WHAT FACTORS AFFECT BEHAVIORAL BIASES?: EVIDENCE FROM TURKISH INDIVIDUAL STOCK INVESTORS

Transcription

1 WHAT FACTORS AFFECT BEHAVIORAL BIASES?: EVIDENCE FROM TURKISH INDIVIDUAL STOCK INVESTORS Bülent Tekçe Yapı Kredi Bank, Business Performance Management Manager This paper uses nationwide individual stock investor transaction data for 244,146 investors with a total of 64 million buy and sell transactions in 2011 to analyze how common overconfidence, familiarity bias, representativeness heuristic and status quo bias are among Turkish individual stock investors and what factors affect these biases. This study is unique in the sense that, up to our knowledge no research focuses on nationwide data to analyze more than one bias. We find that overconfidence and familiarity bias are common among individual investors. Findings of status quo bias are totally in line with overconfidence. Male, younger investors, investors with lower portfolio value and investors in less developed (low income, low education) regions exhibit overconfidence, familiarity bias and status quo bias more. Our findings are robust to the use of different subsamples, bias measures and analysis methods. 1

2 1. Introduction Empirical evidence in the behavioral finance literature show that individuals do not behave rationally. Barberis and Thaler (2003) provide a good summary of models that try to explain the equity premium puzzle, excess volatility, excessive trading, stock return predictability using both Prospect Theory of Kahneman and Tversky (1979) and beliefs. Daniel et al. (2002) support the view that markets are not efficient and investor biases affect security prices substantially. Black (1986), De Long et al. (1990), Shleifer and Vishny (1997), Barberis et al. (2001), Hirshleifer (2001), Daniel et al. (2002), and Subrahmanyam (2007) show that investors are not rational or markets may not be efficient and hence prices may significantly deviate from fundamental values due to existence of irrational investors. Vissing-Jorgensen (2004) uses investor optimism survey data conducted by UBS and Gallup from 1998 to 2002 and find that irrational behavior (such as, representativeness heuristic, self-attribution bias, disposition effect, under-diversification and status quo bias) are weaker for more sophisticated investors (wealth and investor experience used as proxies for investor sophistication). Hence, it can be proposed that behavioral biases affect some investors less than others. As biases may significantly affect stock prices, it is important to understand which factors affect biases. Definition of rationality is unique in the sense that irrespective of personality differences every rational decision maker behaves same. However, there are many ways of being irrational which may depend on individual as well as cultural differences. Hence, individuals may tend to behave differently in their financial decisions from one society to another. Cultural differences may cause differences in biases as cognitive biases can be triggered or suppressed by different life experiences and cultural backgrounds. Degree of individualism/collectivism has significant impact on cognitive styles, risk attitudes and behavioral tendencies of inhabitants. Individuals in collectivist societies tend to be more risk tolerant. As presented by Fan and Xiao (2005) and Statman (2010), individuals in different societies / cultures may have different behavioral biases which may affect their financial decisions. Majority of behavioral finance literature analyzes individual investors in developed markets such as USA, UK and Western Europe. Hofstede (2001) finds that Turkish people are more collectivist compared to USA, UK and Western Europe. Besides, the ambiguity avoidance index, which captures the tolerance of a society for uncertainty and ambiguity, is high among Turkish citizens. As Turkey is an emerging market and there exists cultural differences compared to USA, UK and Western Europe, it is worth analyzing Turkish individual investors in terms of behavioral biases they exhibit. If Turkish individuals differ from those in the developed countries, the behavioral biases of Turkish individual investors may differ from the findings in the literature. Many of the research in behavioral finance literature depend on data that is generally limited to the subsamples of overall investor groups in these countries. This study is unique in the sense that, although there are several studies using nationwide data (either in developed markets such as Finland, or in emerging markets such as Taiwan) to analyze a specific bias, up to our knowledge no research focuses on nationwide data to analyze different biases. It is also interesting to analyze Turkish individual investors as Istanbul Stock Exchange (ISE) has specific characteristics. ISE is a member of World Federation of Exchanges (WFE) and Federation of European Securities Exchanges (FESE). As a leading / advanced emerging market stock exchange, ISE is 2

3 recognized as an investable market according to US Securities and Exchange Commission (SEC) and Japan Financial Services Agency. ISE has one of the highest turnover ratio among world stock markets, which may be related to the biases among Turkish stock investors. According to World Bank, in 2011, ISE is the 5 th highest stock market in terms of turnover ratio after Italy, Republic of Korea, China and USA. Trading volume in ISE is relatively high and provides a liquid market for investors. Although foreign investors hold around 65% of free float in ISE, they constitute only around 15% of the trading volume. Foreign investors mostly prefer ISE30 and ISE100 (a major benchmark) stocks, which have high market capitalization, high liquidity and are representative of sectors they operate. Trading volume and liquidity is mostly provided by local individual investors. This study focuses on four behavioral biases; overconfidence, familiarity bias, representativeness heuristic and status quo bias of all the Turkish individual stock investors and analyzes how prevalent these biases are among investors. We use transaction data and also analyze what factors such as gender, age, wealth, experience and geographical region of residence affect these biases. Due to aggressive trading behavior, overconfident investors may have to pay significant amount of commissions. Besides, overconfident investors may hold riskier portfolios than they should tolerate due to their underestimation of risks. Overconfidence not only affects financial markets and prices, but also individuals in the sense that they make investment mistakes and lose money. Hence it is important to determine overconfidence among investors and factors affecting overconfidence. Familiarity bias is important in the sense that it explains how investors decide to purchase a stock for reasons other than rational motives. The psychology literature shows how representativeness heuristic can explain expectation formation which directly affects investment decisions. There are several studies which focus on how investors extrapolate past price trends to predict future prices and measure representativeness heuristic accordingly. Representativeness heuristic may lead individuals to give investment decisions that harmfully affect their wealth. Such an approach may also distort asset prices. Although overconfident investors trade too much, investors exhibiting status quo bias may refrain from trading at all. 2. Literature Review and Hypothesis Development 2.1. Overconfidence Overconfidence can be defined as the unmerited confidence in self s judgments and abilities. Odean (1998) describes overconfidence as the belief that a trader s information is more precise than it actually is. This is equivalent to narrow confidence intervals in predictions. Daniel et al. (1998) define an overconfident investor as one who overestimates the precision of his private information signal, but not of information signals publicly received by all. Overconfidence may stem from different reasons. Self-attribution bias is attributing successful outcomes to own skill but blaming unsuccessful outcomes on bad luck as discussed in Miller and Ross (1975) and Kunda (1987). Langer (1975) states that illusion of control is the tendency for people to overestimate their ability to control events that they have no influence over. Unrealistic optimism is simply confidence about the future or successful outcome of something. It is the tendency to take a favorable or hopeful view as discussed by Weinstein (1980) and Kunda (1987). Russo and Shoemaker (1992) define confirmation bias as the tendency for people to favor information that confirms their arguments, expectations or beliefs. As discussed by Svenson (1981), better than average effect implies that people think they have superior 3

