Characterizing the Risk of IPO Long-Run Returns: The Impact of Momentum, Liquidity, Skewness, and Investment

Transcription

1 Characterizing the Risk of IPO Long-Run Returns: The Impact of Momentum, Liquidity, Skewness, and Investment RICHARD B. CARTER*, FREDERICK H. DARK, and TRAVIS R. A. SAPP This version: August 28, 2009 JEL Classifications: G12, G24, G30 Keywords: Initial public offering, momentum, liquidity, skewness, investment, new issues puzzle * Corresponding author. Rick Carter may be reached at College of Business, 3333 Gerdin Business Bldg., Iowa State University, Ames, IA , Phone: (515) , Fax: (515) , rbcarter@iastate.edu. Rick Dark may be reached at 3335 Gerdin Business Bldg., Iowa State University, Ames, IA , Phone: (515) , dark@iastate.edu. Travis Sapp may be reached at College of Business, 3362 Gerdin Business Bldg., Iowa State University, Ames, IA , Phone: (515) , trasapp@iastate.edu.

2 Characterizing the Risk of IPO Long-Run Returns: The Impact of Momentum, Liquidity, Skewness, and Investment Abstract We study 6,686 IPOs spanning the period and find that the new issues puzzle disappears in a Fama-French three-factor framework. IPOs do not underperform in the aftermarket on a risk-adjusted basis and do not underperform a matched sample of non-issuers. IPO underperformance is concentrated in the 1980 s and early 1990 s, and IPO s either perform the same as the market, or outperform on a risk-adjusted basis, during We attribute this to the recent shift to alternative offering mechanisms. Factors for momentum, investment, and liquidity help to explain aftermarket returns, whereas skewness does not. IPO investors receive smaller expected returns due to negative momentum and investment exposure and in exchange for higher liquidity.

3 I. Introduction Prior research shows that new issues of equity and debt underperform benchmarks and matched samples of non-issuers (see e.g. Ritter (1991), Loughran (1993), and Loughran and Ritter (1995)). This result has been dubbed the new issues puzzle. We revisit this finding using a sample of more recent data, which encompasses some fundamental changes in the going public landscape. We also examine the impact on initial public offering (IPO) returns of several factors that have been shown to help price securities in the literature, including stock return momentum, share liquidity, return co-skewness, and investment. Momentum has been shown to be a significant and robust factor in pricing securities by Carhart (1997), and Brav, Geczy, and Gompers (2000) show that momentum helps to explain the returns of IPO firms. Eckbo and Norli (2005) find that share liquidity, defined as trading volume divided by number of shares outstanding, is significant in explaining IPO returns. Harvey and Siddique (2000) and Nguyen et al. (2007) show that return skewness is significant in explaining security returns. Lyandres, Sun, and Zhang (2008) find that an investment factor is significant in pricing IPO returns. However, until now no test has been conducted which brings all of these factors together at once. In this paper we explore the explanatory power of each of these four factors in accounting for the aftermarket performance of new equity issues. Using a sample of 6,686 IPOs spanning the period , we implement three tests of long-horizon abnormal performance. We measure abnormal performance using buy-and-hold returns with an attribute-based control sample, factor regressions, and cumulative abnormal returns. We first find that IPOs, particularly in the later years of the sample, do not underperform a sample of non-issuing firms matched on size and book-to-market. This is somewhat surprising, and is in contrast to evidence found in Ritter (1991), among others. When comparing the three- 1

4 year post-ipo performance to that of a market index, there is some evidence of underperformance. However, we find that this underperformance is concentrated in the early part of the sample and disappears over When measured as an alpha from a Fama-French three-factor regression, there is no evidence of IPO underperformance in the 36 months following issuance. This result is also in contrast to numerous prior studies. Further, we find evidence that IPOs outperform during the later years of the sample. Abnormal returns are found to be significantly negative during , flat during , and significantly positive during We hypothesize that differences in sample periods between studies may account for the differences in findings, with the 1980s and early 1990s likely driving the finding of IPO underperformance in other studies. We attribute this development to two recent phenomena: firms are seeking to access the capital markets sooner in their life cycle, and there has accordingly been a recent shift to alternative offering mechanisms, such as reverse mergers and sellouts. 1 For example, as shown in Floros and Sapp (2009), the number of reverse mergers has outpaced that of traditional IPOs in each of the last eight years. This corresponds to the time period over which we find that IPOs do not underperform. In these later years, more firms which are younger, highly information asymmetric, and riskier are choosing to go public using these alternative mechanisms. We believe that this has resulted in a sample of higher quality IPO firms. We further find that momentum and liquidity are significant predictors of IPO initial returns, whereas skewness is not. The evidence shows that IPO investors receive smaller expected returns due to negative momentum exposure. Higher liquidity of new issues is further associated with lower average returns, suggesting that increased trading tends to bid down the 1 A reverse merger is where a private firm acquires a public firm, usually an empty shell company, and inherits its public status. The merger is referred to as reverse because the public firm survives through its public status. A sellout is where a private firm agrees to be acquired by a public firm. 2

