Paper title: Investors characteristics and disposition effect

Transcription

1 Paper title: Investors characteristics and disposition effect Authors information Mariana Oreng (Corresponding author) PhD Student at FGV-EAESP Avenida Nove de Julho, São Paulo-SP. CEP: Claudia Emiko Yoshinaga Associate Professor at FGV-EAESP Avenida Nove de Julho, São Paulo-SP. CEP: William Eid Junior Full Professor at FGV-EAESP Avenida Nove de Julho, São Paulo-SP. CEP: Keywords Disposition effect, individual investors, behavioral finance

2 Investors characteristics and disposition effect ABSTRACT The purpose of this study is to investigate whether individual investors exhibit the disposition effect, which is, if investors sell gains more readily than they sell losses. We use a unique data set on month-end individual portfolios of an asset management firm in Brazil. Using logistic specifications, we find that gender, age and portfolio size are positively associated to this behavioral bias and that male and wealthier investors tend to old on losses for longer. Monte-Carlo permutations tests demonstrate that these variables are still significant after a thousand replications. Keywords: disposition effect, individual investors, investor behavior, behavioral finance JEL Codes: G11, G17, G40, G41, N26 1. Introduction Most part of the studies on behavioral finance provide evidence in accordance to the prospect theory (Kahneman and Tversky, 1979), which essentially demonstrates that individuals are more averse to losses than are pleased by gains. The disposition effect, one of the behavioral biases that academics are increasingly paying attention to, relates to the idea that individual investors are more willing to sell winning assets than to sell the underperformers (Shefrin and Statman (1985), Odean (1998), Dhar and Zhu (2006)). This irrationality occurs because investors are reluctant to accept that the assets they once regarded as best picks are actually underperforming. Few of these studies, however, base their conclusions on real data (Ranguelova, 2001), as most part of the methods consists on laboratory results (Dorow et. al (2017). In this sense, our purpose is to investigate the existence of the disposition effect observing real investor behavior. This study adds to the literature by providing evidence on the disposition effect using a unique data set consisting on month-end individual portfolios of an asset management firm in Brazil. To our knowledge, we use the largest time spam covered in the Brazilian literature to investigate the disposition effect and one of the largest covered in the literature, from June 2007 to February In accordance to the disposition effect theory, our results demonstrate that investors are more willing to sell winning papers than they are ready to give up on their losing choices. We verify that the proportion of gains realized (PGR) is 13%, while the proportion of losses realized (PLR) is merely 3%. A t-test indicates that this difference is indeed significant at 95%. When analyzing the determinants of investor choices, we find evidence that gender, age and portfolio size (total

3 amount invested on the asset management firm) are statistically significant to explain these portfolio turnovers. Our findings indicate that the more money an individual has invested, the more she is likely to incur in irrational behavior and that male investors usually trade more than their female counterparts. Calvet, Campbell and Sodini (2009) argue that most part of the research on the disposition effect investigates mistakes one at a time, often using samples that are non-representative of households. To address this issue, we run 1000 permutation tests to check whether our significance tests are still valid once we pool the groups of observations and permute those using Monte-Carlo simulations. The advantage of this procedure is that it is non-parametrical, making no assumptions regarding the population distribution. Except for the volume of trades, all the remainder variables are still valid to explain investor behavior. One thing that caught our attention is that the investor s risk profile, as assessed by the suitability questionnaire of the asset management firm, has no association to the disposition effect, despite the fact that investors that are more aggressive are the ones who usually trade more, on average. This might indicate that current measures of willingness to risk exposure are not able to capture behavioral biases, despite regulatory efforts to make investment firms offer products that are adequate to their clients needs. Scholars in behavioral finance increasingly advocate in favor of the idea that the profile assessment should be more careful (Rapp (1998), Jansen, Fischer and Hackethal (2008)), and our evidence is in favor of these claims. The paper proceeds as follows. First, we review some of the existing literature on disposition effect. Second, we present the methodology for measuring disposition effect and also the regression we run to understand the association between individual investors characteristics and asset allocation decisions. Third, we present a detailed descriptive statistics sessions. Next we present our results and concluding remarks follow. 2. Literature Review Shefrin and Statman (1985) coined the expression disposition effect, and Odean (1998) is one of the first studies to measure the disposition effect empirically. The general idea behind it is that individual investors are more willing to sell winner assets than to sell losers. To put it simply, investors are reluctant to accept they made mistakes when picking assets to their portfolios, so if one of these assets is underperforming, there is hope it will outperform in the future. On the other hand, a winning asset is a validation that portfolio choices are correct, so selling it is less painful. Although investors can also sell assets for rebalancing or transaction costs purposes, this phenomenon is linked to the prospect theory value function proposed by Kahneman and Tversky (1979), who argue that individuals are more averse to losses than they are pleased by gains. This value function is concave for gains and convex for losses.

