Do Fools Rush in? IPOs and Investor Sophistication

Transcription

1 Do Fools Rush in? IPOs and Investor Sophistication Jinghan Cai August 13, 2013 Abstract What kind of investors flock to an IPO mostly sophisticated or mostly naive? The answer to this question points to explaining the puzzlingly extreme trading volume on the first day after an IPO. Existing explanations rely on institutions such as day trading, short selling and inter-dealer trades, yet IPO frenzies are common even when these are entirely absent. Recent evidence points to the possible importance of sentiment from retail investors, but it is not yet clear what kind of retail investors might be harboring these emotions. I access a unique data set for Chinese IPOs that measures investor experience and trading records. I find that inexperienced investors are initially drawn to the IPO while established investors remain on the sidelines. Over time, investor composition shifts in favor of experienced investors. More importantly, I identify market timing of purchase (together with the timing of selling, the purchase price, etc, which I define as the decision bundle) as the predominant channel for determining heterogeneity in returns for experienced versus for inexperienced investors. Furthermore, I find that investors do learn to be more patient and get better investment performance thereof. Also, I am able to depict the learning curve by documenting that the marginal effect of learning varies across the level of stock of experience, and across heterogeneous investor types. JEL Classification: D19, G11 Keywords: IPO, investor experience, learning PRELIMINARY VERSION, PLEASE DO NOT QUOTE OR CITE. Department of Economics, Boston College. caid@bc.edu. I would like to thank Brent Bundick, Don Cox, Jun (QJ) Qian, Nadya Malenko, Dmitriy Muravyev, Yao Shen, Georg Strasser, Le Xia, Zhiie Xiao, Jialin Yu and seminar participants at Boston College for their valuable comments. 1

2 1 Introduction The simple motivation of this paper starts with the observation that the trading volume on the first day following an initial public offering (IPO) is enormous. In the U.S., the first day trading volume exceeds 70% of the shares sold in the offering (see Ellis et al., 2000, 2002; Aggarwal, 2000, 2003, among others). Existing literature tries to figure out who is trading on the very first days of an IPO. Early papers, e.g, Siconolfi (1998), Siconolfi and McGeehan (1998), Tam and Ewing (2000), Smith and Pulliam (2000), believe that flippers (investors who receive allocations of shares in IPOs and sell these allocations in the first few days of trading) are responsible for this large volume. More recently, Ellis (2006) argues that the composition of trading varies widely with the initial return and not all trading is investor-related. Cold IPOs have a high proportion of interdealer sell trades, whereas hot IPOs have balanced investor buying and selling. Moreover, most of the existing literature focuses on the sell side, and especially, on the institutional behaviors. For example, Aggarwal (2003), and Boehmer, Boehmer, and Fishe (2006) study the selling of IPO allocations by institutions in the short run. Chemmanur, et al. (2010) study the institutional selling of allocations beyond the immediate post-ipo period and find that institutions sell 70.2% of their IPO allocations in the first year, fully realize the money left on the table, and do not dissipate these profits in post-ipo trading. However, these existing explanations are far from comprehensive. First, while flippers may be willing to sell their shares, who is interested in buying the shares? This question remains largely unanswered, and this is especially important for order driven markets, where whenever there is a sell, there must be a buy. Second, the huge initial day volume after IPOs is not unique in the U.S. market. Other markets also exhibit similar situations for different reasons. For example, during January 2007 to 2

3 June 2008, the first day turnover of Shenzhen Stock Exchange IPOs in China is as high as 69.5%. In sharp contrast to this number, the second day turnover is 34.2%, and the tenth day turnover is only 10.6%. Consistently, the average daily number of trades is 28.8 thousand on day 1, while that on day 10 is only 4.0 thousand, as shown in Figure 1. Why is there a buying frenzy on the first day, and why does it quickly cool off ust a couple of days later in the Chinese stock market? These questions need further exploration for at least the following reasons: (Insert Figure 1) First, the Shenzhen stock market is a pure order driven market, with no market maker standing in. All the trades are therefore between investors, as opposed to against dealers, so it is certainly not the case that interdealer trading contributes to the substantial volume after IPO. Second, the Shenzhen stock market has a T+1 settlement schedule, which bales out all the day trading as one source of the first day volume in the U.S market (Geczy et al., 2002). Third, in our sample period, short selling is not allowed in the Shenzhen Stock Exchange 1, so the large trading volume cannot be the consequence of short sellers behavior (Geczy et al., 2002). These features show that the causes of Chinese IPO first day high volume may well be different from those of the U.S market. Also, it provides a clean setting: we are able to answer the question that, with the above institutional settings excluded, what is the reason for the initial volume frenzy? Scrutinizing the Chinese IPO setting, we can easily claim that the high volume of the first days after an IPO must come from flipping, or the other way round, the purchasing behavior of the investors (as opposed to dealers). Unlike previous literature, we prefer to read this story from the buyers side. The reason for this 1 Short selling was introduced to Shenzhen Stock Market in

4 is that purchasing behavior comes from investors own wills, rather than from share allocation (lottery) in the primary market, where luck plays an important role. In order to answer the question of who is buying IPO stocks on the first day, we get access to a remarkable dataset, which contains the complete trading records of all investors following all IPOs in Shenzhen Stock Exchange, China, from January 2007 to June Also, we are able to identify the investors demographic attributes, including their age and the opening date of their accounts in Shenzhen Stock Exchange. Thus we are able to explore the relationship between investors prior experiences and their behavioral differences, and consequently their investment performances. We try to see whether investors with different experience will systematically behave differently after IPO (and whether the difference will lead different investment performance). As a result, we do find that the high initial volume may be driven by the irrational (in the sense that they clearly result in worse investment performance) purchases of the inexperienced investors, which reveals a more in-depth reason to strengthen the investment sentiment explanation of the initial trading volume in the literature More importantly, to elaborate the linkage between investors attributes, their irrational behaviors and their investment performances, we have added to the literature that has been called household finance by Campbell (2006). Recent evidence has shown that some individual investors are more informed or skilled. For example, Coval, Hirshlerfer and Shumway (2005) document significant performance persistence among individuals. Ivkovic, Sialm and Weisbenner (2008) show that individuals with more concentrated portfolios tend to outperform those who are more diversified. Considering these findings, it is natural to ask how skilled or informed investors acquire their advantage. Although most existing literature exclusively links directly investors experience to their investment performances, recently several papers have 4

5 revealed how experiences are transformed into performances, including Nicolasi et al (2009) and Campbell et al (2013) (claiming the performance difference is through better portfolio formation) and Seru, Shumway and Stoffman, (2009) (through the reduction of disposition effect), both of which focus on stock selection. In this paper, we contribute to the literature by revealing one alternative way to stock selection: timing the market, or more specifically, when to purchase an IPO. Literally, even if purchasing a same stock, the decision of when and at what price to buy or sell the stock makes significant difference in terms of the investment performance. In this paper, we utilize a unique setting to measure investors behavior in their timing of purchase, and argue that it is through the timing of buying (along with other elements in the decision bundle, including timing of sales, purchase price, etc.) that more experienced investors gain higher returns. Also, we argue that investors do learn from trading, in the sense that when investors participate in more IPOs, they tend to behave more rationally (waiting for a longer time, i.e., acquiring more secondary market information), and thus obtain higher returns. To summarize, our paper contributes to the literature by three folds: first, we provide evidence to support the effect of investor sentiment on the high initial trading volume after IPO by documenting that the high initial trading volume might be closely related to inexperienced investors (irrational) purchasing behavior: more experienced investors are less likely to buy IPO stocks on the first day while inexperienced investors are more likely to do so. More generally, we construct a novel proxy, the observation time before an investor makes her decision to buy, for the trading information that the investor acquires from patiently observing the secondary market. We indeed find that more experienced investors tend to wait longer, i.e., obtain more trading information. Second, as som eof the research already documents, we find that more experienced investors indeed obtain higher returns, as some of the research does. On top of that, 5

6 we take one step further and try to find out how they manage to do so. Unlike most existing literature, which links directly investors experiences and their return, we use a structural equation model to reveal the contents in the black box of the decision making process of investors, and figure out that more experienced investors realize higher return through acquiring more trading information(along with other variables in the decision bundle, including timing of selling, choosing the buy price, and buy size). The decision bundle explains a large fraction of the cross-sectional return variations. Third, we document that investors do learn to be more patient from participating in the IPO stocks: the more IPOs they participate in, the longer time they are willing to wait before they actually purchase the IPOs, and consequently, the better they perform in terms of their returns. However, the marginal effect of learning varies across investors states of maturity, and across their types, which should be carefully taken care of. For example, the additional participation of IPO trading of an inexperienced investor will lead to a longer waiting time (and higher return) in the next participation than an experienced one. Also, investor type plays a role. Evidences are consistent with the facts that infrequent traders take more seriously about whether or not to participate, and expect a higher return, while frequent traders do not really care one specific IPO: they participate quite a few IPOs non-selectively, and are willing to take a lower return from each IPO. The rest of this paper is arranged as follows: section 2 reviews the literature; Section 3 describes the Chinese IPO mechanism and details of our data; Section 4 studies the cross-sectional difference between investor experience, their choice when purchase an IPO, and their investment performance; Section 5 examines investors learning process; Section 6 discusses alternative explanations, and section 7 concludes. 6

7 2 Literature Review In this paper, we link the IPO frenzy with individual investors, which is a standard assumption in theoretical papers and common findings in empirical papers. Many studies have documented that there are lots of sentiment traders in the market. These sentiment traders are very optimistic about the prospect of the new IPO companies and most of them are individual investors. For example, Lungqvist, Nanda, and Singh (2006) and Ritter and Welch (2002) conecture that the over-enthusiasm of retail investors may drive up an IPO s first-day return. Using trade data from the when-issued market of Germany during 1999 and 2000, Dorn (2003) documents that individual investors consistently pay too much for IPOs in the when-issued market relative to the immediate aftermarket. The initial returns of those IPOs aggressively bought by retail investors on the when-issued market are higher than those benchmarks. Based on an empirical analysis of 62 book-built IPOs, Derrien (2005) finds the demand of individual investors is strongly correlated with market conditions and it has a large impact on the IPO price. Further, the individual demand is positively correlated with initial return and turnover, but it is negatively correlated with the long-term stock price performance. Derrien s conclusions tell us that the noise trader sentiment affects not only the pricing of IPOs, but also its aftermarket behavior. Similar results can be found in other markets. For example, Lee, Shleifer and Thaler (1991) and Lowry (2003) find that the discount rate of closed-end funds are negatively correlated with the IPO volume, which indicates that the investor sentiment is an important factor affecting the IPO volume. After analyzing 410 IPOs on Germany market, Oehler, Rummer and Smith (2004) find that the initial return is caused mainly by sentiment, especially during the internet bubble period. Da et al. (2011) use Google search as a direct proxy for retail investors attention and find 7

8 out that first, Google search frequency has strong incremental predictive power for first-day IPO return; Second, it also predicts long-run underperformance among IPO stocks with high first-day returns, which are consistent with the price pressure in Barber and Odean (2008). Based on the ideas shown above, we take one step further and ask the following question: Among all individual investors, who are more responsible for this initial trading volume frenzy, in terms of their experience? This step links yet another stream of literature about investor experience, trading behavior and learning. Present literature about trading behavior renders mixed conclusions about the trading results against experience. Some researchers document that experience has positive effects during their investments and can improve their performance. For example, List (2003) provides experimental evidence in support of the theory that traders can learn to become rational and eliminate the endowment effect. More recently, Seru, Shumway and Stoffman (2009) study the relationship between individual investors year of experience and disposition effect and find that an extra year of experience decreases the disposition effect of the median investor by about 4 percent. Also, Nicolosi, Peng and Zhu (2009) document that individuals previous forecasting ability positively affects their future trading profitability and activity. Basically, they tend to find that individual investors with richer trading experiences tend to make more correct decisions, in terms of better profits. More recently, Campbell et al (2013) use India stock market data and document that individual investors in Indian equities hold better performing portfolios as they become more experienced in the equity market. However, other researchers point out that experience may result in worse decisions because investors may navely expect that what they have experienced before will happen again in the future and this nave reinforcement learning has been documented in many articles, e.g., Cross (1973), Arthur (1991), Roth and Erev (1995), 8