5 share price. This may indicate that the presence of more noise traders in the market is correlated with increased pessimism about the prospects of new issues. We also test the explanatory power of a factor based on the level of investment by firms, which Lyandres, Sun, and Zhang (2008) find to be significant. Our results show that the investment factor does carry significant explanatory power for IPO returns, even in the presence of momentum and liquidity. However, we stress that the three Fama-French factors alone are found to be sufficient for explaining the underperformance of IPO three-year returns. In sum, our results question the continued relevance of prior findings of long-run underperformance for equity IPOs. The equity new issues puzzle appears to be rather sensitive to the sample period studied as well as to the particular matched sample to which it may be compared. The underperformance of new issues seems to have faded over the recent time period of Based on this limited number of years, we are reluctant to declare this a long-term trend, but as more data becomes available we should gain a better understanding of whether a fundamental shift in IPO performance has occurred. Our study also sheds light on additional factors which help to explain IPO returns. Specifically, including momentum, liquidity, and investment-to-assets factors appears to explain IPO returns better than the three-factor model alone. This paper is organized as follows. The next section introduces the data, giving some descriptives. Section III presents results on long-run IPO performance, including the explanatory power of several characteristic-based factors. Some robustness tests follow, and Section IV concludes. 3

6 II. Data The data consists of IPOs spanning the period We use the Thomson SDC Platinum database to identify the IPOs over the sample period of non-financial firms with offer prices over $2. After eliminating depository shares, spin-offs, REITs, closed-end funds, firms with unit offerings, and those with missing data we are left with a sample of 6,686 firms. The annual distribution of the IPOs is displayed in Figure 1. The heaviest volume periods are and , with the 1990 s clearly being the most active. {Figure 1 about here} The SDC database provides most of the information for the firms, including the offer date, the offer price, the gross proceeds from the offer, and the underwriter. Additional information such as net income, stockholder s equity, total assets, and revenues at the time of the offer, is taken from the Compustat database. All data concerning market prices and outstanding shares are from the Center for Research in Securities Prices (CRSP). Because much of our analysis involves comparison of IPOs to a matched sample, we paid particularly close attention to the construction of the matched sample. Following a process used in previous research, we matched first on market value at the time of the IPO and then on the market-to-book ratio (see for example, Brav, et al. (2000) and Loughran and Ritter, (1995)). The market-to-book ratio is based on the market value at the time of the IPO and the latest book value, including the offering. The resulting matched sample is within 0.6% of the IPO sample in market value and within 15% for the market-to-book ratio on average. The analysis calls for the use of monthly factors in order to characterize firm returns. Four of the factors, market return, high minus low book-to-market, small minus big market value, and momentum were taken from Ken French s website. The liquidity factor is developed 4

7 using the procedure outlined in Eckbo and Norli (2005). Specifically, the factor is constructed as the return on a zero-investment equal-weighted portfolio that is long the least liquid 30% of the market and is short the most liquid 30% of the market, where liquidity is defined as portfolio turnover. The co-skewness factor is implemented as in Harvey and Siddique (2000) and Nguyen et al. (2007), and consists of a squared market return series. Finally, we include an investment factor as in Lyandres, Sun, and Zhang (2008) that is long in low-investment stocks and short in high-investment stocks. The investment factor was obtained from Lu Zhang s website. {Table I about here} Descriptive statistics for data from the IPO firm sample are presented in Table I. The average firm size is $351 million with a median value of $102 million. This figure drops to $218 million with an inflation adjustment. The average run-up in stock price from the offering to the first day bid price is 17.92%. This ranges from a low of 7.7% in the late 1980s to a high of 36% in the late 1990s. This is consistent with similar evidence provided by Ritter and Welch (2002). The three-year buy-and-hold raw return over the entire 25 year period is percent, and the five-year buy-and-hold raw return is percent. However, these returns are highly skewed to the right as the median in both cases is less than zero. Figure 2 displays the cumulative returns of the monthly equal-weighted IPO sample, as well as the CRSP value-weighted market series and the sample of firms matched on size and book-to-market. The significance of any perceived differences is examined in Tables II and III. {Figure 2 about here} III. The Performance of Initial Public Offerings A. Matching Sample Analysis 5

8 In Table II we show the three-year buy-and-hold returns and one-year cumulative abnormal returns (CARs) for annual cohorts of IPOs as well as for cohorts of firms matched on size and book-to-market. In computing the three-year buy-and-hold return we exclude the firstday initial return on the IPO. CARs are measured as the one-year cumulative alphas from monthly Fama-French three-factor regressions. 2 The table also displays the wealth relative between the IPO and matching firm cohorts each year. {Table II about here} We see that for the buy-and-hold returns a slim majority of wealth relatives (52%) are greater than one, indicating that in more years than not the IPOs outperform the matching firms. Over the entire sample, the mean three-year IPO buy-and-hold return is 28.6%. For the matched firms, the average three-year buy-and-hold return is 38.6%, giving an overall wealth relative of In looking at the CARs for the IPO firms, we see that they tend to be negative early in the sample period and positive towards the end of the sample. The CARs for the matching firms follow a similar pattern, but are smaller. The majority of the wealth relatives are once again greater than one, indicating outperformance in most years by the IPO firms. The overall CAR for the IPO firms is 26.6%, while that of the matching sample is %. {Table III about here} In Table III we compare the average monthly return and the three-year buy-and-hold return for the IPO firms to those of the CRSP value-weighted market portfolio and the matched sample. We also divide the sample into two sub-periods based on evidence uncovered later in the paper. We first form time series portfolios of average monthly IPO and match firm returns. A comparison of IPO average monthly returns to the average CRSP market return shows no significant difference over any of the time periods. Similarly, when comparing IPO average 2 The computation of the CARs is described in more detail in Section C below. 6