4 abilities than on average. Hence, individuals tend to believe they are in the best class among peers. Calibration refers to how individuals can assess the correctness of their estimates. Deaves et al. (2010) argue that a miscalibrated agent assumes lower level of mistake than she / he actually makes. Different forms of overconfidence reveal that overconfident investors believe that their decisions will prove to be correct and expect higher returns than average. However, this is not necessarily the case and overconfident investors are exposed to possible losses due to their investment decisions. Fischhoff et al. (1977), Russo and Shoemaker (1992), Griffin and Tversky (1992), Kahneman and Riepe (1998) show that overconfidence is common among decision makers. Odean (1998) presents a good summary of overconfidence in different professional fields such as investment bankers and managers. The author also finds that overconfidence affects financial markets; overconfidence increases expected trading volume, increases market depth and decreases the expected utility of overconfident traders. In line with literature, we hypothesize that overconfidence is common among Turkish individual equity investors. Barber and Odean (2001) test whether men are more overconfident than women by partitioning investors on gender. The authors use data from a nationwide brokerage house for the period by focusing on common stock investments of households. The authors define overconfidence as annual turnover and find that women turn their portfolios almost 53% while men turn 77% annually indicating that men trade 45% more than women annually. Findings of Barber and Odean (1999), Chen et al. (2007), Acker and Duck (2008), Graham et al. (2009), Grinblatt and Keloharju (2009), Hoffmann et al. (2010) also support the view that men are more overconfident than women. In line with literature, we also expect Turkish male investors to be more overconfident than female investors. Chen et al. (2007) use transaction data of a large Chinese brokerage house to analyze overconfidence in Chinese investors. The authors find that individual investors in China trade more frequently than US individual investors. Acker and Duck (2008) use a stock market game and predictions of examination marks to measure overconfidence among Asian and British students. They find that Asian students are more overconfident than British students. These findings imply that level of overconfidence can be different among cultures. In line with literature, we hypothesize that Turkish individual stock investors are more overconfident than US individual investors. Graham et al. (2009) find that wealthier and highly educated investors are more likely to perceive themselves as competent, implying overconfidence. On the other hand, Ekholm and Pasternack (2007) confirm that investors with smaller portfolios are more overconfident compared to investors with larger portfolios as these investors are more experienced and wealthier. Hence, we hypothesize that sophisticated investors are less prone to overconfidence Familiarity Bias According to Fox and Tversky (1995), when people are offered two alternatives, they prefer the one that they are familiar with. This finding is also valid for stock selection. This is because people are better informed about the securities that they are familiar compared to the ones that they are not. According to Huberman (2001) this is the defining property of familiarity. Huberman (2001) argues that due to preference for familiar and distaste for and fear from unfamiliar leads to the basic result that people simply prefer to invest in familiar securities. This is probably due to the fact that investors tend to feel 4

5 they know more about the stocks that they are familiar with. Merton (1987) develops a capital market equilibrium model in which each investor knows only a subset of available securities. The subset differs across investors. The model implies that investors make their investment decisions from the stocks that they are familiar with. According to Merton, increasing analyst coverage for the firm can also help increase investor base and grab their attention. Massa and Simonov (2006) use Swedish data set, including income, wealth, demographic variables, and some other additional control variables and find that familiarity affects individual investors investment decisions. Investors face a challenge when they decide to buy a security among many alternatives that is beyond the capabilities of human capacity to analyze and select. Hence, when deciding what securities to invest, individual investors should simplify the search process. This means that individual investors focus on securities that grab their attention most, implying that investors will be inclined to invest in familiar securities. The literature presents that familiarity whether in the form of more marketing / advertising, media citation, being local to investor or analyst coverage affects investment decisions. Taking search costs into accounts, Sirri and Tufano (1998) hypothesize that consumers purchase equity funds that are easier or less costly for them to identify. These may be among funds with more marketing expenses than competitors and those receiving greater media attention which increases brand awareness. Investors will probably put this fund in a consideration set from which they select products. The authors find that a larger share of media citation is related to faster growth in funds. Although the authors state that they cannot disentangle the direction of causality, the findings indicate that the more familiar the investors are with a security, the more likely they are to buy it as it will be in the consideration set of the investor. Jain and Wu (2000) and Barber et al. (2005) also find that individuals invest in securities that they are familiar with, familiarity being increased through advertising. Grinblatt and Keloharju (2001) argue that home bias may be a part of a larger phenomenon in which investors exhibit a preference for the familiar companies. As the authors mention, familiarity has many facets such as distance of the headquarter of the stock from investor, similarity in culture and / or language of the firm may be the roots for familiarity. Using these facets as proxies for familiarity, authors find that investors in Finland are more likely to buy stocks that are familiar to them. Coval and Moskowitz (1999) show that the preference for investing close to home also applies to domestic stock portfolios. According to authors, investment managers exhibit a strong preference for locally headquartered firms, particularly small, highly levered firms. As the firm size increases, more non local investors add the security to their portfolio. These results suggest that investors prefer the securities that they are more familiar with and have advantage over nonlocal investors due to asymmetric information. Coval and Moskowitz (2001) confirm the findings also for mutual fund managers that fund managers trade local securities at an informational advantage due to familiarity towards these assets. Zhu (2002) analyzes individual investor preference for nearby investments for equities. The author argues that local bias (the tendency to invest in nearby investment alternatives) and home country bias may be a function of the same underlying driving factor, familiarity bias. The results confirm that both institutional and individual investors tend to hold stocks of companies with nearby headquarters (individuals exhibiting higher degree of bias). 5

