% of individual investors are not influenced by the disposition effect and can even behave contrary to its predictions. Investors who are immune to the phenomenon tend to have higher trading frequency, higher income, and financial sector jobs. The latter two are proxies of investors’ sophistication, and Feng and Seasholes (2005) also emphasize the importance of investor sophistication and trading experience and argue that those characteristics together can eliminate the reluctance to realize losses. They contend that even though trading experience alone weakens the disposition effect, it does not eliminate it entirely. Kaustia, Alho, and Puttonen (2008) and da Costa et al. (2013) confirm that higher experience leads to a reduction of the disposition effect. Seru, Shumway, and Stoffman (2010) also find that investors’ performance improves as they become more experienced; investors who make more trades, and thus have the opportunity to learn from their mistakes, are less influenced by the disposition effect.8 The results of various empirical studies indicate that other factors that influence the disposition effect are gender, age, and portfolio diversification. Having a gain or loss at the portfolio level does not seem to contribute to the disposition effect (Talpsepp, Vlcek, and Wang, 2014) but the current performance and past price movement of individual stocks affect the disposition effect instead (Kubiska, Markiewicz, and Tyszka, 2012). The learning process that occurs in the financial markets is an important aspect that can also reduce the disposition effect. Learning can occur in different forms in the financial markets as discussed by Fenton-O’Creevy et al. (2012); Nicolosi, Peng, and Zhu (2009); and others. Some earlier studies have focused mostly on “learning from experience,” like Feng and Seasholes (2005), who show that more trading experience reduces the disposition effect. However, Seru, Shumway, and Stoffman (2010) emphasize the role of “learning about one’s ability to trade.” They find that investors who trade regularly suffer less from the disposition effect but have a lower speed of learning. The effects of “learning about one’s ability to trade” are not strong for the market under investigation in the current paper (Muhl and Talpsepp, 2016). The linkage between mental or academic abilities and the disposition effect has received little attention in the literature before now. One of the main reasons is the lack of good data. Many researchers use various proxies to measure sophistication, but these proxies do not strongly relate to mental and academic abilities. Even so, Goo et al. (2010) show that the disposition effect is dependent on education as investors with a higher level of education and a higher academic degree experience a lower disposition effect. 3. Data Our sample combines two different datasets. The stock market dataset is obtained from Nasdaq OMX Tallinn and contains detailed information about all transactions by all market participants for 2004–2012. We have information about the age, gender, and domicile (domestic or foreign) of every investor and can calculate their portfolio size, stock allocations in the portfolio, trading activity, experience of participation in the market or in making trades, average stock holding period, number of transactions, and transaction size, among various other metrics. There may be some liquidity constraints for active trading as the Estonian stock market is small and had only twenty-three different companies listed during the period. The total number of individuals and institutions active during the period was 33,843. Our second dataset contains exhaustive educational data provided by the Estonian Ministry of Education and Science. We have education information for 10,555 investors. Integrating these two datasets gives us a quite unique combination for our study. We are able to identify investors by whether their level of education is high school, bachelor, master, or doctor, and by their subject background, such as mathematics, statistics, economics, medicine, law, information technology, public administration, chemistry, physics, or psychology. Moreover, we have all the high school grades and all the national high school final examination results for each individual investor. These examinations are identical for all high school graduates and as examinations are used for admission to university, the level of difficulty is aimed to be the same throughout the years. With the detailed information about investors’ educational and academic results, we can draw conclusions about their academic abilities that are closely correlated with mental abilities (Deary and Johnson, 2010) and give a picture about their natural or baseline learning abilities. The people in our sample are relatively young, with an average age of 32.6 years, because the national examinations are taken around the age of 18 years and our educational data start from 1997. However, the sample is in line with the overall market as the average Estonian investor is also relatively young because the history of Estonia’s capital markets is short. The age distribution of our sample is presented in Figure 1. Figure 1. View largeDownload slide Age distribution. The graph shows the age distribution of investors in our sample. Most investors in our sample are in their late 20s and early 30s as at December 31, 2012. Figure 1. View largeDownload slide Age distribution. The graph shows the age distribution of investors in our sample. Most investors in our sample are in their late 20s and early 30s as at December 31, 2012. The typical investor in Nasdaq OMX Tallinn is quite similar to an investor in developed markets like those in the USA or Finland in terms of the proportion of male and female investors, and their trading frequency, relative portfolio diversification, and overall disposition effect. Barber and Odean (2001) document that 78.7% of all stock market investors are male in the sample from the USA, which is comparable with the figures in our sample where 73.7% of investors are male (presented also in Table I). Table I. Number of investors with national high school examination results and with education data The table reports statistics for investors whose results from the national high school examinations are available. The total number of investors who had taken their national high school examinations by 2012 is shown together with the number of male and female investors separately. Statistics are presented for 10 different examinations. The proportions of the total for male and female investors are reported. The total sample size is also shown at the bottom of the table. The total sample includes investors with national examination results data together with investors about whom we also have education type or level data. Sample Total Male Proportion (%) Female Proportion (%) National high school examination sample 6,851 5,346 78.0 1,505 22.0 Mother Tongue examination 6,438 5,016 77.9 1,422 22.1 English examination 5,449 4,284 78.6 1,165 21.4 Mathematics examination 4,648 3,794 81.6 854 18.4 History examination 2,600 1,993 76.7 607 23.3 Chemistry examination 1,553 1,197 77.1 356 22.9 Biology examination 1,502 1,065 70.9 437 29.1 Geography examination 1,223 1,007 82.3 216 17.7 Social science examination 1,086 859 79.1 227 20.9 Physics examination 866 814 94.0 52 6.0 German examination 669 519 77.6 150 22.4 Total sample (investors with education data) 10,555 7,779 73.7 2,776 26.3 Sample Total Male Proportion (%) Female Proportion (%) National high school examination sample 6,851 5,346 78.0 1,505 22.0 Mother Tongue examination 6,438 5,016 77.9 1,422 22.1 English examination 5,449 4,284 78.6 1,165 21.4 Mathematics examination 4,648 3,794 81.6 854 18.4 History examination 2,600 1,993 76.7 607 23.3 Chemistry examination 1,553 1,197 77.1 356 22.9 Biology examination 1,502 1,065 70.9 437 29.1 Geography examination 1,223 1,007 82.3 216 17.7 Social science examination 1,086 859 79.1 227 20.9 Physics examination 866 814 94.0 52 6.0 German examination 669 519 77.6 150 22.4 Total sample (investors with education data) 10,555 7,779 73.7 2,776 26.3 Table I. Number of investors with national high school examination results and with education data The table reports statistics for investors whose results from the national high school examinations are available. The total number of investors who had taken their national high school examinations by 2012 is shown together with the number of male and female investors separately. Statistics are presented for 10 different examinations. The proportions of the total for male and female investors are reported. The total sample size is also shown at the bottom of the table. The total sample includes investors with national examination results data together with investors about whom we also have education type or level data. Sample Total Male Proportion (%) Female Proportion (%) National high school examination sample 6,851 5,346 78.0 1,505 22.0 Mother Tongue examination 6,438 5,016 77.9 1,422 22.1 English examination 5,449 4,284 78.6 1,165 21.4 Mathematics examination 4,648 3,794 81.6 854 18.4 History examination 2,600 1,993 76.7 607 23.3 Chemistry examination 1,553 1,197 77.1 356 22.9 Biology examination 1,502 1,065 70.9 437 29.1 Geography examination 1,223 1,007 82.3 216 17.7 Social science examination 1,086 859 79.1 227 20.9 Physics examination 866 814 94.0 52 6.0 German examination 669 519 77.6 150 22.4 Total sample (investors with education data) 10,555 7,779 73.7 2,776 26.3 Sample Total Male Proportion (%) Female Proportion (%) National high school examination sample 6,851 5,346 78.0 1,505 22.0 Mother Tongue examination 6,438 5,016 77.9 1,422 22.1 English examination 5,449 4,284 78.6 1,165 21.4 Mathematics examination 4,648 3,794 81.6 854 18.4 History examination 2,600 1,993 76.7 607 23.3 Chemistry examination 1,553 1,197 77.1 356 22.9 Biology examination 1,502 1,065 70.9 437 29.1 Geography examination 1,223 1,007 82.3 216 17.7 Social science examination 1,086 859 79.1 227 20.9 Physics examination 866 814 94.0 52 6.0 German examination 669 519 77.6 150 22.4 Total sample (investors with education data) 10,555 7,779 73.7 2,776 26.3 Altogether, we have 6851 investors with at least one examination result in our sample. As mother tongue, English and mathematics examinations may be considered as compulsory examinations9 they are also the most popular. Of the 10,555 investors in our sample, 64.3%, or almost two-thirds, had completed higher education by 2012.10 The number of investors with a master’s or doctoral degree is 608, which is 5.8% of the total sample and 9.0% of those with higher education. A further 1521 investors have completed vocational training, and 2244 investors, or 21.3% of the total sample, have only finished high school. Figure 2 shows the distribution of investors by education type. Figure 2. View largeDownload slide Sample distribution by education level. The graph shows the number of investors with the breakdown between male and female investors in each education group on the left axis. The corresponding proportions are shown on the right axis. Figure 2. View largeDownload slide Sample distribution by education level. The graph shows the number of investors with the breakdown between male and female investors in each education group on the left axis. The corresponding proportions are shown on the right axis. The most popular discipline among university subjects for investors is social sciences, a result which is consistent with the findings of Christiansen, Joensen, and Rangvid (2008). At 23.8%, nearly one quarter of the total sample and 37.0%, or over one-third, of investors with higher education have a degree in economics or a related field. The corresponding figures for investors with a business degree are 17.5% of the total and 27.2% of graduates. This means that about 40% of the investors in our total sample have a degree in economics or business, which is not that surprising given the knowledge needed for participating in the stock market. There is a strong representation of investors with a degree in IT in our sample as well, with 716 investors or 6.8% of the total sample. A smaller number of investors have graduated from law or medicine or hold a degree in natural and exact sciences. 4. Methodology The main three methods used for measuring the disposition effect are PGR–PLR analysis, logit regressions, and survival analysis. PGR–PLR analysis was proposed by Odean (1998) and it counts each realized gain, realized loss, paper gain, and paper loss for each day a position is sold. The counts are then used to calculate the proportion of gains realized, labeled PGR, and the proportion of losses realized, labeled PLR.11 Logit regressions are employed by Grinblatt and Keloharju (2001) among others, and survival analysis was proposed by Feng and Seasholes (2005), who present the advantages of the method over alternative approaches. Following the suggestions of Feng and Seasholes (2005) and the later use of survival analysis by Seru, Shumway, and Stoffman (2010) and others, we choose survival analysis as our main method for measuring the disposition effect. As a robustness check, we also run logit regressions, and they confirm the findings in all cases. We use a Cox proportional hazard model to measure the probability that an investor will sell a current stock position. To interpret the results we calculate the hazard rate, which is the probability of selling at time t conditional on holding a stock until time t−1. The hazard rate h and the vector of coefficients β for the covariates are obtained by maximum-likelihood estimation of the following equation: ht,p,X= pλtp-1+exp Xβ+εt, (1) where pλtp-1 indicates the baseline hazard and X is the vector of fixed and time-varying covariates. One advantage of the survival analysis is that it allows for censored observations, which suits our setup as not all positions are closed by the end of the sample period and the data contain partial liquidation as well. Our independent variables include different educational characteristics such as education type and level and academic results, together with control variables such as gender, investor type, number of trades made, experience, and so forth. We divide the national examination results into deciles and quartiles, so that the weakest results are in the lowest decile or quartile and investors with the highest scores are in the top decile or quartile, and we construct dummy variables for this. We use interaction terms of the covariates with the trading loss indicator (TLI) or gain indicator (TGI) variables in the regressions to capture how particular covariates for characteristics affect the disposition effect. We can do this by multiplying the dummy variable for the top mathematics examination decile by the TLI or TGI dummy variable to identify those investors who have the best academic results in mathematics and whose stock position is in loss or gain. All regressions also include TLI and TGI variables without interaction terms to capture the overall tendency of investors to hold on to or sell positions depending on whether they are in loss or gain. The captured hazard ratios for covariates describe a relative probability for how the hazard varies in response to explanatory covariates. Interpreting the coefficient’s hazard ratio is actually relatively easy as the hazard rate changes when an independent variable changes from zero to one. For brevity, we only present results for covariates that are of interest, on occasion giving only the interaction terms, and we omit from the tables control variables like gender, age, trading experience, etc., or if we present only the interaction terms, in most cases we also show the education-related controls but do not present the coefficient for the TLI/TGI alone (as the TLI/TGI coefficients remain unchanged in different regression setups). The baseline level of the disposition effect measured by the TLI or TGI variable as presented in Section 5.1 remains the same for all regressions and the overall probability change of the position being sold dependent on whether it is in loss or gain can be calculated by multiplying the hazard ratios of all the relevant covariates. Running the regressions requires an appropriate data setup. Like Feng and Seasholes (2005) and Seru, Shumway, and Stoffman (2010) we discard any stock purchases that occurred before January 1, 2004 and any resulting sales. We compute whether a position is in loss or gain for every day, for every stock position, and for every investor. This gives us a total of 19 million observations for the sample of 10,555 investors. We use the volume-weighted average purchase price as the reference price for the position and record a gain or a loss for each position if the reference price is lower or higher than the daily low price. The TLI or TGI indicator takes the value 1 if the position is in loss and 0 if it is in profit. 