9 and Camerer and Ho (1999), Choi, Laibson Madrian and Metrick (2007), Kaustia and Knupfer (2008) etc. Among them, Kaustia and Knupfer (2008) use Finland IPO data and find that investors have a tendency to subscribe to IPOs if they earn on past IPO investments. Choi, Laibson, Madrican and Metrick (2007) find that if an investor has experienced a higher 401(k) portfolio return and/or a lower 401(k) return variance, he will increase his 401(k) contribution rate. As Zhu (2010) comments,... whether and how fast individual investors learn about their abilities hence becomes an important topic in the behavior finance literature. 3 The Chinese IPO Mechanism and the Data 3.1 The Chinese IPO mechanism Shenzhen Stock Exchange is one of the two stock exchanges in China. Shenzhen stock Exchange has two layers of market during our sample period January 2007 to June 2008: The main board and the small and medium-size enterprises board (SME board hereinafter). There are no new IPOs for the main board. So we use only the IPOs of SME board. The IPO mechanism in Shenzhen Stock Exchange is a unique two-step auction in the sample period. Of all the IPO shares, 20% is distributed offline, through which only qualified institutions can subscribe. IPO price is set by the underwriter in this offline part, and known by the public. According to this IPO price, the offline shares are distributed to the subscribers as a fraction of their effective number of shares subscribed (i.e., shares whose subscribed price are higher than or equal to IPO price). Shares allocated through offline procedure have a lock-up period of 3 months. Afterwards, the rest 80% of the IPO stocks are put into an online subscription, which is open to both institutions and individuals. Online subscribers only submit 9

10 the quantity, given they already know the IPO price. Similar to offline situation, if oversubscription occurs, the shares will be distributed according to the effective quantities subscribed. 3.2 The Data The data used in this study come from the Shenzhen Stock Exchange, which is one of the two stock exchanges in China. China s clearing and registration system reveals the actual entity that trades the listed stock, as well as the broker firm information. So our data reflect the official record of all trades of the whole market and therefore of extremely high quality. It covers a sample period of January 2007 to June During this period, the Chinese stock market experienced a bull market and a bear market. From Jan 2007 to Mid Oct 2007, there was a strong bull market, and the Shanghai Composite Index, the representative stock price index in China, rocketed from 2675 points to as high as 6092 points on Oct 16th, 2007, as shown in Figure 2. Following that, the market cooled off and the Shanghai Composite Index drops back to 2736 points on June 30th, 2008, which is the end of our sample period. There are 142 IPOs during the period. However, following Chiang, et al (2009), if a stock s close price hits the price limits on the second trading day, we drop this stock from our sample. 38 of 142 stocks are dropped thereof and the final sample number of IPOs in our paper is 104. (Insert Figure 2) Specifically, we use three datasets. The Trading Record Data contain all investors complete trading records of all the 104 sample IPO stocks. The records include: trade size, trade time (in precision of 0.01 second), trade price, the buyer/seller identity of a trade, the buy order and sell order type of a trade (market order or 10

11 limit order), the submission time of the orders (in precision of 0.01 second). However, an order is not included in the dataset if it does not lead to a trade (e.g., expired or cancelled). The Investor Type Data contain the attritions of the above-mentioned investors, including their birthdate, date of opening the account, as well as their account opening locations, in terms of 8 big zones of China; thus we can calculate the age, as well as the tenure, i.e., the number of years that an investor has been in the market. The IPO Data contain the information about our sample IPOs, including offer price, lot winning rate, offer size etc., as well as the information related to the firms can also get, e.g., leverage, etc. Panel A of Table 1 shows the descriptive statistics of the 104 sample stocks. (Insert Table 1) From Panel A of Table 1, we can see that the mean number of shares is million shares, representing million RMB Yuan (equivalent to million U.S. dollars, using the prevailing exchange rate on January 1st, 2007). The average IPO price is Yuan, with initial return of %. Impressively, the online oversubscription is as huge as times the number of shares offered, which means that, of the online fraction which is open to the public, the demand side is 1450 times of the supply. Panel B of Table 1 shows the composition of buyers on day 1 of IPO trading. In China, the open centralized auction period is from 9:25 to 9:30; the continuous auction period is from 9:30-11:30 and 13:00-14:57, while the close centralized auction period is from 14:57 to 15:00. Not surprisingly, the buys in continuous auction period compose more than 82% of all trading volume. However, among all buys on day 1 of IPO, 93.5% of volume is from individuals and that from institutions composes only 6.5%, which motivates us to study the purchasing behaviors of individual investors 11

12 on the first day(s) after IPO to explore who they are, and, if possible, why they rush into the market. 4 IPO Attraction and Investor Experience As a huge volume of literature states, IPOs attract retail investors and form the trading frenzy since retail investors are more likely to be sentiment traders. Looking ahead from this standing point, the attraction of IPOs should impact differently on investors with different attritions. Intuitively, more inexperienced investors are more sentiment driven and behave more irrationally, yet within our knowledge, there is no direct test in existing literature. Along this line, we are going to check whether more sentiment driven investors (inexperienced ones) indeed contribute more to the initial volume. 4.1 Measuring investor experience In this paper, we try to explore whether investors prior experience plays a role in explaining whether investors are rushing into the market. We measure investors experience using the following two measures: 1) Age is the age of the investor as of It is the natural candidate of an investor s experience, which composes of various dimensions of life experience (not only trading experience). 2) Y ears i measures the tenure of investor in the stock market, which is defined as the number of days from investor s account opening date to the date she purchases IPO i for the first time, and then divided by 365. Age and Y ears i are, not surprisingly, positively correlated with correlation coefficient of Intuitively they differ in the sense that Age focus on the investors 12

13 overall spiritual matureness, while Y ears i is a proxy for the trading or observing experience in the stock market. However, as we will later elaborate, using these two measures to proxy experience renders no significant difference in virtually all the empirical tests presented in this paper. We start by drawing the cross-sectional difference in terms of experience of investors entering the market on different days. If an investor makes her first purchase of IPO i on day t (t 10), we then define this investor day-t buyer of IPO i. We can see from Figure 3 that, the average age of day-1 buyers is 40.9, while that of the day-2 buyers is Note that the day-3 buyers and later investors age are between to 42.95, which is quite stable. Consistent results are documented if we scale investors experience using their tenure in the stock market, i.e., Y ear i : The day-1 buyers mean tenure is 5.33 years, while day-2 buyers mean tenure rises to 5.87 years, and it remains around 5.9 years for day-3 or later buyers. The descriptive statistics show that the first day buyers are more likely to be younger investors and those investors with lower tenure in the stock market. (Insert Figure 3 here) A closer look at the experience of investors and their participation day of IPO shows the intraday pattern, as shown in Figure 4. We can see that the intraday pattern of day-t buyers experience shows some similarity. Taking the example of Panel A, Figure 4, we can see that: first, the lower curve represents the day-1 buyers experience, and the two upper curves are the day-2 and day-3 and above buyers. This is what we have observed in Figure 3. Within each curve, we show the intraday pattern. For day-1 buyers, the investors who begin to purchase in the open centralized auction period are older, with averaged age of In contrast to that, the 9:30-10:00 buyers of day 1 (defined as the investors who make their first purchase between 9:30 13

14 and 10:00 am of day 1, and hereinafter) are aged at on average. The age of later comers are stable until for the 14:00-14:30 buyers of day 1, whose mean age is slightly higher to 41.6 and then drops back again to 41.0 in the close auction period. Similar patterns can be found in day-2 and above buyers. In terms of investors tenure as experience, we observe highly consistent results, as shown in Panel B of Figure 4. In sum, the results from Figures 3 and 4 tell us that, first, less experienced investors seem to rush into the market on day 1; second, there is no evidence that within one trading day, the later comers are more experienced than the early comers 2. (Insert Figure 4 here) 4.2 Investor experience and timing of purchase In this subsection, we first formally test whether experienced investors avoid buying in an IPO on the first day. We introduce the following probit model: 1(x ) = α + β 1 Exp + γcontrol + ε eq (1) where 1(x ) is the indicator function of whether investor is a day 1 buyer or not. To obtain 1(x ), first, define 1(x i, ) as the indicator function which takes the value of 1 if investor makers her first purchase of IPO i on the first day, and 0 otherwise. Then, define: 1 1 (x ) = 0 if 1 IP Ocnt IP Ocnt i 1 (x i, ) > 0 otherwise where IP Ocnt is the total number of IPOs investor participates in our sample period, where participate hereinafter refers to the situation that investor makes 2 To our knowledge, there is no theory explaining why the buyers in the open auction period are more experienced than the continuous auction buyers. We conceive that this is due to fact that the some really young and inexperienced investors do not know how to submit an order in the centralized bidding or even do not know they can submit an order during the period. 14

15 her first purchase of the IPO within the first 10 trading days. Literally, 1(x ) equals to 1 as long as investor has ever purchased an IPO on the first day in the sample period, and 0 otherwise. Exp refers to the experience variable, including Age and Y ears. (Y ears is the average of Y ears i across all IPOs investor participates in the sample period). Control variables include: (1) IPO price. (2) Offer size, which is the total number of shares issued in this IPO. (3) IPO P/E ratio: measured as the offer price over earnings per share. To a certain degree, P/E ratio demonstrates the reasonableness of the offer price, which may affect the demand of the IPO stock. (4) over-subscription multiplier: the reciprocal of the lot winning rate, which also proxy the demand of the IPO. (5) growth rate of main business: calculated as the ratio of previous two years core business income. (6) leverage: the asset-liability ratio which is calculated as the ratio of book value of total liabilities and book value of total assets. (7) market return, previous 10 trading days market return. (8) number of IPOs in the previous month. Furthermore, we also include the industry fixed effect, which is based on the CITIC Industry classification, and geographic fixed effect, which is based on the 8 geographic zones each account belongs to. The results are shown in Table 2. (Insert Table 2 here) Table 2 shows that more experience investors are less likely to rush into the IPO market on the first day. Taking the example of column (a), we can see that one more year of age will tend to reduce the probability of buying the IPOs on the first day by 0.3%, which is significant at 1% level, and consistent results are found for using the tenure as experience (see columns (b) and (c)). As we have speculated earlier, this result is an extension to exising sentiment literature: the purchasing behavior of the more sentiment-driven investors (less experienced ones) are responsible for the high 15

16 day-1 trading volume. On the basis of the results from Table 2, we take one step further by introducing a variable elapse i, which is the number of minutes elapsed between 9:25:00 of trading Day 1 of stock i and investor s first purchase time of stock i. This variable captures the different waiting time before an investor makes her decision to buy a specific IPO, having observed secondary market performance (and only the secondary market performance). By this statement we implicitly make two assumptions: first, all investors start to watch an IPO from 9:25:00 on day 1 of IPO; second, there is no inside information leakage between IPO trading starting point and the investor s first purchase 3. These assumptions are important: suppose that we observe two individual investors and k purchase 1000 shares of a certain stock at 10:00am and 11:00am on any day t respectively (not necessarily immediately following an IPO), we can definitely not claim that investor k is more patient and obtains more information, since (1) we do not know when they start to watch the market; and (2) we do not know whether one gets access to more inside information than the other, which makes their motivation of purchase much complicated. However, in our setting, the inside information driven motivation of buying a stock is excluded by assumption 2. Thus, the reason why investors make their first purchase at different time is not due to exogenous information heterogeneity. The purchase decision is exclusively driven by the investors observing the secondary market (to absorb the endogenous information). In short, 3 The validity of the two assumptions can be defended as follows: For assumption 1, we argue that it tends to misclassify some earlier comers as later ones, if assumption 1 does not hold. For example, if an investor buys and IPO at 11:00am of day 2, yet she actually starts to observe the IPO only from 10:00am of day 2, she is a day-1 buyer and waits only 1 hour, yet we count her as a day-2 buyer, with elapse i = 25 hours and 35 minutes. By doing so, the shown results have actually underestimated the difference between day 1 and day 2 investors, since she would have earned lower return (as shown later), had she started trading at 9:35 of day 1. So the existence of the possibility that assumption 1 does not hold strengthens, rather than weakens our conclusion. For assumption 2, we argue that it is (approximately) true that there are no inside information leakage into the market since we study only 10 trading days after IPO. For robustness sake, we further reduce our sample to up to 2 trading days after IPO, and virtually repeat every exercise in this study, and the results are highly consistent. 16