6 In line with literature, we hypothesize that a significant portion of Turkish individual equity investors invest in stocks that they are familiar with. Barber and Odean (2008) argue that professional investors are less prone to familiarity bias than individual investors. Hence, we expect sophisticated investors to have lesser degree of familiarity bias Representativeness Heuristic Representativeness heuristic describes the degree to which a sample is similar to another sample in all essential characteristics. It is based on stereotypes. Tversky and Kahneman (1971) argue that people have erroneous intuitions about chance. Due to law of small numbers, they view a sample randomly drawn from a population highly representative of the population which can be described as representativeness heuristic. Representativeness can affect the prediction procedure of individuals. Tversky and Kahneman (1974) state that people often predict the future value of a stock based on representativeness. If this is the case, investors will be inclined to buy stocks, which have been increasing recently (extrapolation bias). In an experimental study, Andreassen and Kraus (1990) analyze the effects of stock market trends in investment decisions. Investors extrapolate recent past stock price trends which results in more purchasing after two successive bull periods and more selling after two successive bear periods. Extrapolation of stock price trends to the future may be related to representativeness heuristic since investors may think that recent short period of price movements is derived from a process with bull (bear) characteristics. As presented by Lakonishok et al. (1994), in the long run (3-5 years) value stocks outperform growth stocks which cannot be attributed to riskiness of value stocks. The authors argue that, investors think recent past performance of growth stocks will continue in the future as they extrapolate the return trend of these stocks and invest in growth stocks. When it turns out that return patterns do not realize as investors predict, value stocks outperform growth stocks in the long run. According to authors, investors make judgment errors and extrapolate past growth into the future. Empirical research in finance literature identified two patterns on stock returns: underreaction over shorter periods (1-12 months) and overreaction in longer periods (3-5 years). Barberis et al. (1998) develop a theoretical model to explain these two phenomena. The underlying basics of the model depend on representativeness heuristic as well as conservatism. Extrapolation of past returns is the form of representativeness in the model. Individuals who exhibit representativeness heuristic extrapolate past performance into the future. Representativeness in the model assumes that short term trend in the price movements will be followed in the longer term. Benartzi (2001) uses retirement saving plans of S&P 500 firms. The author finds that there is a positive correlation between past returns and subsequent allocations to company stocks, and that correlation gets stronger as the return accumulation period lengthens. This implies that employees extrapolate past returns into the future. Benartzi confirms the extrapolation hypothesis using a survey conducted on Internet among Morningstar subscribers. According to survey results, past returns of stocks are likely to persist, which is supportive evidence for extrapolation hypothesis. In line with findings from theoretical and empirical research, we hypothesize that representativeness heuristic is common among Turkish individual stock investors. 6

7 Findings presented by Grether (1980) confirm representativeness heuristic for inexperienced or financially unmotivated subjects; the evidence is less clear for other subjects. Chen et al. (2007) find that representativeness heuristic is only applicable to individual investors; institutional investors being unaffected by recent past return performance. Hence we also hypothesize that sophisticated investors are less prone to representativeness heuristic Status Quo Bias Most real decisions have a default alternative of doing nothing. Samuelson and Zeckhauser (1988) define status quo as doing nothing or maintaining one s current or previous decision. In an experimental setting, the authors show that individuals stick to status quo. As Tversky and Shafir (1992) state, choice always produces conflict because individuals have difficulties in trading off costs against benefits or comparing risks against value which makes it difficult to give important decisions. Making decisions becomes more complicated due to uncertainty about the actions. When each alternative has its own advantages and disadvantages or when each alternative has risks, then individuals face difficulties to make decision. This may lead individuals to refrain from making decisions and stick to their current positions or at least delay the decision and exhibit status quo bias. The authors argue that conflicts about the alternatives can increase the tendency to choose the default option (status quo), not only the tendency to defer choice. Samuelson and Zeckhauser (1988) argue that status quo bias may stem from rational decision making as well as biases such as loss aversion, regret aversion and avoiding cognitive dissonance. Similarly, Kahneman and Tversky (1982) and Ritov and Baron (1995) argue that status quo may stem from regret aversion, Kahneman et al. (1991) relate status quo with loss aversion and Ritov and Baron (1992) argue that status quo is a result of omission bias as keeping status quo requires omissions of choices. Since there are numerous alternatives in equity investments, individuals may just omit the alternatives to prevent facing the difficulties of making decisions. According to Madrian and Shea (2001), preference of default contribution rate and plan in 401(k) plan of employees in a large US corporation is related to status quo bias. Agnew et al. (2003) use transaction data of participants from retirement plan of a large firm in US. They find that these investors infrequently rebalance their portfolios and tend to maintain their initial asset allocations, which imply status quo bias. Although a strand of literature shows that individual investors are overconfident and trade excessively, studies with retirement plans reveal that investors may exhibit status quo bias in their investment decisions. The analysis based on brokerage house data yield excessive trading whereas analysis based on retirement plan data reveals infrequent trading. Since equity investments have many alternatives to decide among (whether to buy, sell or hold, when to buy/sell or what to buy/sell), and risks and benefits may not be evaluated easily, equity investors may be inclined to stick to status quo (do nothing) or just defer the decision. As psychology literature suggests, we expect a portion of stock investors to keep their portfolio positions unchanged. Status quo is related to reluctance to trade whereas overconfidence is related to excessive trading. Hence, as argued by Hoffmann et al. (2010), it can be assumed that individuals in the opposite edge of overconfidence scale are subject to status quo bias. 7

8 Madrian and Shea (2001) find that men prefer default plan to a lesser degree than women and default contribution rate declines significantly with compensation. These findings imply that women may have higher degree of status quo bias than men and more sophisticated/experienced individuals have lower degree of status quo bias. We also expect that women exhibit higher degree of status quo bias than men. We hypothesize that sophisticated individuals exhibit status quo bias to a lesser degree than less sophisticated investors. 3. Data and Methodology 3.1. Data The analysis is based on Turkish individual stock investors. The main data set consist of all buy and sell transactions as well as monthly stock only and total portfolio positions (stock, funds, private sector bonds and warrants) of whole Turkish individual investors in The second data set consists of demographic and other information of these investors (age, gender, geographical region of residence, account open date). Pursuant to the permission of Capital Markets Board (CMB) and Istanbul Stock Exchange (ISE), analysis on these data sets has been conducted on Central Registry Agency (CRA) servers due to privacy restrictions stock market performance is slightly bearish. ISE100 index, which consists of the largest 100 companies, decreased from 67,608 at the beginning of year to 51,267. However, out of 253 trading days, ISE100 index had positive returns at 129 days and negative returns at 124 days. According to CRA monthly statistics as of 2011 January, total number of Turkish individual stock investors is around 1 million. However a significant portion of these investors is either dormant or have very low stock portfolio value. When data set is limited to individual stock investors whose total stock portfolio in any month in 2011 is above 5,000 TL (approximately US$ 3000), number of investors reduces to 432,085. Of these, 74,051 investors do not have any buy or sell transactions (dormant investors) during the entire year. Dormant investors are mostly at high ages as 75% of them older than 50 and are in the stock market for a long period of time. 66% opened their accounts before Female investors constitute 41% of the dormant investors and 18% of the active investors. In order not to distort overall results, these investors are excluded from the analysis, reducing number of investors to 358,034 (labeled as expanded investor set). Table 1 shows that total trading volume of these investors is billion TL (buy and sell amounts divided by two), 76% of total trading volume in ISE in 2011, indicating that the sample has significant influence on price formation in the stock market. 15% of remaining trading volume is attributable foreign investors and rest (9%) is attributable to low portfolio value Turkish individual stock investors (investors with 2011 monthly average stock portfolios lower than 5,000 TL) and local institutional investors. Some of these investors do not have any buy or sell transaction. Hence, data set is further limited to those investors who have at least 1 buy and 1 sell transaction, reducing data set to 305,546 investors. However, a portion of these investors have very high annual turnover values such as 50,000 and even increasing to 10 billion levels for a few investors. One possible explanation is that these investors (labeled as abnormally high turnover investors) have their wealth managed by professional money managers and / or they act like day traders and scalpers. As it seems that they have different investment characteristics, in 8

