Informed Trading through the Accounts of Children

Transcription

1 THE JOURNAL OF FINANCE VOL. LXIX, NO. 1 FEBRUARY 2014 Informed Trading through the Accounts of Children HENK BERKMAN, PAUL D. KOCH, and P. JOAKIM WESTERHOLM ABSTRACT This study shows that the guardians behind underaged accounts are successful at picking stocks. Moreover, they tend to channel their best trades through the accounts of children, especially when they trade just before major earnings announcements, large price changes, and takeover announcements. Building on these results, we argue that the proportion of total trading activity through underaged accounts (labeled BABYPIN) should serve as an effective proxy for the probability of information trading in a stock. Consistent with this claim, we show that investors demand a higher return for holding stocks with a greater likelihood of private information, proxied by BABYPIN. THIS STUDY INTRODUCES A novel measure of the probability of information-based trading in a stock, namely, BABYPIN, the proportion of total trading through the accounts of underaged investors. We begin by empirically validating this measure by showing that underaged accountholders are extremely successful at picking stocks, especially when they trade just before large price changes, major earnings announcements, and takeover announcements. We next show that BABYPIN is priced in the cross section of stock returns, consistent with Easley and O Hara (2004). There are two reasons to expect a high proportion of informed trading through underaged investor accounts. First, guardians who open accounts and trade on behalf of young children are likely to be above-average investors. We expect these individuals to have more wealth (to bestow on offspring) and to be more successful at investing, possibly due to superior cognitive skills or Henk Berkman is at the University of Auckland Business School. Paul Koch is at the University of Kansas School of Business. Joakim Westerholm is at the University of Sydney Business School. We acknowledge the helpful comments of Ferhat Akbas, Chris Anderson, Audra Boone, Jennifer Conrad, Bob DeYoung, Doug Foster, Campbell Harvey (Editor), Scott Hein, Felix Meschke, Kelly Welch, Jide Wintoki, an anonymous referee, and Associate Editor, and seminar participants at the University of Auckland, the University of Kansas, the conferences of the Financial Management Association, the Asian Financial Management Association, the Southern Finance Association, and the Australian National University Microstructure Meeting. We also appreciate the excellent research assistance of Evan Richardson and Suzanna Emelio. DOI: /jofi

2 364 The Journal of Finance R comparative advantages in obtaining value-relevant information. 1 These attributes, combined with a basic parental instinct to share the benefits of any information advantage with one s offspring, could lead to a disproportionate number of underaged accounts that bear the fruits of informed trading. Second, underaged accounts can be used to camouflage illegal insider trading by guardians. Although such behavior may seem unlikely, Internet Appendix Section I documents several recent insider trading cases involving the accounts of children. 2 Note that, while we use the term camouflage in the sense of hiding from authorities, an informed guardian could also use the accounts of children to reduce the price impact of trades, in the sense of Kyle (1985). We employ data from Euroclear Finland Ltd. (Euroclear) during the period 1995 to Euroclear records all changes in daily shareholdings for every investor trading on the Nasdaq OMX Helsinki exchange, as well as the age of the investor. We separately analyze the performance of trades made in underaged accounts (defined as accountholders aged 0 to 10 years) versus trades by accountholders in older age categories. 3 We find that underaged accountholders exhibit superior stock-picking skills on both the buy side and the sell side over the days immediately following trades. They significantly outperform older investors by an average of 9 basis points (bp) based on all trades made 1 day earlier, by 7 bp based on trades 2 days earlier, by another 5 bp for trades 3 days earlier, and by an average of 2 bp per day for previous trades made over days 4 through 10. Since this outperformance is especially evident for short horizons, it likely stems from superior private information that is about to become public. We explore this possibility, and find that underaged accountholders perform extremely well when they trade just before major information events. For example, on the day before major earnings announcements, young accountholders trade in the correct direction 57% of the time and outperform older investors by an average cumulative abnormal return on days 0 and +1 (CAR (0,+1)) of 1.1%. Similarly, 1 day ahead of large price changes, young accounts trade in the correct direction 58% of the time and outperform older investors by an average CAR (0,+1) of 2.1%. Finally, on the day before takeover announcements, their proportion of correct trades is 72% and their mean outperformance is a CAR (0,+1) of more than 12%. We extend the analysis by examining the performance of two sets of guardians, who are matched to underaged accounts using either family 1 We confirm empirically that guardians have more wealth, and outperform other adults. Prior work shows that wealth and IQ are associated with superior performance (e.g., see Grinblatt, Keloharju, and Linnainma (2011, 2012)). Superior performance can also stem from many other factors, including a person s position or educational affiliation (e.g., see Cohen, Frazzini, and Malloy (2008, 2010), Cohen, Malloy, and Pomorski (2012)). 2 The Internet Appendix may be found in the online version of this article. 3 Expanding this classification to include an older group of children gives somewhat worse performance. We present evidence that a large proportion of trades by children and young adults between 10 and 20 years old is not initiated by informed guardians, but appears to be motivated by liquidity needs or financial education.

3 Trading through the Accounts of Children 365 surname or similar trades in corporate accounts. Consistent with our expectations, these guardians have significantly more wealth than other adults. In addition, when guardians trade through their own accounts or through corporate accounts, they outperform other adults on the buy side but not on the sell side. This asymmetry suggests that a relatively high proportion of guardian sales are liquidity motivated. In contrast, the fact that underaged accounts outperform on both the buy side and the sell side indicates that liquiditymotivated selling is less prevalent for underaged accounts. 4 Similarly, while guardians outperform when they trade in their own accounts just before major earnings announcements or large price changes, they do so by a wider margin when they trade through underaged accounts. Furthermore, guardians do not outperform when they trade in their own accounts before takeover announcements, but they do so when they trade through underaged accounts. Together, this evidence suggests that informed guardians conduct a higher proportion of informed trading through underaged accounts, relative to their own accounts. The emerging picture points to a broader group of informed guardians who tend to channel their best ideas through the accounts of children. As a result, the most valuable private information from guardians is filtered through underaged accounts to yield a greater degree of outperformance. Based on this evidence, we propose that BABYPIN offers a low-noise proxy for the probability of trading with a privately informed investor. Consistent with this view, we show that BABYPIN is related to future stock returns, confirming the prediction in Easley and O Hara (2004). We find that the tercile of stocks with the highest value of BABYPIN in the previous month significantly outperforms the tercile with the lowest BABYPIN by more than 1% per month, on average. These results are robust to the inclusion of standard risk factors, and they also hold in cross-sectional regression analysis, controlling for firm characteristics such as size, spread, and the traditional measure of the probability of information trading (PIN). Our study contributes to the literature in two areas. First, we build upon the growing number of studies that document that some groups of individual investors are more informed or skilled than others. For example, Grinblatt, Keloharju, and Linnainma (2011, 2012) find that investors with greater IQ have superior stock-picking skills. Seru, Shumway, and Stoffman (2010) show that the disposition effect subsides and performance improves as individual investors become more experienced. Cohen, Frazzini, and Malloy (2008, 2010) and Cohen, Malloy, and Pomorski (2012) show that corporate insiders and investors with access to social networks through educational affiliation enjoy an advantage in obtaining value-relevant information. Finally, Ivkovic and 4 Chan and Lakonishok (1993) and Saar (2001) argue that sales are unlikely to be informative if investors in need of liquidity are forced to sell one of the limited number of stocks they already own. Since there are many more securities to choose from when buying, the authors argue that purchases are more likely to convey firm-specific positive news. In support of this view, Kraus and Stoll (1972), Cohen, Frazzini, and Malloy (2008), and Grinblatt, Keloharju, and Linnainma (2012) find that the market acts as if purchases are more informative than sales.

4 366 The Journal of Finance R Weisbenner (2005) find that individuals who invest in local companies outperform other individual investors a finding confirmed in this study. 5 Our study also contributes to the recent literature that examines whether information asymmetry affects required returns. Unique among proxies for information asymmetry, we directly establish the empirical validity of BABYPIN, and show that a large proportion of trading through underaged accounts is motivated by superior private information. Our evidence that BABYPIN is strongly related to excess returns in the relatively illiquid Finnish market is consistent with the models in Easley and O Hara (2004) and Lambert, Leuz, and Verrecchia (2011). The remainder of this study is organized as follows. Section I briefly reviews the literature and Section II describes the data. Section III presents the analysis of all trades by different groups of investors. Section IV focuses on trades made just before major information events. Section V examines the relation between alternative measures of information asymmetry (BABYPIN and PIN) and stock returns. A final section concludes. I. Literature Review Easley and O Hara (2004) develop a model in which investors require a higher return to hold stocks with a greater probability of private information. This higher return is necessary to compensate uninformed investors for their tendency to overweight stocks with undisclosed bad news and underweight stocks with undisclosed good news, relative to informed investors. In subsequent theoretical work Hughes, Liu, and Liu (2007) and Lambert, Leuz, and Verrecchia (2007) show that, if there is perfect competition (i.e., any single investor s demand has a negligible impact on price), the effect of asymmetric information on expected returns is diversifiable and should not affect the cost of capital. In contrast, Lambert, Leuz, and Verrecchia (2011) show that imperfect competition leads to differential private information having an impact on prices, and that this impact is stronger for illiquid stocks. On the empirical front, Easley, Hvidkjaer, and O Hara (2002) develop their well-known PIN measure and find that stocks with a higher PIN tend to have a higher return. Easley, Hvidkjaer, and O Hara (2010) provide additional evidence of the robustness of PIN in a factor pricing structure, and Li et al. (2009) find a strong positive relation between expected Treasury bond returns and information risk measured by PIN. Armstrong et al. (2011) empirically confirm the main prediction in Lambert, Leuz, and Verrecchia (2011) that expected returns are increasing in the degree of adverse selection when there is a relatively low degree of market competition. On the other hand, Mohanram and Rajgopal (2009) find that the relation between PIN and stock returns is not robust to alternative specifications and time 5 In contrast to Ivkovic and Weisbenner (2005), Seasholes and Zhu (2010) find no evidence of outperformance by locals when they account for contemporaneous correlation of returns across household portfolios. Our finding for Finnish investors thus contributes to the debate on whether local investors outperform.

5 Trading through the Accounts of Children 367 periods, casting doubt on whether PIN is truly a priced risk factor. Similarly, Duarte and Young (2009) show that liquidity effects unrelated to information asymmetry can explain the relation between PIN and expected returns. Finally, Aktas et al. (2007) find that the behavior of PIN around a sample of merger and acquisition announcements in Euronext Paris is inconsistent with other evidence of information leakages during the preevent period. 6 II. Institutional Background, Data, and Sample Characteristics A. Institutional Background Similar to U.S. law, in Finland a person must be 18 years old to set up a brokerage account. For anyone under the age of 18, a guardian must open the account and make any transfer of shares into or out of the account. Banks and brokerage firms have their own rules for opening and trading in underaged accounts, and they typically require a guardian s signature. The Finnish tax code provides some room for wealthy individuals to benefit from opening or trading through a child s account, similar to the U.S. code. Details of the Finnish tax code are discussed in Internet Appendix Section II. Finally, insider trading laws in Finland were passed in 1989 and first enforced in 1993 (see Bhattacharya and Daouk (2002)). B. Data Sources This study focuses on share price performance following trades by investors in different age groups. We obtain the information necessary for this analysis from Euroclear. There are a total of 183 Finnish stocks listed on the Nasdaq OMX Helsinki exchange during our sample period. To trade on this exchange, investors must register with Euroclear. Each investor is given a unique Euroclear account, even if he or she trades through multiple brokers. The Euroclear database records the shareholdings of all registered accounts in Finland and documents daily changes in shareholdings for each registered investor. 7 Our sample includes the transactions of more than a half million individuals over the period January 1, 1995, through May 31, All transactions placed each day are recorded as either purchases or sales. Trades are aggregated for every investor each day, and we use the daily net change in an investor s position of a given stock as our unit of observation. We split all individual investors into six age groups. The youngest group consists of accountholders aged 0 to 10 years. The second group includes older children, aged 11 to 20 years. The next three groups include individuals aged 6 This literature review focuses on PIN as a measure of information asymmetry. Van Ness, Van Ness, and Warr (2001) and Neal and Wheatley (1998) study adverse selection estimates based on several other microstructure models, and conclude that these estimates provide weak measures of information asymmetry. 7 Grinblatt and Keloharju (2000) provide a detailed description of the Euroclear database.