4 Odean (1998) analyzes trading records for 10,000 accounts at a large discount brokerage house covering 4 years of data. He finds evidence that investors demonstrate a strong preference for realizing winners rather than losers. The author investigates whether this behavior is due to taxation, portfolio rebalancing or transaction costs of low priced stocks, but none of these reasons seem to explain investor s irrational behavior. Weber and Camerer (1991) use an experimental setting to study the topic using 6 risky assets and find evidence in favor of the disposition effect. Ranguelova (2001) analyzes daily trading records of 78,000 clients over six years. Using logistic regressions, she documents that the disposition effect is concentrated primarily in large-cap stocks and that stocks at the bottom of the market capitalization exhibit a reverse effect. The author thus argues that these evidences challenge the view that the disposition effect is an implication of the prospect theory. Dhar and Zhu (2006) observe the relationship between individual characteristics and the disposition effect using trading records of more than 50,000 individual investors between 1991 and They report that wealthier individuals and individuals employed in professional occupations exhibit a lower disposition effect, and the same occurs for trading frequency. Studying the disposition effect deals essentially with the investor s reference point, which is, we define asset gains or asset losses based on a certain purchase price. Following most part of the literature, we assume the reference point as the average purchase price, although this can be open to debate. One can argue, for instance, that most part of the individuals do their mental accounting based on the first or the last purchase price, as these figures are easier to remember than the average of all the transactions made. Kahneman and Tversky (1979) argue there are situations in which gains and losses are coded relative to an expectation or aspiration level that differs from the status quo, and Odean (1998) argues that the price path may also affect the level of the reference point. Odean (1998) discusses some of the reasons why investors might display this irrational behavior, namely the shape of the value function or incorrect expectation of mean-reverting prices. Lakonishok and Smidt (1986) discuss two other reasons why investors might exhibit a behavior like the disposition effect but still be rational: either because they do not hold the market portfolio or due to expectation that some favorable information is yet to be incorporated into price. Ranguelova (2011), on the other hand, argues that the disposition effect occurs because of individual beliefs rather than preferences. When confronted with a loss in a certain stock, individuals would become less risk-averse, in an attempt to break-even. In our interpretation, her explanation is in line with Odean (1998) s idea that irrational behavior occurs because of incorrect expectations. In line with Ranguelova (2011), Ben-David and Hirshleifer (2012) argue that tradings based on belief revisions can potentially explain the findings on the disposition effect. They argue that the disposition effect is not