9 monthly returns to average monthly returns on the matched firm sample, there is no significant difference. This measure tends to be noisier than looking at buy-and-hold returns, as there are fewer observations (months) over which to average as opposed to looking at buy-and-hold returns for the individual firms. Therefore, we next turn to three-year buy-and-hold returns. From the table, we see that IPOs underperform the market by an average of 16.9% during the first three years of post-ipo returns. This is slightly less than the evidence described in Ritter and Welch (2002), who report underperformance of 23.4%. The aftermarket performance of IPOs relative to the market in years four and five is negative, but substantially smaller at -3.7%. Thus, consistent with other studies, most of the IPO underperformance is seen to be concentrated in the first 36 months after issuance. In results not tabulated, we find that the underperformance is stronger in the 1980 s with a three-year market-adjusted return of -23.5% over and -13.3% over When looking at the period , which more closely resembles that of most existing studies, we see IPO three-year underperformance of 20.9%. However, in the more recent time period we note that IPO underperformance disappears. The return difference is only -4.7% and is statistically insignificant. When comparing IPO returns to the matched firm sample, we see that the matched sample outperforms the IPO sample by a significant 9.9% over the first 36 months, but there is no significant difference in years four and five. When looking at the early sample period of , match firms significantly outperform the IPO sample by 7.2% over the first three years. In the later period of the matched firms significantly outperform the IPO threeyear buy-and-hold returns by 18.2%. Ritter and Welch (2002) examine the internet bubble years of 2000 and 2001 separately, since these constitute an apparent abnormal period of time. To examine the extent to which the internet bubble impacts the three-year buy-and-hold returns of 7

10 the cohorts for 1999 and 2000, we exclude these two years in the last row of Table III. We find that the outperformance of the match firms is driven by these two years. When excluding these two years, match firms do not significantly outperform IPO firms. Further, we find that IPO firms significantly outperform the matched sample over by 13.5%. Overall, we find no compelling evidence that IPO firms consistently, significantly underperform a matched sample. The fact that IPOs do not underperform a sample of firms that is matched on size and book-to-market is surprising and is in contrast to the findings of previous studies, such as Ritter (1991) and Loughran and Ritter (1995). However, Brav and Gompers (1997) also find that IPOs perform no worse than similar issuing firms. This highlights the sensitivity of results to the particular matched sample that is chosen for comparison to the IPO firms. Ritter and Welch (2002) note that IPO firms tend to be small growth firms and this segment of the market has performed poorly over the last several decades. Thus, a sample that is matched on size and bookto-market is already tilted toward relatively poor performance. As pointed out by Fama (1998) and Mitchell and Stafford (2000), event-time buy-andhold returns tend to magnify small differences in performance and may therefore be less desirable than other performance measures such as time series regressions and cumulative abnormal returns. Accordingly, we next compute alphas based on calendar-time IPO portfolio returns. This is followed by an analysis of cumulative abnormal returns. B. Portfolio Time Series Factor Regressions We next explore the risk-adjusted aftermarket performance of IPOs within a factor regression framework. We regress a rolling portfolio of the monthly firm excess returns for the 36 months following the IPO against seven contemporaneous factors. The factors are market 8

11 excess returns (RMRF), a size factor (SMB), a book-to-market factor (HML), a momentum factor (MOM), a liquidity factor (LIQ), a co-skewness measure (SKEW), and an investment factor (INVEST). The first three factors (RMRF, SMB, HML) and the fourth factor (MOM) are wellknown and discussions can be found in Fama and French (1993) and Carhart (1997), respectively. The liquidity factor is from Eckbo and Norli (2005). The factor is constructed by subtracting the returns of high turnover stocks from those of low turnover stocks, where monthly turnover is defined as the trading volume divided by number of shares outstanding. Co-skewness is from Harvey and Siddique (2000) and is defined as the squared excess return on the market, less its mean. Finally, the investment factor of Lyandres, Sun, and Zhang (2008) is the return on a portfolio long in stocks with the lowest investment-to-assets ratio and short stocks with the highest investment-to-assets ratio. Lyandres et al show that this factor has power to explain IPO returns. The regression formula appears in equation (1): r pt p RMRFt SMBt HMLt MOM t LIQt SKEWt INVESTt (1) t where r pt is the excess return for either IPOs or the matched sample. The intercept in this regression is interpreted as a measure of average abnormal performance. We estimate this equation as a time series portfolio of monthly average IPO returns. As noted by Loughran & Ritter (2000), and confirmed by Brav, Geczy, and Gompers (2000), valueweighted portfolios have low power to detect abnormal performance. We therefore focus on equal-weighted portfolios of returns. Specifically, we form a calendar-time equal-weighted portfolio of 335 monthly average post-ipo returns, from January 1981 through November The risk free rate is subtracted from the monthly raw returns for each month. A single regression is then performed on the resulting time series of monthly average IPO excess returns. This is 9