5. Results 5.1. The Aggregate Disposition Effect The characteristics that our sample shows are very closely comparable to the results presented in earlier empirical studies in terms of the magnitude of the general disposition effect. Our results confirm that there is a tendency for investors to sell winners too early and hold losers too long. The results of the survival analysis are presented in Table II. The hazard ratio of 0.793 from the TLI regression indicates that the investors’ propensity to sell is 20.7% lower when the position is in loss than in the baseline probability12 for selling the stock. Similarly, investors’ propensity to abandon their stocks increases by 24.9%, with a hazard ratio of 1.249, when a stock is trading above the purchase price.13 Table II. Aggregate disposition effect The table reports the results of the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if a position is in loss and the value 0 otherwise. Similarly TGI is the total gain indicator, which takes the value 1 if a stock is trading above the purchase price and zero otherwise. Investor-specific variables are interacted with TLI and TGI to capture the disposition effect. These investor-specific variables are also dummy variables, which take the value 1 if an investor is identified by a specific variable (e.g., the male variable takes the value 1 if an investor is a male) and zero otherwise. Variable Hazard ratio z-Statistic Variable Hazard ratio z-Statistic TLI 0.793*** −25.62 Total gain indicator (TGI) 1.249*** 24.64 Male×TLI 1.190*** 6.75 Male×TGI 0.838*** −6.87 Foreign investor×TLI 1.224** 1.92 Foreign investor×TGI 0.816** −1.93 Variable Hazard ratio z-Statistic Variable Hazard ratio z-Statistic TLI 0.793*** −25.62 Total gain indicator (TGI) 1.249*** 24.64 Male×TLI 1.190*** 6.75 Male×TGI 0.838*** −6.87 Foreign investor×TLI 1.224** 1.92 Foreign investor×TGI 0.816** −1.93 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Table II. Aggregate disposition effect The table reports the results of the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if a position is in loss and the value 0 otherwise. Similarly TGI is the total gain indicator, which takes the value 1 if a stock is trading above the purchase price and zero otherwise. Investor-specific variables are interacted with TLI and TGI to capture the disposition effect. These investor-specific variables are also dummy variables, which take the value 1 if an investor is identified by a specific variable (e.g., the male variable takes the value 1 if an investor is a male) and zero otherwise. Variable Hazard ratio z-Statistic Variable Hazard ratio z-Statistic TLI 0.793*** −25.62 Total gain indicator (TGI) 1.249*** 24.64 Male×TLI 1.190*** 6.75 Male×TGI 0.838*** −6.87 Foreign investor×TLI 1.224** 1.92 Foreign investor×TGI 0.816** −1.93 Variable Hazard ratio z-Statistic Variable Hazard ratio z-Statistic TLI 0.793*** −25.62 Total gain indicator (TGI) 1.249*** 24.64 Male×TLI 1.190*** 6.75 Male×TGI 0.838*** −6.87 Foreign investor×TLI 1.224** 1.92 Foreign investor×TGI 0.816** −1.93 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Our results (see Table II) show that male investors tend to be less affected by the disposition effect and female investors more affected, which is in line with Feng and Seasholes (2005) and the experimental study of Rau (2014). One reason is that relatively young men generally trade clearly more, which lowers the disposition effect, as also noted by Feng and Seasholes (2005), who find that men are 30% more likely to realize a loss than women. That is supported by Rau (2014) who shows that men sell a higher proportion of capital losses but a lower proportion of capital gains than women. After studying the selling behavior of local and foreign investors (see Table II), we conclude that foreign investors, who may be considered more sophisticated14 than domestic investors and who do not suffer from the home bias, exhibit the reverse disposition effect. Local investors tend to hold their losers longer, with a hazard ratio of 0.817, and are more likely to sell their winners, with a hazard ratio of 1.225. This is consistent with the results in Talpsepp (2011) and also with several other studies that ascertain the relationship between investors’ sophistication and the disposition effect.15 In terms of experience as measured by “learning by doing,” our results are once again in line with the earlier literature such as Feng and Seasholes (2005). The results presented in Table III show that the disposition effect starts to decrease when an investor has made at least 10 trades. Table III. The effects of “learning by doing” The table reports results from the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if the position is in loss and the value 0 otherwise. TGI is the total gain indicator, which takes the value 1 if a stock is trading above the purchase price and zero otherwise. The results are reported only for variables showing experience (how many trades an investor has made) and for interacted variables. Experience-related variables are interacted with TLI and TGI to capture the disposition effect. Experience-related variables are dummy variables. The trading data cover the period from 2002 to 2012. The table summarizes the results of individually run regressions. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Number of trades: 1–5×TLI 0.938** −2.18 Number of trades: 1–5×TGI 1.061** 2.02 Number of trades: 1–5 0.597*** −24.86 Number of trades: 1–5 0.560*** −27.79 Number of trades: 6–10×TLI 0.940** −2.05 Number of trades: 6–10×TGI 1.067** 2.13 Number of trades: 6–10 0.667*** −20.63 Number of trades: 6–10 0.626*** −20.23 Number of trades: 11–20×TLI 1.077*** 2.85 Number of trades: 11–20×TGI 0.927*** −2.94 Number of trades: 11–20 0.715*** −20.93 Number of trades: 11–20 0.772*** −12.75 Number of trades: 21–50×TLI 1.152*** 6.74 Number of trades: 21–50×TGI 0.869*** −6.69 Number of trades: 21–50 0.813*** −16.24 Number of trades: 21–50 0.937*** −3.93 Number of trades: 51–100×TLI 1.148*** 5.44 Number of trades: 51–100×TGI 0.870*** −5.53 Number of trades: 51–100 1.199*** 11.90 Number of trades: 51–100 1.377*** 15.86 Number of trades: over 100×TLI 1.161*** 7.36 Number of trades: over 100×TGI 0.863*** −7.27 Number of trades: over 100 2.541*** 75.12 Number of trades: over 100 2.945*** 65.27 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Number of trades: 1–5×TLI 0.938** −2.18 Number of trades: 1–5×TGI 1.061** 2.02 Number of trades: 1–5 0.597*** −24.86 Number of trades: 1–5 0.560*** −27.79 Number of trades: 6–10×TLI 0.940** −2.05 Number of trades: 6–10×TGI 1.067** 2.13 Number of trades: 6–10 0.667*** −20.63 Number of trades: 6–10 0.626*** −20.23 Number of trades: 11–20×TLI 1.077*** 2.85 Number of trades: 11–20×TGI 0.927*** −2.94 Number of trades: 11–20 0.715*** −20.93 Number of trades: 11–20 0.772*** −12.75 Number of trades: 21–50×TLI 1.152*** 6.74 Number of trades: 21–50×TGI 0.869*** −6.69 Number of trades: 21–50 0.813*** −16.24 Number of trades: 21–50 0.937*** −3.93 Number of trades: 51–100×TLI 1.148*** 5.44 Number of trades: 51–100×TGI 0.870*** −5.53 Number of trades: 51–100 1.199*** 11.90 Number of trades: 51–100 1.377*** 15.86 Number of trades: over 100×TLI 1.161*** 7.36 Number of trades: over 100×TGI 0.863*** −7.27 Number of trades: over 100 2.541*** 75.12 Number of trades: over 100 2.945*** 65.27 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Table III. The effects of “learning by doing” The table reports results from the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if the position is in loss and the value 0 otherwise. TGI is the total gain indicator, which takes the value 1 if a stock is trading above the purchase price and zero otherwise. The results are reported only for variables showing experience (how many trades an investor has made) and for interacted variables. Experience-related variables are interacted with TLI and TGI to capture the disposition effect. Experience-related variables are dummy variables. The trading data cover the period from 2002 to 2012. The table summarizes the results of individually run regressions. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Number of trades: 1–5×TLI 0.938** −2.18 Number of trades: 1–5×TGI 1.061** 2.02 Number of trades: 1–5 0.597*** −24.86 Number of trades: 1–5 0.560*** −27.79 Number of trades: 6–10×TLI 0.940** −2.05 Number of trades: 6–10×TGI 1.067** 2.13 Number of trades: 6–10 0.667*** −20.63 Number of trades: 6–10 0.626*** −20.23 Number of trades: 11–20×TLI 1.077*** 2.85 Number of trades: 11–20×TGI 0.927*** −2.94 Number of trades: 11–20 0.715*** −20.93 Number of trades: 11–20 0.772*** −12.75 Number of trades: 21–50×TLI 1.152*** 6.74 Number of trades: 21–50×TGI 0.869*** −6.69 Number of trades: 21–50 0.813*** −16.24 Number of trades: 21–50 0.937*** −3.93 Number of trades: 51–100×TLI 1.148*** 5.44 Number of trades: 51–100×TGI 0.870*** −5.53 Number of trades: 51–100 1.199*** 11.90 Number of trades: 51–100 1.377*** 15.86 Number of trades: over 100×TLI 1.161*** 7.36 Number of trades: over 100×TGI 0.863*** −7.27 Number of trades: over 100 2.541*** 75.12 Number of trades: over 100 2.945*** 65.27 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Number of trades: 1–5×TLI 0.938** −2.18 Number of trades: 1–5×TGI 1.061** 2.02 Number of trades: 1–5 0.597*** −24.86 Number of trades: 1–5 0.560*** −27.79 Number of trades: 6–10×TLI 0.940** −2.05 Number of trades: 6–10×TGI 1.067** 2.13 Number of trades: 6–10 0.667*** −20.63 Number of trades: 6–10 0.626*** −20.23 Number of trades: 11–20×TLI 1.077*** 2.85 Number of trades: 11–20×TGI 0.927*** −2.94 Number of trades: 11–20 0.715*** −20.93 Number of trades: 11–20 0.772*** −12.75 Number of trades: 21–50×TLI 1.152*** 6.74 Number of trades: 21–50×TGI 0.869*** −6.69 Number of trades: 21–50 0.813*** −16.24 Number of trades: 21–50 0.937*** −3.93 Number of trades: 51–100×TLI 1.148*** 5.44 Number of trades: 51–100×TGI 0.870*** −5.53 Number of trades: 51–100 1.199*** 11.90 Number of trades: 51–100 1.377*** 15.86 Number of trades: over 100×TLI 1.161*** 7.36 Number of trades: over 100×TGI 0.863*** −7.27 Number of trades: over 100 2.541*** 75.12 Number of trades: over 100 2.945*** 65.27 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. 5.2. How Does the Education Level Affect the Disposition Effect? So far we have shown that the investors in our sample are very similar to investors in previous studies in the disposition effect and also in factors affecting the disposition effect. Although the time spent at an educational institution may give the investor very little specific knowledge about how to avoid the disposition effect, it can give a direct indication of an investor’s overall intelligence, including academic and learning abilities. Thus, we can assume that an investor who has spent more time studying is probably more intelligent than an investor whose educational path has been shorter or conducted at a lower level. Our results (presented in Table IV) show that investors with a master’s or doctoral degree tend to be less affected by the disposition effect. The probability of investors with a master’s or doctoral degree selling a losing position is 8.6% higher, with a hazard ratio of 1.086, and the probability of them selling a winning position is 7.2% lower, with a hazard ratio of 0.928. It should be noted that on average such investors hold on to their positions with hazard ratios of 0.769 and 0.831, which means that they generally trade less and their overall probability of selling a position is lower than for other investors regardless of the effect of having a winning or losing position. Table IV. Education level and disposition effect The table reports results from the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if the position is in loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above the purchase price and zero otherwise. Results are reported only for education level variables and for interacted variables in the regressions. Education level variables are interacted with TLI and TGI to capture the disposition effect. Education level variables are also dummy variables. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Vocational training×TLI 0.913*** −3.52 Vocational training×TGI 1.089*** 3.32 Vocational training 1.069*** 4.19 Vocational training 0.980 −1.00 High school×TLI 1.036 1.54 High school×TGI 0.967 −1.49 High school 1.070*** 4.65 High school 1.107*** 5.79 Bachelor’s degree×TLI 0.996 −0.20 Bachelor’s degree×TGI 1.004 0.24 Bachelor’s degree 0.987 −1.16 Bachelor’s degree 0.983 −1.19 Master’s or doctoral degree×TLI 1.086** 1.99 Master’s or doctoral degree×TGI 0.928* −1.81 Master’s or doctoral degree 0.769*** −10.43 Master’s or doctoral degree 0.831*** −5.62 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Vocational training×TLI 0.913*** −3.52 Vocational training×TGI 1.089*** 3.32 Vocational training 1.069*** 4.19 Vocational training 0.980 −1.00 High school×TLI 1.036 1.54 High school×TGI 0.967 −1.49 High school 1.070*** 4.65 High school 1.107*** 5.79 Bachelor’s degree×TLI 0.996 −0.20 Bachelor’s degree×TGI 1.004 0.24 Bachelor’s degree 0.987 −1.16 Bachelor’s degree 0.983 −1.19 Master’s or doctoral degree×TLI 1.086** 1.99 Master’s or doctoral degree×TGI 0.928* −1.81 Master’s or doctoral degree 0.769*** −10.43 Master’s or doctoral degree 0.831*** −5.62 *** Significant at 1% the level; ** significant at the 5% level; * significant at the 10% level. Table IV. Education level and disposition effect The table reports results from the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if the position is in loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above the purchase price and zero otherwise. Results are reported only for education level variables and for interacted variables in the regressions. Education level variables are interacted with TLI and TGI to capture the disposition effect. Education level variables are also dummy variables. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Vocational training×TLI 0.913*** −3.52 Vocational training×TGI 1.089*** 3.32 Vocational training 1.069*** 4.19 Vocational training 0.980 −1.00 High school×TLI 1.036 1.54 High school×TGI 0.967 −1.49 High school 1.070*** 4.65 High school 1.107*** 5.79 Bachelor’s degree×TLI 0.996 −0.20 Bachelor’s degree×TGI 1.004 0.24 Bachelor’s degree 0.987 −1.16 Bachelor’s degree 0.983 −1.19 Master’s or doctoral degree×TLI 1.086** 1.99 Master’s or doctoral degree×TGI 0.928* −1.81 Master’s or doctoral degree 0.769*** −10.43 Master’s or doctoral degree 0.831*** −5.62 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Vocational training×TLI 0.913*** −3.52 Vocational training×TGI 1.089*** 3.32 Vocational training 1.069*** 4.19 Vocational training 0.980 −1.00 High school×TLI 1.036 1.54 High school×TGI 0.967 −1.49 High school 1.070*** 4.65 High school 1.107*** 5.79 Bachelor’s degree×TLI 0.996 −0.20 Bachelor’s degree×TGI 1.004 0.24 Bachelor’s degree 0.987 −1.16 Bachelor’s degree 0.983 −1.19 Master’s or doctoral degree×TLI 1.086** 1.99 Master’s or doctoral degree×TGI 0.928* −1.81 Master’s or doctoral degree 0.769*** −10.43 Master’s or doctoral degree 0.831*** −5.62 *** Significant at 1% the level; ** significant at the 5% level; * significant at the 10% level. The hazard ratio showing the disposition effect for investors with vocational training is also statistically significant in our regressions. In the Estonian educational system, vocational training either replaces the last 3 years of high school or follows high school. It can be considered academically less challenging than going to high school and afterward continuing at a university. In contrast to the experience in some countries, vocational training was clearly out of favor for students during the sample period. Our data indicate that investors with this kind of education are prone to the disposition effect as they tend to hold losing positions for longer, with a hazard ratio of 0.913, and to sell winning positions more easily, with a hazard ratio of 1.089. These results are consistent with Goo et al. (2010), who document that investors with a higher level of education exhibit a lower disposition effect. Another interesting result is that even though the higher level of education given by a bachelor’s or higher degree tends to reduce the disposition effect, investors with higher education are more reluctant to sell their positions in general. For example, investors who have only high school education or vocational training tend to have a higher probability than other investors of selling their positions if the effects of being in gain or loss are discounted (see the hazard ratios of the control variables). This means that more highly educated investors trade less frequently than less educated investors. This can have positive effects on their returns from the transaction cost point of view as they are not trying to outsmart the market by believing in their ability always to choose the best stocks. As they are also less influenced by the disposition effect, they let their profits run slightly more. The finding is consistent with Barber and Odean (2000), who show that trading too much can have negative effects on the portfolio performance and the same is true for the same sample we use (Liivamägi, Vaarmets, and Talpsepp, 2014). All in all, we