6 368 The Journal of Finance R 21 to 40 years, 41 to 60 years, and 61 to 80 years, respectively, while the final group is comprised of people 81 years and older. 8 We also identify two subsets of guardians for the youngest two age groups, matched either by family name or by trades through corporate accounts. First, we obtain an identification number for each individual Finnish account that identifies the surname of the accountholder, without revealing the surname. Given this information, we link underaged accounts (aged 0 to 10 and 11 to 20) to their guardians accounts using four screens, where we require the guardian accountholder to: (i) have the same surname, (ii) have the same post code, (iii) be older than 25 years, and (iv) have at least two trades in the same security on the same day with the same sign (buy/sell). Our second set of guardian accounts consists of corporate trading accounts that are linked to underaged accounts by the following three screens: (i) any Finnish-owned domestic private corporation, (ii) with the same post code, and (iii) with at least five identical trades matched to an underaged account. This more stringent requirement for a match of five trades is likely to miss some corporate guardian accounts in exchange for ensuring that the accounts we identify are accurate matches. We obtain earnings announcement dates from Bloomberg. 9 Merger and acquisition announcement dates are from SDC Platinum. Daily share prices and the number of shares outstanding are taken from Compustat Global. The market-to-book ratios for all Finnish firms are obtained from Worldscope. Finally, we estimate the average intraday bid ask spread and the traditional PIN measure using intraday quotes from Thomson Reuters, in combination with Euroclear data and intraday trading data from Nasdaq OMX Helsinki. C. Trading Activity and Sample Characteristics for Age Groups and Guardians C.1. Distribution of Trading Activity across Age Groups and Guardians Panel A of Table I gives descriptive statistics for the distribution of trading across age groups and guardians. The first column reports the total number of trading days across all stocks and accountholders in every age group, while the second column gives the percentage of total trading days attributable to every group. Accountholders in the youngest group make up less than 1% of the total number of stock trading days across all individual accounts. The second age group comprises slightly less than 2%. Not surprisingly, the bulk of trading through individual accounts is done by investors between the ages of 21 and Internet Appendix Section V shows that restricting our youngest age group to range from 0 to 5 years yields similar results, while expanding this classification to range from 0 to 17 years gives worse performance. In Internet Appendix Section IV we analyze the relative performance of older children, aged 11 to 20 years. 9 We checked the earnings announcement dates from Bloomberg against the official source at and found no discrepancies.

7 Trading through the Accounts of Children 369 Table I Sample Characteristics and Trading Activity by Age Groups and Guardians This table summarizes the characteristics of the sample of all trades, and the attributes of the firms traded, by individual accountholders in every age group along with the two sets of guardian accounts. The first column of Panel A reports the total number of trading days across all stocks and individual accountholders in every group. The second column gives the percent of total trading days attributable to every group. The third column shows the percent of trading days in which the accountholder was a net buyer. Columns 4 and 5 report the average value of shares bought and sold per transaction, for every group. Column (6) provides the percent of all trades made by female accountholders. For every group, the last two columns give the number of accounts and the median wealth (i.e., account size in euros) on January 5, The total number of trades (N) incolumn(1)is obtained by summing over all individuals in the six age groups. Panel B reports the mean adjusted ranks of the control variables for every group. Control variables include the firm s β, market-to-book ratio, size, and previous stock returns taken over four time frames. These variables are defined in the text. We transform each control variable into decile ranks by first sorting the cross-section of stocks each day into 10 groups. Next we assign a value to the stocks in each decile, where the values are adjusted to range from 0.5 (for the lowest decile) to +0.5 (for the highest decile). The mean adjusted rank values in Panel B of Table I are then obtained by averaging these adjusted ranks across all stock trading days within every age group. Panel A. Frequency and Volume of Trading by Accountholders in Different Age Groups and Guardians Age Group (i) # of Trades (ni) (1) (2) (3) (4) (5) (6) (7) (8) %oftrades (ni/n) 100 %ofbuysi (#Buysi/ni) 100 Avg Value of Buys ( Value of Buys)i/ni Avg Value of Sales ( Value of Sales)i/ni %oftradesby Females # of Different Accounts Median Wealth 1: 0 to 10 95, % 66.3% 2,533 3, % 21,508 2,844 2: 11 to , % 55.5% 3,558 5, % 42,804 3,387 3: 21 to 40 3,360, % 55.0% 3,912 4, % 216,360 4,840 4: 41 to 60 5,681, % 55.4% 6,121 7, % 298,659 8,060 5: 61 to 80 2,362, % 53.5% 6,540 8, % 177,689 7,413 6: , % 37.5% 7,113 15, % 35,841 5,200 Guardians-Family 333, % 53.5% 10,788 11, % 2,708 25,742 Guardians-Corp 305, % 49.9% 53,622 61, ,836 Total (N): 11,872, % Panel B. Attributes of Firms Traded by Accountholders in Different Age Groups and Guardians (1) (2) (3) (4) (5) (6) (7) Age Group (i) Rank (β) Rank (M/B) Rank (Size) Rank (Ryear) Rank (Rmonth) Rank (Rweek) Rank (Rday) 1: 0 to : 11 to : 21 to : 41 to : 61 to : Guardians-Family Guardians-Corp

8 370 The Journal of Finance R The third column in Panel A shows the percent of trading days in which the accountholder was a net buyer, for every age group. In line with expectations, young investors are net buyers on two-thirds of the days they trade. This proportion drops to roughly 55% for traders between the ages of 11 and 80. Investors older than 80 years tend to be sellers on more than 60% of the days they trade. Columns (4) and (5) in Panel A show that the value of the average purchase and sale steadily increase as investors get older. Column (6) in Panel A provides the percentage of all trades made by female accountholders within every age group. Females account for almost 50% of all trading days for the youngest age group. This outcome is consistent with both of our explanations for abnormal performance by underaged accounts, namely, parental sharing of profits from informed trading and camouflage trading by informed parents. There is no reason to expect the gender of the underaged accountholder to play a role in either explanation. The proportion of female accountholders is lowest for investors between ages 21 and 40 and then increases for older investors. The relatively high percentage of trades among females in the oldest group may reflect their longer life expectancy. Column (7) of Panel A shows that the youngest age group (0 to 10 years) comprises a total of 21,508 accounts, or roughly 3% of all 671,438 individual accounts in the sample. There are approximately twice as many accounts in the second age group (11 to 20 years). Once again, the middle three age groups (21 to 80 years) contain the bulk of all trading accounts. The last column, (8), presents the median value of the stock portfolio for accountholders in every age group as of January 5, The median wealth (i.e., portfolio value in euros) for each group of accountholders increases steadily with age, before declining for investors over 60 years old. The last two rows of Panel A summarize the trading behavior of the two sets of guardian accounts. We are able to match 2,708 adult accounts to underaged accounts in the youngest age group (0 to 10 years) based on family surname, and 330 corporate accounts based on similar trades. These two sets of guardians are relatively active, trading as much as 2.8% and 2.6%, respectively, of the roughly 12 million total stock trading days by all individual investors. Both sets of guardians buy around 50% of the time, and their average purchase has roughly the same value as their average sale. It is noteworthy that the value of the average trade by guardians matched to corporate accounts is much larger than the average value for individual accounts. For guardians matched by family name, the percentage of trades made by females is only 14.5%. This low percentage is likely due to our first matching scheme for guardians, which relies on family surname and thus favors the father s side. Finally, the median portfolio value of name-matched guardians is 25,742 euros, while the median value of corporate guardian accounts is more than 200,000 euros. Both of these values are significantly larger than the median portfolio value of 6,619 euros for all other nonguardian adults at least 25 years old (p-values are less than for both Wilcoxon tests).

9 Trading through the Accounts of Children 371 C.2. Characteristics of Stocks Purchased by Different Age Groups and Guardians Panel B of Table I reports descriptive statistics for our control variables, which include characteristics of the firms held by investors in the different groups. This information is important to investigate whether different age groups tend to focus on different investment styles or firms with certain characteristics, which could lead to superior performance. Our choice of control variables is based on Grinblatt, Keloharju, and Linnainma (2012), who analyze the performance of Finnish individuals based on IQ scores. Similar to these authors, we report the mean adjusted ranks of the firm s β (BETA), market-tobook ratio (M/B), and firm size (Size), as well as four variables that summarize the stock s returns over the past year excluding the last month (RYear), the past month excluding the last week (RMonth), the past week excluding the last day (RWeek), and the past day (RDay). These ranks are adjusted to range from 0.5 to 0.5, and serve to attenuate the influence of outliers. Internet Appendix Section III further describes the construction of these control variables. Results in Panel B of Table I reveal that individual investors tend to be more active in large stocks with relatively high β s and market-to-book ratios. The relation to past returns is less uniform across age groups. For all control variables, the differences in the mean scaled ranks across age groups are small in magnitude. For example, the maximum difference in the average ranks between age groups is only 0.08 (i.e., the difference in the mean rank for firm size between age groups 6 and 3). This difference is less than 0.1, which is the change in scaled ranks between any pair of adjacent decile groups. These results imply that investors from different age groups tend to trade firms with similar characteristics. There is also no evidence that the average stock traded by guardians is substantially different. III. Analysis of All Trades We analyze the investment skills of accountholders in the different age groups, using a Fama MacBeth (1973) regression approach similar to the analysis of Grinblatt, Keloharju, and Linnainma (2012). 10 First, for each trading day (t) in the sample period, we identify all accounts that trade in any given stock (i). Second, we separate accounts that are net buyers from accounts that are net sellers of stock i on day t. This procedure results in two cross-sections for every day that contain the trading activity across all stocks by all buyers and sellers, respectively. We then analyze the return performance on day t for the stocks bought or sold x days earlier, on day t x, by investors in different age groups (where x = 1, 2, 3, and 4 to 10 days earlier). Specifically, we separately estimate 10 This approach is attractive because it documents the marginal effect of investor age on performance, while controlling for other relevant attributes such as the characteristics of investors and firms. Our robustness tests in Internet Appendix Section V show that a calendar-time portfolio approach produces similar results.