9 order not to distort overall analysis and use same sample for all biases / proxies, these investors are also excluded from further analysis. Using trial and error and comparing with international benchmarks, high annual turnover cut off point is set to be 100. Although this cut off can be increased (up to 10,000) or decreased, back of the envelope calculations reveal that overall results do not change significantly. Capping turnover at 100, sample size decreases to 244,146 investor (labeled as analysis investor set) with exclusion of 61,400 abnormally high turnover investors. This data set is used as the data set for detailed analysis. Although analysis investor set is significantly reduced, majority of the results are also confirmed using the expanded data set with 358,034 investors. Table 1 shows that total trading volume of the investors is billion TL (average of buy and sell amounts), which is 22% of the total trading volume in ISE in The investors made 31.9 million buy transactions amounting to 149 billion TL and 31.7 million sell transactions amounting to billion TL. Average buy volume is 4,674 TL, slightly higher than average sell volume of 4,621 TL. Demographic break down of 244,146 investors are presented in Table 2. Due to confidentiality reasons, CRA provided categorized data for age, experience and wealth. Age is the age of investors as of Wealth is the average of 12 end of month portfolios consisting of equity, funds, warrants and corporate bonds. Experience is the date the investor opens the account (if more than one accounts available, opening date of oldest account taken into account). Region is the geographical region of residence of investor registered in CRA database. Male investors constitute 83% of the total investor base age groups constitute 76% of all investors. 76% of the investors have average 2011 wealth of 50,000 TL or less. 90% of the investors have 3 or more years of investment experience in stock exchange. Almost half of the investors (45%) reside in Marmara Region, mostly in Istanbul, which is the largest city in Turkey. Next is Central Anatolia with 17%, probably mostly Ankara, which is the 2 nd largest city. 15% is in Aegean, in İzmir, the 3 rd largest city. Marmara region is the most developed and Southeast Anatolia region is the least developed among the regions in terms of welfare, income, education, etc. Demographics of abnormally high turnover investors (61,400 investor with turnover higher than 100) are slightly different. Compared to analysis investor set, abnormally high turnover investors are mostly male (88% versus 83%), younger (investors up to 35 years old are 21% versus 27%), not as wealthy 1 (investors with wealth up to 10,000 TL - approximately US $6,000 - are 41% versus 34%) and more experienced (account open date 2002 or earlier 49% versus 36%). There is no difference in terms of region of residence. However, as expected, investors with abnormally high turnover have significantly more buy and sell transactions than those in analysis data set. Number of buy trade higher than 1,000 at 12% (versus 2% in analysis investor set), total value of buy trades higher than 1.5m TL (approximately US$ 800K-850K) at 20% (versus 6% in analysis investor set), number of sell trade higher than 1,000 at 12% (versus 2% in analysis investor set) and total value of sell trades higher than 1.5m TL at 20% (versus 6% in analysis investor set). 1 GDP per capita in Turkey is USD 10,469 in

10 3.2. Methodology Using a theoretical model, Harris and Raviv (1993) show that, differences in opinions lead to trading among investors. Hence, trading volume is related to different expectations among investors. Differences in opinions are result of different interpretation of same signal by investors. As they rely on their beliefs and decisions more, overconfident investor s interpretation of the same signal will significantly differ compared to rational investors. This difference should cause increased trading volume for overconfident investors. De Bondt and Thaler (1995) state that the key behavioral factor to understand trading puzzle is overconfidence. Kyle and Wang (1997) and Benos (1998) argue that overconfident investors trade more aggressively as they believe that they have better information. Kahneman and Riepe (1998) propose that overconfidence should be expected from those who do not face similar problems every day, make explicitly probabilistic estimates and obtain feedback on outcomes of their decisions, implying that individual stock investors are likely to be overconfident. Odean (1998) develops a theoretical model in which overconfident investors overestimate the precision of their knowledge about the value of an asset. These investors overestimate the probability that their personal assessment of an asset s value is more accurate than that of others. Thus, overconfident investors believe their valuations more which increases the differences in opinions among individual investors. The author proposes that that trading volume increases when investors are overconfident. Odean (1999) tests this hypothesis using data provided by a nationwide discount brokerage house in US. He argues that if traders are overconfident in precision of information, then average return of securities they sell must outperform average return of securities they buy. He finds that average return of securities sold outperform average return of securities purchased over horizons of four months, one year and two years. The author looks for possible explanations to excessive trading resulting in losses and eliminates meeting liquidity needs, realizing tax losses and rebalancing the portfolio or moving to a less risky portfolio. He concludes that excessive trading resulting in losses may be due to overconfidence. Barber and Odean (1999), Barber and Odean (2000), and Hirshleifer and Luo (2001), Gervais and Odean (2001), Barber and Odean (2001), Barber and Odean (2002), Chuang and Lee (2006), Statman et al. (2006), Glaser and Weber (2007), Graham et al. (2009), Glaser and Weber (2009), Grinblatt and Keloharju (2009), Hoffmann et al. (2010) also confirm that overconfident investors trade more. Barber and Odean (2001) define monthly portfolio turnover as one-half the monthly sales turnover plus one-half the monthly purchase turnover. The monthly sales turnover is calculated as the shares sold in month t times beginning of month price divided by total beginning of month t market value of household s portfolio. The monthly purchase turnover is calculated as the shares purchased in month t-1 times beginning of month t price divided by total beginning of month t market value of household s portfolio. Annual turnover is simply twelve times monthly turnover. Similar to Barber and Odean (2001), we measure overconfidence as annual turnover. Higher turnover implies higher overconfidence. Since both theoretical and empirical findings for turnover are robust, it is used as the main proxy to measure overconfidence while others are used for robustness checks. Josephs et al. (1992) argue that low self esteem individuals take less risk than individuals high in self esteem. As Campbell (1990) shows, high self-esteem people have higher confidence. Hence, it can be inferred that overconfident investors tend to take more risk. Chuang and Lee (2006) find that overconfident investors trade more in riskier securities. They measure riskiness of a security as return volatility and firm specific risk (return volatility minus market component). Glaser and Weber (2009) also 10