can conclude that intelligence plays a role in affecting investor behavior for the disposition effect. The effect is present at the bottom and top ends of the educational path for investors with vocational training and those with Master’s or Doctoral degrees. We can see that the disposition effect is greater for investors with vocational training. These are investors who have chosen a less challenging educational path and are thus likely to exhibit lower learning abilities as well. We do not find any clearly distinguishable effects for the majority of investors, but we see an effect once again for investors who have taken the longest educational path and gained a master’s or doctoral degree, and who thus probably have the highest intelligence, resulting in the disposition effect being reduced. We present the robustness checks along with explanations in Appendix A for all of the results presented in the main sections of the paper. The effects presented in this subsection are also found with statistically significant coefficients when survival models are run with many different covariates for control variables and education level variables at the same time. When experience-related variables are added to the same model, the statistical significance of the education level variables starts to suffer, though the signs of the coefficients remain the same. 5.3. How Does the Type of Education Affect the Disposition Effect? In addition to intelligence that can be signaled by the level of education, other more specific cognitive abilities can also be revealed by the type of higher education. People who have chosen different subjects for study may have clearly different cognitive abilities that may also affect their trading behavior. Of course they also possess different sets of knowledge after the university but we do not find proof that the knowledge taught in university is of great importance in reducing the disposition effect. We test the effects of possible differences in cognitive abilities by studying how the disposition effect is affected by the choice of courses in the broader sense of humanities, social studies or natural sciences, or the narrower sense of particular subjects. Our results (see Table V) show that investors with a degree in humanities exhibit the disposition effect and investors with a degree in natural sciences tend to be more immune to it. Investors with a humanities degree have a hazard ratio of 0.782, meaning a 21.8% lower probability of selling their losing stock, and they have a 28.3% higher probability than the sample average of selling their winning stock. Table V. Disposition effect for specialty groups The table reports results from the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if the position is trading with a loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above purchase price and zero otherwise. The results are reported for specialty groups’ variables as well as for interacted variables. Specialty groups are interacted with TLI and TGI to capture the disposition effect. Specialty groups’ variables are also dummy variables, which take the value 1 if an investor has a degree in a specific discipline and the value 0 otherwise. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Humanities×TLI 0.782*** −6.37 Humanities×TGI 1.283*** 6.48 Humanities 1.226*** 9.16 Humanities 0.957 −1.39 Natural science×TLI 1.056** 2.20 Natural science×TGI 0.913*** −3.81 Natural science 0.992 −0.49 Natural science 1.070*** 3.42 Social science×TLI 0.995 −0.29 Social science×TGI 1.005 0.26 Social science 0.943*** −5.37 Social science 0.938*** −4.54 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Humanities×TLI 0.782*** −6.37 Humanities×TGI 1.283*** 6.48 Humanities 1.226*** 9.16 Humanities 0.957 −1.39 Natural science×TLI 1.056** 2.20 Natural science×TGI 0.913*** −3.81 Natural science 0.992 −0.49 Natural science 1.070*** 3.42 Social science×TLI 0.995 −0.29 Social science×TGI 1.005 0.26 Social science 0.943*** −5.37 Social science 0.938*** −4.54 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Table V. Disposition effect for specialty groups The table reports results from the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if the position is trading with a loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above purchase price and zero otherwise. The results are reported for specialty groups’ variables as well as for interacted variables. Specialty groups are interacted with TLI and TGI to capture the disposition effect. Specialty groups’ variables are also dummy variables, which take the value 1 if an investor has a degree in a specific discipline and the value 0 otherwise. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Humanities×TLI 0.782*** −6.37 Humanities×TGI 1.283*** 6.48 Humanities 1.226*** 9.16 Humanities 0.957 −1.39 Natural science×TLI 1.056** 2.20 Natural science×TGI 0.913*** −3.81 Natural science 0.992 −0.49 Natural science 1.070*** 3.42 Social science×TLI 0.995 −0.29 Social science×TGI 1.005 0.26 Social science 0.943*** −5.37 Social science 0.938*** −4.54 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Humanities×TLI 0.782*** −6.37 Humanities×TGI 1.283*** 6.48 Humanities 1.226*** 9.16 Humanities 0.957 −1.39 Natural science×TLI 1.056** 2.20 Natural science×TGI 0.913*** −3.81 Natural science 0.992 −0.49 Natural science 1.070*** 3.42 Social science×TLI 0.995 −0.29 Social science×TGI 1.005 0.26 Social science 0.943*** −5.37 Social science 0.938*** −4.54 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. These results stand in contrast to those for investors with a degree in natural sciences, whose probability of selling their losing stock is 8.3% higher than the sample average and probability of selling their winning stock is 7.5% lower. Given that the probability of selling a losing position is about 20% smaller than that of selling a winning position in general, this alone does not completely eliminate the disposition effect, but it clearly reduces it when other factors like experience, which can have a combined bigger effect, are not considered. Once again we do not find a significant effect for the majority of the sample, as 74% of the investors with higher education and 49% of the total sample have a degree in social science. However, there are clear differences between investors at the different ends of the education scale of natural sciences and humanities. There are two main sources from which such differences can originate. A possible explanation of the differences between the results for investors with degrees in humanities and natural sciences is that the latter possess more skills for dealing with numbers by analyzing and calculating. This can make them more pragmatic as they can analyze stock market situations better and reach more rational conclusions. We cannot completely exclude the explanation that investors who feel more comfortable with numbers are more interested in stock markets and therefore have more experience, which tends to decrease the disposition effect according to the earlier literature.16 Nor are we able to measure any non-cognitive skills17 that may also play an important role in any investment activities, and we cannot conclude that investors who have studied natural sciences are more intelligent in general. It may be argued18 that studying natural sciences requires more effort as there is less room for relative interpretations and more skills need to be mastered before a student can graduate. Higher dropout rates from natural sciences courses indicate the same, so investors graduating from natural sciences may also possess higher learning abilities or problem-solving abilities, which are certain components of intelligence. We do not find a degree in social sciences to have a statistically significant effect on selling behavior, but we are able to break investors with degrees in social sciences into smaller groups by subject and study those groups in more detail (the results are presented in Table VI). Table VI. University degrees and the disposition effect The table presents hazard ratios associated with an individual investor’s decision to sell or hold stocks at a loss or gain based on the investor’s degree. The hazard ratios together with the z-value and the level of statistical significance are reported for subject group variables and also for interacted variables. This means that we interact each subject variable with the trading loss indicator (TLI) and with the TGI in order to measure cross-sectional differences in investors’ propensities to sell losers and winners. TLI takes the value 1 if a stock is trading below its purchase price, and 0 otherwise. Similarly, the TGI variable takes the value 1 if a stock is trading above its purchase prize and 0 otherwise. The trading data cover the period from 2002 until 2012; educational data are taken as of 2012. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Maths or statistics×TLI 0.994 −0.04 Maths or statistics×TGI 1.015 0.10 Maths or statistics 0.919 −0.89 Maths or statistics 0.908 −0.87 Chemistry, physics, or biology×TLI 1.177** 2.11 Chemistry, physics, or biology×TGI 0.844** −2.21 Chemistry, physics, or biology 0.937* −1.41 Chemistry, physics, or biology 1.107* 1.67 IT×TLI 1.101*** −24.95 IT×TGI 0.905*** −3.18 IT 1.172*** 8.67 IT 1.294*** 10.10 Economics related×TLI 0.943** −2.85 Economics related×TGI 1.066*** 3.13 Economics related 0.808*** −17.23 Economics related 0.759*** −16.93 Finance×TLI 0.773*** −4.23 Finance×TGI 1.302*** 4.34 Finance 1.085** 2.36 Finance 0.836*** −3.60 Law×TLI 1.184*** 4.07 Law×TGI 0.845*** −4.06 Law 0.885*** −4.70 Law 1.046 1.42 Medicine×TLI 1.248** 2.38 Medicine×TGI 0.799** −2.42 Medicine 0.748*** −5.12 Medicine 0.935 −0.92 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Maths or statistics×TLI 0.994 −0.04 Maths or statistics×TGI 1.015 0.10 Maths or statistics 0.919 −0.89 Maths or statistics 0.908 −0.87 Chemistry, physics, or biology×TLI 1.177** 2.11 Chemistry, physics, or biology×TGI 0.844** −2.21 Chemistry, physics, or biology 0.937* −1.41 Chemistry, physics, or biology 1.107* 1.67 IT×TLI 1.101*** −24.95 IT×TGI 0.905*** −3.18 IT 1.172*** 8.67 IT 1.294*** 10.10 Economics related×TLI 0.943** −2.85 Economics related×TGI 1.066*** 3.13 Economics related 0.808*** −17.23 Economics related 0.759*** −16.93 Finance×TLI 0.773*** −4.23 Finance×TGI 1.302*** 4.34 Finance 1.085** 2.36 Finance 0.836*** −3.60 Law×TLI 1.184*** 4.07 Law×TGI 0.845*** −4.06 Law 0.885*** −4.70 Law 1.046 1.42 Medicine×TLI 1.248** 2.38 Medicine×TGI 0.799** −2.42 Medicine 0.748*** −5.12 Medicine 0.935 −0.92 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Table VI. University degrees and the disposition effect The table presents hazard ratios associated with an individual investor’s decision to sell or hold stocks at a loss or gain based on the investor’s degree. The hazard ratios together with the z-value and the level of statistical significance are reported for subject group variables and also for interacted variables. This means that we interact each subject variable with the trading loss indicator (TLI) and with the TGI in order to measure cross-sectional differences in investors’ propensities to sell losers and winners. TLI takes the value 1 if a stock is trading below its purchase price, and 0 otherwise. Similarly, the TGI variable takes the value 1 if a stock is trading above its purchase prize and 0 otherwise. The trading data cover the period from 2002 until 2012; educational data are taken as of 2012. Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Maths or statistics×TLI 0.994 −0.04 Maths or statistics×TGI 1.015 0.10 Maths or statistics 0.919 −0.89 Maths or statistics 0.908 −0.87 Chemistry, physics, or biology×TLI 1.177** 2.11 Chemistry, physics, or biology×TGI 0.844** −2.21 Chemistry, physics, or biology 0.937* −1.41 Chemistry, physics, or biology 1.107* 1.67 IT×TLI 1.101*** −24.95 IT×TGI 0.905*** −3.18 IT 1.172*** 8.67 IT 1.294*** 10.10 Economics related×TLI 0.943** −2.85 Economics related×TGI 1.066*** 3.13 Economics related 0.808*** −17.23 Economics related 0.759*** −16.93 Finance×TLI 0.773*** −4.23 Finance×TGI 1.302*** 4.34 Finance 1.085** 2.36 Finance 0.836*** −3.60 Law×TLI 1.184*** 4.07 Law×TGI 0.845*** −4.06 Law 0.885*** −4.70 Law 1.046 1.42 Medicine×TLI 1.248** 2.38 Medicine×TGI 0.799** −2.42 Medicine 0.748*** −5.12 Medicine 0.935 −0.92 Variables Hazard ratio z-Statistic Variables Hazard ratio z-Statistic Maths or statistics×TLI 0.994 −0.04 Maths or statistics×TGI 1.015 0.10 Maths or statistics 0.919 −0.89 Maths or statistics 0.908 −0.87 Chemistry, physics, or biology×TLI 1.177** 2.11 Chemistry, physics, or biology×TGI 0.844** −2.21 Chemistry, physics, or biology 0.937* −1.41 Chemistry, physics, or biology 1.107* 1.67 IT×TLI 1.101*** −24.95 IT×TGI 0.905*** −3.18 IT 1.172*** 8.67 IT 1.294*** 10.10 Economics related×TLI 0.943** −2.85 Economics related×TGI 1.066*** 3.13 Economics related 0.808*** −17.23 Economics related 0.759*** −16.93 Finance×TLI 0.773*** −4.23 Finance×TGI 1.302*** 4.34 Finance 1.085** 2.36 Finance 0.836*** −3.60 Law×TLI 1.184*** 4.07 Law×TGI 0.845*** −4.06 Law 0.885*** −4.70 Law 1.046 1.42 Medicine×TLI 1.248** 2.38 Medicine×TGI 0.799** −2.42 Medicine 0.748*** −5.12 Medicine 0.935 −0.92 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Surprisingly, a degree in finance and or in any field related to economics including finance does not give an advantage as investors with this type of higher education tend to be even more prone to the disposition effect, and the hazard ratio for selling losing positions is 0.943 for investors with a degree in any field of economics and 0.773 for investors with a degree more narrowly in finance. This unexpected finding may be a result of overconfidence as investors with education in finance and economics may see themselves as better educated about the foundations of the stock market, and so they may lose the necessary focus or try to outsmart the market. Although we can classify business graduates separately and we do not find any statistically significant effect for them, it is very difficult to find major differences between the curricula of various economics and business subjects at undergraduate level, and classifying them into different groups can be somewhat arbitrary. Thus, when we group all economics subjects together, we still see a slightly increased disposition effect for investors with an economics background. Conclusions about investors with a background in natural sciences are confirmed by the results for investors with degrees in chemistry, physics, biology, and IT. The hazard ratios for those subjects indicate a higher probability of selling losing stocks and a lower probability of selling winning stocks. We are able to distinguish some additional subjects but most of them do not have a large enough number of observations to give statistically significant results, so for example we do not find any relevance for a psychology degree. Even so, we see that investors with a degree in law or medicine tend to be affected less by the disposition effect in our sample. One explanation for this finding could be that people graduating from law or medicine tend to have higher academic results when they are admitted to university, but, as will be seen in the next subsection, this cannot be the major reason. Very high dropout rates from medicine courses also indicate that those who graduate may possess other possibly non-cognitive traits such as persistence, which we cannot directly control for. 5.4. How Do Academic Abilities Affect the Disposition Effect? A natural question that emerges from the results presented so far is whether smarter investors are less influenced by the disposition effect. As our results show, better academic results, which should indicate higher intelligence, may reduce the disposition effect but there is no straightforward linkage. Examination results can indicate the strengths of various mental abilities because the skills and knowledge required for the examinations differ from examintion to examination. We divide investors in our sample into quartiles and deciles depending on their national high school final examination results in a particular examination. As the examination results are an important factor for university admissions for students and school rankings for high school teachers, motivation to do well in the examinations is generally high. All the investors in our sample took their final examinations at around the age of 18 years, so before they entered the stock market. Although success in those examinations requires a clear learning effort, overall intelligence along with high mental and academic abilities clearly plays an important role. Those examinations are more of a one-time effort requiring preparation and high mental abilities. Thus, we regard those examinations mostly as an indicator of intelligence and academic abilities.19 Grouping investors into quartiles or deciles (we report results for quartiles only) thus gives us the smartest investors in the top group, the fourth quartile, and the academically weakest in the bottom or first, group. Not every investor has taken all the examinations shown in Table VII but 94% took the mother tongue examination, 80% took the English language examination, and 68% took the mathematics examination. Those three examinations are the most important, as most of the university courses require two or all three of those three examinations as part of their admission criteria. Some of the natural sciences courses also require the chemistry, biology, or physics examination, and some social sciences and humanities require the history examination. There are also other possible choices for the national examinations but we present the results of three examinations for natural sciences and three examinations for humanities. Table VII. Trading losses, gains, and academic results The table presents hazard ratios associated with an individual investor’s decision to sell or hold stocks based on the investors’ national high school examination results. High school examination results are divided into deciles and quartiles for each examination, meaning that the related variable takes the value of 1 if the investors’ examination result is in a specific decile or quartile and zero otherwise. The results for six different examinations are reported. The hazard ratios together with the z-value and the level of statistical significance are reported for interacted variables. This means that each examination group variable is interacted with the TLI or TGI in order to measure cross-sectional differences in investors’ propensities to sell losers. TLI takes the value 1 if a stock is trading below its purchase price, and zero otherwise, and TGI takes the value 1 if a stock is trading above its purchase price, and zero otherwise. The trading data cover the period from 2002 to 2012; the national high school examination results data are from the period 1997 to 2012. Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Quartiles Maths examination Maths examination Mother tongue Mother tongue 1st×TLI (TGI) 0.936** −2.10 1.071* 2.19 0.998 −0.06 1.000 −0.02 2nd×TLI (TGI) 0.943* −1.88 1.058* 1.80 0.952 −1.51 1.050 1.50 3rd×TLI (TGI) 1.051 1.58 0.946* −1.78 1.011 0.46 0.993 −0.31 4th×TLI (TGI) 1.066** 2.05 0.944* −1.85 1.019 0.73 0.979 −0.79 Quartiles Chemistry examination Chemistry examination English examination English examination 1st×TLI (TGI) 0.878** −2.42 1.131** 2.30 1.009 0.30 0.995 −0.17 2nd×TLI (TGI) 0.953 −0.93 1.055 1.03 1.012 0.42 1.048 1.58 3rd×TLI (TGI) 0.923 −1.35 1.083 1.35 0.962 −1.30 0.977 −0.78 4th×TLI (TGI) 1.267*** 4.30 0.791*** −4.27 1.013 0.44 0.995 −0.15 Quartiles Physics examination Physics examination History examination History examination 1st×TLI (TGI) 0.970 −0.42 1.024 0.33 0.887*** −2.91 1.132*** 2.99 2nd×TLI (TGI) 0.952 −0.71 1.042 0.59 1.031 0.76 0.973 −0.70 3rd×TLI (TGI) 1.011 0.19 0.997 −0.06 1.126*** 2.75 0.892*** −2.67 4th×TLI (TGI) 1.086 1.09 0.923 −1.05 0.972 −0.74 1.020 0.51 Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Quartiles Maths examination Maths examination Mother tongue Mother tongue 1st×TLI (TGI) 0.936** −2.10 1.071* 2.19 0.998 −0.06 1.000 −0.02 2nd×TLI (TGI) 0.943* −1.88 1.058* 1.80 0.952 −1.51 1.050 1.50 3rd×TLI (TGI) 1.051 1.58 0.946* −1.78 1.011 0.46 0.993 −0.31 4th×TLI (TGI) 1.066** 2.05 0.944* −1.85 1.019 0.73 0.979 −0.79 Quartiles Chemistry examination Chemistry examination English examination English examination 1st×TLI (TGI) 0.878** −2.42 1.131** 2.30 1.009 0.30 0.995 −0.17 2nd×TLI (TGI) 0.953 −0.93 1.055 1.03 1.012 0.42 1.048 1.58 3rd×TLI (TGI) 0.923 −1.35 1.083 1.35 0.962 −1.30 0.977 −0.78 4th×TLI (TGI) 1.267*** 4.30 0.791*** −4.27 1.013 0.44 0.995 −0.15 Quartiles Physics examination Physics examination History examination History examination 1st×TLI (TGI) 0.970 −0.42 1.024 0.33 0.887*** −2.91 1.132*** 2.99 2nd×TLI (TGI) 0.952 −0.71 1.042 0.59 1.031 0.76 0.973 −0.70 3rd×TLI (TGI) 1.011 0.19 0.997 −0.06 1.126*** 2.75 0.892*** −2.67 4th×TLI (TGI) 1.086 1.09 0.923 −1.05 0.972 −0.74 1.020 0.51 *** Indicates the significance at the 1% level, ** 5% level, and * 10% level. Table VII. Trading losses, gains, and academic results The table presents hazard ratios associated with an individual investor’s decision to sell or hold stocks based on the investors’ national high school examination results. High school examination results are divided into deciles and quartiles for each examination, meaning that the related variable takes the value of 1 if the investors’ examination result is in a specific decile or quartile and zero otherwise. The results for six different examinations are reported. The hazard ratios together with the z-value and the level of statistical significance are reported for interacted variables. This means that each examination group variable is interacted with the TLI or TGI in order to measure cross-sectional differences in investors’ propensities to sell losers. TLI takes the value 1 if a stock is trading below its purchase price, and zero otherwise, and TGI takes the value 1 if a stock is trading above its purchase price, and zero otherwise. The trading data cover the period from 2002 to 2012; the national high school examination results data are from the period 1997 to 2012. Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Quartiles Maths examination Maths examination Mother tongue Mother tongue 1st×TLI (TGI) 0.936** −2.10 1.071* 2.19 0.998 −0.06 1.000 −0.02 2nd×TLI (TGI) 0.943* −1.88 1.058* 1.80 0.952 −1.51 1.050 1.50 3rd×TLI (TGI) 1.051 1.58 0.946* −1.78 1.011 0.46 0.993 −0.31 4th×TLI (TGI) 1.066** 2.05 0.944* −1.85 1.019 0.73 0.979 −0.79 Quartiles Chemistry examination Chemistry examination English examination English examination 1st×TLI (TGI) 0.878** −2.42 1.131** 2.30 1.009 0.30 0.995 −0.17 2nd×TLI (TGI) 0.953 −0.93 1.055 1.03 1.012 0.42 1.048 1.58 3rd×TLI (TGI) 0.923 −1.35 1.083 1.35 0.962 −1.30 0.977 −0.78 4th×TLI (TGI) 1.267*** 4.30 0.791*** −4.27 1.013 0.44 0.995 −0.15 Quartiles Physics examination Physics examination History examination History examination 1st×TLI (TGI) 0.970 −0.42 1.024 0.33 0.887*** −2.91 1.132*** 2.99 2nd×TLI (TGI) 0.952 −0.71 1.042 0.59 1.031 0.76 0.973 −0.70 3rd×TLI (TGI) 1.011 0.19 0.997 −0.06 1.126*** 2.75 0.892*** −2.67 4th×TLI (TGI) 1.086 1.09 0.923 −1.05 0.972 −0.74 1.020 0.51 Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Quartiles Maths examination Maths examination Mother tongue Mother tongue 1st×TLI (TGI) 0.936** −2.10 1.071* 2.19 0.998 −0.06 1.000 −0.02 2nd×TLI (TGI) 0.943* −1.88 1.058* 1.80 0.952 −1.51 1.050 1.50 3rd×TLI (TGI) 1.051 1.58 0.946* −1.78 1.011 0.46 0.993 −0.31 4th×TLI (TGI) 1.066** 2.05 0.944* −1.85 1.019 0.73 0.979 −0.79 Quartiles Chemistry examination Chemistry examination English examination English examination 1st×TLI (TGI) 0.878** −2.42 1.131** 2.30 1.009 0.30 0.995 −0.17 2nd×TLI (TGI) 0.953 −0.93 1.055 1.03 1.012 0.42 1.048 1.58 3rd×TLI (TGI) 0.923 −1.35 1.083 1.35 0.962 −1.30 0.977 −0.78 4th×TLI (TGI) 1.267*** 4.30 0.791*** −4.27 1.013 0.44 0.995 −0.15 Quartiles Physics examination Physics examination History examination History examination 1st×TLI (TGI) 0.970 −0.42 1.024 0.33 0.887*** −2.91 1.132*** 2.99 2nd×TLI (TGI) 0.952 −0.71 1.042 0.59 1.031 0.76 0.973 −0.70 3rd×TLI (TGI) 1.011 0.19 0.997 −0.06 1.126*** 2.75 0.892*** −2.67 4th×TLI (TGI) 1.086 1.09 0.923 −1.05 0.972 −0.74 1.020 0.51 *** Indicates the significance at the 1% level, ** 5% level, and * 10% level. We test how national high school final examination results, which we take to indicate investors’ intelligence, are related to the disposition effect. No similar tests have so far been presented in the literature due to the lack of data, though there are a number of studies about the relationships between investor sophistication and the disposition effect. The studies conclude that more sophisticated investors are less prone to the disposition effect. The examination results used do not directly measure how sophisticated the investors in our sample are, but they can act as a general proxy of intelligence, which can be connected with an investor’s sophistication as well. In general, our results for the impact of mental abilities on the disposition effect are mixed (see Table VII). The results of the survival analysis for the losing positions (TLI = 1) cover six different examinations. The picture is clearer for examinations in the natural or exact sciences. For example, investors with higher score in the mathematics examination in the fourth quartile seem to be less influenced by the disposition effect, with hazard ratios above 1 in Table VII. We also see that being less successful and in the bottom quartile in maths or chemistry seems to increase the disposition effect. We do not find statistically significant effects for the physics examination but at least for quartile results, the magnitude of the effects is the same, with hazard ratios <1 for the 1st and 2nd quartiles and above 1 for the 3rd and 4th quartiles. It should be noted that only 13% of investors for whom we have examination data took the examination in physics, and that is 8% of our total sample. The results for the softer sciences are more mixed. We show the results for the mother tongue, English, and history examinations, which are the three most popular examinations in that category. We do not find any statistically significant results by quartiles for the mother tongue and English examinations, which are the two most popular examinations. We do find some statistically significant coefficients for decile regressions but we basically end up with quite random results, which we do not consider meaningful. Being among the top performers in those examinations clearly requires high intelligence, but it might not require such an effort in preparation. It is very hard to become a top performer in those examinations only by studying hard and doing practice exercises, as it is very difficult to master the art of writing essays or speaking a foreign language without having an innate talent for it.20 We do find some statistically significant coefficients for the history examination but the indication from that is also mixed, as both doing really well and doing quite poorly in the history examination seem to decrease the disposition effect and all in all, we cannot draw any meaningful conclusions about the mental abilities that are manifest in the results of humanities examinations.21 Such effects emerge much more clearly from study of the effect of mental abilities on market participation and stock portfolio performance22 using the same proxies. Thus, we conclude that maths skills may be among the abilities that help to attenuate the disposition effects, but skills in humanities do not play an important role. The stronger effect in natural sciences is more in line with the effects that we see when we compare the effect of different subjects, where natural sciences also played a bigger role. As the top performers in most of the natural science examinations have a lower probability of being affected by the disposition effect, we conclude that certain mental abilities such as being good at numbers help in defeating the disposition effect. But we do not find support for the argument that all areas of intelligence can contribute to reducing the disposition effect. 5.5. What Affects the Speed of Learning? As we argue that the learning abilities of investors are different, we should also see the effects in our trading data. As presented in Section 5.1, learning by doing appears to take place for the investors in our sample, because the investors become less influenced by the disposition effect when they have made more trades. If the level of education and the type of education can be used as proxies for intelligence and learning abilities, we should at least see some differences in the magnitude and speed of learning by doing for different groups as well. And that is indeed the case. We run regressions with different subsamples of the level or field of education. The results (see Figure 3 and Table VIII) confirm our previous findings and show that the disposition effect fades faster for investors with higher education such as master’s or doctoral degrees than for others. The hazard showing the probability of an investor selling a stock position that is in loss shows a 32.7% higher probability for an investor who has made at least 11 trades than for one who has made fewer than 5 trades. The same figure reaches that high for investors with a bachelor’s degree only when they have made over 100 trades, meaning their speed of learning by doing is clearly slower. It would be expected that the more trades an investor has made, the less that investor is influenced by the disposition effect, and we can see such a linearly increasing trend for most groups. Table VIII. Learning by doing by education level and field of studies The table reports the results of the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if a position is in loss and the value zero otherwise. TGI represents the total gain indicator, taking the value 1 if a position is in gain and the value 0 otherwise. Trade dummies (the number of trades made) are interacted with TLI and TGI to capture the disposition effect. Control variables are omitted from the table. Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TLI 1.026 0.13 0.932 −1.33 0.964 −0.35 Number of trades: 11–20×TLI 1.327* 1.67 1.040 0.80 1.030 0.31 Number of trades: 21–50×TLI 1.475*** 2.68 1.108** 2.33 1.151 1.59 Number of trades: 51–100×TLI 2.135*** 4.49 1.201*** 3.87 1.028 0.29 Number of trades: over 100×TLI 1.095 0.60 1.253*** 5.23 1.143 1.50 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 1.161 1.60 0.890** −2.03 1.270 1.20 Number of trades: 11–20×TLI 1.066 0.72 1.096* 1.71 1.101 0.55 Number of trades: 21–50×TLI 1.223** 2.46 1.108** 2.15 1.140 0.84 Number of trades: 51–100×TLI 1.300*** 2.84 1.182*** 3.26 1.383* 1.79 Number of trades: over 100×TLI 1.171* 1.73 1.304*** 5.65 1.331** 1.97 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TGI 0.975 −0.13 1.072 1.31 1.067 0.61 Number of trades: 11 to 20×TGI 0.750* −1.70 0.956 −0.93 0.986 −0.15 Number of trades: 21 to 50×TGI 0.675*** −2.71 0.898** −2.44 0.891 −1.30 Number of trades: 51 to 100×TGI 0.469*** −4.48 0.827*** −4.04 0.996 −0.04 Number of trades: over 100×TGI 0.908 −0.65 0.795*** −5.32 0.896 −1.23 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI 0.869 −1.51 1.125** 2.05 0.788 −1.20 Number of trades: 11–20×TGI 0.948 −0.61 0.912* −1.72 0.908 −0.55 Number of trades: 21–50×TGI 0.830** −2.29 0.898** −2.25 0.874 −0.86 Number of trades: 51–100×TGI 0.777*** −2.72 0.838*** −3.45 0.710* −1.90 Number of trades: over 100×TGI 0.867 −1.56 0.762*** −5.78 0.748** −2.00 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TLI 1.026 0.13 0.932 −1.33 0.964 −0.35 Number of trades: 11–20×TLI 1.327* 1.67 1.040 0.80 1.030 0.31 Number of trades: 21–50×TLI 1.475*** 2.68 1.108** 2.33 1.151 1.59 Number of trades: 51–100×TLI 2.135*** 4.49 1.201*** 3.87 1.028 0.29 Number of trades: over 100×TLI 1.095 0.60 1.253*** 5.23 1.143 1.50 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 1.161 1.60 0.890** −2.03 1.270 1.20 Number of trades: 11–20×TLI 1.066 0.72 1.096* 1.71 1.101 0.55 Number of trades: 21–50×TLI 1.223** 2.46 1.108** 2.15 1.140 0.84 Number of trades: 51–100×TLI 1.300*** 2.84 1.182*** 3.26 1.383* 1.79 Number of trades: over 100×TLI 1.171* 1.73 1.304*** 5.65 1.331** 1.97 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TGI 0.975 −0.13 1.072 1.31 1.067 0.61 Number of trades: 11 to 20×TGI 0.750* −1.70 0.956 −0.93 0.986 −0.15 Number of trades: 21 to 50×TGI 0.675*** −2.71 0.898** −2.44 0.891 −1.30 Number of trades: 51 to 100×TGI 0.469*** −4.48 0.827*** −4.04 0.996 −0.04 Number of trades: over 100×TGI 0.908 −0.65 0.795*** −5.32 0.896 −1.23 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI 0.869 −1.51 1.125** 2.05 0.788 −1.20 Number of trades: 11–20×TGI 0.948 −0.61 0.912* −1.72 0.908 −0.55 Number of trades: 21–50×TGI 0.830** −2.29 0.898** −2.25 0.874 −0.86 Number of trades: 51–100×TGI 0.777*** −2.72 0.838*** −3.45 0.710* −1.90 Number of trades: over 100×TGI 0.867 −1.56 0.762*** −5.78 0.748** −2.00 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Table VIII. Learning by doing by education level and field of studies The table reports the results of the survival analysis—the hazard ratios, z-values, and significance levels. TLI represents the total loss indicator, taking the value 1 if a position is in loss and the value zero otherwise. TGI represents the total gain indicator, taking the value 1 if a position is in gain and the value 0 otherwise. Trade dummies (the number of trades made) are interacted with TLI and TGI to capture the disposition effect. Control variables are omitted from the table. Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TLI 1.026 0.13 0.932 −1.33 0.964 −0.35 Number of trades: 11–20×TLI 1.327* 1.67 1.040 0.80 1.030 0.31 Number of trades: 21–50×TLI 1.475*** 2.68 1.108** 2.33 1.151 1.59 Number of trades: 51–100×TLI 2.135*** 4.49 1.201*** 3.87 1.028 0.29 Number of trades: over 100×TLI 1.095 0.60 1.253*** 5.23 1.143 1.50 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 1.161 1.60 0.890** −2.03 1.270 1.20 Number of trades: 11–20×TLI 1.066 0.72 1.096* 1.71 1.101 0.55 Number of trades: 21–50×TLI 1.223** 2.46 1.108** 2.15 1.140 0.84 Number of trades: 51–100×TLI 1.300*** 2.84 1.182*** 3.26 1.383* 1.79 Number of trades: over 100×TLI 1.171* 1.73 1.304*** 5.65 1.331** 1.97 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TGI 0.975 −0.13 1.072 1.31 1.067 0.61 Number of trades: 11 to 20×TGI 0.750* −1.70 0.956 −0.93 0.986 −0.15 Number of trades: 21 to 50×TGI 0.675*** −2.71 0.898** −2.44 0.891 −1.30 Number of trades: 51 to 100×TGI 0.469*** −4.48 0.827*** −4.04 0.996 −0.04 Number of trades: over 100×TGI 0.908 −0.65 0.795*** −5.32 0.896 −1.23 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI 0.869 −1.51 1.125** 2.05 0.788 −1.20 Number of trades: 11–20×TGI 0.948 −0.61 0.912* −1.72 0.908 −0.55 Number of trades: 21–50×TGI 0.830** −2.29 0.898** −2.25 0.874 −0.86 Number of trades: 51–100×TGI 0.777*** −2.72 0.838*** −3.45 0.710* −1.90 Number of trades: over 100×TGI 0.867 −1.56 0.762*** −5.78 0.748** −2.00 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TLI 1.026 0.13 0.932 −1.33 0.964 −0.35 Number of trades: 11–20×TLI 1.327* 1.67 1.040 0.80 1.030 0.31 Number of trades: 21–50×TLI 1.475*** 2.68 1.108** 2.33 1.151 1.59 Number of trades: 51–100×TLI 2.135*** 4.49 1.201*** 3.87 1.028 0.29 Number of trades: over 100×TLI 1.095 0.60 1.253*** 5.23 1.143 1.50 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 1.161 1.60 0.890** −2.03 1.270 1.20 Number of trades: 11–20×TLI 1.066 0.72 1.096* 1.71 1.101 0.55 Number of trades: 21–50×TLI 1.223** 2.46 1.108** 2.15 1.140 0.84 Number of trades: 51–100×TLI 1.300*** 2.84 1.182*** 3.26 1.383* 1.79 Number of trades: over 100×TLI 1.171* 1.73 1.304*** 5.65 1.331** 1.97 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Hazard ratio z-Statistic Hazard ratio z-Statistic Hazard ratio z-Statistic Number of trades: 6–10×TGI 0.975 −0.13 1.072 1.31 1.067 0.61 Number of trades: 11 to 20×TGI 0.750* −1.70 0.956 −0.93 0.986 −0.15 Number of trades: 21 to 50×TGI 0.675*** −2.71 0.898** −2.44 0.891 −1.30 Number of trades: 51 to 100×TGI 0.469*** −4.48 0.827*** −4.04 0.996 −0.04 Number of trades: over 100×TGI 0.908 −0.65 0.795*** −5.32 0.896 −1.23 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI 0.869 −1.51 1.125** 2.05 0.788 −1.20 Number of trades: 11–20×TGI 0.948 −0.61 0.912* −1.72 0.908 −0.55 Number of trades: 21–50×TGI 0.830** −2.29 0.898** −2.25 0.874 −0.86 Number of trades: 51–100×TGI 0.777*** −2.72 0.838*** −3.45 0.710* −1.90 Number of trades: over 100×TGI 0.867 −1.56 0.762*** −5.78 0.748** −2.00 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. Figure 3. View largeDownload slide The speed and magnitude of learning by doing for different education groups. The graph shows the speed and magnitude of learning by doing. The X-axis shows the number of trades made (interacted trade dummy variables with TLI) and the Y-axis shows the corresponding hazard ratios. A hazard ratio over 1 (marked by the black dotted horizontal line) indicates that investors in the subgroup are less influenced by the disposition effect. The higher the hazard ratios, the less investors are influenced by the disposition effect given that they have made a particular number of trades. The slopes of the lines show the speed of learning. Figure 3. View largeDownload slide The speed and magnitude of learning by doing for different education groups. The graph shows the speed and magnitude of learning by doing. The X-axis shows the number of trades made (interacted trade dummy variables with TLI) and the Y-axis shows the corresponding hazard ratios. A hazard ratio over 1 (marked by the black dotted horizontal line) indicates that investors in the subgroup are less influenced by the disposition effect. The higher the hazard ratios, the less investors are influenced by the disposition effect given that they have made a particular number of trades. The slopes of the lines show the speed of learning. However, we observe that investors with a master’s or doctoral degree seem not to have learnt anything when they have made over 100 trades, as the corresponding hazard ratio is not statistically different from 1. The explanation for this could be that investors with high academic degrees trade less than other investors and given a smaller number of such investors, our sample just does not contain enough investors with high academic degrees to draw any conclusions about how a very large number of trades affects their bias. The same applies for the same variables for certain subject-based subsamples. We do not find any statistically significant effect of learning by doing for investors with vocational training. This supports our previous conclusion that such investors may have a lower level of intelligence. The increasing effect of learning by doing is present for investors holding a bachelor’s or equivalent degree. The picture is quite similar for different subjects of study. We do observe a statistically significant and monotonically increasing effect for investors with a background in social sciences, who clearly become less influenced by the disposition effect when they have made more trades. Given that those investors mostly hold degrees in economics and finance and surprisingly are prone to the disposition effect in general, their learning by doing effect is to be expected. We also observe a similar increasing effect for investors with a background in natural sciences, who are able to learn faster than investors with a social sciences background. This is once again consistent with our previous hypothesis that having a natural science background can be beneficial in the stock market environment. We find a statistically significant learning effect for humanities graduates only when they have made over 100 trades. We run similar regressions for all the top and bottom quartiles for different final examination results. We find a consistent effect only for the mathematics examination, where people with high skills in maths learn relatively fast by trading but investors in the bottom quartile exhibit basically no notable improvement for the disposition effect. The results for all other examinations—which indicate different types of mental abilities—do not show any other meaningful pattern or type of skillset that can be beneficial for an investor to learn faster than the average. All in all, we conclude that the learning by doing effect occurs faster for investors with higher intelligence and is almost absent for investors with lower intelligence. We find consistent effects of learning faster by doing for all the groups whom we expect to exhibit educational characteristics that help to reduce the disposition effect. 5.6. Discussion We would like to highlight the importance of intelligence and certain mental abilities, especially maths, which influence trading behavior and the speed of cognitive learning in the financial markets. When we consider cognitive learning through gaining investment experience by trading, we face a process of a Bayesian expectation revision mechanism (Grossman, Kihlstrom, and Mirman, 1977). Investors have to deal with a large amount of information and must update their beliefs accordingly. Updating beliefs means asking a large number of questions about the investment decision, which are all part of the belief revision mechanism. By repeating the process multiple times, investors gain a better ability to estimate probabilities for Bayesian inference and are thus more successful. However, people are not inherently good at estimating probabilities and the learning process is not easy. It would be safe to assume that some people are able to learn faster than others in that setting. So far the previous literature has mainly focused on various socioeconomic characteristics and different forms of learning but has partly neglected intelligence, mostly because of the lack of appropriate data. The sample that we use lets us distinguish between different mental abilities and after testing different sets of mental abilities we find that intelligence plays an important role. When we consider the constructs of “complex problem solving knowledge acquisition,” “complex problem solving system control,” “learning abilities,” and “intellectual status” [as identified by Beckmann and Guthke (1995)] as cognitive mental abilities or components of overall intelligence, we find the strongest effect is for long-term learning abilities. As we cannot use the results of standard intelligence tests, and our measures of complex problem-solving activities emerge from the academic setting such as maths and physics examiantions, we can only suggest a direction from which the effect mostly comes. We cannot rule out other possible sources, as mathematics skills played some role in our dataset as a part of complex problem solving for example, although some other components of traditional intelligence testing23 like academic results and the examination results that we could use did not become significant in affecting the disposition effect. Our results concerning the importance of intelligence do not rule out any other possible cognitive abilities and learning. Nor do our results contradict any current studies. In combination with financial market research on IQ, such as Grinblatt, Keloharju, and Linnainmaa (2012), we are able to add to the growing amount of literature on how different aspects of mental abilities and intelligence affect investment and trading decisions. If we generalize our results, intelligence be an important reason why only a small proportion of investors are able to learn fast enough to eliminate the emergence of behavioral biases in the financial markets, as learning by doing is not very efficient (Bachmann and Hens, 2015). 6. Conclusion The disposition effect is a well-researched area of finance. We are able to contribute by studying the effects of intelligence and education on the disposition effect. We conclude that educational characteristics and intelligence play an important role in affecting investor behavior. We show that the disposition effect is stronger for investors with lower intelligence by using different proxies to measure their mental abilities in the academic setting. We find a stronger effect at both ends of the education scale. Highly educated investors who have master’s or doctoral degrees are less influenced by the disposition effect and investors who have chosen vocational training instead, which may signal they are less able to cope with more challenging academic tasks than others in our sample, or have only high school education are more influenced by the disposition effect. Moreover, highly educated investors are able to learn faster by doing and by actually making trades, and investors with lower intelligence do not improve their trading even when they become more experienced. We find some beneficial effects for mathematical abilities, as better number skills can be associated with lower levels of the disposition effect. As investors with financial education are clearly affected by the disposition effect despite their supposedly better knowledge, we cannot emphasize the importance of knowledge as a policy measure when trying to avoid behavioral biases. Even good abilities in knowledge memorization, as shown by the results of certain final examinations, do not help. Thus, any one-time campaigns to increase investor awareness and knowledge are probably of low impact, but are not completely useless. Domain knowledge can help investors learn a little faster when they enter the market. However, educating investors must be seen as a long process and it will be inevitable that some investors will just not be able to learn despite the efforts made. Appendix A: Robustness Checks This appendix reports robustness checks for the results presented in the main text of the paper. The robustness checks presented here are obtained by running logit regressions for the same model setups that were used for the survival analysis. To help in comparing the results of the hazard models with the results of the robustness checks, all the tables in this Appendix use the same numbers (though in Arabic form) as the corresponding tables in the main text with the prefix “A.” Table AI. Aggregate disposition effect (in the logit regressions) The table reports the results of the logit regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if a position is in loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and zero otherwise. Investor-specific variables are interacted with TLI and TGI to capture the disposition effect. These investor-specific variables are also dummy variables and take the value 1 if an investor is identified by a specific variable (e.g., the male variable takes the value 1 if an investor is a male) and zero otherwise. Variable Coefficient z-Statistic Variable Coefficient z-Statistic Total loss indicator (TLI) −1.169*** −47.82 Total gain indicator (TGI) 1.159*** 47.47 Male×TLI 0.137*** 5.22 Male×TGI −0.144*** −5.5 Foreign investor×TLI 0.232** 2.15 Foreign investor×TGI −0.244** −2.27 Variable Coefficient z-Statistic Variable Coefficient z-Statistic Total loss indicator (TLI) −1.169*** −47.82 Total gain indicator (TGI) 1.159*** 47.47 Male×TLI 0.137*** 5.22 Male×TGI −0.144*** −5.5 Foreign investor×TLI 0.232** 2.15 Foreign investor×TGI −0.244** −2.27 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. All of the regressions in Table AI confirm exactly the results of the survival analysis regressions presented in Table II. The coefficients are slightly different but the magnitude of the effects is similar. Table AI. Aggregate disposition effect (in the logit regressions) The table reports the results of the logit regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if a position is in loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and zero otherwise. Investor-specific variables are interacted with TLI and TGI to capture the disposition effect. These investor-specific variables are also dummy variables and take the value 1 if an investor is identified by a specific variable (e.g., the male variable takes the value 1 if an investor is a male) and zero otherwise. Variable Coefficient z-Statistic Variable Coefficient z-Statistic Total loss indicator (TLI) −1.169*** −47.82 Total gain indicator (TGI) 1.159*** 47.47 Male×TLI 0.137*** 5.22 Male×TGI −0.144*** −5.5 Foreign investor×TLI 0.232** 2.15 Foreign investor×TGI −0.244** −2.27 Variable Coefficient z-Statistic Variable Coefficient z-Statistic Total loss indicator (TLI) −1.169*** −47.82 Total gain indicator (TGI) 1.159*** 47.47 Male×TLI 0.137*** 5.22 Male×TGI −0.144*** −5.5 Foreign investor×TLI 0.232** 2.15 Foreign investor×TGI −0.244** −2.27 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. All of the regressions in Table AI confirm exactly the results of the survival analysis regressions presented in Table II. The coefficients are slightly different but the magnitude of the effects is similar. Table AII. The effects of “learning by doing” (in the logit regressions) The table reports results from the logit regressions—the coefficients, z-values, and significance levels. TLI taking the value 1 if the position is in loss and the value 0 otherwise. TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and zero otherwise. The results are reported only for the variables showing experience (how many trades an investor has made) and for interacted variables. Experience-related variables are interacted with TLI and TGI to capture the disposition effect. Experience-related variables are dummy variables. The trading data cover the period from 2002 to 2012. The table summarizes the results of individually run regressions. Variables Coefficient z-Statistic Variables Coefficient z-Statistic Number of trades: 1–5×TLI 0.041 1.40 Number of trades: 1–5×TGI −0.046 −1.56 Number of trades: 1–5 −0.694*** −33.42 Number of trades: 1–5 −0.653*** 0.02 Number of trades: 6–10×TLI −0.046 −1.52 Number of trades: 6–10×TGI 0.053* 1.76 Number of trades: 6–10 −0.704*** −35.91 Number of trades: 6–10 −0.756*** 0.02 Number of trades: 11–20×TLI 0.011 0.42 Number of trades: 11–20×TGI −0.019 −0.74 Number of trades: 11–20 −0.537*** −33.50 Number of trades: 11–20 −0.521*** −25.59 Number of trades: 21–50×TLI 0.208*** 9.88 Number of trades: 21–50×TGI −0.207*** −9.85 Number of trades: 21–50 −0.242*** −18.93 Number of trades: 21–50 −0.034** −2.01 Number of trades: 51–100×TLI 0.180*** 7.09 Number of trades: 51–100×TGI −0.181*** −7.15 Number of trades: 51–100 0.420*** 27.47 Number of trades: 51–100 0.601*** 29.84 Number of trades: over 100×TLI 0.393*** 19.27 Number of trades: over 100×TGI −0.39*** −19.15 Number of trades: over 100 1.616*** 132.16 Number of trades: over 100 2.008*** 123.44 Variables Coefficient z-Statistic Variables Coefficient z-Statistic Number of trades: 1–5×TLI 0.041 1.40 Number of trades: 1–5×TGI −0.046 −1.56 Number of trades: 1–5 −0.694*** −33.42 Number of trades: 1–5 −0.653*** 0.02 Number of trades: 6–10×TLI −0.046 −1.52 Number of trades: 6–10×TGI 0.053* 1.76 Number of trades: 6–10 −0.704*** −35.91 Number of trades: 6–10 −0.756*** 0.02 Number of trades: 11–20×TLI 0.011 