10 372 The Journal of Finance R cross-sectional regressions on each day (t) for the samples of purchases and sales. The dependent variable in the regressions is Return i,t, the geometric close-to-close return of stock i on day t. A. Performance of Young versus Older Investors for Trades Made 1 Day Earlier We illustrate this analysis with the following cross-sectional simple regression model that omits the control variables and includes only the dummy variable for the youngest age group: Return i,t = a 0 + a 1 BABY i,t x + ε i,t. (1) The dummy variable, BABY i,t-x, is assigned a value of one for all purchases (or sales) in stock i made on day t x by accountholders aged 0 to 10 years, and zero otherwise. Thus, the intercept (a 0 ) represents the mean return on day t averaged across all stock purchases (or sales) made on day t x by investors in the omitted group, which includes all accountholders older than 10 years. The coefficient of the dummy variable (a 1 ) then reflects the difference between this intercept and the mean return on day t across all purchases (or sales) made on day t x by the group of youngest accountholders. For each day in the sample period, we estimate the cross-sectional regression in equation (1). We then compute the Fama MacBeth (1973) mean coefficients across all daily regressions for purchases and for sales. We use the standard deviation of each time-series mean coefficient to construct the t-ratio. 11 We also provide the mean of the daily differences for each coefficient (a 0 or a 1 ) for purchases versus sales. That is, we calculate the mean and standard deviation of the time series of daily differences: (a 0 for purchases a 0 for sales) and (a 1 for purchases a 1 for sales). We return to the interpretation of these hedge portfolio returns in our discussion later. The results appear in Panel A of Table II for trades made 1 day earlier (i.e., x = 1). These statistics are obtained from 2,904 daily regressions. 12 The first column of Panel A presents the mean coefficients, along with their t-ratios, for all purchases made on day t 1, omitting the control variables. The second column provides the analogous results when the control variables are included. Columns (3) and (4) present the same information for all sales made on day t 1, while columns (5) and (6) show the hedge portfolio results. Consider first the evidence for purchases on day t 1, in the first column of Panel A. The mean intercept (a 0 ) for purchases is small in magnitude (+0.011) and is not significantly different from zero (t-ratio = 0.3). This result indicates that, on the day after purchases by older investors, the average return is 1 bp and insignificant. In contrast, the mean coefficient on the dummy variable (a 1 ) 11 We have also constructed t-statistics based on the time-series standard deviation of coefficient estimates, adjusted for autocorrelation using the Newey West method. The results are nearly identical. 12 We exclude days with fewer than 500 different individual accounts trading at least 25 different stocks.

11 Trading through the Accounts of Children 373 Table II Performance of Young versus Older Accountholders across All Trades Made over the Previous 10 Days This table presents Fama MacBeth mean daily coefficients from estimating models (1) and (2) to analyze the average return realized by young versus older accountholders on day t, based on earlier trades made on day t x, wherethelag(x) is varied from 1 to 10 days: Returni,t = a0 + a1 BABYi,t x + [Control Variables]i,t + εi,t, (1) Returni,t = b0 + b1 BABYi,t x + b2 GUARDIAN-Family i,t x + b3 GUARDIAN-Corp i,t x + b4 MATCHi,t x BABYi,t x + [Control Variable]i,t + εi,t (2) where BABYi,t-x is one for purchases (sales) in stock i made on day t x by accountholders aged 0 to 10 years, or zero otherwise; GUARDIAN-Familyi, t-x is one for all trades made by guardians matched to family name, or zero otherwise; GUARDIAN-Corpi, t-x is one for all trades made by guardians matched to corporate trades, or zero otherwise; and MATCHi,t-x is one for trades by underaged investors matched to either set of guardians, or zero otherwise. Panels A and B provide the results for (1) and (2), respectively, based on trades made 1 day earlier. Panels C and D present analogous results for model (2), omitting the interaction term, for trades made 2 and 3 days earlier. Panel E provides the average results based on earlier trades made during the previous 2 weeks (i.e., days 4 through 10). In each case we provide regression results both including and excluding the adjusted ranks for the following firm-specific control variables: Beta, M/B, Size, RYear, Rmonth, Rweek, and Rday. Control variables are defined in the text. All returns are in percentage points, with the associated t-statistics to the right of each coefficient estimate. * indicates statistical significance at the 0.10 level, ** at the 0.05 level, and *** at the 0.01 level. Panel A. Young versus Older Accountholders, for Trades on Day 1 Purchases Sales Hedge Portfolio (1) (2) (3) (4) (5) (6) Intercept a BABY a *** *** * ** *** *** Control variables No Yes No Yes No Yes Panel B. Young versus Older Accountholders and Guardians, for Trades on Day 1 Intercept b * BABY b *** ** * ** *** *** GUARDIAN-Family b *** *** *** *** GUARDIAN-Corp b *** *** *** *** MATCH BABY b Control variables No Yes No Yes No Yes (Continued)

12 374 The Journal of Finance R Table II Continued Panel C. Young versus Older Accountholders and Guardians, for Trades on Day 2 Purchases Sales Hedge Portfolio (1) (2) (3) (4) (5) (6) Intercept b *** BABY b ** *** ** *** ** GUARDIAN-Family b *** *** ** *** *** GUARDIAN-Corp b *** *** * *** Control variables No Yes No Yes No Yes Panel D. Young versus Older Accountholders and Guardians, for Trades on Day 3 Intercept b *** BABY b ** * ** * GUARDIAN-Family b *** *** ** *** GUARDIAN-Corp b *** *** *** *** Control variables No Yes No Yes No Yes Panel E. Young versus Older Accountholders and Guardians, for Days 4 through 10 Intercept b BABY b * GUARDIAN-Family b ** ** * ** *** *** GUARDIAN-Corp b ** *** ** ** Control Variables No Yes No Yes No Yes

13 Trading through the Accounts of Children 375 is positive (+0.057) and highly significant (t-ratio = 3.2). Thus, stocks bought by young investors (in age group 1) significantly outperform stocks bought by older investors by approximately 6 bp, on average, 1 day later. Column (2) in Panel A of Table II presents the analogous results when we include the control variables. 13 Including the control variables results in a slightly smaller mean difference between the next day s performance for young versus older investors (a 1 )of+4 bp, which remains significant at the 0.01 level (t-ratio = 2.9). Together, the results in columns (1) and (2) are consistent with our hypothesis, indicating that young accountholders are good stock pickers. The stocks they buy significantly outperform those bought by older investors, on average, on the following day. Columns (3) and (4) in Panel A of Table II present the same analysis for sales. Now the stocks sold by underaged investors underperform the stocks sold by older investors by an average (a 1 )of 5.7 bp on the following day when the control variables are excluded (t-ratio= 1.9), or 4.8 bp(t ratio = 2.1) when the controls are included. It is noteworthy that the exceptional stock-picking skills of young investors also manifest themselves on the sell side. This result contrasts with prior evidence in several studies that find that purchases are more informative than sales. 14 A possible explanation for our result is that sales are uncommon for very young accountholders (see Table I), and are less likely to be triggered by liquidity shocks. As a result, a relatively high proportion of underage sales is likely to be informed. Columns (5) and (6) of Panel A in Table II present the mean differences in each coefficient (a 0 or a 1 ) across the subsamples of purchases versus sales. Each mean difference represents the average return from a hypothetical zero-cost hedge portfolio strategy. According to this strategy for older investors (a 0 for purchases a 0 for sales), a hypothetical investor goes long $1/n in all n stocks that are bought on day t x by older investors, and short $1/m in all m stocks that are sold on day t x by older investors. Columns (5) and (6) show that this hedge portfolio yields an insignificant mean return on the following day of +1.8 bp or 0.4 bp, depending on whether the control variables are included (t-ratio = 1.1 or 0.5, respectively). In contrast, the analogous hedge portfolio based on the purchases and sales of young accountholders (a 1 for purchases a 1 for sales) is more impressive, outperforming the hedge portfolio above by an average of +11 bp per day without control variables, or +9 bpperdaywith control variables (t-ratio = 3.2 for both cases). This result is also economically significant. With daily rebalancing, an average daily excess return of 9 bp per day accumulates to a hypothetical excess return of more than 20% on an annualized basis. 13 The coefficients on the control variables for this base case are provided in Internet Appendix Section V. 14 Kraus and Stoll (1972), Cohen, Frazzini, and Malloy (2008), and Grinblatt, Keloharju, and Linnainma (2012) find that buys are more informative than sales. In contrast, Cohen, Malloy, and Pomorski (2012) find that both purchases and sales by insiders are informative.

14 376 The Journal of Finance R B. The Performance of Guardians and Earlier Trades Panel B of Table II provides results for the following expanded version of equation (1): Return i,t = b 0 + b 1 BABY i,t x + b 2 GUARDIAN-Family i,t x + b 3 GUARDIAN-Corp i,t x + b 4 MATCH i,t x BABY i,t x + ε i,t, (2) where GUARDIAN-Family i,t-x equals one for trades by guardians matched to family name, or zero otherwise; GUARDIAN-Corp i,t-x equals one for trades by corporate guardian accounts, or zero otherwise; and MATCH i,t-x equals one for trades by all underaged investors matched to guardians, or zero otherwise. The coefficients, b 2 and b 3, show the differential performance of our two groups of guardian accounts over all other nonguardian adults (b 0 ). The coefficient on the interaction term (b 4 ) distinguishes the performance of one subset of underaged accounts that we can match to guardians (b 1 + b 4 ) versus the remaining subset that we are unable to match to guardians (b 1 ). Panel B of Table II reveals that, like underaged accounts, both sets of guardians significantly outperform older investors on the buy side. However, unlike underaged accounts, guardians do not outperform on the sell side. As a result, when we consider the mean hedge portfolio returns, underaged accounts outperform other nonguardian adults by a greater margin than either set of guardian accounts. It is noteworthy that the results for guardians are similar to the performance of the highest IQ stanine in Finland based on trades made 1 day earlier. For that group of investors, Grinblatt, Keloharju, and Linnainma (2012) report a mean outperformance of 7.5 bp for buys, and no significant outperformance for sells. This evidence suggests that guardians are more likely to make uninformed liquidity trades on the sell side through their own accounts than they are through the accounts of children. The coefficient on the interaction term (b 4 ) is insignificant for purchases, sales, and the hedge portfolio. This result is important because it indicates that guardians operating behind the subset of matched children demonstrate a similar information advantage as guardians trading for the remaining subset of unmatched children. We conjecture that the guardians behind the group of unmatched children include relatives with different surnames or post codes (e.g., mothers or grandparents), family friends, or foreign accountholders. 15 Panels C and D extend the analysis to trades made 2 and 3 days earlier. The results are similar to those in Panel B, although somewhat smaller in magnitude. They show that guardians also outperform with these earlier trades, through both underaged accounts and their own accounts. Once again, this outperformance is concentrated on the buy side when guardians trade in their 15 Since the coefficient, b 4, is always insignificant, hereafter we omit the interaction term for brevity.