5 driven by a simple direct preference of an asset being a winner rather than a loser and that investors are not necessarily myopic, as they are more likely to sell big losers than small ones. Aspara and Hoffmann (2015) use three experiments to examine social and psychological factors that affect the susceptibility to the disposition effect. Specifically, they document that there is less evidence of disposition effect when the cause of poorly performing investments is because of investor s own fault and when they invest other people s money. Another alternative explanation to the existence of the disposition effect is from Calvet, Campbell and Sodini (2009). They construct an index of financial sophistication using data on Swedish households and demonstrate that it can explain the tendency to sell winning stocks and hold losing stocks. In Brazil, Karsten (2005) analyses trading records of individual and institutional investors covering the period from 2001 to The author demonstrates that individual investors behave according to the disposition effect, but finds mixed results for institutional investors. He also shows that tax-purposes, rebalancing, dividend distribution and liquidity do not increase the susceptibility to incur in this behavioral bias. 3. Methodology Barberis and Xiong (2009) demonstrate that defining the disposition effect over realized gains and losses is more reliable than using annual gains and losses. Following Odean (1998), the methodology to measure the disposition effect consists on the following: for each date the investors makes an asset sale, we identify which asset was sold and what was the investor s portfolio then. First, for each asset sold, we compare its sale price to its average purchase price. If the sale price is higher than the average purchase price, we compute it as a Realized Gain (RG). In case it is lower, we compute it as a Realized Loss (RL). Second, for each asset composing the portfolio on that date, we compare their month s highest and lowest prices to their average purchase price. If the highest and the lowest prices are both above the average purchase price, we compute it as a Paper Gain (PG). If they are below, we compute it as a Paper Loss (PL). Using these measures, we compute the Proportion of Gains Realized (PGR) and the proportion of Losses Realized (PLR), given by the following expressions: (1) PGR = RG RG+PG (2) PLR = RL RL+PL

6 In case the PGR is higher than the PLR, individual investors are indeed more prone to sell winning assets and to hold on losing ones, which is evidence in favor of the disposition effect. It is important to mention that most part of the studies analyzing the disposition effect have at their disposal daily trading records. For our study, however, we were only able to collect information on month-end portfolios of individual investors. From these data, we re-built portfolios so we could check whether an asset was purchased, sold or maintained during a specific month. To determine if a transaction occurred, we based our calculations on information gathered for each one of the assets. In case of a stock, for instance, we calculated the number of shares in a portfolio based on the closing price of that stock during that period. If for the following period the number increased (or decreased), we count it as a negotiation. The same applies for funds, except that in this case we used the price of quotas. After calculating the PGR and the PLR, we create a dummy representing the investor s behavior, which assumes a value of 1 if the investor behaves according to the disposition effect theory (which is, if PGR is higher than PLR) and zero if the investors exhibits a rational behavior. We then run logistic regressions to verify the association between investors behavior and personal characteristics, according to the following equation. (3) Disposition = f(gender, Education, Age, Balance, Ntrades) Gender is a categorical variable that is equal 1 if the investor is male and zero if she is female. Education is another categorical variable indicating whether the investor went to college. We also investigate whether the disposition effect has any association to the amount invested. For this purpose, we include the total portfolio size of each investor as a control, and we use it as a proxy for wealth. We note that although this is not the most proper measure of wealth as we are covering only the amount invested on this specific asset management firm, we do not have individual information regarding total net worth or any other wealth measure. Finally, Vtrades corresponds to the amount of portfolio transactions each investor did during the period we cover. We also run an ordered logistic model to see how the above-mentioned controls can explain the existence of the disposition effect. In order to do this analysis, we first calculated the difference between the PGR and the PLR. We call this difference Disposition_2. From the initial 5402 individual-period observations, we exclude all the zeros and ones (as these extremes often refer to transactions in a single direction, thus not making it possible to identify whether there are more sales than purchases or vice versa) and all the negative values, which leaves us with 344 observations. This drop in the number of observations has to do with the fact that most part of the monthly transactions involve a single purchase or a single sale. We also excluded the individuals for which we had no information on personal characteristics, which finally