12 done for both the IPO sample and the matched sample. The results are presented in Table IV. 3 {Table IV about here} We first note that none of the alphas from the IPO regressions is significant. Model I consists of only the three Fama-French factors and each of the factor coefficients is significant in explaining IPO returns. The monthly alpha of % has a t-statistic of and shows no evidence of significant IPO underperformance. Hence, the new issues puzzle disappears in a factor regression setting. We note that this result is in contrast to the findings of several other studies, such as Lyandres, Sun, and Zhang (2008), who report a significantly negative threefactor alpha of -0.43% over the period Loughran and Ritter (1995) report a significant three-factor alpha of over Brav and Gompers (1997) report a significant three-factor alpha of -0.52% for small firm IPOs over Brav, Geczy, and Gompers (2000) report an insignificant three-factor alpha of -0.37% for a sample of Nasdaq IPOs over the period Ritter and Welch (2002) report an insignificant three-factor alpha of -0.21% over the period They note that long-run IPO performance results can be rather sensitive to the exact time period analyzed, and we apparently see evidence of that here. We explore this issue in more detail in the next section. Model II includes the investment factor of Lyandres, Sun, and Zhang (2008). Results show that investment is significantly negatively correlated with IPO returns, confirming the finding of Lyandres, Sun, and Zhang (2008). Model III consists of the Fama-French three-factor model plus a momentum factor, where we see that the coefficient on the momentum factor is significantly negative. This is in agreement with the finding of Brav, Geczy, and Gompers (2000) who report a significant negative loading on the momentum factor in a four factor 3 We also implement a Fama-MacBeth (1973) monthly firm-level regression approach. The results for the full model are qualitatively similar to those of the portfolio time series regression approach and are therefore not reported. However, a subset of results using this approach is reported in Table V. 10

13 regression of long-run IPO returns. Model IV includes the liquidity factor, the co-skewness factor, and the investment factor. Results show that liquidity helps explain IPO aftermarket returns. The coefficient on liquidity is negative and significant, indicating that increased trading is associated with lower aftermarket performance. The coefficient on co-skewness is also negative, but insignificant. The investment factor is seen to be still negative and significant. In Table IV we also analyze a sample of non-issuing firms that has been matched to the IPO sample based on firm size and book-to-market. The four regression models analyzed for the IPO sample are repeated for the matched firm sample. The matched firm sample shows underperformance of 0.22% according to the Fama-French three-factor benchmark (Model I), but is only significant at the 10% level. Model II adds the investment factor which, in contrast to the IPO sample, is not significant for the matched firms. Model III consists of the three Fama- French factors augmented with a momentum factor. We see that momentum is a significant predictor of the matched firm returns and negative momentum exposure succeeds in explaining away the underperformance of the matched firm sample. Specifically, the alpha moves from % to an insignificant 0.02% when a momentum factor is added. In Model IV both liquidity and momentum are seen to be negative and significant in explaining the matched firm returns, consistent with our findings for IPO returns. In all four match firm regressions, the coefficient for HML is seen to be significantly positive, whereas for the IPO firms it is consistently significantly negative. Finally, the last two models of Table IV are estimated on the difference between the time series of IPO returns and that of the matching firm returns. To the extent that the IPO firms and matching firms have similar factor exposures, the estimated coefficients should be zero. We first note that the monthly three-factor alpha is an insignificant 0.17% (t-statistic = 0.99), indicating 11

14 no significant difference in performance between the IPOs and the matching firms. There is no significant difference in exposure to the size and value factors. However, HML exposure is significantly more negative for the IPO firms. In the Model IV regression which includes momentum, liquidity, skewness, and investment, we see that IPO firms also have significantly greater exposure to the investment factor. C. Firm-Level Regression Analysis Loughran & Ritter (2000) and Brav, Geczy, and Gompers (2000) point out that the timeseries regression approach weights each month equally instead of taking into account firm-level effects and the number of offerings in a given month. This may have the effect of understating the underperformance of new equity issues. To address this concern, we next implement a firmlevel regression analysis. Each month, three-year post-ipo monthly excess returns are regressed on the three Fama-French factors for all firms issuing that month and the equal-weighted average alpha is saved. The equal-weighted cumulative sum of this time series of abnormal returns is then computed. The resulting cumulative abnormal returns (CARs) are displayed graphically in Figure 3. {Figure 3 about here} The result is quite revealing. As the figure shows, new issues have consistently negative abnormal performance in the early part of the sample period, up to about The CAR over this period is %. This levels off over , where the CAR is an insignificant 2.41%. The graph up to this point is roughly consistent with a similar graph in Brav, Geczy, and Gompers (2000) whose sample ends in 1995, but after 1995 we begin to see a significant change. During the last part of the sample period, , we see substantial positive abnormal performance. The CAR over this later period is 50.00%. This graph helps to explain the finding 12