17 the heterogeneity in elapse i measures investor patience through proxying the length of time an investor spends in observing the secondary market after IPOs, and thus captures any lumpsum information difference from secondary market. We now try to check whether the waiting time correlates with the investors experience. Specifically, we use the following regression: elapse = α + β 1 Exp + γcontrol + ε eq (2) where the variables are defined as follows: first, define rawelapse as the average minutes elapsed between 9:25:00 of trading Day 1 of stock i and investor s first purchase time of stock i, across all i s. Then, elapse is the logarithm of 1 plus rawelapse. Other variables are defined as earlier. The results are shown in Table 3. (Insert Table 3) Panel A of Table 3 shows the descriptive statistics for rawelapse, and Panel B of Table 3 shows the regression results, which are consistent with that from Table 2: we can see that more experienced investors tend to wait for longer time before she makes her mind to buy an IPO. More specifically, one more year of age will lead to 1.8% longer waiting time, holding other variables at their means. The results are signficant and robust if Y ears is used to proxy experience. In columns (c) and (d) of Panel B, Table 3, we also include in the regression IP Ocnt, (the total number of IPOs investor has participated in our sample period, as defined earlier), as well as the square term of IP Ocnt. Presently, a simple look at this variable tells us that the more IPOs an investor participates in the sample period, the longer time they are willing to wait before purchase, but the rate of increase in waiting time is decrease, implying a concave relationship between waiting time and number of participated IPOs We are going to discuss in more details about IP Ocnt in Section 5. 17

18 Combining the results from Tables 2 and 3, we can conclude that experienced investors are indeed more patient than the inexperienced ones. However, why does patience make a difference? What is wrong with simply rushing into the market? Why do we consider the impatient guys irrational? All these questions need to be tackled with the consideration of reurn, i.e., investment performance, as we will later formally test. But before that, we might have to look at other elements an investor needs to determine when she actually wants to buy a stock. In addition to the timing of purchase, as we have discussed ust now, the timing of selling, the purchase price, as well as the number of shares they are willing to buy, among others, need to be determined simultaneously. We therefore define these variables together the decision bundle an investor has to decide simultaneously when she buys an IPO. Next, we will continue with the other elements in the decision bundle, including timing of selling, the purchase price and the number of shares she chooses. 4.3 Investors experience and other elements in the decision bundle Turnover Among the choices in the decision bundle, we first explore the selling behavior of investors by defining a variable turnover. Conceptually, turnover is the average fraction of shares sold within the 20-trading day window after a specific investor makes her first purchase, to the total number of shares she has purchased in the window. Formally, it is defined as follows. First, define turnover i, as turnover i = Mi m=1 q im Ki k=1 q ik where K i (M i ) is the total number of buy (sell) trades investor has made during 18

19 the 20-trading day window since investor s first purchase of stock i s IPO. Then, turnover is defined as: turnover = 1 IP Ocnt IP Ocnt i=1 turnover i We show the descriptive statistic of turnover and the relationship between turnover and experience in Table 4. (insert Table 4) As shown in Panel A of Table 4, the mean turnover is, surprisingly, as high as 77.21%, which means that over only 20 trading days, investors sell on average 77% of all shares they have purchased during the period. Another number to notice is the percentile. The 25th percentile is 62.50% and the median is 100%, which means that over 50% of investors sell all the shares they have purchased. This number is extremely high. As a comparison, Chemmanur, et al (2010) study the institutional selling of allocations from primary market beyond the immediate post-ipo period and find that institutions sell 70.2% of their IPO allocations in the first year. In our case, the investors who purchase from the secondary market (rather than from primary market allocation) sell 77% in 20 trading days (rather than 1 year), which is even more impressive. A more direct comparison shows that the investors who get share allocation sell 69.5% on the first day, as stated earlier, indicating a more acute trading frenzy in Chinese stock market, compared with the U.S. market. Panel B of Table 4 shows the results of the following model turnover = α + β 1 Exp + γcontrol + ε eq (3) It is shown that investors with higher experience level tend to sell less than the inexperienced ones within the 20-trading day period. For example, from column (a), 19

20 we can see that one additional year in age will lead to 0.27 percent lower in terms of turnover. This pattern is consistent, regardless of whether the Age or Y ears are used to proxy the experience. So in summary, more experienced investors tend to sell a smaller fraction than the inexperienced investors. Moreover, the coefficients of IP Ocnt show that investors participating more IPOs tend to have higher turnover rate. This is somehow hard to explain if we consider IP Ocnt as another dimension of trading experience. We will keep this as a puzzle presently and discuss this topic in section 5 in a more detailed manner Purchase price Another important dimension investors want to seriously consider is the price they are willing to pay. We explore the relationship between investors experience and purchase price in this subsection. We define P as follows: First, define P i as investor s first purchase price of IPO i over IPO i s open price on the first trading day after IPO. Then, P is defined as: P = 1 IP Ocnt IP Ocnt i=1 P i Literally P is the standized initial price investor pays on average in the sample period. The results are shown in Table 5. Panel A of Table 5 shows the descriptive statistics of P, while Panel B of Table 5 uses the following model: P = α + β 1 Exp + γcontrol + ε eq (4) (Insert Table 5) It is shown in Panel B of Table 5 that cross-sectionally, more experienced investors tend to (start to) buy an IPO stock with lower price. Judging from column (a) and 20

21 (b) of Panel B, Table 5, one additional year in age will lead to 3.3 basis points lower in terms of standardized purchasing price, and one additional year in tenure of the stock market will lead to 11 basis points lower in standardized purchasing price. The evidence is consistent with the facts that more experienced investors can time the market to buy at lower price than the in experienced investors. However, the estimated coefficient of IP Ocnt is positive and significant, meaning that investors trading more IPOs tend to pay higher price. Like the turnover in previous section, this seems to contradict the fact that investors with more IPO traded should be more sophisticated and pay lower price. Again, we will discuss this in later part Size of purchase Yet another dimension in the decision bundle of an investor s purchase is the size. i.e., how many shares (or RMB Yuan value) she would like to buy. Admittedly, the buy size is bounded by the investors budget constraint, which is further highly correlated with their experience. However, for completeness sake, we still include the size of purchase into consideration. We denote investor s size of purchase as value, which is defined as follows: First, we define value i as the sum of all RMB Yuan value investor has purchased in the 20-trading days after her first purchase of IPO i. value i = K i k=1 q ik where K i is the total number of buy trades investor has made during the 20- trading day window since investor s first purchase of stock i s IPO, and q ik means the RMB Yuan value of the kth purchase investor makes on IPO i in the 20-day window. Then, value is defined as 21

22 value = 1 IP Ocnt IP Ocnt i=1 value i The relationship between size of purchase and experience is then shown in the following regression: log value = α + β 1 Exp + γcontrol + ε eq (5) where log value is the logrithm of value, and the results are shown in Table 7. Not surprisingly, we can see that one additional year of age leads to 1.7% more in terms of RMB Yuan value, keeping other variables at their means. Also, the investors who buy more IPOs in the sample period tend to be larger buyers, if we read the estimated coefficient of IP Ocnt. (Insert Table 7) So far, we have shown that higher level of investor experience will lead to longer waiting time to observe the secondary market, lower turnover, lower purchasing price, and larger purchase quantity. Naturally the ensuing question would be: why do investors systematically differ in their above trading behavior? Will these differences lead to higher or lower return? We will discuss these in the following subsection Investment performance The exploration of trading experience and investment performance has been mounting, but the conclusion is still mixed. List (2003), Dhar and Zhu (2006), and Nicolasi, Peng and Zhu (2009) find evidence supporting that investors with richer trading experiences tend to make more correct decisions, eliminate udgment errors, and result in better investment performance. On the other hand, Cross (1973), Arthur (1991), Roth and Erev (1995), and Camerer and Ho (1999), Choi, Laibson Madrian 22

23 and Metrick (2007), Kaustia and Knupfer (2008), among others, raise evidence that experience may result in worse decisions by increasing the degree of overconfidence. There is room for future investigation in this stream. One caveate in the literature is to use directly investors return from one or more stocks to represent their performance. It is not ideal ex-ante. The reason comes from the motivation of portfolio optimization, which is extremely hard to identify. An investor may hold a portfolio that contains stocks, bonds, commodities, real estates, among others, yet the return from a single stock, or even from all stocks held from the stock market may only represent a fraction of the portfolio. So we cannot rule out the possibility that the observed return(s) provides a certain type of risk to the complete portfolio of the investor, and thus cannot udge whether her overall performance is good or not. However, our setting rules out the portfolio optimization motivation from an ex-post perspective: since the investors sell 80% of the IPOs within only 20 trading days, it is not possible that they are constructing an optimized portfolio. Our setting thus provides again a clean setting that we can use the return from a single stock to proxy the investor s performance The naive model We start our investigation of experience and return with the naive model, i.e., linking experience directly to investment performance, without exploring how the performance is realized. We introduce the abnormal return to measure investor performance. First, we define the abnormal return investor obtains from IPO i as: rtn i = Ki where L l=1 q0 il k=1 q ikp ik ( L i l=1 q0 il K i Li l=1 q0 il p0 ik k=1 q ik)p 20 i 1 Mktrtn i is the total shares of stocks investor buys during the 20-trading day window after her first purchase of stock i; p 0 il is the actual purchase price for 23

24 investor to buy stock i for the lth time. q ik is the number of stock i investor sells for the kth time during the 20-trading day window; p ik is the actual trading price of investor when she sells q ik of stock i. p 20 i is the closing price of stock i on the 20th trading day after the investor s first purchase. K i is the number of times investor sells stock i during the 20-day window after her first purchase of the stock. L i is the number of times investor buys stock i during the 20-day window. Mkttrtn i is the Shanghai Composite Index return of the 20-day window. as: We then define the average return investor obtains throughout the sample period rtn = 1 IP Ocnt IP Ocnt i=1 rtn i After defining the average return, we use the following model to study the relationship between experience and return: rtn = α + β 1 Exp + γcontrol + ε eq (6) The results are shown in Table 8. (Insert Table 8) The results in Table 8 show that one additional year of age leads to 3.27 basis points higher abnormal return in the 20-trading days, which corresponds to an annualized return of 0.43%. Consistent results are found if tenure is used to measure investors experience, and all results are statistically significant at 1% level. Present results are consistent with the theory that investors may learn from prior experience, providing supporting evidence to List (2003), Dhar and Zhu (2006), and Nicolasi, Peng and Zhu (2009), among others. 24

25 In section 3.2, we show that more experienced investors tend to wait for longer time before they make the purchase decision. Based on the results, we will simultaneously study the impact of experience and waiting time on investment performance. We first sort within each stock by the experience (Age or Y ears) and categorize the investors into 10 groups, then categorize day-1 to day-10 buyers within each experience group by their first day of purchasing IPO. We then have groups and calculate their corresponding abnormal return. The results are shown in Figure 6. (Insert Figure 6 and Table 9) Figure 6 and Table 9 convey the following information: first, fixing an experience group (Age or Y ears), to wait for more days will translate to significantly higher abnormal return. Second, for day-1 buyers, Age group 10 investors outperformance the age group 1 investors by 1.269% in the 20-trading days (significant at 1% level), while for day-10 buyers, the difference is only 0.124%, and is not significant at acceptable level. The difference in difference is as high as 1.146%, and is significant at 1% level. The results are consistent if we use value-weighted average return or if Y ears i is used to proxy experience. We further use the following regression (results shown in Table 10) to control for other potential impacts: rtn i = α+β 1 ExpDum i +β 2 DayDum i +β 3 ExpDum i DayDum i +γcontrol i + ε i eq (7) (Insert Table 10) The results imply that, if an inexperienced investor is willing to wait for, say, 10 days longer, they will be rewarded with better absolute return (compared with the 25

26 case of purchasing on day 1), and relative return (compared with the more experienced investors). Again, in this model, we first sort within each stock by the experience (Age or Y ears) and categorize the investors into 10 groups, then categorize within each experience group by their first day of purchasing IPO. We then keep only the lowest and the highest experience group, and day 1 and day 10 buyers groups. ExpDum i represents the experience dummy (which includes AgeDum and Y eardum i ) which takes the value of 1 if the investor is from the highest Age (Y ears) group, and 0 otherwise. DayDum i takes the value of 1 if investor is a Day 10 buyer and 0 if the investor is a Day 1 buyer on IPO i. Firm control variables include the total Yuan value purchased within the 20-trading-day window after the first purchase of investor on IPO i, IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10- day market return, and number of IPOs in the past month of an IPO. Industry fixed effect is based on the CITIC Industry classification. Geographic fixed effect is based on the 8 geographic zones each account belongs to. We can see from the results of Table 10 that the interaction term of the ExpDum and DayDum are significantly negative in all four cases, implying that the return gap of experienced and inexperienced investors are significantly lower for day-10 buyers, after controlling for various firm-specific or industry level factors. The results are highly consistent with those from Figure 6 and Table 9, and the story is: for an inexperienced day-1 buyer, if she is willing to be more patient and to wait for 10 days, her investment performance will increase in both the absolute sense (versus day 10 group return) and comparative sense (versus more experienced group). It is thus intriguing to ask: since everyone in the market can easily see this pattern, why are there yet so many (inexperienced) investors rushing into the IPO secondary market on 26