11 find that individuals buy high risk stocks. These findings imply that overconfidence can also be measured by using portfolio riskiness. Consistently, percentage of stocks from ISE 30 (as these stocks have high market capitalization and high liquidity, they are assumed to be less risky) and percentage of small stocks in the portfolio (assuming smaller firms are riskier) are used as proxies of portfolio riskiness in this study. For all month ends, number of different stocks from ISE 30 divided by total number of different stocks in the portfolio is calculated. Average of 12 months (ISE30 ratio) is used to measure portfolio riskiness. The lower the percentage, the riskier the portfolio is. For example, suppose a portfolio consists of stocks A, B and C and suppose A and B are in ISE 30. For this portfolio, ISE 30 ratio is calculated to be 67%. Likewise, for all month ends, number of different stocks labeled as small based on market capitalization divided by total different number of stocks is calculated. Average of 12 months (small Mcap ratio) is used to measure portfolio riskiness. The higher the percentage, the riskier the portfolio is. Firms with market capitalization lower than USD 100m are labeled as small. As of 2011 year end, almost 50% of stocks have Mcap lower than USD 100m. Maximum Mcap is USD 13,119m. Using return data, we found that volatility of small stocks is on average larger than rest of the stocks. Besides, average volatility of stocks in ISE30 is smaller than rest of the stocks. Hence, taking also return volatility into account, ISE30 stocks turn out to be less risky and small stocks are more risky as expected. Heath and Tversky (1991) argue that as explained by competence hypothesis, overconfident investors may forego the advantage of diversification and concentrate on a small number of companies with which they are more familiar with. Odean (1998) finds that overconfident traders hold under-diversified portfolios. Goetzmann and Kumar (2008) find that high portfolio turnover, which is a sign of overconfidence is related to under-diversification. According to authors, this finding implies the more overconfident investors hold under-diversified portfolios along with investors with a tendency in local stocks (familiarity bias). Glaser and Weber (2009) argue that, with increased portfolio turnover, individuals reduce number of stocks in their portfolio. These findings imply that overconfidence can be measured using diversification. In line with literature, average number of stocks in the portfolio is used as a naïve way of measuring diversification level. Odean (1999) suggests that securities that have performed unusually poor or well are more likely to be discussed in the media, more likely to be considered by individual investors and as a result more likely to be purchased. He finds that the investors tend to buy securities that have risen or fallen more over the previous six months than the securities they sell. Gervais et al. (2001) find that stocks experiencing high trading volume over a day or week tend to appreciate over the following month. The findings imply that shocks to trading activity increase a stock s visibility and demand in the upcoming days increase. Hirshleifer et al. (2008) use transaction data of individual investors from a brokerage house and find that investors are net buyers after both negative and positive extreme earnings surprises, consistent with an attention effect. This can be interpreted as stocks with extreme positive or negative earnings grab attention of investors, whose familiarity towards these stocks increase and tendency to invest in these stocks increase. Barber and Odean (2008) argue that buying behavior of individual investors is heavily influenced by stocks that draw their attention. Authors use stock news in the media, unusual trading volume and extreme returns as proxies for attention grabbing factors. The authors find that abnormal 11

12 trading volume is the best indicator of attention while return and news metric follow abnormal trading volume. Findings imply that familiarity bias can be measured by looking at the relation between stock purchases and factors increasing familiarity towards these stocks. The more the investor is exposed to the stock, the more familiar he or she becomes. From this standpoint, previous ownership is expected to be a good measure for familiarity bias. After an investor buys a stock, it becomes more familiar. Following this argument, all purchase transactions are flagged if the stock has been purchased previously in Number of flagged purchase transactions divided by total number of purchase transactions is used as a proxy (previous ownership ratio) to measure familiarity bias. Higher previous ownership ratio indicates higher familiarity bias. Previous ownership ratio is used as the primary proxy to measure familiarity bias as it is the most direct indicator of familiarity towards a stock whereas others will be used for robustness checks. Similar to Barber and Odean (2008), extreme return can also be used to measure familiarity bias. Number of stock purchase transactions with absolute abnormal return (positive or negative) divided by total number of stock purchase transactions (absolute abnormal return ratio) is used as a proxy to measure familiarity bias. Higher ratio indicates higher familiarity bias. A purchase transaction is counted to have absolute abnormal return if absolute value of previous day return of stock divided by previous day ISE100 (index composed of largest 100 companies in ISE) return is above 125%. This cut off point is determined based on the absolute return of stocks and ISE100. In 2011, average of mean absolute return of stocks is 2.03% whereas mean absolute return of ISE100 is 1.27%. On average, 123 days of 253 trading days, stocks' absolute return is higher than 125% of ISE100 absolute return (minimum 0 days, maximum 182 days). As presented in Barber and Odean (2008), unusual trading volume can also be used to measure familiarity bias. Number of stock purchase transactions with abnormal volume change divided by total number of stock purchase transactions (abnormal volume ratio) is used as a proxy to measure familiarity bias. Higher ratio indicates higher familiarity bias. A purchase transaction is counted to have absolute volume change if value of previous day volume change (versus 2 days ago) of stock divided by previous day ISE100 volume change (versus 2 days ago) is above 150%. As proposed by Merton (1987), analyst coverage can be used as another proxy to measure familiarity bias. It has been hypothesized that the more analyst covers a stock, the more likely that it will grab attention of investors. Hence, average number of analysts covering stocks purchased can be used to measure familiarity bias. Higher analyst coverage indicates higher familiarity bias. Chan et al. (2004) argue that representativeness heuristic may lead investors to extrapolate past performance of assets into the future and thus, set prices too low or too high which in turn generates return reversals. This argument implies that representativeness heuristic can be measured by using the relation between stock purchases and recent past performance of stocks. Chen et al. (2007) use transaction data of a large Chinese brokerage house to analyze representativeness heuristic in Chinese investors. The authors use extrapolation as a form of representativeness heuristic. They find that 4 month prior performance of stocks purchased is surprisingly high whereas past 1-year return is almost normal. This finding indicates that investors extrapolate recent past returns of stocks they purchase. Barber et al. (2009) 12