15 of no abnormal performance on average from the portfolio times series regressions over the entire sample period. We note that most other studies on the new issues puzzle which find average underperformance of IPOs do not contain data for this later period. {Figure 4 about here} Figure 4 displays the cumulative abnormal returns for the matching firm sample. The CARs are negative across the entire sample period, emphasizing that small growth firms have done poorly over the last couple of decades on a risk-adjusted basis. The performance begins to improve after about However, the CARS do not become positive in the later years as in the case of the IPO sample. Overall, the matching firm CARs indicate that the positive abnormal performance of the IPOs over is not due to a secular shift in the performance of small growth firms in general. {Table V about here} Table V displays Fama-French three-factor regression results for the entire sample period as well as the three sub-periods. Coefficients are the time-series averages of monthly regression coefficients, and due to the obvious serial correlation in the estimates, Newey-West t-statistics are given in parentheses. Not surprisingly, the average monthly alpha over the entire sample period of 0.09% is seen to be small and insignificantly different from zero. However, for the subsample the average alpha is -0.31% (t-statistic = -3.40) and is similar to the figure of % (t-statistic = -1.23) reported by Ritter and Welch (2002) who examine the period The average alpha over is an insignificant 0.02%. For the later period , the average alpha of 0.54% is both positive and significant (t-statistic = 2.79). This is surprising 13

16 and represents a significant departure from the underperformance documented in prior studies for earlier sample periods. 4 Informed by the results of this analysis, we partitioned the sample into two sub-periods in Table III. Recall there we saw that the three-year buy-and-hold IPO return versus the CRSP value-weighted index over the sub-period is an insignificant -4.7% (t-statistic = ). In conjunction with our results from matching firm buy-and-hold returns and our regression analysis, we conclude that the new issues puzzle has disappeared in recent years. Most existing studies of the new issues puzzle do not include data from this recent time period. An exception is Lyandres, Sun, and Zhang (2008) whose sample extends back to 1970 and ends in Their long sample may mask the lack of underperformance of IPOs in the later time period. Ritter and Welch (2002) study the sample period and report monthly benchmark-adjusted underperformance of 0.21%, but with an insignificant t-statistic. Based on the limited number of years, we are reluctant to declare the disappearance of IPO underperformance a long-term trend. However, as more data becomes available we should gain a better understanding of whether a fundamental shift in IPO performance has occurred. {Table VI about here} D. Robustness Tests We next conduct robustness tests by analyzing the sample based on firm size, and then by including additional control variables in our regression model. In Table VI we report regression results after dividing the sample into small and large firms. Small firms are those whose market capitalization is below the median in our sample. First, we note that the Fama-French threefactor alphas are not significant for the IPO sample. Also, there is a noticeable decrease in the 4 An exception is Gompers and Lerner (2003) who study a sample of IPOs over the pre-nasdaq period and find some evidence of outperformance by IPOs. 14

17 SMB factor loadings for the large firms as opposed to the small firms, as we might expect. Small issuing firms are seen to have insignificant loadings on HML, whereas the HML factor loadings are significantly negative for large issuing firms. Momentum and liquidity are negative and significant for both small and large issuing firms, consistent with results for the full sample. For the large matched firms, the Fama-French three-factor alpha is -0.35% and significant. Momentum is seen to be negative and significant for small and large non-issuers. Finally, we note that skewness and the investment factor are not significant in any of the regressions from the partitioned samples. We next re-estimate Model IV from Tables IV and VI for both the IPOs and the matched sample while adding three control variables. These variables are the log of the average market value of the firms in each calendar time portfolio (LNVAL) and their average market-to-book ratios (MTB), both estimated at the time of the IPO, and the average standard deviation of the daily return from market day IPO +50 through market day IPO +150 (STD). These variables, among others, have been argued to either influence the timing of equity offerings or their subsequent returns and may be instrumental in explaining long-run returns. 5 While STD is estimated during the buy-and-hold excess return sampling period, it only constitutes less than 15% overlap. We use three-year buy-hold returns for the dependent variable. In non-tabulated results, we find that none of the control variables are significant. Overall, the seven-factor model explains IPO returns at least as well as it does for similar seasoned firms. E. Predicting Firm Failure The failure rate for new public firms tends to be higher than for the general population and varies over time. We use the CRSP delisting codes to identify IPO firm failure. Figure 5 5 See for example, Baker and Wurgler (2002), Jain and Kini (1994), and Johnson and Miller (1989). Loughran and Ritter (1995) also use firm size and book-to-market in their Table VII to help explain IPO long-run returns. 15