27 the very first days, and resulting in a statistically significant and economically huge negative return? Why aren t they willing to wait simply for, say, 10 days and enoy an annualized 13% higher return? We argue that this is the evidence showing that the the rusher-ins behaviors are indeed irrational, and there is no way to validate their behaviors by a reasonable motivation. We will try to rule out some other potential explanations in Section 6 (Robustness checks). But before that, we are going to further scrutinize the investors experience and their investment return, and try to answer this question: Through what channels can the experienced investors realized higher return? In other words, we are looking into the decision bundle of investors and exploring the function of the decision bundle in the process Structural equation model Within our knowledge, most previous research links experience and investment performance directly, as what we have done in Section Even Seru, Shumway and Stoffman (2009) are silent on the direct impact of reducing dispostion effect on their return (since their focus is differentiating learning by trading from learning the ability ). In this Section, we are going to take one step further to open the decision making black box of investors purchasing behavior. Specifically, we ask the question: what is the function of the decision bundle? Therefore, we introduce a structural equation model to explore the inter-relationship among experience, decision bundle and investment performance, as shown in Figure 7.1. Literally, the experience-return relationship we observe in Section may be through the decision bundle. (Insert Figure 7) In Figure 7.1, the upper arrow marked with χ 1 and the lower arrow marked with χ 2 are what most existing literature tries to document. In additional to that, we 27

28 include the timing of purchase, along with other elements in the decision bundle into the model. Investors with different experience will make different decisions (timing of buy, timing of sale, price and quantity), which are documented earlier in this paper, and are represented by coefficients γ 11 to γ 14, respectively. The investors final goal by making these decisions is for the return, and the relationship is represented by θ 1 to θ 4, respectively. So γ 11 to γ 14 as well as θ 1 to θ 4 represent the black box of decision making process, which has not been thoroughly explored in the related literature. Also, we control for other firm-specific or industry and market level factors, which are recorded by the dotted arrows with coefficients γ 21 to γ 24. Overall, the equation system contains: elapse = γ 11 Exp + γ 21 Control + ε 11 eq (8) turnover = γ 12 Exp + γ 22 Control + ε 12 eq (9) P = γ 13 Exp + γ 23 Control + ε 13 eq(10) value = γ 14 Exp + γ 24 Control + ε 14 eq(11) rtn = θ 1 elapse + θ 2 turnover + θ 3 P + θ 4 value + χ 1 Exp + χ 2 Control + ν eq(12) We now estimate the structural equation model, with everything put together. The results are shown in Table 11 and Figure 7.2. (Insert Table 11) The results in Table 11 and Figure 7.2 show that, first, consistent with previous results, higher experience lead to (1) longer waiting time before buying; (2) lower turnover rate; (3) lower purchasing price and (4) larger quantity. Second, we further document that longer waiting time, lower turnover rate, lower purchasing price will consequently lead to higher rate of return, consistent that more experienced investors indeed behave more rationally, which in turn lead to better investment performance. One variable that shows some ambiguity is value, which shows that although more 28

29 experienced investors tend to spend more, the large purchase value will result in worse returns. We are going to discuss it more in next section. More importantly, we can see from column (a), Panel E of Table 11 that, although the coefficient of Age (see χ 1 ) is still significantly positive, the size of the coefficient is now 0.95 basis points, which is a lot smaller than that from the simple regression in Table 8 (3.27 basis points). Furthermore, from column (b), Panel E of Table 11, the coefficient of Y ears is now insignificant. These results show that, the inclusion of the decision bundle in the structural equation model explains a large fraction of the return variation.in plain English, the above findings are: 1) more experienced investors obtain better investment results; 2) they obtain the higher results exclusively through being more patient and/or through obtaining more secondary market information before purchasing, as well as other elements in the decision bundle. 4.5 Discussions Up to now we have documented the following cross-sectional evidence: first, the relationship between investor experience and the investment behavior (decision bundle), including timing of buying, timing of selling, purchase price and buy size. We can conclude that more experienced investors tend to: (1) wait for longer time before buying; (2) have a lower turnover rate (sell a smaller fraction); (3) obtain lower purchasing price and (4) purchase more shares, compared with the inexperienced investors. Everything supports the speculation that more experienced investors behave more patiently (wait for longer time in buying and selling), more sophisticatedly (they are able to obtain a lower purchasing price), and with more strength (they buy more shares). In short, evidences show that experienced investors exhibit more patience (wait for longer time before buying or selling) and more sophistication (buy at lower price). 29

30 Second, we have documented investors experience and return are postively related. Consistent with List (2003), among others, we document that more experienced investors obtain higher returns. Moreover, we document that it is through the decision bundle that the investors obtain the return. The inclusion of the decision bundle explains a large fraction of return variation. The patience and sophistication during decision making process lead to better investment performance, confirming that experienced investors are more rational. However, if we consider the number of IPOs investors participate in the sample period as a proxy of their learning, we may face some ambiguous results: investors who learned more (through more IPO participations) wait for longer time (Table 3), and buy more shares (Table 7), which makes sense. However, evidence seems to indicate that investors who learn more have higher turnover rate (Table 4) and higher purchasing price (Table 5). Furthermore, the structural equation model results in Table 11 also show that investors participating more IPOs, as well as those purchasing with larger value, result in lower return, which contradicts the intuition that investors who learn more get more rational. We argue that the mixed results from IPO participation of investors come from the complexity of investors learning behavior, their stock of experience, as well as their intrinsic type (i.e., whether they are frequent/larger traders or not), which cannot be clearly shown in the cross-sectional settings as in this section. In the following section, we are going to discuss the learning process in details. 5 Learning by Trading? Literature about investor learning behavior has been mounting. Early classical learning literature focus on the learning-by-doing models (Arrrow, 1962, Grossman, Kihlstrom, 30

31 and Mirman, 1977), in which investors might improve their ability as they trade. A second type of learning is called learning about ability : as investors trade, they might realize that their inherent level of ability is low and decide to stop trading (Mahani and Bernhardt, 2007). With the appearance of high quality proprietary datasets, researchers are gradually able to directly test different types of models. For example, Seru, Shumway and Stoffman (2009) find that, if learning about ability is overlooked, the effect of learning is substantially over-estimated. In Seru, Shumway and Stoffman (2009), the tenure in the market, as well as the cumulative number of trades are used as experience measures to proxy learning. That is, experience and learning, as well as tenure and cumulative number of trades, are used de facto interchangeably in their paper. But if we scrutinize the cumulative trade and tenure, we can see that, other than the difference between actual trading v.s. hypothetical trading, there are other slight differences between them: first, cumulative trades are the results of subective decisions of an investor. She may voluntariely decide to trade, or rather not to trade at all. When she makes the decision to trade, she pays due effort to take the fundamental and/or technical analyses, and enoys (or suffers) the trading results. Contrary to the in-person participation, tenure, on the other hand, does not depend on investors own will. It simply documents the time elapse since the account opening date. A second difference is: in addition to the trading/observing behavior of stock market, tenure may also include life experience, social interactions, etc., while cumulative trades are limited to the trading experience from stock market. Third, tenure can be compared on a cross-sectional basis to proxy the maturity of different investors. For example, an investor with 30 years of tenure in the market is for sure more mature than an investor with only 1 year tenure in terms of their stage of life-cycle of experience: the former may have more trading experience, more life wisdom, and is possibly older in terms of age. However, an 31

32 investor with 3000 trades this year may not be more mature than the one with 100 trade, in terms of their life-time experience, or state of spiritual maturity. For example, if the first investor makes 30 million trades in her life time of trading, yet 3000 trades are only such a small fraction of all her position that she may not really take seriously. However, for the latter person, she may only have 1000 trades in her life time trading, and the documented 100 trades compose one-tenth of her life long trading experience (and she may be really careful about the investment since this means a big fortune for her and thus may learn more from the 100 trades). In short, investor type heteogeneity makes it convoluted to direct explain the cross sectional difference of cumulative trades across different investors. To quickly summarize, culmulative trades represent the active part of how investors learn and are not measures of state of maturity of investors, while tenure captures the passive part of experience and is able to scale the state of maturity. Admittedly, as sample period is getting longer, tenure and culmulative trades are converging, in terms of what they are measuring. However, unlike Seru, Shumway and Stoffman (2009), which use a sample period of 8 years (not extremely long but still may contain enough variation about tenure and cumulative trades), we have only one and a half years sample for this paper, which means that even if we do not collapse the time-series of individual investors trading to the cross-sectional mean, we are still not able to capture the time-varying elements measured by Age or Y ears. These facts motivate us to separate learning from experience : we use the former to capture the active leaning procedure and use the latter to measure the state of maturity. This setting will allow us to fix the level of experience and study the effect of learning over time, as we will discuss in the following section. 32

33 5.1 Important proxies and hypotheses We have so far studied the cross-sectional relationship between investors experience, their behavior difference (waiting time, turnover, etc), and investment performance. However, since we are able to track one specific investor s behavior when she participates different IPOs sequentially, we are able to capture how the investor s behavior evolves time-seriesly as she continues to trade. We therefore call this sequential participation the active learning process of an investor. Slightly different from the experience variable, which depicts the level of experience stock one investor has at some time, and acts as state variable, active learning variables act like control variables. Experience variables measure the stage of maturity in an investor s life cycle, in terms of their age or their tenure in stock market, while active learning variables measure the voluntary participation (or quitting) investors actively choose. It captures the learning process that whether investors actively get involved in trading, regardless of the investors stage of experience (young or old). To proxy active learning, we introduce the variable: IP Oseq i, which is defined as follows: As said earlier, IP Ocnt is the total number of IPOs investor participates altogether in the sample period. Based on that, define the ith IPO investor participates as IP Oseq i (1 i IP Ocnt ). In other words, IP Oseq i is the sequence of IPOs investor participates in our sample period. As of the experience variables, we continue to use Age and Y ears i, which captures the stock of knowledge and the state of maturity in the investor life-cycle. That being said, the relationship between experience and (active) learning can be expressed as follows: Lower Age/Y ears mean that the investors are at early stage in their life cycle and lower state of maturity, thus the effect of learning is most prominent. However, when an investor is getting older/more tenured in the market, she enters a more mature state of her life cycle, in which her level of ability 33

34 is high, yet the marginal effect of learning is low. It is like the fact that a baby knows nothing but has enormous growing and learning potential, while adults are a lot more knowledgeable, yet they do not learn new stuff as fast as babies do. In previous section, we have proved that more experienced (higher Age/Y ears) investors behave more rationally, and are rewarded with better performance, which is the level effect. In this section, we will first focus on the active learning process by testing the following hypothesis: H1: Investors learn to be more rational, yet the marginal effect of learning decreases with Age/Y ears. Another dimension that convolutes the understanding of investors learning behavior is through the variable IP Ocnt, i.e., the number of IPOs investor participates in the sample period. The question is: what can we learn from this variable? (Insert Table 12) On the one hand, IP Ocnt measures the overall learning effect from participating IPOs. The larger the IP Ocnt is, the more investor learns from actively participating IPOs. This is an ex post measure. On the other hand, IP Ocnt measures another dimension of investor attrition: the type of investor, i.e., whether investors trade frequently or not. This is from an ex ante perspective. From Table 12 we can see that, the maximum number of IP Ocnt is as high as 89. Considering the total number of sample IPO is only 104, that investor participates almost every IPO in the 18 months. Let s consider the situation that we are standing at the very beginning of the sample period, and ask the investor with 89 IP Ocnt : how many IPOs are you going to participate in the following 18 months? She may probably answer: Well, quite a few! If we ask instead an investor who ends up with only a few IPO participations the same question, her answer might be: Maybe a couple of IPOs only. That is, in 34