13 use extrapolation as a form of representativeness heuristic and measure 3 year prior market adjusted return of stocks purchased by individual investors. The authors find that, individual investors buy stocks with strong past returns. This relation peaks in 1-year prior to purchase and lasts till 3 years prior to purchase. Hence, as presented in Chen et al. (2007) and Barber et al. (2009), representativeness heuristic can be measured by the degree to which investors make their buy decisions according to recent past trend of stock prices. Chen uses prior 4-month and 1-year returns. Barber et al. (2009) finds that representativeness heuristic peaks with 1 year prior returns and diminishes in the longer periods. As stated in Bildik and Gülay (2007), Turkish individual stock investors are more myopic. Hence, we employ shorter time period. For each buy transaction, positive return trend is calculated to be number of positive returns in last 90 trading days prior to purchase date divided by 90. For each investor, representativeness heuristic is measured as average positive return trend for all purchases. Representativeness heuristic is also measured for last 30 as well as 150 trading days before purchase date using the same calculation methodology. The status quo bias is related to doing nothing or maintaining current decisions, implying that status quo bias involves reluctance to trade. Hence, individuals exhibiting status quo bias are expected to keep their current portfolios unchanged. The more the portfolio of an individual changes, the more decisions he/she has given implying lesser degree of status quo bias. Using all buy and sell transactions in 2011, end of day portfolios for each investor are formed. Average percentage of change in number of stocks in the portfolios for each day is used to measure status quo bias (portfolio percentage change). The higher the portfolio percentage change, the lower the status quo bias. For example, suppose in day 1, portfolio consists of 2 A and 4 B stocks and suppose in day 2, portfolio consists of 4 A, 2 B and 2 C stocks, daily percentage change in the portfolio is 67% (50% change in A, 50% change in B and 100% change in C divided by 3 representing number of stocks A, B, C). Correlation of proxies for each bias is presented in Table 3. Turnover is negatively correlated to ISE30 ratio and diversification and positively correlated to small stock ratio. Small stock ratio is by definition negatively correlated to ISE30 ratio and not correlated to diversification. Correlation of diversification with other overconfidence proxies is either low or insignificant. Similarly, correlation among familiarity bias proxies is either insignificant or too low. All correlations among representativeness heuristic are statistically significant, positive and high. For all proxies, using histograms and descriptive statistics, level of prevalence of each bias among Turkish individual stock investors is assessed. Then regression analysis is conducted to determine how each demographic factor affects behavioral biases taking others into account. In this regression model, bias(es) are overconfidence, familiarity bias, representativeness heuristic and status quo bias. Age is the age of investor and is a continuous variable. Male is a dummy variable, which equals one for male investors. Experience is the date account is opened and is a continuous variable. Wealth_Low is a dummy variable, which equals one for wealth levels up to 10,000 TL and Wealth_High is a dummy variable and is equal to one for wealth levels higher than 100,000 TL. Marmara is a dummy 13

14 variable, which equals one for Marmara (most developed) region and Southeast is a dummy variable, which equals one for Southeast (least developed) region. Coefficients are expected change sign between Wealth_Low and Wealth_High and between Marmara and Southeast. Turnover, previous ownership ratio, 90 day positive return trend and portfolio percentage change are used as main measures of overconfidence, familiarity bias, representativeness heuristic and status quo bias respectively. Other proxies such as ISE30 ratio, small Mcap ratio, and absolute abnormal return ratio are also used for robustness checks. Since explanatory variables are categorical, three additional regression models have been utilized for robustness checks. In these models, wealth is a dummy which equals one either for each wealth level presented in Table 2 or for low and high wealth levels presented above. In these models, experience is either continuous or is a dummy variable which equals one for each experience level presented in Table Results 4.1. Overconfidence i. Turnover Turkish individual stock investors have high turnover. As presented in Table 4, on average, an investor shifts his or her portfolio 11 times annually which is high compared to similar studies. When we compute the mean annual turnover including those with turnover higher than 100, mean turnover increases to 1.15 million mainly due to a small set of investors (around 4, 000 investors) whose turnover is above 1 million, which is extremely high for a typical individual investor. Both standard deviation presented in Table 4 and histogram in Figure 1 confirm that turnover level is highly dispersed. Barber and Odean (2001) find that for a sub sample of US investors, mean turnover ratio is 0.77 for men and 0.53 for women, implying that Turkish individual stock investors have higher turnover than US investors. Chen et al. (2007) find that for Chinese investors, mean turnover is 3.27, significantly higher than US investors, yet still lower than Turkish investors. Taking into account abnormally high turnover investors and international benchmarks, it can be stated that overconfidence is common among Turkish individual stock investors. Table 5 shows that turnover is higher for male investors. Age is nonlinearly related to turnover, increasing up to age group, decreasing afterwards. Turnover decreases with wealth with only exception at the lowest wealth group which has 2nd lowest turnover. This is probably mainly due to low available funds to trade. Investors with annual buy and sell amounts up to 30,000TL constitute 66% of the lowest wealth group investors, reducing to ~52% for second lowest wealth group and further decreasing to ~30% for all investors excluding lowest wealth group. This finding shows that lowest wealth group investors buy and sell low amount of stocks, implying lower overconfidence. Investors in Marmara region have lowest turnover and investors in Southeast Anatolia region have highest turnover. 14

15 ii. Robustness Checks 1. ISE30 Ratio ISE30 stocks constitute 30% of the mean investor portfolio as presented in Table 13. This seems to be high and inconsistent with the overconfidence hypothesis. However as presented in Figure 5, dispersion of ISE30 ratio indicates that 60,369 investor (25% of total investor base) do not have ISE30 stocks in their portfolios and 107,616 (44% of total investor base) have 10% or less ISE30 stocks in their portfolios. Average diversification level of investors with 10% or less ISE30 stock in their portfolios is Number of investors who have only ISE30 stocks in their portfolios is very low at 13,786 (6% of total investor base). These figures reveal that a significant portion of investors have no or very low level of ISE30 stock in their portfolios. This finding supports the hypothesis that a significant portion of investors prefers riskier stocks. Table 14 shows that ISE30 ratio is lower for male investors. Age is nonlinearly related to ISE30 ratio, decreasing up to age group, increasing afterwards. ISE30 ratio increases with wealth and experience. Investors in Marmara region have highest ISE30 ratio and investors in East Anatolia and Southeast Anatolia region have lowest ISE30 ratio. 2. Small Mcap Ratio Stocks with Mcap lower than US $100 million are labeled as small Mcap. As presented in Table 13 2, on average, small stocks constitute 28% of investor portfolios. Although mean small stock ratio is not very high, histogram in Figure 6 shows that there is high amount of investors holding small stocks. 68,361 investors do not have any small stock in their portfolios (75% of these 68,361 investors have on average only 1 or less stock in their portfolios). 54,009 investors (22% of total investors) have 50% or higher small stock ratio in their portfolios. Besides 6,346 (3% of total investors) have only small stocks in their portfolios (with mean diversification of 0.73). These figures reveal that a significant portion of investors have high level of small stocks in their portfolios. Table 15 shows that small Mcap ratio is higher for male investors. Age is nonlinearly related to small Mcap ratio, increasing up to age group, decreasing afterwards. Small Mcap ratio decreases with wealth and experience. Investors in Marmara region have lowest small Mcap ratio and investors in Southeast Anatolia region have highest small Mcap ratio. 3. Diversification On average, investors diversify their portfolios with 3.43 stocks, as presented in Table 13. The median investor holds 2 stocks. Chen et al. (2007) find that for Chinese individual investors, mean diversification is 2.6, lower than Turkish individual investors. Goetzman and Kumar (2008) find that mean diversification in US investors is in the range of , with a monotonic increase between 1991 and 1996, which is by far higher compared to Turkish investors. Barber and Odean (2000) median investor holds 2.61 stocks for the same data set, higher than Turkish investors. 48% of investors hold two or lower 2 82 investors purchased only stocks with new ISIN code for existing stocks (due to reasons such as stock splits etc.), hence analysis based on 244,064 investors. 15