7 leaves us with 328 observations. After this procedure, we categorize the disposition effect according to the following scale: Disposition_2 = 1 if 0 < (PGR-PLR) 0.33 Disposition_2 = 2 if 0.33 < (PGR-PLR) 0.66 Disposition_2 = 3 if 0.66 < (PGR-PLR) < 1 The number of individual-period observations with a PGR-PLR difference higher than 0.66, however, is very small, as we found only 3 of those cases. In this sense, we also divide the observations in two sub-groups as a robustness check: individuals with (PGR-PLR) difference smaller than 0.5 are included in the group 1, while the remainder cases fall into the group Permutation tests Permutation tests are statistical significance non-parametric tests that use random shuffles of data to check whether results are asymptotic valid. Monte-Carlo simulations are used to do the resampling. The purpose of this procedure is to derive robust estimates of confidence intervals of some parameters of interest. In this study, this test are of interest as our database contains information on Brazilian investors negotiating on the Brazilian market, so in order to get rid of any idiosyncrasy that may cause results to be different than what would be observed in other financial markets. 4. Data Our initial database contains information on monthly portfolios of 189 investors. Portfolios are from June 2007 to February 2017 and contain 124 different types of assets. One of the largest asset management firms in Brazil provided us this information, as well as data on individual characteristics, such as age, gender, educational level and investor profile (based on their suitability questionnaire). The variable Profile equals 1 if the investor is conservative, 2 if she is moderate without volatility, 3 if she is moderate with volatility and 4 if she is aggressive. Our gender dummy equals 1 if the investor is male and zero if she is female. We excluded 8 assets and 51 investors due to missing information, so our final database contains information on 138 investors and 116 different assets. Using the methodology explained in the previous section, we detected 5092 transactions. Portfolio assets fall into three main categories: fixed income, funds and equity. In the funds category, we have multi-asset (high and low volatility), equity, interest rates, fixed income and real estate funds. We raise a note of caution here, as

8 most part of the assets in our portfolios is related to the real estate sector (due to the very nature of the asset management firm that provided us the information). For the computation of the disposition effect, however, this does not seem to be an issue, as it deals essentially with asset prices and is not dependent on the nature of the asset per se. 5. Descriptive Statistics In our database, 63% of the investors are male and 89% of them have superior education. The average age is 55 years old. Although this may not correspond to the profile of the average Brazilian, we consider it a representative sample of the average Brazilian investor. Comparing it to the investor profile released by the Brazilian Stock Exchange (B 3 ), our sample is even more balanced in terms of gender, as the average proportion of male investors from 2002 to 2016 is 77%. In terms of age, the investors in our sample are relatively younger, because the majority of investors in B 3 (43.56%) is older than 66 years. Table 1 brings descriptive statistics for each asset class. The average asset position in this asset management firm is R$107, From the 5092 portfolio transactions we identify, investors incur in gains or losses in From these transactions, we identify that the disposition effect is present in 10% of them, or 437 transactions. The average PGR is 13.1%, while average PLR is 3.5%. These results demonstrate that, on average, investors do exhibit the tendency to sell winners and hold on to losers. Table 3 reports tests on the equality of means. As expected, the difference between the investors who exhibit the disposition effect and the ones that do not is statistically significant for all the variables we use as controls except for age. Contrary to the common sense idea that educated investors are usually more prepared to rebalance their portfolios, our t- tests indicate that, on average, investors who went to college are more willing to sell gains than losses. Besides, the disposition effect correlates positively to the volume of portfolio transactions. Altogether, these results indicate that wealthier and more educated investors are more likely to exhibit the disposition effect. One explanation for such a behavior has to do with overconfidence (Odean, 1998), which is, these investors may be overconfident about their abilities, knowledge and future prospects because they are too certain about their opinions and because they perceive their actions to be less risky (Barber and Odean, 1999). The correlation between the disposition effect dummy and the gender dummy is 21%, which indicates that in terms of portfolio transactions, men are, on average, more irrational than women are. Barber and Odean (1999) present evidence that men trade more actively than women do and detract from their net return performance. The correlation between the disposition effect dummy and the dummy for education is 18.5%, which also suggests a positive relation between irrational investment behavior and the fact that individuals have higher education. The correlation between the disposition