35 expectation an investor knows what type of investor she herself is, and has a rough idea of how many IPOs she is going to choose even before the sample period. This ex ante variation is about investor type (her intrinsic trading preference or wealth level), and not about learning at all. By scrutinizing these two types of investors, we may infer that there are some maor differences between these types. For the infrequent traders (who are only enrolled in only a couple IPOs), they are putting all the eggs in one basket, and thus they may want to be really cautious. Also, they may expect a higher rate of return since this is what they select from a big pool, and the reason they choose this specific one or two stocks (and not others) is the udgment that the target stocks are better than the rest of the pool. On the other extreme of the spectrum are the ones with many IPO participation records, which means that they are not really selective in choose specific stocks: they literally participate a lot. So it is reasonable to assume that this type of frequent traders do not expect that each stock will render a high rate of return, and may be less cautious when trading the stocks. Also, frequent traders marginal effect of learning is lower than infrequent traders. We thus provide the following testable hypotheses: H2: The marginal effect of learning is higher for infrequent traders. Summarizing the content so far in this section, we also have a third hypothesis that is about the level effect of patience, investor type and experience. H3: Experienced investors are more patient; Frequent traders are less cautious in trading. 5.2 Patience: Type vs Experience We start with the level effect of hypothesis 3. The level effects are shown in Figure 9. (Insert Figure 9) 35

36 From previous Table 3, we see ambiguous results: it seems that, although more experienced investors are more patient (wait longer before their first purchase), more frequent traders exhibit more patience, according to Table 3. This observations, together with other wierd observations on the estimated coefficients of IP Ocnt from Tables 4, 5, 6 and 11, may come from the dual meaning of IP Ocnt : As we argued earlier, IP Ocnt measures both the learning effect through participating IPOs (which can also be captured by IP Oseq i ), and investor type. In order to measure investor type, we need to control the learning effect by using IP Oseq i simultaneously. Also, we follow Seru, Shumway and Stoffman (2009) and use the Heckman model in order to address the selection problem. We start by defining: s i = I(α + βexp i + δx i + ν i ) > 0 eq(12) where I() is an indicator variable. Here, s i equals zero if investor exists in the sample after IPO i and one otherwise. That is, s i equals zero only for the last IPO of investor s trading. This is the first stage. The second stage of the model direct examines the effect of learning, after controlling for ceasing to trade: Elapse i+1, = α + β 1 IP Ocnt + β 2 Exp i + β 3 IP Ocnt Exp i +θ IP Oseq i + γ Control i+1, + δ IMR + ε i+1, eq(13) In the equation, subscript i + 1, refers to the IPO investor participates after the ith IPO. For example, Elapse i+1, refers to investor s waiting time in the IPO next to the ith one. IMR is the inverse Mills ratio of the selection model (the Probit model). All other variables are defined earlier. As instruments, we use the proportion of frequent traders in the zone of the individual (Area rate), which is based on the proportion of accounts in an individual s area that participate at least 10 IPOs in the sample period. This instrument satisfies the exclusion restriction, since having active traders in ones area is unlikely to directly affect changes in performance, this 36

37 variable is likely to satisfy the necessary exclusion restrictions: it is likely to affect the probability of remaining in the sample, but is unlikely to affect changes in an individuals timing or performance except through their effect on survival. The results of the selection model are shown in Table 13. (Insert Table 13) Column 1 of Table 13 shows that the estimate on return is positive and significant, which says that as low-ability investors trade, they learn about their inherent ability and cease trading when they get negative return, and more successful investors remain in the market and participate trading IPO. Economically, the estimate means that, keeping other explanatory variables at their mean levels, a decrease in returns of one standard deviation increases the probability that the investor will cease to trade in the next IPO by around 5 percent. The other coefficient estimates reported in the first column of Table 13 also seem plausible: investors are more like to remain in the sample and trade IPOs if they have more experienced; Having higher trading activity in the zone of an investor increase the probability that the investor will continue trading in the next period. The results in columns 2-7 of Table 13 support our story about patience, experience and investor type. We observe from the second stage regression (columns 2-4) that, first, after controlling the learning effect (IP Oseq i ), all the estimated coeffocients of IP Ocnt are significantly negative (unlike previous results in Table 3); Second, the coefficients of experience variables are all significantly postive; Third, the interaction terms of IP Ocnt and experience variables are significantly negative. These results imply that: patience (waiting time) increases with experience, and decreases with the frequency of traders (i.e., frequent traders are less cautious), which is exactly what is depicted in Figure 9. 37

38 5.3 Learning effect: model without selection We now start with H1 and introduce the following regression to test the relationship between learning (IP Oseq i ) and experience (Age /Y ears i ): Elapse i+1, = α + β 1 IP Oseq i + β 2 Exp i + β 3 IP Oseq i Exp i +θ IP Ocnt + γ Control i+1, + ε i+1, eq(14) All the variables from this equation are defined as earlier. The results are shown in Table 14. (Insert Table 14) Table 14 shows active learning s marginal effect on investors patience. In all three columns (a) to (c), we can see that the coefficients of IP Oseq are significantly positive. Taking the example of column (a), one additional participation of IPO will make an investor more patient and wait for 1.9 Another dimension of learning effect comes from the rate of return. To test this, we introduce the following equation: rtn i+1, = α + β 1 IP Oseq i + β 2 Exp i + β 3 IP Oseq i Exp i +θ IP Ocnt + γ Control i+1, + ε i+1, eq(15) All the variables from this equation are defined as earlier. The results are shown in Table 15. (Insert Table 15) The first take home message out of Table 15 is that participating one more IPO will significantly increase the rate of return in the next IPO, which is true for all three settings. Furthermore, we can also see that, this learning effect is more prominent for inexperienced investors, in the sense that all three interaction terms of IP Oseq i 38

39 and Experience variables (Age/Y ears) are significantly negative. These evidences are consistent with our hypotheses Learning effect: model selection In order to control the situation that endogenous attrition may affect learning estimates through the decision of whether investor will continue trading, we again use a Heckman selection model. s i = I(α + βexp i + δx i + ν i ) > 0 eq(12) where I() is an indicator variable. Here, s i equals zero if investor exists in the sample after IPO i and one otherwise. That is, s i equals zero only for the last IPO of investor s trading. This is the first stage. The second stage of the model direct examines the effect of learning, after controlling for ceasing to trade: Elapse i+1, = α + β 1 IP Oseq i + β 2 Exp i + β 3 IP Ocnt Exp i +θ IP Oseq i + γ Control i+1, + δ IMR + ε i+1, eq(16) All the variables are defined as earlier. Also, we use the proportion of frequent traders in the zone of the individual (Area rate) as instrument. The results of the selection model are shown in Table 16. (Insert Table 16) The results of the second stage regression are shown in columns 2-7. First, the inverse Mills ratio in all the 6 settings are significantly positive, suggesting that the factors that predict with investors stay in the sample are positively correlated with waiting time and future performance. The choice of ceasing to trade excludes two potential future possibilities: On the one hand, the choice of stop trading will exclude the probability of getting more rational (waiting for longer time) and obtaining higher return. On the other hand, it excludes the possibility of being a frequent trader (who 39

40 waits for less time and accepts a lower return). The two effects cancel each other, and thus lead to ambiguous oint direction. So in this paper, we only control for the selection process, without formally testing the direction of the selection. Second, the results from Table 14 directly provide supports to our H2. Specifically, all the estimated coefficients of IPOseq in columns 2-7 are significantly positive. For example, according to columns 2 and 5, as an investor participates in one additional IPO, she tends to wait for 2.7% longer before her actual purchase, and it will increases average 20-day post-purchase returns by 44 bps, or approximately 5.7% at an annualized rate, keeping other variables at their mean values. Moreover, the estimates of coefficients on learning*experience are all significantly negative in all six cases, implying that the marginal effect of learning is decreasing with experience (Age/Y ears). Overall, H1 is supported. Next, we further test H2, which is about investor type and their behavior. We begin with Figure 8, which draws the relationship between investor type (frequent vs non-frequent traders), patience and rate of return. We first classify investors according to their total number of IPOs in our sample into 15 categories, namely number =1, 2, 3,.., 10, (10, 20], (20, 30], (30, 40], (40, 50], >50, and then calculate the mean waiting time, as well as 20-day post-purchase return against the 15 categories. From Figure 8, we can see from a univariate perspective that: 1) waiting time monotonically decreases with trading frequency, implying that investors are getting more and more impatient as they trade more IPOs; 2) the return is distributed as an inverse-u shape: when the total number of IPOs starts to increase from, say 2 to 3, the 20-day postpurchase return increases, but it begins to decrease after the total number is 10 or larger. It seems that, when an investor starts to trade more IPOs, she becomes more experienced and getting better return, while when she becomes a frequent trader, she actually participates many IPOs and requires from each IPO a lower rate of 40

41 return. These evidences do not see to be consistent with each other, probably due to the multicolinearity between IPOcnt (type) and IPOseq (learning). So we need to formally test the hypothesis. We use the above-mentioned Heckman model again, but for this time, we categorize the investors by their total number of IPOs into four groups, namely most infrequent traders (IPOcnt 5), infrequent traders (5<IPOcnt 10), frequent traders (10<IPOcnt 30), and most frequent traders (IPOcnt>30). By controlling for the effect of IP Ocnt, we would like to study the impact of IP Oseq alone and also, we are interested in comparing the impact of IP Oseq across groups. The results are shown in Table 17. (Insert Table 17) Table 17 uses the same setting as in Table 16, but we show only the second stage model. The results show that, after controlling for the selection bias and controlling for IP Ocnt, there is still evidence that investors are learning to be more patient, and resulting in better returns. More importantly, we can see that the learning effect occurs more prominently in the most infrequent traders, and the effect decreases as investors become frequent traders. This directly supports our hypothesis 2. To summarize the finding in this section, we find that investors do learn to be more patient and obtain higher return thereof. after controlling for the selection problem documented in Seru, Shumway and Stoffman (2009), we do find evidence that investors become more patient and enoy higher rate of return through active learning, i.e., participating more IPOs. However, the learning effect varies with investors state of maturity, and with their types. Specifically, when an investor is immature (young or not staying in the market for long), her marginal effect of active learning is high, while when she gets older or stays in the market for longer time, her marginal effect of learning is getting thinner. As to types, if an investor is an infrequent trader, she 41

42 cares a lot about each of her investment behavior. Thus the active learning effect is significant (waiting for significantly longer time, and get significantly higher return, as a result of learning). If she is a frequent trader, she participates quite a few IPOs, and she is not as careful in each IPO, and is willing to accept a lower rate of return. The attrition heterogeneity makes the learning behavior complicated, and needs further attention. 6 Robustness In this section, we will discuss the potential robustness issues. 6.1 Misclassification bias By introducing elapse i to capture the waiting time before the decision to buy, we have made the assumption that all investors start observing the secondary market at 9:25 am on the first day after an IPO. However, what if some investors start to observe later, and this assumption does not hold? This bias might be more severe for the day-10 buyers, since it is more likely that investors may start to observe the market at any time before day 10 and after 9:25 of day 1. We do not have further information about the precise time when investors actually start observing the market. However, we argue that the possibility that investors start to observe the market later than 9:25 of day 1 does not weaken our presented results. For example, if a random fraction λ of investors actually start to observe the market at 10:00 am on day 3, and make their first purchase at 10:10 am on the same day, we classify these investors as day 3 buyers, although they observe the market only for 10 minutes. Had they noticed the stock on day 1, they would have been day 1 buyers, rather than day 3. This type of misclassification bias will definitely exist. However, the misclassifi- 42

43 cation bias tends to recognize some day-t buyers as day t + k buyers (k>0), but not the other way round, which results in underestimation of the behavioral difference between day groups. Therefore, we argue that the existence of the misclassification bias strengthens, rather than weakens, the empirical results in this paper. In order to reduce the misclassification bias, we further conduct the follow exercise. We keep only day 1 and day 2 buyers, which are the least possibly subect to the misclassification bias, and virtually redo every exercise in this paper. By doing so we obtain highly consistent results with present ones (results are not reported). 6.2 Too busy to trade? One possible counter-argument to our results is that, the experienced investors do not intentionally wait for more days after IPO. It may be the case that experienced investors tend to be older, and busier in terms of their daily work, which prevents them from watching the secondary market anytime they wish to. In other words, the more experienced investors tend to be systematically too busy to trade as freely or frequently as the inexperienced ones, who are younger in age with less restrictive daily workload. This may be the reason why more experienced investors systematically wait for longer time before purchasing IPO stocks. We tackle this problem in two folds. First, we use a subsample and keep only those people who are either below 30 or above 60. In China, the legal retirement age is 60 for males and 55 for females. Since we do not have gender information of the investors, we use the criteria of 60 as the retirement age. By this way, we can compare the young investors against those old yet retired investors, who are not subect to the too busy to trade condition. We, again, redo all previous exercises and find again consistent results, which reect the too busy to trade hypothesis. A second way we try to argue against is the following: if more experienced investors 43