18 shows the percentage of firms by cohort year delisting within five years of the IPO. The figure reveals relatively high failure rates for IPOs in the mid eighties and the late nineties. In particular, in the late nineties during the internet bubble, the five-year failure rate hovers around 20%. After that, the failure rate returns to historic lows, reflecting higher quality firms going public. Our final analysis predicts firm failure using estimated factor coefficients from the sevenfactor model in equation (1). Any firm that was dropped from CRSP in the first three years with a code equal to or greater than 500 was given a value of 2. Any firm that was dropped from CRSP in the 4 th or 5 th year with a code equal to or greater than 500 was given a value of 1. All other firms were assigned a value of zero. We use a logistic regression model where the independent variables are the same as those in the previous analysis, and Model I is without the three control variables. The results are presented in Table VII. {Table VII about here} The significant intercepts in three of the four models suggest that there is considerable information about delisting (failure) that is not captured in the seven-factor model. The only predicted factor that is consistently significant for the IPOs is investment. However, addition of the control variables, even though they were estimated at least four years before delisting, improves the 2 considerably. The coefficient estimates for firm size and volatility are seen to be significant for both the IPO sample and the matching sample in predicting delisting. Hence, smaller more volatile firms that tend to have sensitivity to the investment factor are more likely to fail. 16

19 IV. Conclusion It has become a stylized fact in finance that IPOs underperform the market during the first 36 months after issuance. We study the long-run performance of 6,686 IPOs spanning the period using three alternative measures. We find that the new issues puzzle disappears in a simple Fama-French three-factor framework. IPOs do not underperform in the aftermarket on a risk-adjusted basis. We also find that, in recent years, IPOs do not underperform a similar sample of firms that are matched on size and book-to-market. This result stands in stark contrast to studies by Ritter (1991), Loughran and Ritter (1995), and Lyandres, Sun, and Zhang (2008), among others. Long-run performance results thus appear to be highly sensitive to both the time period studied and the particular matching sample that is chosen. We find evidence that IPO underperformance is concentrated in the 1980 s and early 1990 s, and IPO s either perform the same as the market, or outperform on a risk-adjusted basis, during The phenomenon of IPO underperformance appears to have faded in recent years. We attribute this development to two recent phenomena: firms are seeking to access the capital markets sooner in their life cycle, and there has accordingly been a shift to alternative offering mechanisms, such as reverse mergers and sellouts. In recent years, more firms which are younger, highly information asymmetric, and riskier are choosing to go public using these alternative mechanisms. We believe that this has resulted in a sample of IPO firms that is more likely to perform well. We also find that additional factors for momentum, liquidity, and investment all help to explain IPO aftermarket returns. The significance of the liquidity factor suggests that the performance of heavily traded IPO stocks tends to suffer, and the momentum factor shows that investors sacrifice some return due to negative momentum exposure. Finally, unlike momentum 17

20 and liquidity, the explanatory power of the investment factor of Lyandres, Sun, and Zhang (2008) for IPO returns is not shared by the matched sample firms and suggests that this factor is capturing an element of risk that is unique to IPO firms. In sum, our results present fresh evidence on IPO performance, shedding light on additional factors that help to explain IPO returns, while challenging the established paradigm that new issues underperform. 18

21 References Baker, M., and J. Wurgler (2002) Market timing and capital structure, Journal of Finance 57, Brav, A., and P. A. Gompers (1997) Myth or reality? The long-run underperformance of initial public offerings: evidence from venture and nonventure capital-backed companies, Journal of Finance 52, Brav, A., C. Geczy, and P. A. Gompers (2000) Is the abnormal return following equity issuances anomalous? Journal of Financial Economics 56, Carhart, M. (1997) On persistence in mutual fund performance, Journal of Finance 52, Eckbo, E. and O. Norli (2005) Liquidity risk, leverage and long-run IPO returns, Journal of Corporate Finance 11, Fama, E. (1998) Market efficiency, long-term returns, and behavioral finance, Journal of Financial Economics 49, Fama, E., and K. French (1993) Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, Fama, E., and K. French (2004), New lists and seasoned firms: Fundamentals and survival rates, Journal of Financial Economics, Vol. 73, pp Fama, E., and J. MacBeth (1973) Risk, return and equilibrium: Empirical tests, Journal of Political Economy 81, Floros, I., and T. Sapp (2009) Shell Games: On the Stock Price Performance of Shell Companies, Iowa State University Working Paper. Gompers, P., and J. Lerner (2003) The really long-run performance of initial public offerings: The pre-nasdaq evidence, Journal of Finance 58, Harvey, C., and A. Siddique (2000) Conditional skewness in asset pricing tests, Journal of Finance 55, Jain, B., and O. Kini (1994). The post-issue operating performance of IPO firms, Journal of Finance 49, Johnson, J., and R. Miller (1989) Investment banker prestige and the underpricing of initial public offerings, Financial Management 17. Loughran, T. (1993) NYSE vs. NASDAQ returns: Market microstructure or the poor performance of IPOs? Journal of Financial Economics 33,