9 effect and the natural logarithm of age, however, is minimal. Interestingly, the correlation between the disposition effect dummy and the natural logarithm of portfolio size each investor has in their investment account is 55%, suggesting that the more money an individual has invested, the more he is likely to incur in irrational behavior. This correlation may be more pronounced, however, because investors with higher portfolio sizes also tend to make a larger number of transactions: the correlation between the natural logarithm of portfolio size and the natural logarithm of the amount of transactions each investor made is 78.7%. For this reason, we also include the natural logarithm of the number of transactions as a control in our regressions. The correlation between gender and the amount of negotiations (in natural logarithm), however, is small (10%), which suggests that male investors are more prone to behave irrationally not because of the volume of trades they make, but because of less risk aversion. For information only, we also point out that 90% of the assets in our sample are real estate investment funds. Lakonishok and Smidt (1986) argue that this can partially explain the existence of the disposition, as individual investors do not hold the market portfolio. For the computation of the PGR and PLR, however, this does not seem to be an issue. In this sense, the analysis of such a topic is beyond the scope of this paper. We also had access to the suitability of the investors, which is essentially a measure of willingness to risk exposure. The asset management firm that provided us the data categorizes individuals in four main groups, as detailed in the Data section. In Brazil, risk assessment is an obligation of the investment firms registered at the Securities and Exchange Comission of Brazil (CVM), and offering products that are adequate to their clients risk profile is required. Despite regulatory efforts to match the clients needs, however, some preliminary analysis of our data indicates that these profiles are not significant to explain individuals allocation on risky assets (results not reported). The correlation between the investor s profile and the number of negotiations is moderate, 32%. The correlation between investor s profile and the disposition effect dummy, however, is smaller, around 17%. The pure correlation analysis, however, does not enable us to state any association between profiles that are more aggressive and the susceptibility to incur in irrational losses. In this sense, we perform a test on the equality of means comparing two profile groups: the conservatives and the moderates with high volatility. We do so because we only have 3 aggressive individuals in our database. To compare them, we use the natural logarithm of the volume of trades. The average for the moderates is 0.6 while the average for conservatives is 0.6, and the t-stat for the difference is This indicates that, on average, more aggressive individuals trade more. To further explore the relationship between these variables and the disposition effect, both variables are included in our regression models. A wide range of papers study the phenomenon of month and holiday effects on stock returns. A well-known occurrence, for instance, is the January effect, which relates to the idea that the increase in stock prices during this month can be

10 attributable to an increase in buying, following the drops in the previous month as investors realize their capital gains in December. Some other studies also find evidence that this phenomenon can be related to company size, most noticeably for small stocks. We observed all negotiations made by all investors during the period we cover in order to verify if there are more negotiations during a specific month. For the period we cover, we had a total of 5092 portfolio transactions (Graph 1). Indeed, the average number of transactions is 424 with a standard deviation of 102. The peak occurs in November with 693 transactions, which is more than two standard deviations above the average. The peak of transactions occurs in 2016, with 361 registered transactions. If we observe the number of negotiations in November as a proportion of the total number of negotiations during the year, however, the maximum occurs in 2009, with 10% of the negotiations made over the entire year (the average is 4%). This can be so because this period is right after the subprime crisis, when the Brazilian Stock Exchange went from points at the beginning of the year to more than at the end. Most part of the transactions along this year were asset purchases, and not asset sales. In all the transactions we verify in this year, 6% of them represent realization of gains and in none of them investors realized losses, perhaps in the hope that some of these assets would recover in value. The only literature we could identify that discusses the November effect is from Gibson, Safieddine and Titman (2000) and He and He (2011). Both of them argue that this phenomenon is related to the 1986 Tax Reform Act (TRA) in United States, which apparently resembles no similarity to what we observe here. 6. Results Table 4 describes our logistic results. Except for the amount of portfolio transactions, all the variables we include are significant in explaining the disposition effect. Specifically, as we already noted in the previous section, male investors tend to hold on losses for longer. Contrary to common beliefs, investors who went to college are also more likely to engage in irrational behavior when it comes to personal investments. Actually, from the 90 investors who exhibit such a tendency in our sample, 84 of them have superior education. This result, however, may be not very accurate, as we are dealing with a sample where the majority of investors have higher education. In this sense, to see if we can find more robust results, we also run ordered logistic regressions, presented in Table 5. Indeed, the Education variable is no longer significant, but we still find significance for gender, age and portfolio size. 6.1 Suitability Besides the fact that we control for variables such as gender, age and education, we believe that the risk profile of a certain investor consider a wider range of individual characteristics. In this sense, we also check whether the investor risk profile, as determined by the asset management firm, has any association to the existence of the disposition effect. Regarding suitability, we verify that the correlation between the volume of trades and the investor s risk profile is 23%. If we use