44 are too busy to trade, we would expect that on average, more experienced investors may trade on fewer days in the 20-trading day window in our sample. Although they are likely to have greater wealth and trade in larger size, the number of days they are able to trade should be fewer. In order to test this, we calculate the correlation between experience variables and trade frequency, which include the number of buydays (the number of days on which the investor makes at least one purchase), and the number of sell-days (the number of days on which the investor makes at least one sale). We first average the Age, Y ears i, number of buy-days and number-of sell-days for each investor, and then calculate the correlation of the four measures across investors. The results are shown in Table 18. (Insert Table 18) Table 18 shows that the experience measures are actually positively correlated with the number of buy-(sell-) days, which means that, on average, more experienced investors trade more frequently. The too-busy-to-trade hypothesis is not supported. 6.3 Portfolio optimization and risk consideration One argument about why investors purchase the IPO stock is that they are unlikely to hold their optimal allocation of shares, creating the possibility of a high volume trading period as long-term investors adust their portfolios to optimal positions (see Ellis, 2006). However, this hypothesis cannot explain why some investors are rushing into the market on the very first days in our paper. As we have shown earlier, within the 20-traing-day window after investors first purchase of an IPO, investors sell on average 81.11% of all shares they have purchased during the period. Our evidence indicates that the investors are clearly not long-run investors who try to optimize their portfolio. Rather, they are more likely to be short-run investors profiting from 44

45 quick capital gains. The extremely high turnover rate in the 20 trading day window rules out the possibility that investors are optimizing their portfolios, or other risk considerations. 6.4 Order submission time and transaction time One other consideration about our story is the trading time. An investor is able to decide the precise time to place an order, yet the trading time is not exactly controlled by herself. In order to test whether this discrepancy makes any difference, we use the order placing time, rather than the transaction time and redo all the exercises. The results are highly consistent with present ones. 7 Conclusion In this paper, using the specific IPO setting and a novel dataset, we demonstrate that experienced investors and inexperienced investors have different decisions when buying an IPO stock. We present that, first, IPOs attract inexperienced investors more, in the sense that inexperienced investors are more likely to buy IPO stocks on the first day while experienced investors are less likely to do so. More generally, we find that more experienced investors tend to wait longer time before they make their first purchase after IPOs. Second, using the waiting time (before the first purchase of an IPO stock) as a novel proxy of secondary market information that investors acquire, we find that it is through acquiring more trading information i.e., waiting longer before purchase, (along with other elements in the decision bundle, including timing of selling, choosing purchasing price, etc) that more experienced investors obtain a higher return, than the inexperienced investors. The decision bundle explains a large fraction of the cross-sectional return variations. Third, investors do learn to be more 45

46 patient through participating more IPOs (and therefore learn to obtain higher return). However, the marginal effect of learning varies across the level of stock of experience, and across heterogeneous investor types. 46

47 References [1] Aggarwal, R., Stabilization activities by underwriters after initial public offerings. Journal of Finance 55, [2] Aggarwal, R., Allocation of initial public offerings and flipping activity. Journal of Financial Economics 68, [3] Arrow, K. 1962, the economic implications of learning by doing, Review of Economic Studies 29, [4] Ang, A. and Maddaloni, A. (2005) Do demographic changes affect risk premiums? Evidence form international data, Journal of Business 78(1), [5] Arthur, B. W., 1991, Designing Economic Agents That Act Like Human Agents: A Behavioral Approach to Bounded Rationality, American Economic Review, 81, [6] Barber, B. and Odean, T. (2001) Boys will be Boys: gender, overconfidence, and common stock investment, Quarterly Journal of Economics 116(1), [7] Barber, Brad M., and Terrance Odean, 2008, All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors, Review of Financial Studies 21, [8] Barber, Brad M., Terrance Odean, Ning Zhu, 2009, systematic noise, Journal of Financial Markets,12, [9] Barber, B. M., T. Odean, and M. Strahilevitz, Once Burned, Twice Shy: Nave Learning,Counterfactuals, and the Repurchase of Stocks Previously Sold. working paper, University of California, Berkeley. [10] Boehemer, B., E. Boehemer and R. Fishe (2006), Do Institutions Receive Favorable Allocations in IPOs with Better Long-Run Returns, Journal of Financial and Quantitative Analysis, 41 (4), [11] Camerer, C. F., and T.-H. Ho., 1999, Experience-weighted Attraction Learning in Normal Form Games, Econometrica, 67, [12] Campbell, J. Y Household Finance. Journal of Finance 61,

48 [13] Campbell, J., T. Ramadorai and B. Ranish, 2013, Getting better: Learning to invest in an emerging stock market, working paper [14] Chemmanur, T., G. Hu and J. Huang, (2010) The Role of Institutional Investors in Initial Public Offerings, Review of Financial Studies 23, [15] Chiang, Y.M., D. Hirshleifer, Y. Qian, and A. E. Sherman, 2009, learning to fail? Evidence from frequent IPO investors, working paper [16] Choi, J. J., D.Laibson, B. C. Madrian, and A. Metrick, 2007, Reinforcement Learning and Invesor Behavior, working paper. [17] Coval, J. D., D. A. Hirshleifer, and T. Shumway Can Individual Investors Beat the Market?Working Paper, University of Michigan. [18] Cross, J. G., 1973, A Stochastic Learning Model of Economic Behavior, Quarterly Journal of Economics, [19] Da, Z., J., Engelberg, and P. Gao, 2011, In Search of Attention, Journal of Finance, Vol LXVI, no 5, [20] De Long, B. J., A. Shleifer, L. H. Summers and R.J.Waldmann, 1990(a), noise trader risk in financial markets, Journal of Political Economy, 98(4), [21] De Long, B. J., A. Shleifer, L. H. Summers and R.J.Waldmann, 1990(b), positive feedback investment strategies and destabilizing rational speculation, Journal of Finance, 45, [22] Derrien, F., 2005, IPO Pricing in Hot Market Conditions: Who Leaves Money on the Table? Journal of Finance, 60, [23] Dhar, Ravi, and N. Zhu, 2006, up close and personal: an individual level analysis of the disposition effect, Management Science 52 (5), [24] Dorn, D., 2009, Does sentiment drive the retail demand for IPOs? Journal of Financial and Quantitative Analysis 44, [25] Ellis, K., Michaely, R., OHara, M., When the underwriter is the market maker: an examination of trading in the IPO aftermarket. Journal of Finance 55,

49 [26] Ellis, K., Michaely, R., OHara, M., The making of a dealer market: from entry to equilibrium in the trading of Nasdaq stocks. Journal of Finance 57, [27] Ellis, K., 2006, Who trades IPOs? A close look at the first days of trading, Journal of Financial Economics 79, [28] Feng, L., and M. S. Seasholes, 2005, Do Investor Sophisitication and Trading Experience Eliminate Behavioral Biased in Financial Markets?, Review of Finance, 9, [29] Geczy, C., Musto, D., Reed, A., Stocks are special too: an analysis of the equity lending market. Journal of Financial Economics 66, [30] Goyal, A. (2004) Demographics, stock market flows, and stock returns, Journal of Finance and Qualitative Analysis 39(1), [31] Grossman, S. J., R. E. Kihlstrom, and L. J. Mirman, 1977, A Bayesian Approach to the Production of Information and Learning by Doing. Review of Economics Studies 44, [32] Ivkovic, Z., C. Sialm, and S. J. Weisbenner Portfolio Concentration and the Performance of Individual Investors. Journal of Financial and Quantitative Analysis 43: [33] Kaustia, M., and S. Knupfer, 2008, do investors overweight personal experience? Evidence form IPO subscriptions, Journal of Finance, 63(6), [34] Korniotis, George and Kumar, Alok (2007) Does Investment Skill decline due to Cognitive Aging or Improve with Experience? [35] Lee, C. M. C., A. Shleifer, and R. H. Thaler, 1991, Investor sentiment and the closed-end fund puzzle, Journal of Finance 46, [36] List, J. (2003) Does market experience eliminate market anomalies?, Quartely Journal of Economics 118(1), [37] Linnainmaa, J., 2009, Why Do Households Trade So Much?, working paper 49

50 [38] Lungqvist, A.,Vikram Nanda and Radeep Singh, Hot Markets, Investor Sentiment, and IPO Pricing, The Journal of Business, vol. 79(4), pages [39] Lowry, M., 2003, Why does IPO volume fluctuate so much? Journal of Financial Economics, 67, 3-40 [40] Mahani, R., and D. Bernhardt, 2007, Financial Speculators Underperformance: Learning, Self-Selection, and Endogenous Liquidity, Journal of Finance, 62, [41] Nicolosi, G., L.Peng, and N. Zhu, 2009, do individual investors learn from their trading experience?, Journal of Financial Markets 12, [42] Oehler, Andreas, M. Rummer and P. Smith,2004. IPO Pricing and the Relative Importance of Investor Sentiment: Evidence from Germany, Discussion Papers 26, University of Bamberg, Chair of Finance. [43] Ritter. J., 1984, the hot issue market of 1980, Journal of Business 57, [44] Ritter, J., and I. Welch, 2002, A Review of IPO Activity, Pricing and Allocations, Journal of Finance 57, [45] Roth, A. E., and I. Erev. 1995, Learning in Extensive-Form Games: Experimental Data and Simple Dynamic Models in the Intermediate Term, Games and Economic Behavior, 8, [46] Seru, Amit, Shumway, Tyler and Stoffman, Noah, 2009, learning by trading, Review of Finaicial Studies, forthcoming. [47] Shleifer, A., 2000, Inefficient Markets. An Introduction to Behavioral Finance. Oxford University Press, Oxford and New York. [48] Siconolfi, M., Spinning of hot IPOs is probed-us Attorney begins criminal investigation. Wall Street Journal April 16, C1. [49] Siconolfi, M., McGeehan, P., Flip side: Wall street brokers press small investors to hold IPO shares-big institutions can cash out quickly; the little guy cant without penalties-rife with double standards? Wall Street Journal June 26, A1. 50

51 [50] Tam, P.-W., Ewing, T., Heres a flip: buying and then quickly selling an IPO stock is no longer such a bad thing. Wall Street Journal February 2, C1. [51] Pulliam, S., Smith, R., Trade-offs: seeking IPO shares, investors offer to buy more in after-market-and pledges can be a factor, underwriters say, though they deny quid pro quo-trying to avoid the flippers. Wall Street Journal December 6, A1. [52] Smith, R., Pulliam, S., US probes inflated commission for hot IPOs. Wall Street Journal December 7, C1. [53] Zhu, N., (2010) Individual Investor Trading, forthcoming, Chapter 28, Behavioral Finance (edited by Kent Baker and John Nofsinger), The Wiley. 51

52 Table 1: Descriptive Statistics Table 1 shows the descriptive statistics for the IPO stocks, as well as buyer distribution on the first day of IPO. If a stock s close price hits the price limits on the second trading day, we drop this stock from our sample. 38 of 142 stocks are dropped thereof. The final sample size is 104. mean std. min max Panel A: IPOs (N=104) Shares offered (Mil.) Offer amount (Mil. Yuan) IPO price (Yuan) Initial return (%) Over Subscription IPO P/E ratio Panel B: Intraday distribution of buys on the first day # of shares bought (% in parentheses) individual institutional Sum limit order Open centralized auction (15.530) (1.530) (17.061) market order (0) (0) (0) limit order Continuous auction (77.379) (4.911) (82.290) market order (0.030) (0.001) (0.031) limit order Close centralized auction (0.572) (0.046) (0.618) market order (0) (0) (0) Sum (93.512) (6.488) ( ) 51