15 Trading through the Accounts of Children 377 own accounts. On the other hand, this outperformance appears more on the sell side when guardians make earlier trades through underaged accounts. 16 Panel E of Table II presents the average daily results for the performance of earlier trades made during the previous 2 weeks (i.e., on days 4 through 10). The mean daily differential performance of guardians is now smaller in magnitude, at approximately 1 to 2 bp on both the buy side and the sell side, regardless of whether guardians make earlier trades through the accounts of children or through their own accounts. For each group of accounts, this performance combines to form a mean daily hedge portfolio return that is small in magnitude, at 2 to 3 bp per day, although it is significantly different from zero. Using the information in Table II, we can calculate the hypothetical annualized return of a hedge portfolio that is rebalanced every 2 weeks for each group of investors. For example, the excess return of a fortnightly rebalanced portfolio replicating trades by underaged investors would accumulate to 9.2% per annum (p.a.). Similarly, a replicating portfolio based on the trades of guardians with the same family name would yield 7.0% p.a., and a portfolio based on trades by corporate guardians would yield an excess return of 6.9% p.a. 17 C. Extensions and Robustness Tests for Analysis of All Trades C.1. (Internet Appendix Section IV) Young Children versus Older Children and Other Older Age Groups In Internet Appendix Section IV we investigate the performance of young children relative to different sets of older investors. We find that the outperformance of young investors (aged 0 to 10) extends to each older age group separately, including the second group of older children (aged 11 to 20). At first glance, this latter result is surprising since guardians likely control most of the second age group as well. However, Table I shows that there are roughly twice as many accounts for older children (aged 11 to 20) as there are for younger children (aged 0 to 10). We conjecture that guardians who wait to open accounts until the child is older are likely to have less wealth and be less informed than those who open accounts for very young children. 16 Combining days 1, 2, and 3, we find that underaged investors are significantly better than both sets of guardians on the sell side (p = 0.01), and are not significantly worse on the buy side. The hedge portfolios for underaged investors significantly outperform both family guardians and corporate guardians when control variables are not included (p = 0.05 and p = 0.06, respectively). When control variables are included, the significance of this differential performance declines somewhat (to p = 0.15 and p = 0.21, respectively). 17 These figures are obtained by summing the hedge portfolio coefficients for trades made by every age group during the previous 2 weeks. For example, a hedge portfolio constructed at the end of trading day t based on the trades by young investors would earn 10.2 bp on day t + 1, 6.5 bp on day t + 2, 4.7 bp on day t + 3, and 2.0 bp on each day during the 7-day period from day t + 4tot If this portfolio were rebalanced every 2 weeks, the annualized excess return would be 26 ( ), which equals 9.2% per annum.

16 378 The Journal of Finance R We examine this conjecture by splitting all accountholders aged 11 to 20 into two categories: (i) the subset of old accounts that were opened earlier when the child was 0 to 10 years old, and (ii) the remaining subset of new accounts that were not opened until the child reached at least 11 years of age. We show that the median account size for guardians of new accounts is significantly smaller than that for guardians of old accounts. Furthermore, the subset of new accounts (aged 11 to 20) significantly underperforms young children (aged 0 to 10), but the subset of old accounts (aged 11 to 20) does not. This evidence supports our conjecture that guardians behind the subset of recently opened new accounts for older children have less wealth, and are less likely to be motivated by superior private information and more by other motives such as financial education or liquidity needs. We confirm this result by further analyzing the performance of earlier trades in old accounts when the child was young, versus later trades in the same account when the child was older, adjusting for investor-level fixed effects. We find that earlier trading activity when the child was young (aged 6 to 10) does not significantly outperform later activity in the same account when the child was slightly older (aged 11 to 15 or 16 to 20), when the informed guardian was likely still in charge of the account. However, earlier trading when the child was young does significantly outperform trading in the same account much later, when the child was 21 to 25 years old and more likely to make his or her own trades. C.2. (Internet Appendix Section V) Additional Robustness Tests for the Analysis of All Trades Internet Appendix Section V provides a series of additional extensions to the analysis of all trades. We first demonstrate that the results in this section are generally robust: (i) before and after the Finnish wealth tax was removed in 2005; (ii) for the subset of all trades that excludes days with major earnings announcements, takeover announcements, and large price changes; (iii) for the subset excluding all trades in Nokia, Finland s largest firm; (iv) using alternative age ranges to define young investors; (v) for arithmetic returns; and (vi) for rank regressions. We also document similar results if we use a calendar-time portfolio approach. This battery of tests establishes that the mean outperformance of young versus older investors is robust. Internet Appendix Section V also extends the analysis of all trades by investigating whether there is a greater degree of outperformance for situations in which there is likely to be greater information asymmetry (i.e., for investments in small stocks, growth firms, or local stocks). Our findings can be summarized as follows:

17 Trading through the Accounts of Children 379 (i) young investors outperform older investors by a significantly greater amount for trades made in small stocks relative to large stocks; (ii) young investors outperform older investors by a larger amount for trades made in growth firms relative to value firms, but this differential is not significant; (iii) young investors outperform older investors by a larger amount when they invest in local versus nonlocal stocks, but this differential outperformance is not significant; and (iv) the extent of outperformance by young versus older investors is not significantly different across females versus males. Overall, the results in this section strongly indicate that the class of underaged accountholders possesses significant short-term informational advantages that result in superior stock returns on the days immediately following their trades. Given the short-term nature of this apparent information advantage, we expect the superior performance of underaged investors to manifest itself to a greater extent around large price changes or major corporate events that are commonly associated with increased information asymmetry, such as takeover and earnings announcements. This conjecture is the subject of the next section. IV. Trades before Major Announcements and Large Price Changes This section uses an event study approach to focus on trades made in the days prior to takeover and earnings announcements. In addition, we analyze trades in the days before large price changes, which presumably reflect the arrival of substantive value-relevant information. Our tests compare two aspects of investors ability based on the trades by each age group in the days and weeks before an event. First, we examine the mean market-adjusted cumulative abnormal return on the day of and the day after each type of event (CAR(0,+1)). 18 Second, we investigate the mean frequency of trades made in the correct direction. In our event studies of earnings announcements and large price changes we select the sample of events in such a way that, under the null hypothesis that all traders are uninformed, the expected CAR(0,+1) equals zero and the expected proportion of trades in the correct direction is 50%. We accomplish this goal by creating a matched sample with an equal number of good news and bad news events that are similar in terms of the distribution of the absolute CAR(0,+1). By designing such a fair experiment, we attempt to neutralize any potential biases that may arise from the tendency for underaged accountholders to buy more frequently than older accountholders (see Table I). In contrast to earnings announcements and large price changes, takeovers are typically good news events for the target firm. Because we are unable to 18 The abnormal return of a stock is defined as the actual return minus the return on the valueweighted All-Share Index (where the maximum weight of one stock is limited to 10% of the total market value of the index).

18 380 The Journal of Finance R construct a similar fair experiment for takeovers, we include an alternative test for these events that compares the realized mean CAR(0,+1) and frequency of correct trades with the expected value given the overall tendencies for each age group to buy or sell. A. Event Studies for Earnings Announcements and Large Price Changes A.1. Sample Selection Criteria for Earnings Announcements and Large Price Changes We generate our sample of earnings announcements by first obtaining from Bloomberg a total of 4,136 quarterly announcements made by all Finnish firms over the period 1999 to We then focus on major events that are likely to be characterized by substantial information asymmetry by selecting only those earnings announcements with a CAR(0,+1) of at least 4% in absolute value. This screen results in a sample of 964 major negative earnings announcements and 760 major positive earnings announcements. 20 We next further restrict this sample to have an equal number of good news and bad news events that have similar distributions in terms of the absolute CAR(0,+1). To do so, we match (without replacement) the 760 good news events with their nearest neighbor for the absolute CAR(0,+1) in the larger sample of 964 bad news events, where we also require the difference in the absolute CAR(0,+1) of the matched pair to be smaller than 1%. This exercise results in a final sample of 1,492 events, comprised of 746 good news and 746 bad news earnings announcements with similar distributional properties. Descriptive statistics for these good news and bad news earnings announcements are provided in Panel A of Table III. We follow a similar procedure to create a matched sample of good news and bad news large price changes. First, for each stock-year, we select the 2 days with the largest and smallest market-adjusted abnormal returns. This selection yields an initial sample of 2,347 large price increases and 2,347 large price decreases for all Finnish firms across all 16 years in the sample period. Second, we retain a major price change event if it is not within 5 days of an earnings announcement or acquisition announcement, and if it is not within 1 month of another large price change event for the same stock. Third, as before, we require the large price change event to have a CAR(0,+1) of at least 4% in absolute value. These screens result in 993 positive price change events and 870 negative price change events. Finally, we match (without replacement) the 870 negative price change events with the nearest neighbor for CAR(0,+1) from the larger sample of 993 positive price change events, where we also require the difference in absolute CAR(0,+1) to be smaller than 1%. This screen results in our final 19 Finnish earnings announcement dates are not available from Bloomberg before When we limit the analysis to earnings announcements or large price changes with an absolute CAR(0,+1) larger than 3% or 5%, we obtain similar results.

19 Trading through the Accounts of Children 381 Table III Descriptive Statistics for Samples of Events This table presents descriptive statistics for CAR(0,+1) for the different samples of events analyzed in this study. Panels A and B provide results for the samples of earnings announcements and large absolute price changes, respectively. In each row of these first two panels, we separately present the summary statistics for the matched subsamples of good news events (where CAR(0,+1) > 0) and bad news events (where CAR(0,+1) < 0). Panel C provides the results for the entire sample of takeover announcements. All returns are in percentage points. Panel A: Descriptive Statistics for CAR(0,+1) for Sample of Earnings Announcements Mean St dev Min Max q3 Median q1 # > 0 # < 0 Good news Bad news Panel B: Descriptive Statistics for CAR(0,+1) for Sample of Large Price Changes Good news Bad news Panel C: Descriptive Statistics for CAR(0,+1) for Sample of Takeover Announcements All events sample of 771 large price increases and 771 large price decreases. Descriptive statistics are provided in Panel B of Table III. A.2. Event Study Design for Earnings Announcements and Large Price Changes We first compute the stock s market-adjusted cumulative abnormal return on the event day and the next day, CAR(0,+1). We then sign this CAR for each stock trading day by every accountholder, depending on whether that account was a net buyer or seller x days before the event. If an account was a net purchaser (i.e., shares bought exceed shares sold on day t x), then the event period return for that account equals the stock s CAR(0,+1). Alternatively, if an account was a net seller (i.e., shares sold exceed shares bought), then the event period return for that account equals the stock s CAR(0,+1) multiplied by 1. For every event, we calculate the mean signed CAR(0,+1) across all accounts in each age group that bought or sold the stock on day t x. We then average these mean signed CARs across all events for trades made by young and older accountholders. The standard error of this mean signed CAR across all events is used to construct a t-test of the null hypothesis that the signed CAR(0,+1) is zero for trades by each age group. We also use a difference-in-means t-test to examine the null hypothesis that the mean signed CAR of young accountholders equals that of older accountholders. Our analysis of performance based on the mean frequency of trades in the correct direction follows a similar procedure. For every event, we begin by

20 382 The Journal of Finance R computing the proportion of investors in each age group who trade in the direction of the subsequent price move reflected in CAR(0,+1). That is, we compute p n, the proportion of total trades on day t x for which young or older investors correctly buy (or sell) a stock before the nth good (or bad) news earnings announcement or large price change. We then calculate the mean of the proportions of correct trades across all N events (Frequency). The standard error of the mean frequency is used to test the null hypothesis that the proportion of correct trades is 50%, and to investigate the difference in mean frequencies across young versus older investors. In addition to the mean frequency of correct trades (Frequency), we also report the mean proportions of correct buys and sells (P b and P s ), and the numbers of good news and bad news events (N 1 and N 2 ) with at least one trade on day t x by accountholders in each age group. A.3. Event Study Results for Major Earnings Announcements Panel A of Table IV provides results for trades made in each of the 3 days before major earnings announcements. Each set of results includes three columns. The first column summarizes the average performance of young investors across the subset of all announcements in our sample where at least one young accountholder trades on day t x. The second column documents the mean performance of older investors across the larger sample of events where at least one older accountholder trades on day t x. The third column reports the difference in this mean performance across the two age groups. The first column in Panel A of Table IV shows that at least one young investor trades on day t 1 for 252 of the 1,492 major earnings announcements in the sample. For these events, young traders have a significant mean signed CAR(0,+1) of 1.36% (p-value = 0.04). Also, 57% of the trades by young investors are in the correct direction, on average. This mean relative frequency is significantly greater than 50% (p-value = 0.03). Note that the mean frequency of correct trades is the weighted average of the mean proportions of correct buys and sells (P b and P s ), respectively, where the weights reflect the relative proportions of positive and negative information events with at least one trade on day t x by accountholders in each age group (N 1 /(N 1 + N 2 )andn 2 /(N 1 + N 2 )). We present these numbers in the bottom two rows of Panel A in Table IV. The first column shows that young accountholders correctly buy 1 day before good news events 72% of the time. Similarly, underaged accounts correctly sell 1 day before bad news events 39% of the time. The second column in Panel A of Table IV presents the analogous results for older accountholders trading on the day before major earnings announcements. For 1,474 of the 1,492 events in this sample, at least one older accountholder trades on the day before the announcement. Based on these trades, older investors also significantly outperform the market, with a mean signed CAR(0,+1) of 0.25% (p-value = 0.04) and an average of 52% of their trades in the correct direction (p-value = 0.00). This latter result is due to older