22 Loughran, T., and J. Ritter (1995) The new issues puzzle, Journal of Finance 50, Loughran, T., and J. Ritter (2000) Uniformly least powerful tests of market efficiency, Journal of Financial Economics 55, Lyandres, E., Le Sun, and Lu Zhang (2008) The new issues puzzle: Testing the investment based explanation, The Review of Financial Studies 21, Lyon, J., B. Barber and C. Tsae (1999) Improved methods for tests of long-run abnormal stock returns, Journal of Finance 54, Mitchell, M. L., and E. Stafford (2000) Managerial decision making and long-term stock price performance, Journal of Business 73, Newey, W. K., and K. D. West (1987), A Simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix, Econometrica 55, Nguyen, D., S. Mishara, A. Prakash, and D. K. Ghosh (2007) Liquidity and asset pricing under the three-moment CAPM paradigm, Journal of Financial Research 30, Ritter, J. (1991) The long-run performance of initial public offerings, Journal of Finance 46, Ritter, J. and I. Welch. (2002) A review of IPO activity, pricing and allocations," Journal of Finance 57,

23 Table I Descriptive Statistics The table presents statistics for 6,686 IPO firms that went public from 1981 through 2005 as well as firms matched to the IPO sample based on size and book-to-market. IPO firm market value at the offering ($ millions) IPO firm market value at the offering ($ millions, inflation adjusted) First day run-up from offer price to closing bid (%) Gross proceeds from the offering ($ millions) Gross proceeds from the offering inflation adjusted ($ millions) Age of the IPO firm at the offering (years) Offer price ($) Variables Mean Median Standard Deviation Market-to-book ratio of the IPO firm at the time of the IPO Standard deviation of the IPO firm (Days IPO +50 to +150) (%) IPO 3 year buy-and-hold return (%) IPO buy-and-hold return for years 4 & 5 (%) IPO 5 year buy-and-hold return (%) Matched firm market value at the time of the IPO ($ millions) Market-to-book ratio of the matched firm at the time of the IPO

24 Table II The Long-Run Performance of IPOs by Cohort Year The table shows the buy-and-hold returns and cumulative abnormal returns (CARs) of 6,686 new equity issues with offer dates between 1981 and Each IPO is matched with a seasoned firm according to industry, book-tomarket, and size. CARs are measured as the one-year cumulative alphas from monthly firm-level Fama-French three-factor regressions. Wealth relatives are computed as 1 RiT / 1 R mt where R it is the IPO cohort return, and R mt is the return on a matching firm cohort over the same period. 3-Year Mean Buy-and- Hold Returns (%) 1-Year CARs (%) Cohort Number Matching Wealth Matching Wealth Year of IPOs IPOs Firms Relative IPOs Firms Relative ,

25 Table III Difference Tests The table presents the mean and t-statistics for the difference in 6,686 IPOs compared to the contemporaneous value-weighted CRSP market return and compared to seasoned firms matched to each IPO firm by industry, size, and book-to-market. Mean differences are in percent. T-statistics are in parentheses. Variables IPO average monthly portfolio return less the market return IPO average monthly portfolio return less the matched sample return IPO 3-year buy-and-hold return less the market return IPO buy-and-hold return for years 4 & 5 less the market return IPO 3-year buy-and-hold return less the matched sample return IPO buy-and-hold return for years 4 & 5 less the matched sample return IPO 3-year buy-and-hold return less the matched sample return (minus ) (-0.17) (-0.46) (-7.54) (-1.98) (-3.65) (-0.96) (-1.88) (-0.68) (0.93) (-7.52) (-3.22) (-2.16) (-0.48) (-2.16) 0.27 (0.56) (-0.49) (-1.44) (2.93) (-4.46) (-1.22) (2.24) 23

26 Table IV Determinants of IPO Monthly Returns This table presents regression coefficients where the dependent variable is the equal weighted monthly percentage excess return for a portfolio of IPOs that have gone public in the prior 36 months. All regressions use 335 monthly observations. The independent variables are market excess returns (RMRF), a size factor (SMB), a book-to-market factor (HML), a momentum factor (MOM), a liquidity factor (LIQ), a co-skewness measure (SKEW), and an investment factor (INVEST). The LIQ factor is the premium on a portfolio of low turnover stocks versus high turnover stocks. SKEW is the squared excess return on the market, less its mean. The INVEST factor is the premium on a portfolio of low-investment stocks versus high investment stocks. Each IPO is matched with a seasoned firm according to industry, book-to-market, and size. Newey-West t-statistics are in parentheses. IPO Sample Matched Sample IPO Matched Explanatory Variables I II III IV I II III IV I IV Intercept (Alpha) RMRF SMB HML MOM LIQ SKEW INVEST (-0.29) 1.01 (23.21) (6.56) (-3.30) (0.56) (23.32) (7.37) (-2.00) (-4.16) (0.88) (21.74) (8.67) (-4.63) (-3.44) (1.26) (18.66) (7.82) (-2.69) (-2.77) (-2.04) (-1.56) (-2.84) (-1.86) (32.10) (8.70) (3.51) (-1.35) (33.70) (9.56) (4.04) (-1.84) (0.14) (31.46) (13.49) (3.79) (-5.86) (0.09) (28.77) (13.94) (4.58) (-6.42) (-2.07) (-1.36) (-0.40) (0.99) (0.16) (0.31) (-8.74) (1.11) (-0.60) (-0.09) (-6.25) (1.77) (-1.53) (-0.60) (-2.38) Adj. R