0.42 Number of trades: 11–20×TGI −0.019 −0.74 Number of trades: 11–20 −0.537*** −33.50 Number of trades: 11–20 −0.521*** −25.59 Number of trades: 21–50×TLI 0.208*** 9.88 Number of trades: 21–50×TGI −0.207*** −9.85 Number of trades: 21–50 −0.242*** −18.93 Number of trades: 21–50 −0.034** −2.01 Number of trades: 51–100×TLI 0.180*** 7.09 Number of trades: 51–100×TGI −0.181*** −7.15 Number of trades: 51–100 0.420*** 27.47 Number of trades: 51–100 0.601*** 29.84 Number of trades: over 100×TLI 0.393*** 19.27 Number of trades: over 100×TGI −0.39*** −19.15 Number of trades: over 100 1.616*** 132.16 Number of trades: over 100 2.008*** 123.44 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. All of the regressions in Table AII confirm exactly the results of the survival analysis regressions presented in Table III. The results of survival analysis are statistically more significant and clearer for some regressions (e.g., Number of trades: 11–20 × TLI). Table AII. The effects of “learning by doing” (in the logit regressions) The table reports results from the logit regressions—the coefficients, z-values, and significance levels. TLI taking the value 1 if the position is in loss and the value 0 otherwise. TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and zero otherwise. The results are reported only for the variables showing experience (how many trades an investor has made) and for interacted variables. Experience-related variables are interacted with TLI and TGI to capture the disposition effect. Experience-related variables are dummy variables. The trading data cover the period from 2002 to 2012. The table summarizes the results of individually run regressions. Variables Coefficient z-Statistic Variables Coefficient z-Statistic Number of trades: 1–5×TLI 0.041 1.40 Number of trades: 1–5×TGI −0.046 −1.56 Number of trades: 1–5 −0.694*** −33.42 Number of trades: 1–5 −0.653*** 0.02 Number of trades: 6–10×TLI −0.046 −1.52 Number of trades: 6–10×TGI 0.053* 1.76 Number of trades: 6–10 −0.704*** −35.91 Number of trades: 6–10 −0.756*** 0.02 Number of trades: 11–20×TLI 0.011 0.42 Number of trades: 11–20×TGI −0.019 −0.74 Number of trades: 11–20 −0.537*** −33.50 Number of trades: 11–20 −0.521*** −25.59 Number of trades: 21–50×TLI 0.208*** 9.88 Number of trades: 21–50×TGI −0.207*** −9.85 Number of trades: 21–50 −0.242*** −18.93 Number of trades: 21–50 −0.034** −2.01 Number of trades: 51–100×TLI 0.180*** 7.09 Number of trades: 51–100×TGI −0.181*** −7.15 Number of trades: 51–100 0.420*** 27.47 Number of trades: 51–100 0.601*** 29.84 Number of trades: over 100×TLI 0.393*** 19.27 Number of trades: over 100×TGI −0.39*** −19.15 Number of trades: over 100 1.616*** 132.16 Number of trades: over 100 2.008*** 123.44 Variables Coefficient z-Statistic Variables Coefficient z-Statistic Number of trades: 1–5×TLI 0.041 1.40 Number of trades: 1–5×TGI −0.046 −1.56 Number of trades: 1–5 −0.694*** −33.42 Number of trades: 1–5 −0.653*** 0.02 Number of trades: 6–10×TLI −0.046 −1.52 Number of trades: 6–10×TGI 0.053* 1.76 Number of trades: 6–10 −0.704*** −35.91 Number of trades: 6–10 −0.756*** 0.02 Number of trades: 11–20×TLI 0.011 0.42 Number of trades: 11–20×TGI −0.019 −0.74 Number of trades: 11–20 −0.537*** −33.50 Number of trades: 11–20 −0.521*** −25.59 Number of trades: 21–50×TLI 0.208*** 9.88 Number of trades: 21–50×TGI −0.207*** −9.85 Number of trades: 21–50 −0.242*** −18.93 Number of trades: 21–50 −0.034** −2.01 Number of trades: 51–100×TLI 0.180*** 7.09 Number of trades: 51–100×TGI −0.181*** −7.15 Number of trades: 51–100 0.420*** 27.47 Number of trades: 51–100 0.601*** 29.84 Number of trades: over 100×TLI 0.393*** 19.27 Number of trades: over 100×TGI −0.39*** −19.15 Number of trades: over 100 1.616*** 132.16 Number of trades: over 100 2.008*** 123.44 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. All of the regressions in Table AII confirm exactly the results of the survival analysis regressions presented in Table III. The results of survival analysis are statistically more significant and clearer for some regressions (e.g., Number of trades: 11–20 × TLI). Table AIII. Education level and disposition effect (in the logit and OLS regressions) The upper panel of the table reports results from the logit regressions and the lower panel of the OLS regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if the position is in loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and 0 otherwise. Results are reported only for education level variables and for interacted variables in the regressions. Education level variables are interacted with TLI and TGI to capture the disposition effect. Education level variables are also dummy variables. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Logit regressions Variables Coefficient z-Statistic Variables Coefficient z-Statistic Vocational training×TLI −0.032 −1.25 Vocational training×TGI 0.021 0.84 Vocational training 0.106*** 6.64 Vocational training 0.080*** 4.02 High school×TLI 0.100*** 4.37 High school×TGI −0.098*** −4.29 High school 0.166*** 11.38 High school 0.264*** 15.04 Bachelor’s degree×TLI −0.040** −2.18 Bachelor’s degree×TGI 0.042** 2.30 Bachelor’s degree −0.046*** −4.02 Bachelor’s degree −0.087 −6.06 Master’s or doctoral degree×TLI −0.0670 −1.61 Master’s or doctoral degree×TGI 0.077* 1.86 Master’s or doctoral degree −0.415*** −16.47 Master’s or doctoral degree −0.488*** −14.86 OLS regressions Variables Coefficient t-statistic Variables Coefficient t-statistic Vocational training×TLI −0.00053*** −5.37 Vocational training×TGI 0.00048*** 4.87 Vocational training 0.00069*** 8.72 Vocational training 0.00018*** 3.11 High school×TLI −0.00047*** −5.20 High school×TGI 0.00047*** 5.12 High school 0.00110*** 14.95 High school 0.00063*** 11.67 Bachelor’s degree×TLI 0.00010 1.50 Bachelor’s degree×TGI −0.00009 −1.30 Bachelor’s degree −0.00029*** −5.28 Bachelor’s degree −0.00019*** −4.69 Master’s or doctoral degree×TLI 0.00133*** 10.68 Master’s or doctoral degree×TGI −0.00129*** −10.30 Master’s or doctoral degree −0.00218*** −21.77 Master’s or doctoral degree −0.00086*** −11.61 Logit regressions Variables Coefficient z-Statistic Variables Coefficient z-Statistic Vocational training×TLI −0.032 −1.25 Vocational training×TGI 0.021 0.84 Vocational training 0.106*** 6.64 Vocational training 0.080*** 4.02 High school×TLI 0.100*** 4.37 High school×TGI −0.098*** −4.29 High school 0.166*** 11.38 High school 0.264*** 15.04 Bachelor’s degree×TLI −0.040** −2.18 Bachelor’s degree×TGI 0.042** 2.30 Bachelor’s degree −0.046*** −4.02 Bachelor’s degree −0.087 −6.06 Master’s or doctoral degree×TLI −0.0670 −1.61 Master’s or doctoral degree×TGI 0.077* 1.86 Master’s or doctoral degree −0.415*** −16.47 Master’s or doctoral degree −0.488*** −14.86 OLS regressions Variables Coefficient t-statistic Variables Coefficient t-statistic Vocational training×TLI −0.00053*** −5.37 Vocational training×TGI 0.00048*** 4.87 Vocational training 0.00069*** 8.72 Vocational training 0.00018*** 3.11 High school×TLI −0.00047*** −5.20 High school×TGI 0.00047*** 5.12 High school 0.00110*** 14.95 High school 0.00063*** 11.67 Bachelor’s degree×TLI 0.00010 1.50 Bachelor’s degree×TGI −0.00009 −1.30 Bachelor’s degree −0.00029*** −5.28 Bachelor’s degree −0.00019*** −4.69 Master’s or doctoral degree×TLI 0.00133*** 10.68 Master’s or doctoral degree×TGI −0.00129*** −10.30 Master’s or doctoral degree −0.00218*** −21.77 Master’s or doctoral degree −0.00086*** −11.61 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. For the results shown in Table AIII, which are comparable to those in Table IV, we run the logit regressions as well as OLS regressions for the robustness checks. Given the signs of the coefficients in the logit regressions, the results support survival analysis results in most of the cases. The only difference comes in the variables for Master’s or Doctoral degrees. To address this issue we also run OLS regressions, which confirm our survival results. In addition, when we consider the statistical significance of the variables connected with the Master’s or Doctoral degree and also vocational training, the OLS regressions results support our survival analysis results. The problem with logit regressions is that with the setup where we use the level of education as an explanatory variable, the regressions do not converge in some cases (we report the results obtained with just a small number of iterations to obtain the coefficients) and thus the results obtained are not reliable. This unreliability of the logit regressions due to convergence issues is the reason why we also employ OLS regressions. The OLS regressions match the results of the survival analysis even in cases when the results of logit regressions differed (e.g., Master’s or doctoral degree × TLI). Table AIII. Education level and disposition effect (in the logit and OLS regressions) The upper panel of the table reports results from the logit regressions and the lower panel of the OLS regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if the position is in loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and 0 otherwise. Results are reported only for education level variables and for interacted variables in the regressions. Education level variables are interacted with TLI and TGI to capture the disposition effect. Education level variables are also dummy variables. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Logit regressions Variables Coefficient z-Statistic Variables Coefficient z-Statistic Vocational training×TLI −0.032 −1.25 Vocational training×TGI 0.021 0.84 Vocational training 0.106*** 6.64 Vocational training 0.080*** 4.02 High school×TLI 0.100*** 4.37 High school×TGI −0.098*** −4.29 High school 0.166*** 11.38 High school 0.264*** 15.04 Bachelor’s degree×TLI −0.040** −2.18 Bachelor’s degree×TGI 0.042** 2.30 Bachelor’s degree −0.046*** −4.02 Bachelor’s degree −0.087 −6.06 Master’s or doctoral degree×TLI −0.0670 −1.61 Master’s or doctoral degree×TGI 0.077* 1.86 Master’s or doctoral degree −0.415*** −16.47 Master’s or doctoral degree −0.488*** −14.86 OLS regressions Variables Coefficient t-statistic Variables Coefficient t-statistic Vocational training×TLI −0.00053*** −5.37 Vocational training×TGI 0.00048*** 4.87 Vocational training 0.00069*** 8.72 Vocational training 0.00018*** 3.11 High school×TLI −0.00047*** −5.20 High school×TGI 0.00047*** 5.12 High school 0.00110*** 14.95 High school 0.00063*** 11.67 Bachelor’s degree×TLI 0.00010 1.50 Bachelor’s degree×TGI −0.00009 −1.30 Bachelor’s degree −0.00029*** −5.28 Bachelor’s degree −0.00019*** −4.69 Master’s or doctoral degree×TLI 0.00133*** 10.68 Master’s or doctoral degree×TGI −0.00129*** −10.30 Master’s or doctoral degree −0.00218*** −21.77 Master’s or doctoral degree −0.00086*** −11.61 Logit regressions Variables Coefficient z-Statistic Variables Coefficient z-Statistic Vocational training×TLI −0.032 −1.25 Vocational training×TGI 0.021 0.84 Vocational training 0.106*** 6.64 Vocational training 0.080*** 4.02 High school×TLI 0.100*** 4.37 High school×TGI −0.098*** −4.29 High school 0.166*** 11.38 High school 0.264*** 15.04 Bachelor’s degree×TLI −0.040** −2.18 Bachelor’s degree×TGI 0.042** 2.30 Bachelor’s degree −0.046*** −4.02 Bachelor’s degree −0.087 −6.06 Master’s or doctoral degree×TLI −0.0670 −1.61 Master’s or doctoral degree×TGI 0.077* 1.86 Master’s or doctoral degree −0.415*** −16.47 Master’s or doctoral degree −0.488*** −14.86 OLS regressions Variables Coefficient t-statistic Variables Coefficient t-statistic Vocational training×TLI −0.00053*** −5.37 Vocational training×TGI 0.00048*** 4.87 Vocational training 0.00069*** 8.72 Vocational training 0.00018*** 3.11 High school×TLI −0.00047*** −5.20 High school×TGI 0.00047*** 5.12 High school 0.00110*** 14.95 High school 0.00063*** 11.67 Bachelor’s degree×TLI 0.00010 1.50 Bachelor’s degree×TGI −0.00009 −1.30 Bachelor’s degree −0.00029*** −5.28 Bachelor’s degree −0.00019*** −4.69 Master’s or doctoral degree×TLI 0.00133*** 10.68 Master’s or doctoral degree×TGI −0.00129*** −10.30 Master’s or doctoral degree −0.00218*** −21.77 Master’s or doctoral degree −0.00086*** −11.61 *** Significant at the 1% level; ** significant at the 5% level; * significant at the 10% level. For the results shown in Table AIII, which are comparable to those in Table IV, we run the logit regressions as well as OLS regressions for the robustness checks. Given the signs of the coefficients in the logit regressions, the results support survival analysis results in most of the cases. The only difference comes in the variables for Master’s or Doctoral degrees. To address this issue we also run OLS regressions, which confirm our survival results. In addition, when we consider the statistical significance of the variables connected with the Master’s or Doctoral degree and also vocational training, the OLS regressions results support our survival analysis results. The problem with logit regressions is that with the setup where we use the level of education as an explanatory variable, the regressions do not converge in some cases (we report the results obtained with just a small number of iterations to obtain the coefficients) and thus the results obtained are not reliable. This unreliability of the logit regressions due to convergence issues is the reason why we also employ OLS regressions. The OLS regressions match the results of the survival analysis even in cases when the results of logit regressions differed (e.g., Master’s or doctoral degree × TLI). Table AIV. Disposition effect for specialty groups (in the logit regressions) The table reports results from the logit regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if the position is trading with a loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and 0 otherwise. The results are reported for course group variables and for interacted variables. Course groups are interacted with TLI and TGI to capture the disposition effect. Course group variables are also dummy variables, which take the value 1 if an investor has a degree in a specific discipline and the value 0 otherwise. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Variables Coefficient z-Statistic Variables Coefficient z-Statistic Humanities×TLI −0.336*** −8.72 Humanities×TGI 0.343*** 8.91 Humanities 0.221*** 9.91 Humanities −0.119*** −3.80 Natural science×TLI 0.071*** 2.71 Natural science×TGI −0.076*** −2.91 Natural science −0.037** −2.26 Natural science 0.037* 1.82 Social science×TLI −0.025 −1.40 Social science×TGI 0.027 1.50 Social science −0.118*** −10.74 Social science −0.144*** −10.27 Variables Coefficient z-Statistic Variables Coefficient z-Statistic Humanities×TLI −0.336*** −8.72 Humanities×TGI 0.343*** 8.91 Humanities 0.221*** 9.91 Humanities −0.119*** −3.80 Natural science×TLI 0.071*** 2.71 Natural science×TGI −0.076*** −2.91 Natural science −0.037** −2.26 Natural science 0.037* 1.82 Social science×TLI −0.025 −1.40 Social science×TGI 0.027 1.50 Social science −0.118*** −10.74 Social science −0.144*** −10.27 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. All of the regressions in Table AIV confirm exactly the results of the survival analysis presented in Table V. All of the signs and the statistical significance of the coefficients correspond to the hazard ratios reported in Table V. Table AIV. Disposition effect for specialty groups (in the logit regressions) The table reports results from the logit regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if the position is trading with a loss and the value 0 otherwise. Similarly, TGI is the total gain indicator, which takes the value 1 if a stock is trading above its purchase price and 0 otherwise. The results are reported for course group variables and for interacted variables. Course groups are interacted with TLI and TGI to capture the disposition effect. Course group variables are also dummy variables, which take the value 1 if an investor has a degree in a specific discipline and the value 0 otherwise. Trading data cover the period from 2002 to 2012; educational data are taken as of 2012. Variables Coefficient z-Statistic Variables Coefficient z-Statistic Humanities×TLI −0.336*** −8.72 Humanities×TGI 0.343*** 8.91 Humanities 0.221*** 9.91 Humanities −0.119*** −3.80 Natural science×TLI 0.071*** 2.71 Natural science×TGI −0.076*** −2.91 Natural science −0.037** −2.26 Natural science 0.037* 1.82 Social science×TLI −0.025 −1.40 Social science×TGI 0.027 1.50 Social science −0.118*** −10.74 Social science −0.144*** −10.27 Variables Coefficient z-Statistic Variables Coefficient z-Statistic Humanities×TLI −0.336*** −8.72 Humanities×TGI 0.343*** 8.91 Humanities 0.221*** 9.91 Humanities −0.119*** −3.80 Natural science×TLI 0.071*** 2.71 Natural science×TGI −0.076*** −2.91 Natural science −0.037** −2.26 Natural science 0.037* 1.82 Social science×TLI −0.025 −1.40 Social science×TGI 0.027 1.50 Social science −0.118*** −10.74 Social science −0.144*** −10.27 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. All of the regressions in Table AIV confirm exactly the results of the survival analysis presented in Table V. All of the signs and the statistical significance of the coefficients correspond to