21 Trading through the Accounts of Children 383 Table IV Performance of Young versus Older Accountholders before Major Earnings Announcements and Large Price Changes This table presents the mean CAR on the day of, and the day after, major earnings announcements or large price changes (CAR(0,+1)) that pertains to trades made by young versus older accountholders in the 3 days ahead of the event. For each stock-trading day by every accountholder, the CAR is signed depending on whether that account was a net buyer or seller x days before the event. In addition, we provide the mean frequency of trades in the correct direction for each age group. This frequency is equivalent to a weighted average of the mean proportions of correct buys and sells, where the weights reflect the relative proportions of positive and negative information events with at least one trade on day t x by accountholders in the relevant age group: Frequency = (1/N) N n=1 pn = (N1/N)(Pb) + (N2/N)(Ps), where pn is the proportion of correct trades (buys or sells) before the nth event; P b is the mean of the proportions of correct buys across all N1 positive information events; N1 is total number of positive information events (where CAR(0,+1) > 0) with at least one trade on day t x by accountholders in the relevant age group; Ps is the mean of the proportions of correct sells across all N2 negative information events; N2 is the total number of negative information events (where CAR(0,+1) < 0) with at least one trade on day t x by accountholders in the relevant age group; and N = N1 + N2 is the total number of information events; finally, we also provide the mean proportions of correct buys and sells (P b and Ps), along with the numbers of positive and negative information events (N1 and N2), respectively, for each age group. p-values highlighted in bold and with * represent significance at the 0.10 level or better. Panel A. Trades during the 3 Days before Major Earnings Announcements (1) (2) (3) (1) (2) (3) (1) (2) (3) Trades on Day 1 Trades on Day 2 Trades on Day 3 Young Older Diff Young Older Diff Young Older Diff N 252 1, , ,472 CAR(0,+1) % (p-value) 0.04* 0.04* 0.01* * 0.05* * Frequency (p-value) 0.03* 0.00* 0.01* 0.05* * 0.05* * P b ; N1 0.72; ; ; ; ; ; 737 Ps; N2 0.39; ; ; ; ; ; 735 (Continued)

22 384 The Journal of Finance R Table IV Continued Panel B. Trades during the 3 Days before Large Price Changes (1) (2) (3) (1) (2) (3) (1) (2) (3) Trades on Day 1 Trades on Day 2 Trades on Day 3 Young Older Diff Young Older Diff Young Older Diff N 192 1, , ,451 CAR(0,+1) % (p-value) 0.03* * * Frequency (p-value) 0.01* * * * P b ;N1 0.72; ; ; ; ; ; 732 Ps; N2 0.39; ; ; ; ; ; 719

23 Trading through the Accounts of Children 385 investors buying before good news events 59% of the time, while they sell before bad news events 45% of the time. The third column in Panel A of Table IV then compares the average performance across the two age groups. The mean difference t-test indicates that the 1.10% difference in the average signed CAR(0,+1) is significant at the 0.01 level. Similarly, the 5% difference in mean relative frequencies across the two age groups is also significant at the 0.01 level. The second set of three columns in Panel A of Table IV presents analogous results for trades made 2 days before the announcement. Now the mean signed CAR(0,+1) is 0.89% (p-value = 0.23) for young accountholders and 0.00% (p-value = 0.99) for older investors. The difference between these mean signed CARs equals 0.88% and is significantly different from zero (p-value = 0.04). Similarly, the difference in the mean frequency of correct trades is 6%, which is significantly different from zero (p-value = 0.01). In the last three columns, for trades made 3 days before the announcement, the mean signed CAR(0,+1) for young accountholders is 1.43% (p-value = 0.05), while the analogous result for older accountholders is 0.11% (p-value = 0.36). The difference in these mean signed CARs is a significant 1.54% (p-value = 0.01). Similarly, the 6% difference in mean relative frequencies across age groups is also significant (p-value = 0.01). 21 A.4. Event Study Results for Large Price Changes Panel B of Table IV documents the performance of trades prior to large absolute price changes. The most dramatic results appear in the first three columns for trades made 1 day before these events, where young accountholders have a significant mean signed CAR(0,+1) of 2.00% (p-value = 0.03), with an average of 58% of their trades in the correct direction (p-value = 0.01). This exceptional performance is comprised of correct buys 72% of the time, and correct sells 39% of the time. In contrast, older investors slightly underperform the market with an insignificant mean signed CAR(0,+1) of 0.09%, (p-value = 0.67) and an average of 50% of their trades in the correct direction (p-value = 0.85). The 2.1% difference in the mean signed CAR(0,+1) across the two age groups is significant at the 0.01 level, as is the 8% difference in the mean frequency of correct trades. The second set of three columns in Panel B of Table IV reports the analogous results based on trades made 2 days before large price changes. In this set of results, neither age group significantly outperforms the other in terms of either the mean signed CAR(0,+1) or the mean frequency of correct trades. In the third set of columns, based on trades made 3 days before the event, young investors once again significantly outperform older investors. The difference in the mean signed CAR(0,+1) is 1.18% (p-value = 0.08), and young 21 Similar analysis of earlier trades up to 6 weeks before the earnings announcement reveals no substantive evidence of earlier outperformance by either young or older accountholders.

24 386 The Journal of Finance R investors trade in the correct direction 4% more frequently than older investors, on average (p-value = 0.10). 22 We conclude from Table IV that a relatively large proportion of the trades by young accountholders during the few days ahead of both major earnings announcements and large price changes is motivated by superior private information that is about to become public. B. Event Study for Merger and Acquisition Announcements Table V presents the results for trades made by young versus older accountholders prior to takeover announcements. Panel C of Table III presents the descriptive statistics for this sample of announcements, and confirms that takeovers are typically good news events for the target firm. The mean CAR(0,+1) for the target firm across all takeover announcements is 10.31%, and more than 75% of these events have a positive CAR(0,+1). To account for the potential bias due to the tendency of young accountholders to buy more frequently than older investors, we include a new test in Table V that compares the mean performance of trades by young and older investors relative to alternative normal benchmarks for both groups, based on their own respective average propensities to buy and sell. B.1. Alternative Benchmark Analysis for Takeover Announcements If young accountholders are uninformed and buy 66% of the time on average (see Table I), then our alternative benchmark CAR for their expected return performance is: Normal CAR-young = [0.66 (CAR (0, +1) for buys) 0.34 (CAR (0, +1) for sells)] = 0.32 CAR (0, +1). Thus, we can define an alternative measure for the adjusted (abnormal) performance for young accountholders for each event as the difference between the actual signed CAR based on trades by young investors and this Normal CAR-young as follows: Adjusted CAR-young = [actual signed CAR (0, +1) 0.32 CAR (0, +1)]. For older investors we follow the same procedure, but now the likelihoods of buying and selling are 55% and 45%, respectively, resulting in the following measure of adjusted performance for older investors: Adjusted CAR-old = [actual signed CAR (0, +1) 0.1 CAR (0, +1)]. Similarly, the expected frequency of correct trades for uninformed young investors, Normal Frequency-young, is 0.66 for positive events and 0.34 for 22 Once again, similar analysis of trades up to 6 weeks prior to the large price change reveals no evidence of significant outperformance by any age group.

25 Trading through the Accounts of Children 387 Table V Performance of Young versus Older Accountholders before Takeover Announcements This table presents the mean CAR around takeover announcements (CAR(0,+1)) for trades by young or older investors before the announcement. We also provide the mean frequency of trading in the correct direction (Frequency = P b (N1/N) + Ps(N2/N)) for each age group, along with the mean proportions of correct buys and sells (P b and Ps), and the numbers of good news and bad news events (N1 and N2), respectively. Diff-1 represents the differential mean performance of young versus older investors. Diff-2 is the differential adjusted performance of young versus older investors relative to alternative benchmarks representing the expected performance of young and older accountholders, based on their respective tendencies to buy 66% and 55% of the time. All returns are in percentage points. p-values highlighted in bold and with * represent significance at the 0.10 level or better. Panel A. Trades during the 3 Days before Takeover Announcements (1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4) Trades on Day 1 Trades on Day 2 Trades on Day 3 Young Older Diff-1 Diff-2 Young Older Diff-1 Diff-2 Young Older Diff-1 Diff-2 n CAR(0,+1) % (p-value) * 0.02* * * 0.01* Frequency (p-value) 0.02* * 0.01* 0.03* * 0.01* 0.01* * 0.02* P b ; N1 : positive events 0.73; ; ; ; ; ; 107 Ps; N2: negative events 0.67; ; ; ; ; ; 33 (Continued)

26 388 The Journal of Finance R Table V Continued Panel B. Trades during the 3 Weeks before Takeover Announcements Trades during Week 1 Trades during Week 2 Trades during Week 3 Young Older Diff-1 Diff-2 Young Older Diff-1 Diff-2 Young Older Diff-1 Diff-2 N CAR(0,+1) % (p-value) * 0.05* 0.07* * 0.02* Frequency (p-value) 0.00* * 0.01* P b ; N1: positive events Ps; N2: negative events Panel C. Trades during the 4 to 6 Weeks before Takeover Announcements Trades during Week 4 Trades during Week 5 Trades during Week 6 Young Older Diff-1 Diff-2 Young Older Diff-1 Diff-2 Young Older Diff-1 Diff-2 N CAR(0,+1) % (p-value) 0.05* * 0.08* 0.06* * 0.08* 0.01* * 0.01* Frequency (p-value) 0.00* * * * * 0.33 P b ; N1: positive events 0.68; ; ; ; ; 109 Ps; N2: negative events 0.56; ; ; ; ; 33