16 number of stocks in their portfolios indicating that a majority of investors do not diversify their portfolios. Both standard deviation and histogram in Figure 7 show that diversification is widely dispersed. Table 16 shows that diversification is lower for male investors. Age is nonlinearly related to diversification, decreasing up to age group, increasing afterwards. Diversification increases with wealth and experience. Investors in Marmara region have highest diversification and investors in Southeast Anatolia region have lowest diversification. Table 3 displays that turnover is negatively correlated to ISE30 ratio and diversification and positively correlated to small stock ratio. Small stock ratio is by definition negatively correlated to ISE30 ratio and not correlated to diversification. Correlation of diversification with other proxies is either low or insignificant, implying that diversification is not as good as other proxies to measure overconfidence or not necessarily measuring overconfidence. Hence, further analysis for overconfidence robustness checks is based on portfolio riskiness (measured by ISE30 ratio and small stock ratio). iii. Regression Results Results are presented in Table 6. As expected, overconfidence decreases with age. Male investors are more overconfident than female investors, which confirm the vast majority of findings in literature. Experience increases overconfidence contrary to expectations. However, this finding is valid only for low wealth investors. Experience decreases overconfidence for high wealth investors. Hence, it is probable that experience per se is not related to overconfidence and possible interactions with other factors should be factored in the analysis. Another possible explanation is the definition of experience. Account opening date does not necessarily imply high experience. An investor may gain experience in a shorter period of time with high frequency trading. Hence, a better measure for experience is needed to better understand the relation between experience and overconfidence. Wealth decreases overconfidence. Wealth may be related to financial sophistication as wealthier investors have better access to information and can leverage on professional portfolio management. Investors in Marmara region have lower and investors in Southeast Anatolia region have higher overconfidence. Turnover difference between regions is not related to gender, age, experience or wealth. Marmara region is economically more developed than Southeast Anatolia region. Besides, percentage of university graduates is higher in Marmara region (13% versus 6% in Southeast region) 3. These two factors indicate that financial literacy in Marmara region is most probably higher than that of in Southeast Anatolia region, implying that increase in financial literacy decreases overconfidence. Wealth and regional results imply that sophisticated investors are less prone to overconfidence. Regression results are confirmed for sub samples (male only, female only, low / high age, low / high experience, low / high wealth regressions). Our findings are also robust to different regression models and different proxies, results of which are presented in Tables Although not presented here, results do not change for ISE30 and small Mcap regressions when data set is expanded to 358,034 investors. Although our findings are robust to different measures and models, excluding return data from the analysis imposes a limitation as high turnover does not necessarily imply overconfidence. Lower returns 3 Based on Turkish Statistical Institute data 16

17 should accompany turnover to confirm overconfidence. Yet, as Barber and Odean (2000) and Barber et al. (2009) show, individual investors have poor trading performance. Besides, ISE30 and small Mcap results which are independent of return data, confirm turnover results. These two factors mitigate the limitation imposed by lack of return data Familiarity Bias i. Previous Ownership As presented in Table 4, our findings demonstrate that almost 50% of the stocks purchased by the investors in 2012 have also been previously purchased by the same investors in Histogram in Figure 2 shows that 42% of investors have 50% or lower previous ownership ratio. 32,628 (13% of investor base) investor purchased stocks which they did not purchase in 2011 previously. However, of these 32,628 investors, 71% made 1 to 5 purchase transactions, which shows that previous ownership ratio for lower end of histogram should be read carefully. These findings imply that a significant amount of investors purchase stocks that they are familiar with through previous ownership. Table 7 shows that previous ownership ratio is higher for male investors. Age is nonlinearly related to previous ownership ratio, increasing up to age group, decreasing afterwards. Previous ownership ratio increases with wealth and experience. Investors in Marmara region have lowest previous ownership ratio along with Black Sea and Aegean regions and investors in Southeast Anatolia region have highest previous ownership ratio. ii. Robustness Checks 1. Absolute Abnormal Return Absolute abnormal return ratio is simply number of stock purchase transactions with absolute abnormal return divided by total number of stock purchase transactions where abnormal return is defined to be returns higher than 125% of ISE100 return. Our findings presented in Table 13 shows that on average 59% of all purchase transactions has abnormal previous day absolute abnormal return. Barber and Odean (2008) find that individual investor attention display attention-driven buying behavior and net buyers of extreme negative and positive one day return stocks, which is in line with our finding. 9,890 investors (4% of investors) have only purchased stocks with no previous day abnormal return. However, of these 9,890 investors, 81% made 1 to 5 purchase transactions. Besides, histogram in Figure 8 shows that 79% of investors have higher than 0.5 absolute abnormal return ratio. These figures along with mean ratio show that a significant portion of investors buys stocks which have high absolute abnormal previous day return. Table 21 shows that absolute abnormal return ratio is higher for male investors. Age is nonlinearly related to absolute abnormal return ratio, increasing up to age group, decreasing afterwards. Absolute abnormal return ratio decreases with wealth and experience. Investors in Marmara region have lowest absolute abnormal return ratio and investors in Southeast Anatolia region have highest absolute abnormal return ratio along with Black Sea and East Anatolia regions. 17