27 Table V IPO Abnormal Returns by Time Period The table presents the average coefficients from 296 regressions of monthly returns from January 1981 through December 2005, as well as for sub-periods. The dependent variable is three-year post-ipo monthly excess returns for all firms that have gone public in a given month. The independent variables are market excess returns (RMRF), and factors for firm size (SMB) and book-to-market (HML). Four months are missing because there were no IPOs in those months. Newey-West t-statistics are in parentheses. Alpha (1.03) RMRF (39.63) SMB (19.14) HML (-3.23) Sample Period (-3.40) (36.54) (14.64) (-2.09) (0.24) (29.12) (22.50) (-2.60) (2.79) (20.51) (9.22) (-1.74) Average Adj. R

28 Table VI Determinants of IPO Returns by Firm Size This table presents regression coefficients where the dependent variable is the equal weighted monthly percentage return for a portfolio of IPOs that have gone public in the prior 36 months. All regressions use 335 monthly observations. The independent variables are market excess returns (RMRF), a size factor (SMB), a book-to-market factor (HML), a momentum factor (MOM), a liquidity factor (LIQ), a co-skewness measure (SKEW), and an investment factor (INVEST). The LIQ factor is the premium on a portfolio of low turnover stocks versus high turnover stocks. SKEW is the squared excess return on the market, less its mean. The INVEST factor is the premium on a portfolio of low-investment stocks versus high investment stocks. Each IPO is matched with a seasoned firm according to industry, book-to-market, and size. Small and large firms are separated at the median firm size. Newey-West t-statistics are in parentheses. Intercept RMRF SMB HML MOM LIQ SKEW INVEST IPO Sample Matched Sample Small Issuers Large Issuers Small Non-issuers Large Non-issuers I IV I IV I IV I IV (-1.26) (20.51) (7.87) (0.53) (-0.22) (14.94) (8.41) (1.16) (-2.01) (-2.16) ( -0.73) (-0.61) 0.15 (0.83) (25.23) (6.06) (-2.90) (2.63) (25.53) (7.24) (-3.82) (-4.15) (-2.24) (0.65) (-1.80) (-0.89) (23.35) (14.59) (2.50) (-0.29) (14.89) (11.81) (2.90) (-3.00) (-1.64) (-1.63) (0.72) (-3.03) (39.74) (8.57) (3.60) (-1.02) 1.02 (42.23) (13.72) (4.09) (-6.09) (-1.22) (0.40) (-1.16) Adj. R

29 Table VII Logistic Regressions: Predicting Firm Failure The table presents the coefficients from a logistic regression for categorical dependent variables for 6,686 firms conducting IPOs between January 1981 and December The dependent variable is two if the firm failed in the first three years and one if the company failed in the 4 th or 5 th year post IPO, respectively and zero otherwise. The independent variables are the estimated coefficients from the model depicted in Table IV * the following coincident factors: market excess returns (RMRF), small less large firm size (SMB), high minus low book-to-market (HML), a momentum (MOM), liquidity (LIQ), co-skewness (SKEW) and investment (INVEST). The variable LNVAL is the log of the value of the firm at the time of the IPO, MTB is the market-to-book ratio at the time of the offering and STD is the standard deviation of the daily return from days IPO +50 to IPO+150. p-values are in brackets. IPO Sample Matched Sample Explanatory Variables I II I II Intercept Predicted RMRF Predicted SMB Predicted HML Predicted MOM Predicted LIQ Predicted SKEW Predicted INVEST MTB LNVAL STD [0.11] [0.02] [0.17] [0.22] [0.34] Wald [0.66] [0.66] [0.07] [0.41] [0.41] [0.15] [0.66] [0.04] [0.39] [0.19] [0.78] [0.01] [0.34] [0.81] [0.17] [0.01] [0.45] [0.06] [0.61]

30 Figure 1 Annual IPO Volume The figure shows the annual distribution of 6,686 new equity issues with offer dates between 1981 and Number of IPOs Year 28

31 Figure 2 Cumulative Returns The figure shows the cumulative continuously compounded monthly returns over of the CRSP valueweighted market portfolio, a rolling sample of equal-weighted IPOs held for 36 months after issuance, the contemporaneous equal-weighted sample of matched firms, and the riskfree asset % Return (100s) Market IPOs Matches Riskfree 29

32 Figure 3 Calendar-Time IPO Cumulative Abnormal Returns The figure shows the cumulative abnormal returns from 296 monthly regressions of IPO returns on the three Fama- French factors. The dependent variable is the three-year post-ipo monthly excess returns for firms that have gone public in a given month. The abnormal return is defined as the regression intercept (alpha). Four months are missing because there were no IPOs in those months Cumulative Alpha (%) Year 30