27 Trading through the Accounts of Children 389 negative events. Similarly, for older investors Normal Frequency-old is 0.55 for positive events and 0.45 for negative events. Thus, an alternative measure of the adjusted frequency of correct trades equals the difference between the actual mean frequency of correct trades by young or old investors and the respective normal frequency for each age group, as follows: Adjusted Frequency-young = actual Frequency 0.66, for positive events, and = actual Frequency 0.34, for negative events; Adjusted Frequency-old = actual Frequency 0.55, for positive events, and = actual Frequency 0.45, for negative events. As before, we average these alternative measures for the adjusted CAR and the adjusted frequency of correct trades across all events to obtain the mean adjusted CAR and the mean adjusted frequency of correct trades by investors in each age group. B.2. Event Study Results for Takeover Announcements Table V provides the results for takeover announcements. The first three columns in each set of results apply the same analysis as in Table IV. As before, the third column, labeled Diff-1, compares the actual mean performance of young accountholders with that of older investors trading just before the event. The fourth column, labeled Diff-2, provides our comparison of the new adjusted performance measures for young versus older accountholders. The first column of Panel A in Table V is based on the 23 takeover announcements where young investors trade on day 1. This column reveals an average signed CAR(0,+1) that is very large in magnitude, at 12.7%, although it is not statistically significant (p-value = 0.15). Similarly, this column documents an extraordinary overall success rate, showing that young investors trade in the correct direction 72% of the time on the day before takeover announcements (p-value = 0.02). This mean overall frequency is due to young accountholders correctly buying just before good news takeover announcements an average of 73% of the time, and correctly selling just before bad news events 67% of the time. In comparison, the second column of Panel A in Table V is based on the 140 takeover announcements for which at least one older investor trades on day t 1. This column reveals that older investors earn a mean signed CAR(0,+1) of 0.32% (p-value = 0.77), while the mean frequency of correct trades is only 47% (p-value = 0.18). This extraordinary performance by young investors also extends to their earlier trades made during the 6 weeks prior to the takeover announcement. For all nine time frames examined in Table V, the mean signed CAR(0,+1) for young investors is consistently positive and large in magnitude, whereas the mean signed CAR(0,+1) for older investors is close to zero. Similarly, the

28 390 The Journal of Finance R mean overall frequency of correct trades is consistently far above 50% for young accountholders while remaining close to 50% for older investors. The third column (Diff-1), shows that the mean differences in signed CAR(0,+1) are significantly greater than zero at the 10% level or better, for eight of the nine time frames analyzed in Table V. Similarly, for the frequency of correct trades, seven of these nine mean differences are significantly greater than zero at the 10% level or better. The fourth column for each set of results in Table V demonstrates that this extraordinary performance by young accountholders prevails when we use our alternative benchmark to adjust for the normal behavior expected for each age group. The mean difference between adjusted signed CARs for young versus older accountholders (Diff-2) is significantly positive for seven of the nine time frames examined, for trades up to 6 weeks before the takeover event. 23 In addition, the mean adjusted frequency of correct trades for young accountholders is larger than that for older investors most of the time, and is significantly larger for trades made in the first week before the announcements. This evidence establishes that young accountholders dramatically outperform older investors when they trade ahead of takeover announcements. This evidence corroborates the results in Table IV, indicating that a large proportion of trades by young accountholders before major information events is motivated by private information that is about to become public. C. Performance of Guardians before Major Information Events We repeat the event study analysis of Tables IV and V for trades made by guardians matched to underaged accountholders by family name or by similar trades in corporate accounts. Results are provided in Table VI for trades made by each set of guardians in the 3 days before major earnings announcements and large price changes. Table VII presents analogous results for trades made in the 6 weeks before takeover announcements. Panels A and B of Table VI indicate that the set of family-matched guardians significantly outperforms other older investors when trading 1 day before earnings announcements, and marginally outperforms for trades made 3 days ahead of large price changes. In contrast, Panels A to C of Table VII show that this set of guardians does not significantly outperform when they trade ahead of takeover announcements. Similar results appear in Panels C and D of Table VI, which show that corporate guardians outperform other adults on the 3 days before earnings announcements, and on days 1 and 3 ahead of large price changes. Similarly, Panels D to F of Table VII indicate that this set of guardians does not outperform other adults ahead of takeover announcements. 23 Considered on its own, the mean adjusted signed CAR for young investors is positive for eight of the nine time frames, and is significantly different from zero for week 6. Similarly, the median adjusted signed CAR is positive for all time frames and, based on the signed rank test, is significantly larger than zero (at the 10% level or better) for days 1, 2, and 3 and weeks 1, 4, and 6.

29 Trading through the Accounts of Children 391 Table VI Performance of Guardians versus Other Older Accountholders before Major Earnings Announcements and Large Price Changes This table presents the mean CAR on the day of, and the day after, major earnings announcements or large price changes (CAR(0,+1)), that pertains to trades made by each set of guardians versus other investors older than 10 years, in the 3 days ahead of the event. In addition, we provide the mean frequency of correct trades (Frequency) for each group. In Panels A and B, we provide results for guardians matched to underaged accounts by family name for each collection of events. In Panels C and D, we present the analogous results for guardians matched with similar trades through a corporation. All returns are in percentage points. p-values highlighted in bold and with * represent significance at the 0.10 level or better. Panel A. Guardians Matched to Family Name versus Other Adults before Earnings Announcements Trades on Day 1 Trades on Day 2 Trades on Day 3 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n 759 1, , ,470 CAR(0,+1) % (p-value) 0.02* * Frequency (p-value) 0.01* 0.03* 0.07* Panel B. Guardians Matched to Family Name versus Other Adults before Large Price Changes n 563 1, , ,443 CAR(0,+1) % (p-value) Frequency (p-value) * 0.04* * * Panel C. Guardians Matched to Corporation versus Other Adults before Earnings Announcements n 761 1, , ,472 CAR(0,+1) % (p-value) 0.00* 0.04* 0.01* 0.05* * * Frequency (p-value) 0.00* 0.00* 0.02* 0.01* * Panel D. Guardians Matched to Corporation versus Other Adults before Large Price Changes n 622 1, , ,451 CAR(0,+1) % (p-value) 0.02* * * * * Frequency (p-value) 0.00* * * * * D. Trading on Inside Information versus Broader Stock-Picking Ability The outstanding performance by underaged accounts just ahead of major information events that we document could result from guardians legally obtaining public information in a manner that enables them to consistently predict forthcoming major events. However, this performance might also be due to insider trading. We shed additional light on this issue by examining whether the superior performance of guardians trading through underaged accounts is revealed only through multiple trades in an individual stock or industry, or is also

30 392 The Journal of Finance R Table VII Performance of Guardians versus Other Adults before Takeover Announcements This table presents the mean CAR (CAR(0,+1)) and the mean frequency of trading in the correct direction (Frequency) for trades made by guardians versus other adults in the 6 weeks before the announcement. In Panels A to C, we provide results for guardians matched to underaged accounts by family name. In Panels D to F, we present results for guardians matched with similar trades through a corporation. All returns are in percentage points. p-values highlighted in bold and with * represent significance at the 0.10 level or better. Panel A. Guardians Matched to Family Name versus Other Adults in the 3 Days before Trades on Day 1 Trades on Day 2 Trades on Day 3 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n CAR(0,+1) % (p-value) Frequency (p-value) Panel B. Guardians Matched to Family Name versus Other Adults in the 3 Weeks before Trades during Week 1 Trades during Week 2 Trades during Week 3 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n CAR(0,+1) % (p-value) 0.09* * Frequency (p-value) Panel C. Guardians Matched to Family Name versus Other Adults in the 4 to 6 Weeks before Trades during Week 4 Trades during Week 5 Trades during Week 6 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n CAR(0,+1) % (p-value) * * Frequency (p-value) Panel D. Guardians Matched to Corporation versus Other Adults in the 3 Days before Trades on Day 1 Trades on Day 2 Trades on Day 3 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n CAR(0,+1) % (p-value) Frequency (p-value) (Continued)

31 Trading through the Accounts of Children 393 Table VII Continued Panel E. Guardians Matched to Corporation versus Other Adults in the 3 Weeks before Trades during Week 1 Trades during Week 2 Trades during Week 3 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n CAR(0,+1) % (p-value) 0.05* Frequency (p-value) Panel F. Guardians Matched to Corporation versus Other Adults in the 4 to 6 Weeks before Trades during Week 4 Trades during Week 5 Trades during Week 6 Guardians Other Adults Diff Guardians Other Adults Diff Guardians Other Adults Diff n CAR(0,+1) % (p-value) Frequency (p-value) revealed through individual trades across multiple stocks or industries. One might expect that a corporate insider would have special access to valuable inside information about just one company, or perhaps a few companies within the same industry, but not about multiple companies in different industries. Therefore, such access might be expected to result in multiple insider trading opportunities over time in the same stock or industry, but not outside that industry. In Table VIII, we compare the performance of two subsets of underaged accountholders (aged 0 to 10): (1) those who trade before more than one event for any given firm (or industry), and (2) those who trade before only one event for any given firm (or industry), but do so for two or more different firms (or industries). We focus this new analysis on the broad sample of information events that combines our three samples of earnings announcements, large price changes, and takeover announcements. If an investor benefits from private information that is concentrated at the level of a firm or industry, we would expect this advantage to appear around all three types of events. We focus on trades made by young accountholders in the 3 days before these events, since our results indicate that these trades are most informative. The question is whether young accountholders are only successful when they trade before multiple events for an individual stock or industry, or perhaps are also successful through trades before individual events across multiple stocks or industries. In total our sample comprises 1,860 events for which one or more young accountholders trade in the 3 days prior to the event. First, across all these events, we identify the two subsets of young accountholders described earlier. Next, for each event, we calculate the average signed CAR(0,+1) and frequency of correct trades for each group of young accountholders. Finally, for both groups,

32 394 The Journal of Finance R Table VIII Performance of Two Subsets of Underaged Accountholders Who Trade: (1) before Multiple Information Events for an Individual Firm or Industry; (2) before an Individual Information Event for Multiple Firms or Industries This table presents the mean CAR on the day of and the day after the event (CAR(0,+1)), and the mean frequency of correct trades (Frequency), for two subsets of underaged accountholders (aged 0 to 10 years) who trade in the 3 days before more than one information event. We present these performance measures for: (1) the group that trades before more than one information event for any given firm (or industry); (2) the group that trades before only one information event for any given firm (or industry), but does so for two or more firms (or industries); and (3) the difference in means across these two groups. In Panel A, we provide results for the entire combined sample including all three types of events (earnings announcements, large price changes, and takeover announcements). In Panel B, we present the analogous results for the subset of this combined sample that includes only major information events, where the absolute CAR(0,+1) > 4%. All returns are in percentage points. p-values highlighted in bold and with * represent significance at the 0.10 level or better. Panel A. Underaged Trading before All Events of All Three Types Underaged Accountholders Who Trade before: Underaged Accountholders Who Trade before: (1) (2) (3) (1) (2) (3) Multiple Events for Individual Firm Individual Event for Multiple Firms Diff: (1) (2) Multiple Events for Individual Industry Individual Event for Multiple Industries Diff: (1) (2) n 346 1, CAR(0,+1) % (p-value) 0.09* 0.01* * 0.04* 0.10* Frequency (p-value) 0.01* 0.01* * 0.01* 0.38 Panel B. Underaged Trading before Only Major Events with Absolute Value of CAR(0,+1) > 4% Underaged Accountholders Who Trade before: Underaged Accountholders Who Trade before: (1) (2) (3) (1) (2) (3) Multiple Events for Individual Firm Individual Event for Multiple Firms Diff: (1) (2) Multiple Events for Individual Industry Individual Event for Multiple Industries Diff: (1) (2) n CAR(0,+1) % (p-value) 0.01* 0.02* 0.05* 0.01* 0.09* 0.01* Frequency (p-value) 0.01* 0.01* * 0.01* 0.13