53 Table 2: Experience and first-day participation We introduce the following probit model: 1 ( x ) α + β Exp + γ Control + ε = 1 where 1(x ) is the indicator function of whether investor is a day 1 buyer or not. First, define 1(x i, ) as the indicator function which takes the value of 1 if investor makers her first purchase of IPO i on the first day, and 0 otherwise. Then, define: m 1 1, if 1( xi ( x ) = m i= 0, otherwise 1 1, ) > 0, where m is the total number of IPOs investor participates in our sample period. Literally, 1(x ) equals to 1 as long as investor has ever participated an IPO on the first day in the sample period, and 0 if an investor has never participated trading an IPO on the first day. Of all Exp. refers to the experience variable, including Age and Years.(Years is the average of Years i across all IPOs investor participates in the sample period). Control variables include: (1) IPO price. (2) Offer size, which is the total number of shares issued in this IPO. (3) IPO P/E ratio: measured as the offer price over earnings per share. To a certain degree, P/E ratio demonstrates the reasonableness of the offer price, which may affect the demand of the IPO stock. (4) over-subscription multiplier: the reciprocal of the lot winning rate, which also proxy the demand of the IPO. (5) growth rate of main business: calculated as the ratio of previous two years core business income. (6) leverage: the asset-liability ratio which is calculated as the ratio of book value of total liabilities and book value of total assets. (7) market return, previous 10 trading days market return. (8) number of IPOs in the previous month. Furthermore, we also include the industry fixed effect, which is based on the CITIC Industry classification, and geographic fixed effect, which is based on the 8 geographic zones each account belongs to. z-statistics are provided in parentheses. Dep: 1(x ) (a) (b) (c) Age (-84.29) (-69.00) Years (-54.25) (-22.51) Constant (-5.95) (-31.21) (-7.22) Firm Control Yes Yes Yes Market Control Yes Yes Yes Industry Fixed Effect Yes Yes Yes Geographic Fixed Effect Yes Yes Yes Pseudo R2 (%) Prob>Chi # of Obs

54 Table 3: Experience and waiting time Panel A of this table shows the descriptive statistics of rawelapse, which is the average minutes elapsed between 9:25:00 of trading Day 1 of stock i and investor s first purchase time of stock i., across all i's. In Panel B of this table we use the following regression: elapse α + β Exp + γ Control + ε = 1 where elapse is the logarithm of 1 plus rawelapse. Exp represents different experience measurements, including Age, which investor s age as of Years is the average number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i across all i's. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Industry fixed effect is based on the CITIC Industry classification. Geographic fixed effect is based on the 8 geographic zones each account belongs to, and heteroskadasticity adusted t-values are provided in parentheses. Panel A Number of minutes elapsed (rawelapse ) Mean 5874 median th pct th pct 9208 Panel B Dep: elapse (a) (b) (c) (d) Age (104.7) (66.30) Years (115.2) (76.39) IPOcnt (46.03) (41.92) IPOcnt-sq (10-2 ) (-20.72) (-20.26) Constant (306.6) (351.2) (351.1) (304.0) Firm Control Yes Yes Yes Yes Market Control Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Geographic Fixed Effect Yes Yes Yes Yes R-square (%) Prob>F # of Obs

55 Table 4: Experience and turnover Panel A of this table shows the descriptive statistics of turnover, literally the 20-trading-day turnover of investors after IPOs, which is defined as follows: First, define the window of investor on stock i as 20-trading day window after investor s first purchase on stock i. Within this window, we sum up all the buy volume and sell volume, and define the ratio of total sell volume over total buy volume as turnover i. If turnover is larger than 1 (which is possible since some investors get IPO share allocation from the primary market), we define turnover i as 100%. turnover is the average turnover for investor across all IPO stocks. Formally, turnover i = M m= 1 K k= 1 q q im ik where K (M) is the total number of buy (sell) trades investor has made during the 20-trading day window since investor s first purchase of stock i s IPO. Then, turnover is defined as: I 1 turnover = turnover i I i= 1, where I is the total number of IPOs investor participates in our sample period. In Panel B of this table, we use the following regression: turnover α + β Exp + γ Control + ε = 1 where Exp represents different experience measurements, including Age, which investor s age as of Years is the number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i, across all i's. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Geographic fixed effect is based on the 8 geographic zones each account belongs to, and heteroskadasticity adusted t-values are provided in parentheses. Panel A turnover (%) Mean median th pct th pct 100 Panel B Dep: turnover (a) (b) (c) (d) Age (10-2 ) (-95.83) (-64.99) Years (10-2 ) (-94.56) (-72.00) IPOcnt (67.59) (68.80) IPOcnt-sq (-22.82) (-22.92) 54

56 Constant (282.55) (272.4) (247.8) (262.6) Firm Control Yes Yes Yes Yes Market Control Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Geographic Fixed Effect Yes Yes Yes Yes R-square (%) Prob>F # of Obs

57 Table 5: Experience and purchasing price Panel A of this table shows the descriptive statistics of P, literally the standardized price investor purchases IPOs, which is defined as follows: First, define P i as investor s first purchase price of IPO i over IPO i's open price on day 1. Then P is defined as: P 1 = IPOcnt IPOcnt i= 1 P i, Panel B of this table uses the following regression: P α + β Exp + γ Control + ε = 1 where Exp represents different experience measurements, including Age, which investor s age as of Years is the average number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i. across all i's. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Industry fixed effect is based on the CITIC Industry classification. Geographic fixed effect is based on the 8 geographic zones each account belongs to, and heteroskadasticity adusted t-values are provided in parentheses. Panel A Mean 1.00 median th pct th pct 1.07 Panel B Dep: P (a) (b) (c) (d) Age (10-3 ) (-41.31) (-21.96) Years (10-3 ) (-54.84) (-42.81) IPOcnt (10-2 ) (16.55) (23.17) IPOcnt-sq (10-4 ) (-10.96) (-14.06) Constant (1022) (1079) (1073) (1014) Firm Control Yes Yes Yes Yes Market Control Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Geographic Fixed Effect Yes Yes Yes Yes R-square (%) Prob>F # of Obs P 56

58 Table 7: Experience and buy size In this table, we use the following regression: value α + β Exp + γ Control + ε = 1 where value is the average logged RMB Yuan value of investor s all purchase of IPO stock i in the 20 trading days after the her first purchase, across all IPO stocks. Exp represents different experience measurements, including Age, which investor s age as of Years is the average number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i., across all i's. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Industry fixed effect is based on the CITIC Industry classification. Geographic fixed effect is based on the 8 geographic zones each account belongs to, and heteroskadasticity adusted t-values are provided in parentheses.. Dep: value (a) (b) (c) (d) Age (176.8) (87.14) Years (252.8) (186.9) IPOcnt (61.08) (55.90) IPOcnt-sq (10-2 ) (-20.77) (-20.04) Constant (685.5) (-772.1) (760.0) (684.5) Firm Control Yes Yes Yes Yes Market Control Yes Yes Yes Yes Geographic Fixed Effect Yes Yes Yes Yes R-square (%) Prob>F # of Obs

59 Table 8: Experience and return In this table, we use the following regression: rtn α + β Exp + γ Control + ε = 1 Where rtn is defined as follows: first, define rtn i as the abnormal return for investor i, in buying stock within 20 trading days: N M N 0 20 q, k p, k ( q, l q, k ) p k = 1 l= 1 k = 1 i, = 1 M 0 0 q, l p, l l= 1 rtn mktrtn, i where N k = 1 M q q, k l = 1 0, l, M 0 q l l = 1. is the total number of stocks investor i buys during the 20-trading day window after her first purchase of stock ; 0 p,l is the actual trading price for investor i to buy stock for the l th time. q, k is the number of stock investor i sells for the k th time during the 20-trading day window; p, is the actual trading price of k investor i when he sell q, stock. k 20 p is the closing price of stock on the 20 th trading day after her first purchase. N is the number of times investor i sells stock during the 20 day window after his buying on the first day or the first non-hit day. M is the total number of times investor i buys stock during the 20-day window. mktrtn i is the Shanghai A-share Composite Index return of the 20-day window. The define rtn : rtn 1 = IPOcnt IPOcnt i= 1 rtn i, Exp represents different experience measurements, including Age, which investor s age as of Years is the average number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i., across all i's. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Industry fixed effect is based on the CITIC Industry classification. Geographic fixed effect is based on the 8 geographic zones each account belongs to, and heteroskadasticity adusted t-values are provided in parentheses. 58

60 Dep: return (a) (b) (c) (d) Age (10-3 ) (38.28) (29.80) Year (10-5 ) (28.83) (15.87) IPOcnt (10-2 ) (-28.31) (-31.02) IPOcnt-sq (10-4 ) (16.00) (16.75) Constant (23.60) (36.93) (41.66) (26.54) Firm Control Yes Yes Yes Yes Market Control Yes Yes Yes Yes Geographic Fixed Effect Yes Yes Yes Yes R-square (%) Prob>F # of Obs

61 Table 9: two-way comparison between experience, waiting time and return This Table shows the difference-in-difference regression for the return distribution for experience groups and waiting days groups. We first sort within each stock by the experience (Age or Years) and categorize the investors into 10 groups, then categorize within each experience group by their first day of purchasing IPO. We then keep only the lowest and the highest experience group, and day 1 and day 10 buyers groups. Panel A: Age as experience Return gap between Experience Group 10 Simple mean (%) Weighted mean (%) and Group 1 On IPO Day 1 (t-value for equality) On IPO Day 10 (t-value for equality) Diff: Day 1 Day 10 (t-value for equality) (6.07) (0.89) (4.90) (5.88) (1.60) (2.12) Panel B: Years as experience Return gap between Experience Group Simple mean (%) Weighted mean (%) 10 and Group 1 On IPO Day 1 (t-value for equality) On IPO Day 10 (t-value for equality) Diff: Day 1 Day 10 (t-value for equality) (6.08) (0.85) (4.92) (4.51) (0.49) (2.66) 60

62 Table 10: Difference-in-Difference regression for continuous variables This Table shows the difference-in-difference regression for the return distribution for experience and waiting days groups. return i, = α + β1experiencei, + β 2Elapsei, + β 3Elapsei, Experiencei, + γ Control + ε i, where return i is the abnormal return for investor i, in buying stock within 20 trading days, and is defined earlier. Experience i represents different experience measurements, including Age, which investor s age as of 2008 and Years i, the number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i. Firm control variables include the total Yuan value purchased within the 20-trading-day window after the first purchase of investor on IPO i, IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Industry fixed effect is based on the WIND Industry classification. Geographic fixed effect is based on the 8 geographic zones each account belongs to, and heteroskadasticity adusted t-values are provided in parentheses. Dep: return (a) (b) (c) elapse (66.97) (155.5) (62.05) Age (10-3 ) (22.76) (8.24) Year (10-5 ) (37.29) (29.26) Age*elapse (10-4 ) (-16.07) (-2.77) Year*elapse (10-6 ) (-32.49) (-27.54) Constant (-22.04) (-20.94) (-17.92) Firm Control Yes Yes No Market Control Yes Yes No Industry Fixed Effect Yes Yes No Geographic Fixed Effect Yes Yes No R-square (%) Prob>F # of Obs

63 Table 11: Structural Equations This Table shows the results of the structural equation model as shown in Figure X. The structural equations are: Elapse = γ Exprience+ Control + (9) 11 γ 21 ε11 turnover = γ i Control+ 12Exprience + γ 21 ε12 P = γ 13Expriencei + γ 21Control + ε12 value = γ i Control+ 14Exprience + γ 21 ε14 (10) (11) (12) return = θ Elapse + θ 2turnover + θ3p + θ 4value + χ1exp + χ 2Control + ε 1 (13) where all variables are defined as earlier. Panel A: Equation (9) Dep: Elapse (a) (b) (c) γ 11 Age (103.6) (66.34) Years(10-3 ) (111.2) (76.27) IPOcnt (54.90) (49.80) (49.96) IPOcnt-sq (10-2 ) (-35.95) (34.08) (-34.10) Control/Fixed Effect Yes Yes Yes Panel B: Equation (10) Dep: turnover (a) (b) (c) γ 12 Age(10-2 ) (-101.4) (-65.75) Years(10-4 ) (-106.3) (-72.39) IPOcnt (129.9) (134.6) (134.5) IPOcnt-sq (10-3 ) (-63.83) (-68.83) (-70.74) Control/Fixed Effect Yes Yes Yes Panel C: Equation (11) 62

64 γ 13 Dep: P (a) (b) (c) Age(10-3 ) (-41.38) (-21.79) Years(10-5 ) (-55.10) (-42.51) IPOcnt (10-2 ) (10.18) (12.92) (12.91) IPOcnt-sq (10-4 ) (-4.71) (-5.75) (-5.78) Control/Fixed Effect Yes Yes Yes Panel D: Equation (12) Dep: value (a) (b) (c) γ 14 Age (175.6) (89.92) Years(10-3 ) (246.2) (191.6) IPOcnt (108.7) (97.42) (97.34) IPOcnt-sq (10-2 ) (-55.55) (-51.10) (050.98) Control/Fixed Effect Yes Yes Yes Panel E: Equation (13) Dep: return (a) (b) (c) β 1 Elapse (10-2 ) (200.7) (202.4) (200.7) β 2 β turnover (-154.7) (-156.1) (-154.7) P (-26.78) (-27.37) (-26.90) β 4 Value(10-3 ) (-11.81) (-10.01) (-11.10) Age(10-3 ) (11.09) (11.76) Years(10-5 ) (1.16) (-3.93) χ 1 IPOcnt (10-3) (-11.24) (-11.04) (-11.00) IPOcnt-sq (10-4 ) (6.77) (6.69) (6.69) Control/Fixed Effect Yes Yes Yes No. obs