11 the natural logarithm of the volume of trades, this correlation jumps to 32%. This might indicate that more aggressive investors are the ones that usually trade the most. In this sense, we check whether the investor s profile, as reported by the asset management firm s suitability questionnaire, plays any role in explaining the disposition effect. Although the tests on the equality of means indicate that more aggressive investors are usually the ones who trade the most and, thus, are the ones more likely to incur in irrational behavior, when we include this variable in the logistic regression we find that when we consider all the other controls, profile is not significant for explaining the disposition effect. Although we cannot directly control for investor experience, we consider age to be a good proxy for it. 6.2 Robustness Checks We use our categorical variable for the disposition effect and apply some permutation tests in order to perform significance tests. The purpose of this tool is to identify whether the significance tests are still valid once we pool the groups of observations and permute them using Monte-Carlo simulations. We run 1000 repetitions and report the results in Table 6. Except for the number of negotiations, all the other variables are still significant to explain the existence of the disposition effect. These results address the concerns raised by Calvet, Campbell and Sodini (2009), namely that research on the disposition effect often uses nonrepresentative samples of households. As a second robustness check, we run the logistic models using subsamples for portfolio size and volume of trades (Table 7). For each case, we divide the sample in two groups, using the median as the threshold point. For the sub-sample with portfolio size above the median, education and age are no longer significantly associated to the existence of the disposition effect, and the volume of trades becomes positively significant to explain this association. This is the only case in which this variable is significant. We observe the same pattern for the sub-sample with the volume of trades above the median. For the sub-sample with portfolio size below the median, gender is no longer significantly associated with the existence of the disposition effect, and the same occurs for the sub-sample with the volume of trades below the median. In accordance to Dhar and Zhu (2006), trading frequency tends to reduce the disposition effect, but the association is only statistically significant for the sub-sample of portfolio size above the median value. 6.3 Alternative explanations Ranguelova (2011) argues that most part of the literature in behavioral finance explain the disposition effect in the following way: [ ]When confronted with a loss in a specific stock, individuals become risk-seeking and hold on to that stock for a chance to break even. She argues, however, that this behavior varies strongly with firm size, which is, individuals are more irrational when it comes to large caps. In this sense, she states that the answer to what is causing the disposition effect may lie outside prospect theory, as individuals trade based on beliefs rather than preferences.

12 To see if this is the case in our database, we will further analyze the 437 cases in which we verify the existence the disposition effect to see which assets the investors bought and sold on the period. So far, we analyzed information on 96 investor-period observations, which corresponds to 21% of the total. The vast majority of transactions, however, occurs with funds. In this sense, we do not find evidence supporting the author s explanation, as we cannot attribute the performance of a fund to a specific asset of its portfolio. To do so, we should gather information on each one of these specific funds and on the assets composing their portfolios, which is beyond the scope of this study. 7. Conclusion This study uses a unique data set provided by one of the largest asset management firms in Brazil to investigate the disposition effect, which is, whether individual investors sell winning assets more readily than they sell losers. We cover the period from June 2007 to February 2017 and our portfolios contain 124 different types of assets. We use the methodology proposed by Odean (1998) to calculate the disposition effect. Initial t-tests indicate that the difference between the proportion of gains realized (PGR) and the proportion of losses realized (PLR) is statistically significant. We run logistic and ordered logistic regressions and our results demonstrate that gender, age and portfolio size play a positive and significant role in explaining this investment behavior. Monte-Carlo simulations indicate that our results are still valid once we run 1000 permutation test after pooling the groups of observations. Comparing our evidence to the literature, we find some mixed results. While some authors argue that wealthier individuals are usually more rational, our results challenge this view. Our findings indicate that the more money an individual has invested, the more he is likely to incur in irrational behavior and male investors usually trade more than their female counterparts. Besides, most part of the studies deal with data on stocks, and our set of information contain not only stocks but also funds and some classes of fixed income assets. To our knowledge, this is the study that covers the largest time frame when it comes to disposition effect, and also the first to use Monte-Carlo simulations as robustness checks. The Securities and Exchange Comission of Brazil (CVM) has made efforts in the last decade to incentive investment firms to offer adequate products to their clients risk profile. Despite the fact that more aggressive individuals usually trade more, on average, our results demonstrate that the investor profile, as considered by the asset management firm that provided us the data, is not associated to the disposition effect.