the hazard ratios reported in Table V. Table AV. University degrees and the disposition effect (in the logit regressions) The table presents coefficients associated with an individual investor’s decision to sell or hold stocks at a loss or gain based on the investor’s degree. The coefficients together with the z-value and the level of statistical significance are reported for subject group variables and also for interacted variables. This means that we interact each subject variable with the TLI and with the TGI in order to measure cross-sectional differences in investors’ propensities to sell losers and winners. TLI takes the value 1 if a stock is trading below its purchase price, and 0 otherwise. Similarly, the TGI variable takes the value 1 if a stock is trading above its purchase prize and 0 otherwise. The trading data cover the period from 2002 until 2012; educational data are taken as of 2012. Variables Coefficient z-Statistic Variables Coefficient z-Statistic Maths or statistics×TLI 0.374** 2.55 Maths or statistics×TGI −0.358** −2.44 Maths or statistics −0.408*** −4.26 Maths or statistics −0.0440 −0.40 Chemistry, physics, or biology×TLI 0.026 0.33 Chemistry, physics, or biology×TGI −0.0330 −0.43 Chemistry, physics, or biology −0.030 −0.64 Chemistry, physics, or biology 0.0010 0.01 IT×TLI 0.069** 2.18 IT×TGI −0.072** −2.28 IT 0.253*** 13.76 IT 0.325*** 12.72 Economics related×TLI −0.065*** −3.17 Economics related×TGI 0.071*** 3.450 Economics related −0.17*** −13.69 Economics related −0.238*** −14.650 Finance×TLI −0.385*** −6.32 Finance×TGI 0.395*** 6.48 Finance 0.003 0.09 Finance −0.388*** −7.76 Law×TLI 0.121*** 2.92 Law×TGI −0.117*** −2.83 Law −0.152 −5.82 Law −0.0330 −1.04 Medicine×TLI 0.058 0.62 Medicine×TGI −0.0590 −0.63 Medicine −0.497*** −8.75 Medicine −0.439*** −5.98 Variables Coefficient z-Statistic Variables Coefficient z-Statistic Maths or statistics×TLI 0.374** 2.55 Maths or statistics×TGI −0.358** −2.44 Maths or statistics −0.408*** −4.26 Maths or statistics −0.0440 −0.40 Chemistry, physics, or biology×TLI 0.026 0.33 Chemistry, physics, or biology×TGI −0.0330 −0.43 Chemistry, physics, or biology −0.030 −0.64 Chemistry, physics, or biology 0.0010 0.01 IT×TLI 0.069** 2.18 IT×TGI −0.072** −2.28 IT 0.253*** 13.76 IT 0.325*** 12.72 Economics related×TLI −0.065*** −3.17 Economics related×TGI 0.071*** 3.450 Economics related −0.17*** −13.69 Economics related −0.238*** −14.650 Finance×TLI −0.385*** −6.32 Finance×TGI 0.395*** 6.48 Finance 0.003 0.09 Finance −0.388*** −7.76 Law×TLI 0.121*** 2.92 Law×TGI −0.117*** −2.83 Law −0.152 −5.82 Law −0.0330 −1.04 Medicine×TLI 0.058 0.62 Medicine×TGI −0.0590 −0.63 Medicine −0.497*** −8.75 Medicine −0.439*** −5.98 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. The regressions in Table AV confirm the results of the survival analysis results presented in Table VI. The only difference is that the results of logit regressions show that the investors with a mathematics background (variable Maths or Statistics × TLI/TGI) are less affected by the disposition effect, which confirms our conclusion that being good at maths helps to reduce the disposition effect. The results of the survival analysis for the effect of the same variable were statistically not significant and had an unexpected sign. However, logit regression results do not show a statistically significant coefficient (variable Chemistry, Physics, or Biology × TLI/TGI) for the investors with a chemistry or physics or biology background, which the results for the survival analysis did, although the sign is the same as for the results of the survival analysis. Table AV. University degrees and the disposition effect (in the logit regressions) The table presents coefficients associated with an individual investor’s decision to sell or hold stocks at a loss or gain based on the investor’s degree. The coefficients together with the z-value and the level of statistical significance are reported for subject group variables and also for interacted variables. This means that we interact each subject variable with the TLI and with the TGI in order to measure cross-sectional differences in investors’ propensities to sell losers and winners. TLI takes the value 1 if a stock is trading below its purchase price, and 0 otherwise. Similarly, the TGI variable takes the value 1 if a stock is trading above its purchase prize and 0 otherwise. The trading data cover the period from 2002 until 2012; educational data are taken as of 2012. Variables Coefficient z-Statistic Variables Coefficient z-Statistic Maths or statistics×TLI 0.374** 2.55 Maths or statistics×TGI −0.358** −2.44 Maths or statistics −0.408*** −4.26 Maths or statistics −0.0440 −0.40 Chemistry, physics, or biology×TLI 0.026 0.33 Chemistry, physics, or biology×TGI −0.0330 −0.43 Chemistry, physics, or biology −0.030 −0.64 Chemistry, physics, or biology 0.0010 0.01 IT×TLI 0.069** 2.18 IT×TGI −0.072** −2.28 IT 0.253*** 13.76 IT 0.325*** 12.72 Economics related×TLI −0.065*** −3.17 Economics related×TGI 0.071*** 3.450 Economics related −0.17*** −13.69 Economics related −0.238*** −14.650 Finance×TLI −0.385*** −6.32 Finance×TGI 0.395*** 6.48 Finance 0.003 0.09 Finance −0.388*** −7.76 Law×TLI 0.121*** 2.92 Law×TGI −0.117*** −2.83 Law −0.152 −5.82 Law −0.0330 −1.04 Medicine×TLI 0.058 0.62 Medicine×TGI −0.0590 −0.63 Medicine −0.497*** −8.75 Medicine −0.439*** −5.98 Variables Coefficient z-Statistic Variables Coefficient z-Statistic Maths or statistics×TLI 0.374** 2.55 Maths or statistics×TGI −0.358** −2.44 Maths or statistics −0.408*** −4.26 Maths or statistics −0.0440 −0.40 Chemistry, physics, or biology×TLI 0.026 0.33 Chemistry, physics, or biology×TGI −0.0330 −0.43 Chemistry, physics, or biology −0.030 −0.64 Chemistry, physics, or biology 0.0010 0.01 IT×TLI 0.069** 2.18 IT×TGI −0.072** −2.28 IT 0.253*** 13.76 IT 0.325*** 12.72 Economics related×TLI −0.065*** −3.17 Economics related×TGI 0.071*** 3.450 Economics related −0.17*** −13.69 Economics related −0.238*** −14.650 Finance×TLI −0.385*** −6.32 Finance×TGI 0.395*** 6.48 Finance 0.003 0.09 Finance −0.388*** −7.76 Law×TLI 0.121*** 2.92 Law×TGI −0.117*** −2.83 Law −0.152 −5.82 Law −0.0330 −1.04 Medicine×TLI 0.058 0.62 Medicine×TGI −0.0590 −0.63 Medicine −0.497*** −8.75 Medicine −0.439*** −5.98 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. The regressions in Table AV confirm the results of the survival analysis results presented in Table VI. The only difference is that the results of logit regressions show that the investors with a mathematics background (variable Maths or Statistics × TLI/TGI) are less affected by the disposition effect, which confirms our conclusion that being good at maths helps to reduce the disposition effect. The results of the survival analysis for the effect of the same variable were statistically not significant and had an unexpected sign. However, logit regression results do not show a statistically significant coefficient (variable Chemistry, Physics, or Biology × TLI/TGI) for the investors with a chemistry or physics or biology background, which the results for the survival analysis did, although the sign is the same as for the results of the survival analysis. Table AVI. Trading losses, gains, and academic results (in the logit regressions) The table presents coefficients associated with an individual investor’s decision to sell or hold stocks and the investors’ national high school examination results. High school examination results are divided into deciles and quartiles for each examination, meaning that the related variable takes the value of 1 if the investor’s examination result is in a specific decile or quartile and zero otherwise. The results for two different examinations are reported. The coefficients together with the z-value and the level of statistical significance are reported for interacted variables. This means that each examination group variable is interacted with the TLI or TGI in order to measure cross-sectional differences in investors’ propensities to sell losers. TLI takes the value 1 if a stock is trading below its purchase price, and 0 otherwise, and TGI takes the value 1 if a stock is trading above its purchase price, and 0 otherwise. The trading data cover the period from 2002 to 2012; the national high school examination results data are from the period 1997 to 2012. Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Quartiles Maths examination Chemistry examination 1st×TLI (TGI) −0.018 −0.59 0.024 0.77 0.011 0.05 −0.019 −0.35 2nd×TLI (TGI) 0.030 0.96 −0.038 −1.22 −0.160*** −3.06 0.166*** 3.18 3rd×TLI (TGI) −0.070** −2.22 0.062** 1.98 −0.080 −1.35 0.080 1.36 4th×TLI (TGI) 0.037 1.21 −0.028 −0.9 0.205*** 3.73 −0.203*** −3.7 Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Quartiles Maths examination Chemistry examination 1st×TLI (TGI) −0.018 −0.59 0.024 0.77 0.011 0.05 −0.019 −0.35 2nd×TLI (TGI) 0.030 0.96 −0.038 −1.22 −0.160*** −3.06 0.166*** 3.18 3rd×TLI (TGI) −0.070** −2.22 0.062** 1.98 −0.080 −1.35 0.080 1.36 4th×TLI (TGI) 0.037 1.21 −0.028 −0.9 0.205*** 3.73 −0.203*** −3.7 *** Indicates the significance at the 1% level, ** 5% level, and * 10% level. Logit regressions with variables related to examination results mainly confirm our survival analysis results presented in Table VII. We present the results of the robustness checks of the mathematics and chemistry examination in Table AVI as they affect the disposition effect the most in our survival analysis regressions. Chemistry examination results are clearly in line with the conclusion that the natural science-related thinking and the maths-related skills are beneficial for reducing the disposition effect. The maths examination results are more mixed but the coefficients obtained for the bottom and top quartiles have the expected signs that correspond to the results of the survival analysis, though the coefficients are not statistically significant. The coefficients of the results of the other examinations, with the small exception of the history examination, turned out to be not statistically significant in the logit regressions either. Table AVI. Trading losses, gains, and academic results (in the logit regressions) The table presents coefficients associated with an individual investor’s decision to sell or hold stocks and the investors’ national high school examination results. High school examination results are divided into deciles and quartiles for each examination, meaning that the related variable takes the value of 1 if the investor’s examination result is in a specific decile or quartile and zero otherwise. The results for two different examinations are reported. The coefficients together with the z-value and the level of statistical significance are reported for interacted variables. This means that each examination group variable is interacted with the TLI or TGI in order to measure cross-sectional differences in investors’ propensities to sell losers. TLI takes the value 1 if a stock is trading below its purchase price, and 0 otherwise, and TGI takes the value 1 if a stock is trading above its purchase price, and 0 otherwise. The trading data cover the period from 2002 to 2012; the national high school examination results data are from the period 1997 to 2012. Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Quartiles Maths examination Chemistry examination 1st×TLI (TGI) −0.018 −0.59 0.024 0.77 0.011 0.05 −0.019 −0.35 2nd×TLI (TGI) 0.030 0.96 −0.038 −1.22 −0.160*** −3.06 0.166*** 3.18 3rd×TLI (TGI) −0.070** −2.22 0.062** 1.98 −0.080 −1.35 0.080 1.36 4th×TLI (TGI) 0.037 1.21 −0.028 −0.9 0.205*** 3.73 −0.203*** −3.7 Regressions with TLI Regressions with TGI Regressions with TLI Regressions with TGI Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Quartiles Maths examination Chemistry examination 1st×TLI (TGI) −0.018 −0.59 0.024 0.77 0.011 0.05 −0.019 −0.35 2nd×TLI (TGI) 0.030 0.96 −0.038 −1.22 −0.160*** −3.06 0.166*** 3.18 3rd×TLI (TGI) −0.070** −2.22 0.062** 1.98 −0.080 −1.35 0.080 1.36 4th×TLI (TGI) 0.037 1.21 −0.028 −0.9 0.205*** 3.73 −0.203*** −3.7 *** Indicates the significance at the 1% level, ** 5% level, and * 10% level. Logit regressions with variables related to examination results mainly confirm our survival analysis results presented in Table VII. We present the results of the robustness checks of the mathematics and chemistry examination in Table AVI as they affect the disposition effect the most in our survival analysis regressions. Chemistry examination results are clearly in line with the conclusion that the natural science-related thinking and the maths-related skills are beneficial for reducing the disposition effect. The maths examination results are more mixed but the coefficients obtained for the bottom and top quartiles have the expected signs that correspond to the results of the survival analysis, though the coefficients are not statistically significant. The coefficients of the results of the other examinations, with the small exception of the history examination, turned out to be not statistically significant in the logit regressions either. Table AVII. Learning by doing education level and field of studies (in the logit regressions) The table reports the results of the logit regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if a position is in loss and the value 0 otherwise. TGI represents the total gain indicator, taking the value 1 if a position is in gain and the value 0 otherwise. Trade dummies (the number of trades made) are interacted with TLI and TGI to capture the disposition effect. Control variables are omitted from the table. Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TLI −0.078 −0.40 −0.165*** −3.10 −0.053 −0.51 Number of trades: 11–20×TLI −0.078 −0.47 −0.144*** −2.95 −0.071 −0.74 Number of trades: 21–50×TLI 0.328** 2.28 0.068 1.54 −0.026 −0.29 Number of trades: 51–100×TLI 0.271* 1.62 0.205*** 4.34 −0.400*** −4.22 Number of trades: over 100×TLI 0.300** 2.02 0.347*** 8.05 0.383*** 4.29 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 0.018 0.13 −0.162*** −2.81 −0.015 −0.07 Number of trades: 11–20×TLI 0.010 0.09 −0.147*** −2.76 0.184 1.06 Number of trades: 21–50×TLI 0.254** 2.24 0.051 1.07 0.010 0.06 Number of trades: 51–100×TLI 0.109 0.92 0.146*** 2.86 0.257 1.43 Number of trades: over 100×TLI 0.566*** 5.14 0.398*** 8.46 0.276** 1.93 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TGI 0.078 0.40 0.165*** 3.10 0.096 0.91 Number of trades: 11–20×TGI 0.052 0.31 0.132*** 2.70 0.083 0.87 Number of trades: 21–50×TGI −0.333** −2.32 −0.077* −1.75 0.058 0.65 Number of trades: 51–100×TGI −0.271* −1.62 −0.216*** −4.58 0.436*** 4.59 Number of trades: over 100×TGI −0.307** −2.06 −0.354*** −8.22 −0.349*** −3.91 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI −0.043 −0.32 0.166*** 2.89 0.015 0.07 Number of trades: 11–20×TGI −0.047 −0.38 0.142*** 2.66 −0.183 −1.06 Number of trades: 21–50×TGI −0.287** −2.53 −0.058 −1.20 −0.024 −0.15 Number of trades: 51–100×TGI −0.121 −1.02 −0.158*** −3.09 −0.287 −1.60 Number of trades: over 100×TGI −0.576*** −5.23 −0.407*** −8.65 −0.286** −2.00 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TLI −0.078 −0.40 −0.165*** −3.10 −0.053 −0.51 Number of trades: 11–20×TLI −0.078 −0.47 −0.144*** −2.95 −0.071 −0.74 Number of trades: 21–50×TLI 0.328** 2.28 0.068 1.54 −0.026 −0.29 Number of trades: 51–100×TLI 0.271* 1.62 0.205*** 4.34 −0.400*** −4.22 Number of trades: over 100×TLI 0.300** 2.02 0.347*** 8.05 0.383*** 4.29 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 0.018 0.13 −0.162*** −2.81 −0.015 −0.07 Number of trades: 11–20×TLI 0.010 0.09 −0.147*** −2.76 0.184 1.06 Number of trades: 21–50×TLI 0.254** 2.24 0.051 1.07 0.010 0.06 Number of trades: 51–100×TLI 0.109 0.92 0.146*** 2.86 0.257 1.43 Number of trades: over 100×TLI 0.566*** 5.14 0.398*** 8.46 0.276** 1.93 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TGI 0.078 0.40 0.165*** 3.10 0.096 0.91 Number of trades: 11–20×TGI 0.052 0.31 0.132*** 2.70 0.083 0.87 Number of trades: 21–50×TGI −0.333** −2.32 −0.077* −1.75 0.058 0.65 Number of trades: 51–100×TGI −0.271* −1.62 −0.216*** −4.58 0.436*** 4.59 Number of trades: over 100×TGI −0.307** −2.06 −0.354*** −8.22 −0.349*** −3.91 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI −0.043 −0.32 0.166*** 2.89 0.015 0.07 Number of trades: 11–20×TGI −0.047 −0.38 0.142*** 2.66 −0.183 −1.06 Number of trades: 21–50×TGI −0.287** −2.53 −0.058 −1.20 −0.024 −0.15 Number of trades: 51–100×TGI −0.121 −1.02 −0.158*** −3.09 −0.287 −1.60 Number of trades: over 100×TGI −0.576*** −5.23 −0.407*** −8.65 −0.286** −2.00 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. The regressions in Table AVII confirm the results of the survival analysis results presented in Table VIII. In both cases, investors with a higher academic degree (who can be considered to be more intelligent) or with a natural sciences background (who are stronger in maths) learn faster. Less intelligent investors and investors with a background in humanities seem to start to learn how to avoid the disposition effect only when they have made over 100 trades. It should be noted that though the survival analysis didn’t give a statistically significant effect for experience of over 100 trades for holders of Master’s and Doctoral degrees, the expected effect is clearly present and is statistically significant in the logit regressions. Table AVII. Learning by doing education level and field of studies (in the logit regressions) The table reports the results of the logit regressions—the coefficients, z-values, and significance levels. TLI is the total loss indicator, taking the value 1 if a position is in loss and the value 0 otherwise. TGI represents the total gain indicator, taking the value 1 if a position is in gain and the value 0 otherwise. Trade dummies (the number of trades made) are interacted with TLI and TGI to capture the disposition effect. Control variables are omitted from the table. Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TLI −0.078 −0.40 −0.165*** −3.10 −0.053 −0.51 Number of trades: 11–20×TLI −0.078 −0.47 −0.144*** −2.95 −0.071 −0.74 Number of trades: 21–50×TLI 0.328** 2.28 0.068 1.54 −0.026 −0.29 Number of trades: 51–100×TLI 0.271* 1.62 0.205*** 4.34 −0.400*** −4.22 Number of trades: over 100×TLI 0.300** 2.02 0.347*** 8.05 0.383*** 4.29 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 0.018 0.13 −0.162*** −2.81 −0.015 −0.07 Number of trades: 11–20×TLI 0.010 0.09 −0.147*** −2.76 0.184 1.06 Number of trades: 21–50×TLI 0.254** 2.24 0.051 1.07 0.010 0.06 Number of trades: 51–100×TLI 0.109 0.92 0.146*** 2.86 0.257 1.43 Number of trades: over 100×TLI 0.566*** 5.14 0.398*** 8.46 0.276** 1.93 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TGI 0.078 0.40 0.165*** 3.10 0.096 0.91 Number of trades: 11–20×TGI 0.052 0.31 0.132*** 2.70 0.083 0.87 Number of trades: 21–50×TGI −0.333** −2.32 −0.077* −1.75 0.058 0.65 Number of trades: 51–100×TGI −0.271* −1.62 −0.216*** −4.58 0.436*** 4.59 Number of trades: over 100×TGI −0.307** −2.06 −0.354*** −8.22 −0.349*** −3.91 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI −0.043 −0.32 0.166*** 2.89 0.015 0.07 Number of trades: 11–20×TGI −0.047 −0.38 0.142*** 2.66 −0.183 −1.06 Number of trades: 21–50×TGI −0.287** −2.53 −0.058 −1.20 −0.024 −0.15 Number of trades: 51–100×TGI −0.121 −1.02 −0.158*** −3.09 −0.287 −1.60 Number of trades: over 100×TGI −0.576*** −5.23 −0.407*** −8.65 −0.286** −2.00 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TLI −0.078 −0.40 −0.165*** −3.10 −0.053 −0.51 Number of trades: 11–20×TLI −0.078 −0.47 −0.144*** −2.95 −0.071 −0.74 Number of trades: 21–50×TLI 0.328** 2.28 0.068 1.54 −0.026 −0.29 Number of trades: 51–100×TLI 0.271* 1.62 0.205*** 4.34 −0.400*** −4.22 Number of trades: over 100×TLI 0.300** 2.02 0.347*** 8.05 0.383*** 4.29 Natural and real science Social sciences Humanities Number of trades: 6–10×TLI 0.018 0.13 −0.162*** −2.81 −0.015 −0.07 Number of trades: 11–20×TLI 0.010 0.09 −0.147*** −2.76 0.184 1.06 Number of trades: 21–50×TLI 0.254** 2.24 0.051 1.07 0.010 0.06 Number of trades: 51–100×TLI 0.109 0.92 0.146*** 2.86 0.257 1.43 Number of trades: over 100×TLI 0.566*** 5.14 0.398*** 8.46 0.276** 1.93 Master’s or doctoral degree Bachelor equivalent Vocational training Variables Coefficient z-Statistic Coefficient z-Statistic Coefficient z-Statistic Number of trades: 6–10×TGI 0.078 0.40 0.165*** 3.10 0.096 0.91 Number of trades: 11–20×TGI 0.052 0.31 0.132*** 2.70 0.083 0.87 Number of trades: 21–50×TGI −0.333** −2.32 −0.077* −1.75 0.058 0.65 Number of trades: 51–100×TGI −0.271* −1.62 −0.216*** −4.58 0.436*** 4.59 Number of trades: over 100×TGI −0.307** −2.06 −0.354*** −8.22 −0.349*** −3.91 Natural and real science Social sciences Humanities Number of trades: 6–10×TGI −0.043 −0.32 0.166*** 2.89 0.015 0.07 Number of trades: 11–20×TGI −0.047 −0.38 0.142*** 2.66 −0.183 −1.06 Number of trades: 21–50×TGI −0.287** −2.53 −0.058 −1.20 −0.024 −0.15 Number of trades: 51–100×TGI −0.121 −1.02 −0.158*** −3.09 −0.287 −1.60 Number of trades: over 100×TGI −0.576*** −5.23 −0.407*** −8.65 −0.286** −2.00 *** Significant at the 1% level; ** significant at the 5% level; and * significant at the 10% level. The regressions in Table AVII confirm the results of the survival analysis results presented in Table VIII. In both cases, investors with a higher academic degree (who can be considered to be more intelligent) or with a natural sciences background (who are stronger in maths) learn faster. Less intelligent investors and investors with a background in humanities seem to start to learn how to avoid the disposition effect only when they have made over 100 trades. It should be noted that though the survival analysis didn’t give a statistically significant effect for experience of over 100 trades for holders of Master’s and Doctoral degrees, the expected effect is clearly present and is statistically significant in the logit regressions. Footnotes 1 First documented by Shefrin and Statman (1985) with prominent work by Odean (1998), Grinblatt and Keloharju (2001), Feng and Seasholes (2005), Dhar and Zhu (2006), etc. 2 Barberis and Xiong (2009) show that prospect theory is more likely to fail to explain the disposition effect when the expected risky asset return is high (once the expected return exceeds a certain level) and when the number of trading periods is low (in contrast to when the number of trading periods is high and the expected risky asset return is relatively low). Kaustia (2010) argues that the propensity to sell a stock jumps when the return exceeds zero, but it is approximately constant over a wide range of losses and increasing or constant over a wide range of gains (prospect theory predicts that the propensity to sell the stock declines as its price moves away from the purchase price in either direction). Hens and Vlcek (2011) conclude that the explanation based on prospect theory is sound expost, assuming that the investment was made, but would not hold exante because investors who are affected by the disposition effect would not have made the investment in the first place. 3 Being influenced by the disposition effect has been identified as a costly bias, see, for example, Goulart et al. (2015). 4 We also focus on learning ability as one of the constructs of intelligence. We use the term “baseline learning abilities” to indicate the capacity to learn in a general fashion. Some people are able to learn new things faster than other people, by mastering more difficult concepts, processing, and memorizing more information, etc. We use overall learning abilities as a synonym for “baseline learning abilities” without explicitly distinguishing between the sources or means of learning. Learning can occur in many different ways, such as reading information from books, which is a more abstract way of learning, or alternatively by personally experiencing the world. We distinguish “learning by doing” in our paper as one particular type of learning where an investor is able to learn only by making trades themselves. “Learning about one’s abilities” is used for a more abstract way of learning in which an investor has to be able to make generalizations about their ability to trade and then adjust by stopping or continuing their trading accordingly. 5 Barberis and Xiong (2009) develop an alternative implementation of explanations based on the prospect theory of the disposition effect. Some of the alternative explanations of the disposition effect include the Barber and Odean (1999) hypothesis that investors have a belief that all stocks revert to the mean, which is related to the contrarian strategy and assumes that past winners tend to underperform past losers. Lakonishok and Smidt (1986) propose another explanation that investors restore previous diversifications by rebalancing their portfolios after large price fluctuations. Ferris, Haugen, and Makhija (1988) hypothesize that trading costs play a substantial role in investors not selling stocks at lower prices. 6 Including, for example, Odean (1998), Shapira and Venezia (2001), Grinblatt and Keloharju (2001), Coval and Shumway (2005), Locke and Mann (2005), and Feng and Seasholes (2005). 7 Including Talpsepp (2011) and Cici (2012). 8 Bachmann and Hens (2015) argue that learning from one’s own mistakes may not be very effective, and indeed they point out that learning by doing is irrational and it is more effective to seek advice in order to avoid investment mistakes like the disposition effect. 9 Mother Tongue, English, and Mathematics are required examinations for admission to most courses at Estonian universities. 10 This compares to 34% for the whole population according to the 2011 Census. Thus, investors tend to be significantly better educated than the average person within the same age bracket. 11 PGR–PLR analysis is also used by Frazzini (2006), Dhar and Zhu (2006), Chen et al. (2007), and Kumar and Lim (2008). 12 The Cox proportional hazard model allows the coefficients for covariates to be estimated without the baseline hazard being estimated. The following example running the Cox proportional hazard model on our sample illustrates the interpretation of the hazard ratios. The estimated survivor function shows that 7% of positions still open after 30 days will be sold during the next 5 days, and this can be considered the baseline hazard. If those positions are in loss, only 7% × 0.79 = 5.5% of positions will be sold during the same period. 13 Feng and Seasholes (2005) find that the hazard rate of a sale decreases by 36.8% if a stock is in loss. At the same time, Grinblatt and Keloharju (2001) conclude that Finnish households are 21% less likely to sell their positions with a moderate loss and 32% less likely to do so with an extreme loss. 14 As argued by Talpsepp (2011), investing in a foreign market usually requires more knowledge, capital, and experience, as the first steps in the investment world are usually made in the domestic market. 15 For example, Grinblatt and Keloharju (2001) and Dhar and Zhu (2006). 16 For example, Feng and Seasholes (2005) and Dhar and Zhu (2006). 17 See, for example, Heckman and Rubinstein (2001) for emphasis of the importance of non-cognitive skills. 18 See, for example, Hofer and Pintrich (1997) for an overview of epistemological beliefs. 19 See, for example, Deary and Johnson (2010) for discussion of the correlation between intelligence and academic abilities. 20 At least according to popular belief (Horwitz, 1988). 21 The same applies to the examination results for social sciences and other foreign languages, which we do not present here. 22 See Vaarmets, Liivamägi, and Talpsepp (2014) for market participation and Liivamägi, Vaarmets, and Talpsepp (2014) for performance results. 23 See, for example, Guthke and Stein (1996). References Bachmann K. , Hens T. ( 2015 ): Investment competence and advice seeking , Journal of Behavioral and Experimental Finance 6 , 27 – 41 . Google Scholar CrossRef Search ADS Barber B. M. , Odean T. ( 1999 ): The courage of misguided convictions , Financial Analysts Journal 55 , 41 – 55 . Google Scholar CrossRef Search ADS Barber B. M. , Odean T. ( 2000 ): Trading is hazardous to your wealth: the common stock investment performance of individual investors , Journal of Finance 55 , 773 – 806 . Google Scholar CrossRef Search ADS Barber B. M. , Odean T. ( 2001 ): Boys will be boys: gender, overconfidence, and common stock investment , Quarterly Journal of Economics 261 – 292 . Barberis N. , Xiong W. ( 2009 ): What drives the disposition effect? An analysis of a long-standing preference-based explanation , The Journal of Finance 64 , 751 – 784 . Google Scholar CrossRef Search ADS Beckmann J. F. , Guthke J. ( 1995 ): Complex problem solving, intelligence, and learning ability, in: Frensch P. A. , Funke J. (eds.), Complex Problem Solving: The European Perspective, Psychology Press, New York, pp. 177–200. Chen G. , Kim K. A. , Nofsinger J. R. , Rui O. M. ( 2007 ): Trading performance, disposition effect, overconfidence, representativeness bias, and experience of emerging market investors , Journal of Behavioral Decision Making 20 , 425 – 451 . Google Scholar CrossRef Search ADS Christiansen C. , Joensen J. S. , Rangvid J. ( 2008 ): Are economists more likely to hold stocks?, Review of Finance 12 , 465 – 496 . Google Scholar CrossRef Search ADS Cici G. ( 2012 ): The prevalence of the disposition effect in mutual funds’ trades , Journal of Financial and Quantitative Analysis 47 , 795 – 820 . Google Scholar CrossRef Search ADS Coval J. D. , Shumway T. ( 2005 ): Do behavioral biases affect prices?, The Journal of Finance 60 , 1 – 34 . Google Scholar CrossRef Search ADS Da Costa N. , Goulart M. , Cupertino C. , Macedo J. , Da Silva S. ( 2013 ): The disposition effect and investor experience , Journal of Banking and Finance 37 , 1669 – 1675 . Google Scholar CrossRef Search ADS Deary I. J. , Johnson W. ( 2010 ): Intelligence and education: causal perceptions drive analytic processes and therefore conclusions , International Journal of Epidemiology 39 , 1362 – 1369 . Google Scholar CrossRef Search ADS PubMed Dhar R. , Zhu N. ( 2006 ): Up close and personal: investor sophistication and the disposition effect , Management Science 52 , 726 – 740 . Google Scholar CrossRef Search ADS Feng L. , Seasholes M. S. ( 2005 ): Do investor sophistication and trading experience eliminate behavioral biases in financial markets?, Review of Finance 9 , 305 – 351 . Google Scholar CrossRef Search ADS Fenton-O’Creevy M. , Lins J. T. , Vohra S. , Richards D. W. , Davies G. , Schaaff K. ( 2012 ): Emotion regulation and trader expertise: heart rate variability on the trading floor , Journal of Neuroscience, Psychology, and Economics 5 , 227. Google Scholar CrossRef Search ADS Ferris S. P. , Haugen R. A. , Makhija A. K. ( 1988 ): Predicting contemporary volume with historic volume at differential price levels: evidence supporting the disposition effect , The Journal of Finance 43 , 677 – 697 . Google Scholar CrossRef Search ADS Frazzini A. ( 2006 ): The disposition effect and underreaction to news , The Journal of Finance 61 , 2017 – 2046 . Google Scholar CrossRef Search ADS Goo Y.-J. , Chen D.-H. , Chang S.-H. S. , Yeh C.-F. ( 2010 ): A study of the disposition effect for individual investors in the Taiwan stock market , Emerging Markets Finance and Trade 46 , 108 – 119 . Google Scholar CrossRef Search ADS Goulart M. , da Costa N. C. , Andrade E. B. , Santos A. A. ( 2015 ): Hedging against embarrassment , Journal of Economic Behavior and Organization 116 , 310 – 318 . Google Scholar CrossRef Search ADS Grinblatt M. , Keloharju M. ( 2001 ): What makes investors trade?, The Journal of Finance 589 – 616 . Grinblatt M. , Keloharju M. , Linnainmaa J. T. ( 2012 ): IQ, trading behavior, and performance , Journal of Financial Economics 104 , 339 – 362 . Google Scholar CrossRef Search ADS Grossman S. J. , Kihlstrom R. E. , Mirman L. J. ( 1977 ): A Bayesian approach to the production of information and learning by doing , The Review of Economic Studies 44 , 533 – 547 . Google Scholar CrossRef Search ADS Guthke J. , Stein H. ( 1996 ): Are learning tests the better version of intelligence tests?, European Journal of Psychological Assessment 12 , 1. Google Scholar CrossRef Search ADS Heckman J. J. , Rubinstein Y. ( 2001 ): The importance of noncognitive skills: lessons from the GED testing program , American Economic Review 91 , 145 – 149 . Google Scholar CrossRef Search ADS Hens T. , Vlcek M. ( 2011 ): Does prospect theory explain the disposition effect?, Journal of Behavioral Finance 12 , 141 – 157 . Google Scholar CrossRef Search ADS Hofer B. K. , Pintrich P. R. ( 1997 ): The development of epistemological theories: beliefs about knowledge and knowing and their relation to learning , Review of Educational Research 67 , 88 – 140 . Google Scholar CrossRef Search ADS Horwitz E. K. ( 1988 ): The beliefs about language learning of beginning university foreign language students , Modern Language Journal 72 , 283 – 294 . Google Scholar CrossRef Search ADS Kahneman D. , Tversky A. ( 1979 ): Prospect theory: an analysis of decision under risk , Econometrica: Journal of the Econometric Society 47 , 263 – 291 . Google Scholar CrossRef Search ADS Kaustia M. ( 2010 ): Prospect theory and the disposition effect , Journal of Financial and Quantitative Analysis 45 , 791 – 812 . Google Scholar CrossRef Search ADS Kaustia M. , Alho E. , Puttonen V. ( 2008 ): How much does expertise reduce behavioral biases? The case of anchoring effects in stock return estimates , Financial Management 37 , 391 – 412 . Google Scholar CrossRef Search ADS Kubiska E. , Markiewicz L. , Tyszka T. ( 2012 ): Disposition effect among contrarian and momentum investors , Journal of Behavioral Finance 13 , 214 – 225 . Google Scholar CrossRef Search ADS Kumar A. , Lim S. S. ( 2008 ): How do decision frames influence the stock investment choices of individual investors?, Management Science 54 , 1052 – 1064 . Google Scholar CrossRef Search ADS Lakonishok J. , Smidt S. ( 1986 ): Volume for winners and losers: taxation and other motives for stock trading , The Journal of Finance 41 , 951 – 974 . Google Scholar CrossRef Search ADS Liivamägi K. , Vaarmets T. , Talpsepp T. ( 2014 ): Masters of the stock market. Unpublished working paper, TUT Economic