65 Table 12: Descriptive Statistics for Total Number of Participated IPO (IPOcnt ) IPOcnt mean 1.56 median 1 stdev 1.70 min 1 max 89 No of obs

66 Table 13: Patience, Investor Experience and Type This table reports the estimates of selection model regression whose second stage regression in the following form Elapse = α β β β θ γ δ ε i+ 1, + 1IPOcnt + 2Expi, + 3IPOcnt Experiencei, + 1IPOseqi, + Controli+ 1, + IMR + i+ 1, where IMR is the inverse Mill ratio of the selection model (the Probit model). All other variables are defined earlier. Z-values are in parentheses. Selection model 2 nd Stage Dependent variable (1 st stage) Elapse i+1, In-sample i+1, =1 IPOcnt (-28.93) (-49.12) (-29.08) Age (-15.86) (62.48) (39.05) Years (-15.33) (68.72) (47.75) Age*IPOcnt (10-3 ) (-5.50) (-0.53) Years*IPOcnt (10-2 ) (-11.13) (-8.91) IPOseq (810.43) (23.11) (22.55) (21.51) return (51.65) (29.06) (29.13) (27.29) Area_rate (18.64) 65

67 IMR (54.09) (59.41) (59.40) Firm Control Yes Yes Yes Yes Market Control Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Time Fixed Effect Yes Yes Yes Yes Prob>F (Prob>Wald) # of Obs

68 Elapse Table 14: Past IPO Participation and waiting time In this table, we use the following regression: = α β β β γ ε i+ 1, + 1IPOSeqi, + 2Experiencei, + 3IPOSeqi, Experiencei, + Controli+ 1, + i+ 1, In this equation, subscript refers to a specific investor, and subscript i represents IPO i. In the sample period, investor altogether participates in buying M IPOs altogether, where participate is defined as make at least one purchase in the first 10 trading days after IPO. Sort all IPOs investor participates according to the time of her first purchase of that IPO, and define m as the m th number of IPO investor participates ( ), and name m as IPOseq i. Elapse i is the waiting time, which is defined as the logarithm of 1 plus the minutes elapsed between 9:25:00 of trading Day 1 of stock i and investor s first purchase time of stock i. and Elapse i+1, is defined as the waiting time th mi 1 + IPO investor participates. elapse investor, and Experience i represents different experience measurements, including Age, which investor s age as of Years i is the number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income, return of the previous IPO. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO. Dep: Elapse (a) (b) (c) IPOSeq (6.27) (4.20) (3.54) Age (64.78) (44.37) Years (64.07) (41.93) IPOSeq*Age (10-3) (-2.66) (-1.81) IPOSeq*Years (10-3) (1.18) (2.66) IPOcnt (-99.77) (-103.0) (-102.6) Constant (170.4) (206.1) (174.8) Firm Control Yes Yes No Market Control Yes Yes No Industry Fixed Effect Yes Yes No Geographic Fixed Effect Yes Yes No R-squared (%) Prob>F # of Obs

69 return Table 15: Past IPO Participation and return In this table, we use the following regression: = α β β β γ ε i+ 1, + 1IPOSeqi, + 2Experiencei, + 3IPOSeqi, Experiencei, + Controli+ 1, + i+ 1, In this equation, subscript refers to a specific investor, and subscript i represents IPO i. In the sample period, investor altogether participates in buying M IPOs altogether, where participate is defined as make at least one purchase in the first 10 trading days after IPO. Sort all IPOs investor participates according to the time of her first purchase of that IPO, and define m as the m th number of IPO investor participates ( ), and name m as IPOseq i. return i+1, is defined as the return of the th mi 1 + IPO investor participates. Experience i represents different experience measurements, including Age, which investor s age as of Years i is the number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i. Firm control variables include IPO price, IPO size, IPO P/E, subscription multiplier, leverage, growth rate of main business income. Market control variables include previous 10-day market return, and number of IPOs in the past month of an IPO... Dep: return (a) (b) (c) IPOSeq (73.48) (120.4) (75.18) Age (10-3 ) (21.71) (6.62) Years (10-2 ) (43.46) (37.54) IPOSeq*Age (10-3 ) (-10.92) (-2.96) IPOSeq*Years (10-3 ) (-25.25) (-22.87) IPOcnt (10-2 ) (-128.1) (-103.0) (-129.9) Elapse (10-2 ) (76.46) (75.98) (75.18) Constant (46.45) (56.23) (48.96) Firm Control Yes Yes Yes Market Control Yes Yes Yes Industry Fixed Effect Yes Yes Yes Geographic Fixed Effect Yes Yes Yes R-squared (%) Prob>F # of Obs

70 Table 16: learning, experience, waiting time and return This table reports the estimates of selection model regression whose second stage regression in the following form y i+ 1, = α + β1iposeqi, + β 2Experiencei, + β3iposeqi, Experiencei, + θ ε 1 IPOcnt + θ 2IPOseqi, IPOcnt + γ Controli+ 1, + δ IMR + i+ 1, Where yt+1, represents Elapse i+1, and return i+1,. IMR is the inverse Mill ratio of the selection model (the Probit model). All other variables are defined earlier. Z-values are in parentheses. Dependent variable Selection model (1 st stage) In-sample i+1, =1 2 nd Stage Elapse i+1, 2 nd Stage return i+1, IPOSeq (13.17) (16.68) (11.37) (46.18) (57.08) (44.41) Age (10-2 ) (-15.86) (67.47) (42.79) (51.76) (25.60) Years (10-2 ) (-15.33) (73.10) (50.76) (74.62) (59.52) IPOcnt (810.43) (-76.80) (-79.98) (-78.60) (-34.92) (-38.61) (-38.15) Age*IPOseq (10-3 ) (-5.72) (-1.26) (-34.92) (-9.96) Years*IPOSeq (10-2 ) (-10.98) (-8.91) (-22.00) (-17.59) IPOcnt*IPOseq (10-3 )

71 (-0.39) (3.41) (2.73) (-25.82) (-18.57) (-18.95) Elapse (10-2 ) (79.96) (78.32) (76.61) return (51.65) (27.52) (27.29) (24.70) (62.48) (61.70) (60.52) Area_rate (18.64) IMR (45.12) (48.44) (49.25) (54.81) (58.47) (58.83) Firm Control Yes Yes Yes Yes Yes Yes Yes Market Control Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Time Fixed Effect Yes Yes Yes Yes Yes Yes Yes Prob>F (Prob>Wald) # of Obs

72 Table 17: learning and investor type This table reports the estimates of selection model regression whose second stage regression in the following form y i+ 1, = α + β1iposeqi, + β 2Experiencei, + β 3IPOSeqi, Experiencei, + θ θ γ δ ε 1 IPOcnt + 2IPOseqi, IPOcnt + Control i+ 1, + IMR + i+ 1, Where yt+1, represents Elapse i+1, and return i+1,. IMR is the inverse Mill ratio of the selection model (the Probit model). All other variables are defined earlier. Z-values are in parentheses. Panel A: Dependent Variable: Elapse Dep var: Elapse i+1, Type=1: IPOcnt<=5 Type=2: 5<IPOcnt<=10 Type=3: 10<IPOcnt<=30 Type=4: 30<IPOcnt IPOSeq (5.56) (5.08) (5.08) (3.22) (1.97) (2.51) (4.12) (5.41) (3.65) (-2.37) (-5.39) (-2.69) Age (29.46) (18.84) (17.87) (12.29) (13.07) (10.62) (3.87) (4.17) Years (32.01) (21.88) (18.13) (12.47) (9.48) (5.76) (-1.04) (-1.97) IPOcnt (-2.74) (-2.92) (-3.28) (-10.60) (-11.37) (-11.20) (-19.63) (-20.31) (-19.94) (-12.75) (-12.23) (-12.22) Age*IPOseq (10-3 ) (1.32) (0.43) (-2.47) (-1.76) (0.44) (-0.66) (-1.51) (-2.14) Years*IPOSeq (10-2 ) (1.49) (1.33) (-2.59) (-1.89) (-0.49) (-0.67) (2.95) (3.38) IPOcnt*IPOseq (-2.67) (-2.68) (-2.42) (0.70) (1.29) (1.25) (-2.53) (-2.09) (-2.11) (4.90) (3.99) (3.83) 71

73 return (31.00) (30.40) (28.68) (7.72) (27.29) (24.70) (1.11) (1.26) (0.85) (-2.26) (-2.43) (-2.24) IMR (16.66) (17.06) (16.66) (5.46) (6.05) (6.87) (6.72) (6.96) (6.73) (5.36) (5.35) (5.37) Firm Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Market Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Time Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Prob>F (Prob>Wald) # of Obs (total) # of Obs (unsensored) Marginal effect (%) (14.88) (13.76) (13.21) (12.39) (11.15) (10.41) (9.12) (8.70) (8.18) (-3.10) (-3.18) (-2.98) 72

74 Panel B: Dependent Variable: return Dep var: return i+1, Type=1: Type=2: Type=3: Type=4: IPOcnt<=5 5<IPOcnt<=10 10<IPOcnt<=30 30<IPOcnt IPOSeq (19.99) (22.35) (19.19) (6.45) (1.97) (6.11) (16.52) (18.80) (16.11) (6.50) (8.69) Age (10-3) (21.72) (11.38) (6.09) (3.18) (5.93) (4.23) (-1.75) Years (10-2) (30.10) (23.10) (9.02) (7.26) (6.09) (4.45) (-0.42) IPOcnt (10-2) (11.54) (10.95) (10.42) (-7.62) (-8.05) (-8.04) (-11.06) (-11.36) (-11.30) (1.21) (1.19) Age*IPOseq (10-3 ) (-0.21) (-0.83) (0.30) (-0.71) (-2.53) (-2.37) (0.21) Years*IPOSeq (10-2 ) (0.64) (0.96) (2.45) (2.57) (-0.82) (-0.03) (-0.11) IPOcnt*IPOseq (10-2 ) (-13.85) (-13.59) (-13.31) (0.06) (0.26) (0.12) (-8.69) (-8.32) (-8.36) (-7.22) (-6.93) return (55.50) (54.50) (53.18) (28.68) (28.19) (27.77) (14.45) (14.28) (14.23) (0.16) (0.20) Elapse (10-2 ) (74.77) (73.09) (71.52) (27.15) (26.53) (25.95) (15.28) (15.40) (15.05) (3.94) (3.84) IMR (28.38) (28.37) (27.84) (18.69) (18.54) (18.13) (11.31) (11.38) (11.25) (3.50) (3.55) Firm Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Market Control Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Time Fixed Effect Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes (6.50) (-1.71) (0.01) (1.19) (0.24) (-0.18) (-6.92) (0.16) (3.95) (3.48) 73

75 Prob>F (Prob>Wald) # of Obs (total) # of Obs (unsensored) Marginal effect (%) (37.33) (35.75) (35.38) (34.05) (32.56) (32.43) (30.10) (29.34) (29.22) (8.61) (8.65) (8.56) 74

76 Table 18: Correlation between experience and no of trading days This table shows the correlation between the experience measures and the trading frequency measures. Age is investor s age as of Years is the number of years which have elapsed since the opening of the investor s account till the specified day on which investor buy sample IPO stock i. #buy-day is the number of days investor has made purchases on during the 20-trading day after her first purchase of IPO i, #sell-day is the number of days investor has made sales on during the 20-trading day after her first purchase of IPO i. We first calculate the mean of the four measures for each investor, and then calculate the correlation. t-statistics are shown in parentheses. age Years # buy-day # sell-day age Years (0.000) # buy-day (0.000) (0.000) # sell-day (0.000) (0.000) (0.000)

77 Figure 1: Trading volume on day 1 to day 10 after IPO Figure 2: Shanghai Stock Exchange Composite Index 76

78 Figure 3: Age and days in market for investors with different entering days Figure 4: Intraday pattern of investors experience Panel A: Investor s age 77

79 Panel B: Years in the market 78

80 Figure 6.1: First sort on waiting days then on Age Figure 6.2: First sort on waiting days then on Years 79

81 Figure 6.3: First sort on Age, then on buy-in date Figure 6.4: First sort on Years, then on buy-in date 80