13 Our next steps include calculating the index of investor sophistication proposed by Calvet, Campbell and Sodini (2009) to see if it explains the disposition effect. Further, we can also check whether investor in our database suffer from inertia in risk taking. References ASPARA, J., HOFFMANN, A.O.I. (2015) Cut your losses and let your profits run: How shifting feelings of personal responsibility reverses the disposition effect. Journal of Behavioral and Experimental Finance. BARBER, B.M., ODEAN, T. (1999) The Courage of Misguided Convictions: The Trading Behavior of Individual Investors. Financial Analyst Journal, 55(6) BARBERIS, N., XIONG, W. (2009) What Drives the Disposition Effect? An Analysis of a Long-Standing Preference- Based Explanation. Journal of Finance, 64(2), BEN-DAVID, I., HIRSHLEIFER, D. (2012) Are Investors Really Reluctant to Realize Their Losses? Trading Responses to Past Returns and the Disposition Effect The Review of Financial Studies, 25(8), CALVET, L.E., CAMPBELL, J.Y., SODINI, P. (2009) Measuring the Financial Sophistication of Households The American Economic Review, 99(2), DHAR, R., ZHU, N. (2006) Up Close and Personal: Investor Sophistications and the Disposition Effect Management Science, 52(5), DOROW, A., DA COSTA JR., N., TAKASE, E., PRATES, W. (2017) On the neural substrates of the disposition effect and return performance. Journal of Behavioral and Experimental Finance. GIBSON, S.; SAFIEDDINE, A.; TITMAN, S. (2000) Tax-Motivated Trading and Price Pressure: An Analysis of Mutual Fund Holdings Journal of Financial and Quantitative Analysis, 35(3), HE, L.T.; HE, S.C. (2011) Has the November Effect Replaced the January Effect in Stock Markets? Managerial and Decision Economics, 32,

14 JANSEN, C., FISCHER, R., HACKETHAL, A. (2008) The influence of financial advice on the asset allocation of individual investors. EFA 2008 Athens Meeting. Available at SSRN: KAHNEMAN, D., TVERSKY, A. (1979) Prospect Theory: An Analysis of Decision under Risk Econometrica, 47(2), KARSTEN, J.G. (2005) O efeito disposição: um estudo empírico no Brasil (Master disseration). São Paulo School of Economics, Fundação Getulio Vargas LAKONISHOK, J., SMIDT, S. (1986) Volume for winners and losers: Taxation and other motives for stock trading. Journal of Finance, 41, ODEAN, Terrance. (1998) Are Investors Reluctant to Realize Their Losses? The Journal of Finance, 53(5), RANGUELOVA, E. (2001) Disposition Effect and Firm Size: New Evidence on Individual Investor Trading Activity. Harvard University Working Paper. Available at SSRN: RAPP, R.N. (1998) Rethinking Risky Investments for that Little Old Lady: A Realistic Role for Modern Portfolio Thoery in Assessing Suitability Obligations of Stockbrokers. Ohio Northern University Law Review, 24(1), SHEFRIN, H.; STATMAN, M. (1985) The Disposition to Sell Winners Too Early and Ride Losers Too Long: Theory and Evidence The Journal of Finance, 40(3), WEBER, M.; CAMERER, C. (1991) The disposition effect in securities trading: an experimental analysis Manuskripte aus den Instituten für Betriebswirtschaftslehre der Universität Kiel, No. 276

15 Tables and Figures Table 1 Asset allocation: Descriptive Statistics Variable N. Obs Mean Std. Dev. Min Max Multi-Asset funds (low volatility) , ,085, ,787,57 10,939, Multi-Asset Funds (high volatility) , , , , Equity Funds , , , , Fixed Income 42 85, , , Real Estate Funds , , , ,961, Interest Rate Funds , , , , Equity 1 4, , , This table describes the asset allocation of the individual investors in our sample. Values are in Brazilian Reais (BRL). Table 2 Investor Profile: Descriptive Statistics Variable N. Mean Std. Dev. Median Min Max Obs Age Gender Higher Education Profile Portfolio size , , , ,298,00 438, Volume of trades Gender is a dummy that equals 1 if the investor is male and zero if it is female. Higher education is a dummy indicating whether investors have higher education. The asset management firm gave us information on the investor profile according to their suitability assessment, so the variable Profile equals 1 if the investor is conservative, 2 if she is moderate, 3 if she is moderate and 4 if she is aggressive. Portfolio Size indicates the amount invested by investors on the asset management firm, and values are in Brazilian Reais (BRL). Volume of trades indicates the volume of trades made during the entire period.