33 Trading through the Accounts of Children 395 we calculate the average performance measures across all events and compare the group means. Panel A of Table VIII provides the results for the combined sample of all events, while Panel B presents analogous results for the combined subsamples of most informative events where the absolute CAR(0,+1) exceeds 4%. Each panel presents the mean signed CAR(0,+1) and the frequency of correct trades for the two groups of underaged accounts that (1) trade before multiple events for an individual firm (or industry), or (2) trade before an individual event for multiple firms (or industries). The panel also presents the mean differences across these two groups. Table VIII shows that the mean signed CAR(0,+1) and the frequency of correct trades are significantly greater than zero and 50%, respectively, for both groups of young accountholders. Moreover, these results are uniformly larger in magnitude for the subsample of young accountholders who trade before multiple events in the same stock or industry. However, the difference in CARs across the two groups is significant at the 0.05 level or better only in Panel B, when we restrict the sample to major events with an absolute CAR(0,+1) > 4%. Similarly, the difference in the mean frequency of correct trades across the two groups approaches standard levels of statistical significance only in Panel B (where the p-values are 0.15 and 0.13 for trades in the same stock or the same industry, respectively). These results suggest that some guardians have greater access to valuable private information concentrated at the level of a firm or industry, which leads to significantly greater outperformance when they trade several times before major information events involving the same stock or industry. This evidence lends further credence to the view that access to inside information is one of the sources of superior performance behind underaged accounts around major information events. On the other hand, we also find that other guardians also significantly outperform when they trade before only one major event for any given firm, but do so across multiple stocks (or industries). This evidence adds to the case that the outstanding performance of guardians reflects broader stock-picking ability, and not just concentrated inside information. V. Trading through Underaged Accounts and Stock Returns This section tests the proposition in Easley and O Hara (2004) that investors demand a higher return for holding stocks with a greater likelihood of private information. These tests focus on our proxy for stock-specific information asymmetry in month m (BABYPIN i,m ), which, for each stock i, is defined as the proportion of total trading activity through accounts of children aged 0 to 10 years measured over the current and previous 2 months (m, m 1andm 2). Our joint hypothesis is that: (i) a higher proportion of total trading through underaged accounts predicts a higher risk of information-based trading, and (ii) this higher risk translates into a higher required return for that stock. We first describe the data and then examine the relation between BABYPIN and

34 396 The Journal of Finance R Table IX Descriptive Statistics for Monthly Control Variables Panels A and B provide summary statistics and correlations, respectively, for the monthly control variables in our analysis of the relation between BABYPIN and stock returns. A stock s BABYPIN for a given month (m) is defined as the proportion of total trading activity through underaged accounts over a 3-month period including the current month and the previous 2 months (m, m 1, and m 2). BETA, Size, M/B, andryear are defined in the text. Finally, for each month we estimate the average intraday bid ask spread (Spread) and the traditional measure of the probability of informed trading (PIN) using intraday quotes from Thomson Reuters, in combination with Euroclear data and intraday trading data from the Nasdaq OMX Helsinki exchange. The PIN measure is estimated following the procedure in Easley, Kiefer, and O Hara (1997). These statistics are first calculated cross-sectionally each month, and then are averaged across all months in the sample. Correlations highlighted in bold and with * are significantly different from zero at the 0.05 level. Panel A. Summary Statistics BABYPIN Spread BETA Size (Millions of ) M/B RYear PIN MEAN 0.8% 1.1% , % 23.1% STD 0.4% 1.5% , % 15.3% Avg # firms/month Panel B. Correlations across Variables BABYPIN 1 Spread 0.04* 1 BETA 0.16* 0.21* 1 ln (Size) 0.03* 0.88* 0.21* 1 M/B * 0.23* 0.06* 1 Ryear 0.07* 0.11* * 0.23* 1 PIN 0.04* 0.11* 0.07* 0.11* 0.04* future stock returns, after controlling for standard risk factors (including the traditional PIN measure). A. Data Panels A and B of Table IX present summary statistics for the monthly variables used in our tests. These summary statistics are first calculated cross-sectionally each month and then averaged across the months in our sample period. We only include months with trading in at least 30 of the stocks listed on the Nasdaq OMX Helsinki exchange, leaving 127 months in our sample. Since intraday transactions data are only available beginning in 1999, descriptive statistics for the intraday quoted spread and the traditional PIN measure are averaged over 110 months. Measurement of BABYPIN depends on the relatively infrequent trading activity of young accountholders over a 3-month period. To help ensure the reliable measurement of this proxy, we require at least 500 total trades in a given stock during this 3-month period before estimating BABYPIN. For each

35 Trading through the Accounts of Children 397 stock in our sample, we also estimate the traditional PIN measure based on all trades in the previous 3 months, following the procedure in Easley, Kiefer, and O Hara (1997). The other variables are measured as the average daily value during each month. Panel A of Table IX shows that trading through underaged accounts (BABYPIN) comprises less than 1% of all trades in a stock, consistent with the results in Table I. In addition, the typical firm has an average daily proportional quoted spread of approximately 1.1%, a market β of 0.82, a market capitalization of 1.6 billion euros, a market-to-book ratio of 2.8, and a previous 12-month return of 4.9%. The average PIN measure is 23%, similar to that in Easley, Hvidkjaer, and O Hara (2002). Panel B of Table IX provides the time-series means of the monthly crosssectional correlations among these variables. We are particularly interested in the correlation of BABYPIN with the other variables. Consistent with expectations, BABYPIN is positively correlated with the spread (ρ = 0.04, p-value = 0.01) and PIN (ρ = 0.04, p-value = 0.01). Further, BABYPIN is negatively correlated with β, firm size, and the prior 12-month return. B. Portfolio Tests Single Sort Based on BABYPIN As a first step toward testing whether the probability of information trading is related to future returns, we compute the average returns in month m for tercile portfolios of stocks sorted each month by our proxy for information asymmetry in the previous month, BABYPIN m-1. The first column in Table X presents the mean return for these monthly BABYPIN tercile portfolios averaged across all months in the sample period. As before, the t-statistics are based on the standard errors of the time-series means. The first column in Table X shows that the tercile of stocks with the lowest value of BABYPIN m-1 generates a mean return of 0.2% per month, whereas the tercile with the highest BABYPIN m-1 has a mean return of 0.9% per month. The difference between these mean returns represents an average raw BABYPIN hedge portfolio return of 1.1%, which is significant at the 0.05 level. The last two rows of Table X report the α s from time-series regressions of the return on these monthly rebalanced BABYPIN hedge portfolios against: (i) the three Fama French factors (FF (H-L)), and (ii) the three Fama French factors plus the world market excess return over the risk-free rate (FF + World (H-L)). The values of the Fama French α s in the first column of Table X are both approximately 1.1% per month, and are significantly different from zero at the 10% level Using monthly data for all Finnish stocks, we follow the procedures in Fama and French (1993) to calculate their three factors. The world market return is from Global Financial Data and is in local currency.

36 398 The Journal of Finance R Table X The Relation between BABYPIN and Stock Returns: Portfolio Approach This table presents the results from two portfolio approaches and two separate Fama-French regression analyses. In the first approach, we perform a single sort, in which we rank all stocks each month according to BABYPIN in the previous month (m 1). We then present the mean return during the following month (m), for each BABYPIN-Tercile Portfolio, along with the mean return on the BABYPIN-Hedge Portfolio across terciles with high versus low BABYPIN m-1 (Raw (H L)). In the second approach, we perform a series of double sorts in which we first rank all stocks each month according to one of six firm characteristics during the previous month (m 1): Size, BETA, M/B, RYear, Spread, orpin (all variables are defined in the text). Then, within each tercile for every firm characteristic, we sort into finer tercile portfolios by BABYPIN during the previous month (m 1). The result is a set of Adjusted BABYPIN-Tercile Portfolios, where we control for each firm characteristic. We then present the mean return during the following month (m), for each Adjusted BABYPIN-Tercile Portfolio along with the mean return on the monthly rebalanced Adjusted BABYPIN-Hedge Portfolio (Raw (H L)). Finally, we also present the α from a timeseries regression of the monthly rebalanced return of each Adjusted BABYPIN-Hedge Portfolio against the three Fama French factors (FF (H L)), and against the Fama French factors plus the excess return on the world market portfolio (FF + World (H L)). All mean returns are in percentage points. * indicates significance at the 0.10 level, ** at the 0.05 level, and *** at the 0.01 level. Sorted by: Single Sort BABYPIN Size- Adjusted BABYPIN BETA- Adjusted BABYPIN Double Sorts M/B- Adjusted BABYPIN RYear- Adjusted BABYPIN Spread- Adjusted BABYPIN PIN- Adjusted BABYPIN Low t-stat Medium t-stat High t-stat * 1.6* 1.9* Raw (H L) t-stat 2.1** 2.1** 2.0** 1.9* 3.0*** 2.4** 2.6*** FF (H L) t-stat 1.9* 2.1** 1.7* 2.2** 2.7*** 2.4** 2.3** FF + World (H L) t-stat 1.8* 2.1** 1.6* 2.1** 2.3** 2.1** 2.1** Double Sorts Based on Firm Characteristics and BABYPIN The remaining columns of Table X present the average portfolio returns for double sorts based on BABYPIN and several firm characteristics known to be associated with returns. Each month, we first sort the sample stocks into terciles based on the previous month s value of each characteristic (Size m-1, BETA m-1, M/B m-1, Ryear m-1, Spread m-1,orpin m-1 ). We then sort every tercile by each lagged characteristic into finer terciles based on BABYPIN m-1. This procedure results in a set of adjusted BABYPIN tercile portfolios that differ

37 Trading through the Accounts of Children 399 in BABYPIN but are similar in terms of the firm characteristic used in the first sort. The results for this double sorting procedure are consistent with those for the single sort. For example, in the second column of Table X portfolios with low values of size-adjusted BABYPIN have low mean returns ( 0.2%, t-statistic = 0.2), whereas portfolios with high values of size-adjusted BABYPIN have high mean returns (0.9%, t-statistic= 1.4), resulting in a mean hedge portfolio return of 1.1% (t-ratio = 2.1). The same pattern appears for every other double sorted scheme provided in Table X. The average raw adjusted BABYPIN hedge portfolio return, based on high minus low values for every adjusted BABYPIN measure, varies from 0.9% to 1.4% per month across the columns of Table X, andis always significant at the 10% level or better. Similarly, the Fama French α s of the monthly rebalanced adjusted BABYPIN hedge portfolios range from 0.9% to 1.4%, and are significant at the 10% level or better across all of the double sorted schemes, regardless of whether we include the excess world market return. C. Cross-Sectional Regression Tests This section follows Easley, Hvidkjaer, and O Hara (2002) and Fama and French (1992) to specify a cross-sectional regression model in which the stock return in the current month (R i,m ) depends upon several firm characteristics measured over the previous month. The general specification includes our proxy for information asymmetry (BABYPIN i,m-1 ), the firm s β with respect to the local market return (BETA i,m-1 ), the natural log of market capitalization (ln(size i,m-1 )), the market-to-book ratio (M/B i,m-1 ), the return over the previous 12 months (RYear i,m-1 ), the traditional PIN measure (PIN i,m-1 ), the percentage bid ask spread (Spread i,m-1 ), and the β relative to the world market return (BETA World i,m-1 ) as follows: R i,m = c 0 +c 1 BABY PIN i,m 1 +c 2 BETA i,m 1 +c 3 ln(size i,m 1 ) + c 4 M/B i,m 1 + c 5 RYear i,m 1 +c 6 PIN i,m 1 +c 7 Spread i,m 1 +c 8 BETA World i,m 1 +ε i,m. (3) Table XI presents the Fama MacBeth mean coefficients from estimating several alternative specifications that include different subsets of the control variables in equation (3). We first estimate this cross-sectional regression for each of the 127 months in our sample (or 110 months if PIN or SPREAD is included in the specification). The coefficients from these monthly crosssectional regressions are then averaged over time. The results in Table XI show that, during our sample period, larger stocks tend to perform better in Finland. In addition, there is weak evidence that value stocks tend to outperform growth stocks. There is also some evidence of momentum in stock returns. Columns (2) to (5) of Table XI present the results when the model includes one or both measures of information asymmetry (BABYPIN and PIN). The