18 2. Abnormal Volume We find that on average 42% of all purchase transactions have abnormal volume ratio, as presented in Table 11. Histogram in Figure 9 shows that 29% of investors have 0.5 or higher abnormal volume ratio. 77% of investors are concentrated in region. Both descriptive statistics and histogram shows that a significant portion of investors buy stocks who have previous day abnormal volume. Table 22 shows that abnormal volume ratio is lower for male investors. Abnormal volume ratio increases with age, wealth and experience. Investors in East Anatolia region have lowest abnormal volume ratio and investors in Aegean region have highest previous day abnormal volume ratio, yet means are not statistically different between any of the regions. 3. Analyst Coverage We also analyze the analyst coverage of the stocks purchased by investors in our sample. Analyst coverage data is obtained from Bloomberg. As presented in Table 11, our findings show that maximum number of analysts covering a stock is 29 (DOCO) and minimum number of analyst covering a stock is 0 (for 195 stocks). 152 of these 195 stocks have small MCap (lower than USD 100m). Besides, correlation between analyst coverage and size is 0.588, statistically significant and high. These findings indicate that analysts are covering larger stocks as expected. On average, stocks that were purchased by investors have been covered by 7.7 analysts. Histogram in Figure 10 shows that, 10,315 (4% of investors) purchased stocks which have not been covered by any analyst. However, of these 10,315 investors, 48% have made 5 or less purchase transactions. 29% of investors have purchased stocks covered by 10 or more analysts. Table 23 shows that analyst coverage is lower for male investors. Age is nonlinearly related to analyst coverage, decreasing up to age group, increasing afterwards. Analyst coverage increases with wealth and experience. Investors in Marmara region have highest analyst coverage and investors in East Anatolia region have lowest analyst coverage followed by Southeast Anatolia region. Familiarity bias may stem from any attention grabbing event, which are hard to capture with one specific measure. Previous ownership is a direct indicator of familiarity. Hence, it is used for further familiarity bias analysis. As a secondary measure, although not correlated much, absolute abnormal return is used for robustness check as abnormal return changes is more likely to be attention grabbing and seems more related to previous ownership compared to abnormal volume. Table 3 shows that correlation among proxies is either insignificant or too low. Analyst coverage is low yet negatively correlated to previous ownership. Analyst coverage may be increasing investor's information set about stocks and hence serve as a de-biasing tool rather than triggering familiarity bias. Besides, messages shared with investors are important as negative messages for a stock may lead investors refrain from the stock rather than purchasing it. Hence, abnormal volume and analyst coverage are not used for further familiarity bias robustness checks. 18

19 iii. Regression Results Results are presented in Table 8. As expected, familiarity bias decreases with age. Male investors exhibit familiarity bias more than female investors. Experience increases familiarity bias contrary to expectations. This is probably due to high correlation between age (0.458) and experience. When age is omitted from the model, experience turns out to negatively and significantly affect familiarity bias. Age confounds with experience as interaction term between age and experience is statistically significant. Hence, age and experience may be measuring same underlying factor affecting familiarity. Another possible explanation why experience positively affects familiarity bias is the definition of experience. Account opening date does not necessarily imply high experience. An investor may gain experience in a shorter period of time with high frequency trading. Wealth increases familiarity bias contrary to expectations. Due to definition, previous ownership ratio is positively correlated with number of buy transactions (0.607) which is also positively correlated with wealth (0.304), leading to wealth positively affecting familiarity bias. When number of buy transactions is added as a control variable, wealth turns out to negatively affect familiarity bias. Negative effect of wealth on familiarity bias is also confirmed with absolute abnormal return regression analysis presented in Table 24. Investors in Marmara region have lower and investors in Southeast Anatolia region have higher familiarity bias. Difference between these two regions is not related to gender, age, experience or wealth. Marmara region is economically more developed than Southeast Anatolia region. Besides, percentage of university graduates higher in Marmara region (13% versus 6% in Southeast Anatolia region). Similar to findings in overconfidence, financial literacy decreases familiarity bias. Wealth and region results imply that sophisticated investors are less prone to familiarity bias. Regression results are confirmed for sub samples (male only, female only, low / high age, low / high experience, low / high wealth regressions). Results are also fully confirmed for age, gender and wealth and partially confirmed for experience and Southeast Anatolia region using different proxies and regression models as presented in Tables Although not presented here, results do not change when data set is expanded to 358,034 investors. Stock prices in buy transactions may affect familiarity bias, as investors perception to high price stocks may be different than low price stocks. Hence omitting stock prices imposes a limitation on our results. Familiarity bias may also arise due to many different factors (investor being employee of the company, investor living within proximity of the company, advertising & marketing efforts of the company, word of mouth, stock specific or investor specific any other attention grabbing emotional or rational factor). Hence, it is extremely difficult to find proxies to measure familiarity bias confirming each other Representativeness Heuristic Correlation among 30, 90 and 150 trading day positive return trends is presented in Table 3. All correlations are statistically significant, positive and high. Hence, only results for 90 trading day positive return trend are presented. 19

20 i. 90 Trading Day Positive Return Trend Table 4 shows that on average, stocks purchased have positive returns 42% of the days in 90 trading days prior to purchase. Mean positive return trend is 43.2% for 30 trading days and 41.2% for 150 trading days, economically not different from 90 trading day return trend, although statistically different. Histogram in Figure 3 shows that 72% of investors have purchased stocks whose returns in last 90 days prior to purchase were positive between 40% and 50% of the time. ISE100 index is positive on 52.4% in last 90 trading days for each trading day in These findings reveal that, investors are not very positive trend chasers consistent with the findings of Bildik and Gülay (2007). Table 9 shows that 90 trading day positive return trend is lower for male investors. Age is nonlinearly related to 90 trading day positive return trend, decreasing up to age group, increasing afterwards. 90 trading day positive return trend increases with wealth (means are not significantly different in lower wealth levels). 90 trading day positive return trend increases with experience (decreasing for 30 day trend). Investors in Marmara region have highest 90 trading day positive return trend and investors in Southeast Anatolia region have lowest 90 trading day positive return trend. Although our findings are statistically significant, as means are very close to each other, they are not economically significant. ii. Regression Results Results are presented in Table 10 show that representativeness heuristic increases with age. Male investors exhibit representativeness heuristic less than female investors. Experience decreases representativeness heuristic. Wealth increases representativeness heuristic. Investors in Marmara region have higher and investors in Southeast Anatolia region have lower representativeness heuristic. Difference between regions is not related to gender, age, experience or wealth. Our findings are robust to different proxies and regression models as presented in Tables Although not presented here our results do not change when data set is expanded to 358,034 investors. Relation between demographic factors and representativeness heuristic are just the opposite of relation between overconfidence and familiarity bias, implying that proxies may not be measuring representativeness heuristic. Besides, as means are not economically different from each other and explanatory power of the regression models is very low, we strongly suggest using new measures in further studies. Due to restrictions on access to data, it was not possible to perform further analysis. Market adjusted 90 day positive return trends can be used to further analyze representativeness heuristic. Additionally, consistent to findings of Bildik and Gülay (2007), Turkish individual investors might be more myopic, implying that shorter time periods might perform better in explaining representativeness heuristic. 20