16 Table 3 Tests on the equality of means Variable G1(0) Mean1 G2(1) Mean2 MeanDiff Gender ** Education ** Ln_Age Ln_Portsize *** Ln_Vtrades *** Profile ** This table presents t-tests for the main variables used in the study. Number 1 refers to male investors and number 0 refers to female investors. G1(0) refers to female investors and Mean1 represents the mean for female investors. G2(1) refers to male investors and Mean2 represents the mean for male investors. This test on the equality of means considers a 95% confidence interval. Table 4 Logistic Regressions VARIABLES LOGIT LOGIT_profile Gender 1.075** 1.095** (0.474) (0.478) Education 1.642** 1.662** (0.737) (0.739) Ln_Age 1.546* 1.551* (0.887) (0.885) Ln_Portsize 2.445*** 2.411*** (0.553) (0.554) Ln_Vtrades (0.628) (0.641)

17 Profile (0.335) Constant * *** (7.145) (7.145) Observations 138 Pseudo R *** p<0.01, ** p<0.05, * p<0.1 Dependent variable is a categorical variable that indicates whether the investor exhibits the disposition effect, which is the susceptibility to sell gains more readily than losses. For comparability, we use the natural logarithm on the variables age, portfolio size and amount of portfolio transactions. Table 5 Ordered logistic regressions VARIABLES OLOGIT_3 OLOGIT_2 Gender 0.639** (0.277) (0.282) Education (0.492) (0.523) Ln_Age ** ** (0.514) (0.529) Ln_Portsize 0.591* 0.581* (0.309) (0.320) Ln_Vtrades (0.316) (0.325) Constant cut (4.034) Constant cut (3.997) (4.125)

18 Observations Pseudo R *** p<0.01, ** p<0.05, * p<0.1 In this table, the dependent variable measures the level of disposition effect. We calculate is as the difference between the proportion of gains realized and the proportion of losses realized. In the first column, we sub-divide results in three groups (if the difference is up to 0.33, we allocate individuals in group 1; if it is up to 0.66, in group 2, and the remainder individuals we allocate in group 3) and in the second column we use only two groups (the cut value is 0.5). Table 6 Monte Carlo permutation statistics T T(obs) c n P=c/n SE(p) 95% confidence interval Gender Education Ln_Age Ln_Portsize Ln_Vtrades Constant We run permutation tests using Monte Carlo simulations as a robustness check. Our resampling method consists on doing 1000 replications with 138 observations. Note 1: Confidence intervals are with respect to p=c/n. Note 2: c = #{ T >= T(obs) } Table 7 Logistic regressions using sub-samples Portfolio Size Portfolio Size Number of trades Number of trades (above median) (below median) (above median) (below median) VARIABLES LOGIT LOGIT LOGIT LOGIT gender 2.503** *** 0.360

19 (0.994) (0.603) (0.834) (0.714) education * ** (1.376) (1.083) (1.246) (1.376) ln_age ** * (1.286) (1.351) (1.159) (1.526) ln_portsize 4.981*** 2.167*** 2.604*** 3.161*** (1.887) (0.816) (0.933) (0.995) ln_vtrades ** (0.924) (1.187) (0.845) (1.786) profile Constant *** *** *** *** (21.65) (11.79) (11.36) (13.57) Observations *** p<0.01, ** p<0.05, * p<0.1 Graph 1 Negotiations per month Negotiations per month jan feb mar apr may jun jul aug sep oct nov dec This graph describes the sum of the number of negotiations for each month of the year. The peak occurs in November, with a value more than two standard deviations above average.

20