38 400 The Journal of Finance R Table XI The Relation between BABYPIN and Stock Returns: Fama MacBeth Regressions This table presents the results from estimating the following regression model: R i,m = c 0 + c 1 BABY PIN i,m 1 + c 2 BETA i,m 1 + c 3 ln( Size i,m 1 ) + c 4 M/B i,m 1 + c 5 RYear i,m 1 + c 6 PIN i,m 1 + c 7 Spread i,m 1 + c 8 BETA World i,m 1 + ε i,m. (3) All variables are defined in the text. The dependent variable is the firm s monthly return in percentage points. In columns (1) to (5), we present the mean Fama MacBeth coefficients across monthly cross-sectional regressions for five specifications that include different subsets of the right-hand side variables in equation (3). In column (6) we present the same model specified in column (3), except that we replace the single BABYPIN measure with two measures, one based on purchases and one based on sales by underaged investors, both as a proportion of total trading activity. Finally, column (7) reports the analogous results for the same model specified in column (3), using a rank regression in which all right-hand side variables are first converted to decile ranks each month, and then the decile ranks are scaled to range between 0.5 and In each column, the t-statistic appears directly below every Fama MacBeth mean coefficient. The standard error of every time-series mean coefficient is used to construct every t-statistic. * indicates significance at the 0.10 level, ** at the 0.05 level, and *** at the 0.01 level. (1) (2) (3) (4) (5) (6) (7) BETA-Local Mkt ln (Size) ** 2.8*** 2.7*** 3.0*** ** 2.3** M/B * Ryear * 1.7* 1.9* * 0.9* Spread BETA-World Mkt PIN BABYPIN *** 3.6*** 3.1*** 3.0*** BABYPIN-Buy ** BABYPIN-Sell * Avg # Firms Avg Adj R firm s traditional PIN measure is never significantly related to the return in the following month, regardless of whether BABYPIN is included in the specification. Given that most of the sample firms are small by U.S. standards, this result is somewhat surprising since U.S. evidence suggests that smaller stocks

39 Trading through the Accounts of Children 401 with high PIN generate excess returns. 25 The most important result is that a higher risk of informed trading, as captured by BABYPIN, is consistently associated with a significant increase in the return. This result does not change when we include PIN, or when we add additional explanatory variables such as the spread or the world-market β. 26 In column (6) of Table XI, we modify the model in column (3) by splitting BABYPIN into two BABYPIN measures: one based on purchases by young investors and one based on sales by young investors (both computed as a proportion of total trading). The results reveal similar coefficients for both measures of BABYPIN, based on either purchases or sales in underaged accounts. The coefficient on the purchase-based measure of BABYPIN equals 1.3 (t-statistic= 2.0), while the coefficient on the sales-based measure is slightly larger at 1.7 (tstatistic= 1.6). This result is important, since it indicates that both buying and selling activity through underaged accounts reinforce a signal for other market participants to demand a higher return to compensate for the increased risk of trading with an informed investor. In the final column of Table XI, we provide the results from estimating the model in column (3), but we first transform all independent variables into decile ranks each month and then scale these decile ranks to range between 0.5 and This transformation renders the regression results less sensitive to measurement errors or outliers, and facilitates the interpretation of the coefficients (Nagel (2005)). The coefficient estimate on BABYPIN in column (7) implies that the average difference in returns between the top and bottom deciles in terms of BABYPIN is 2.0% per month (t-statistic is 3.0). From the evidence in Table XI, we conclude that our proposed measure for the probability of informed trading based on recent trading activity through underaged accounts provides substantive incremental information regarding the return required for stocks. These results contribute strong corroborating evidence to support theoretical models in which the risk of informed trading is an important determinant of the cross-section of stock returns (Easley and O Hara (2004), Lambert, Leuz, and Verrecchia (2011)). VI. Summary and Conclusions This paper proposes that the proportion of total trading activity through underaged accounts (BABYPIN) serves as a useful measure of the probability of information-based trading in a stock. In support of this view, we show that the accounts of young children, aged 0 to 10 years, exhibit superior stock-picking skills. Over the days immediately following their trades, these underaged accounts significantly outperform older investors. Further analysis 25 For example, see Armstrong et al. (2011), Easley, Hvidkjaer, and O Hara (2002), and Mohanram and Rajgopal (2009). 26 The world market β is estimated using the same procedure as the local β (see Internet Appendix Section III). We have also considered alternative measures of liquidity in this analysis, including the firm s average daily share turnover and the daily number of trades. These alternative measures do not alter the results or conclusions.

40 402 The Journal of Finance R reveals that underaged accountholders perform especially well when they trade in the days before major earnings announcements, large absolute price changes, and takeover announcements. We extend the analysis by examining the trading performance of two separate sets of guardians who are matched to underaged accounts using either family name or similar trades in corporate accounts. Consistent with our expectations, these guardians are wealthier than other adults, and they outperform other adults when they trade through either their own accounts or through corporate accounts. However, our results show that guardians outperform other adults by a greater margin when they trade through underaged accounts, especially on the sell side and in the days before major information events. This evidence indicates that, when guardians trade through underaged accounts, there is a relatively high probability that they are trading on private information. It appears that underaged accounts are an effective mechanism for filtering out private information from informed guardians. We analyze the association between BABYPIN and future stock returns, and find strong empirical support for theoretical models in which investors demand a higher return for holding stocks with a greater likelihood of private information (Easley and O Hara (2004), Lambert and Verrecchia (2010)). Our evidence is based on both portfolio tests and cross-sectional regression analysis, and is robust to controlling for other firm characteristics such as size, β, bookto-market ratio, momentum returns, spread, and the traditional PIN measure. The novel measure of information asymmetry proposed in this study (BABYPIN) is easy to calculate and available in a timely fashion, given data on the age and trading activity of individual accountholders. Furthermore, unlike other commonly used proxies for the probability of information-based trading, its validity is directly established by empirical evidence that a relatively large proportion of underaged trading is motivated by superior private information. Because of the obvious relevance of this analysis to regulators and market participants, we are hopeful that more databases will begin reporting a variety of accountholder characteristics, such as age. This development would facilitate exciting research opportunities in important areas such as asset pricing and behavioral finance. Initial submission: August 7, 2011; Final version received: February 20, 2013 Editor: Campbell Harvey REFERENCES Aktas, Nihat, Eric de Bodt, Fany Declerck, and Herve Van Oppens, 2007, The PIN anomaly around M&A announcements, Journal of Financial Markets 10, Armstrong, Christopher, John Core, Daniel Taylor, and Robert Verrecchia, 2011, When does information asymmetry affect the cost of capital? Journal of Accounting Research 49, Bhattacharya, Utpal, and Hazem Daouk, 2002, The world price of insider trading, Journal of Finance 57, Chan, Louis, and Josef Lakonishok, 1993, Institutional trades and intra-day stock price behavior, Journal of Financial Economics 33,

41 Trading through the Accounts of Children 403 Cohen, Lauren, Andrea Frazzini, and Christopher Malloy, 2008, The small world of investing: Board connections and mutual fund returns, Journal of Political Economy 116, Cohen, Lauren, Andrea Frazzini, and Christopher Malloy, 2010, Sell-side school ties, Journal of Finance 65, Cohen, Lauren, Christopher Malloy, and Lukasz Pomorski, 2012, Decoding inside information, Journal of Finance 67, Duarte, Jefferson, and Lance Young, 2009, Why is PIN priced? Journal of Financial Economics 91, Easley, David, Soeren Hvidkjaer, and Maureen O Hara, 2002, Is information risk a determinant of asset returns? Journal of Finance 57, Easley, David, Soeren Hvidkjaer, and Maureen O Hara, 2010, Factoring information into returns, Journal of Financial and Quantitative Analysis 45, Easley, David, Nicolas Kiefer, and Maureen O Hara, 1997, One day in the life of a very common stock, Review of Financial Studies 10, Easley, David, and Maureen O Hara, 2004, Information and the cost of capital, Journal of Finance 59, Fama, Eugene F., and Kenneth R. French, 1992, The cross section of expected stock returns, Journal of Finance 47, Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, Fama, Eugene F., and James MacBeth, 1973, Risk, return and equilibrium: Empirical tests, Journal of Political Economy 81, Grinblatt, Mark, and Matti Keloharju, 2000, The investment behavior and performance of various investor types: A study of Finland s unique data set, Journal of Financial Economics 55, Grinblatt, Mark, Matti Keloharju, and Juhani Linnainma, 2011, IQ and stock market participation, Journal of Finance 66, Grinblatt, Mark, Matti Keloharju, and Juhani Linnainma, 2012, IQ, trading behavior, and performance, Journal of Financial Economics 104, Hughes, John S., Jing Liu, and Jun Liu, 2007, Information asymmetry, diversification, and cost of capital, The Accounting Review 82, Ivkovic, Zoran, and Scott Weisbenner, 2005, Local does as local is: Information content of the geography of individual investors common stock investments, Journal of Finance 60, Kraus, Alan, and Hans R. Stoll, 1972, Price impacts of block trading on the New York Stock Exchange, Journal of Finance 27, Kyle, Albert S., 1985, Continuous auctions and insider trading, Econometrica 53, Lambert, Richard A., Christian Leuz, and Robert E. Verrecchia, 2007, Accounting information, disclosure, and the cost of capital, Journal of Accounting Research 45, Lambert, Richard A., Christian Leuz, and Robert E. Verrecchia, 2011, Information asymmetry, information precision, and the cost of capital, Review of Finance 16, Lambert, Richard, and Robert Verrecchia, 2010, Cost of capital in imperfect competition settings, Working paper, University of Pennsylvania, Wharton. Li, Hai Tao, Junbo Wang, Chunchi Wu, and Yan He, 2009, Are liquidity and information risks priced in the Treasury bond market? Journal of Finance 64, Mohanram, Partha, and Shiva Rajgopal, 2009, Is PIN priced risk? Journal of Accounting and Economics 47, Nagel, Stefan, 2005, Short sales, institutional investors and the cross section of stock returns, Journal of Financial Economics 78, Neal, Robert, and Simon Wheatley, 1998, Adverse selection and bid ask spreads: Evidence from closed-end funds, Journal of Financial Markets 1, Saar, Gideon, 2001, Price impact asymmetry and block trades: An institutional trading explanation, Review of Financial Studies 14, Seasholes, Mark, and Ning Zhu, 2010, Individual investors and local bias, Journal of Finance 65,

42 404 The Journal of Finance R Seru, Amit, Tyler Shumway, and Noah Stoffman, 2010, Learning by trading, Review of Financial Studies 23, Van Ness, Bonnie, Robert Van Ness, and Richard Warr, 2001, How well do adverse selection components measure adverse selection? Financial Management 30, Supporting Information Additional Supporting Information may be found in the online version of this article at the publisher s web site: Appendix S1: Internet Appendix