Journal of Corporate Finance

Transcription

1 Journal of Corporate Finance 19 (2013) Contents lists available at SciVerse ScienceDirect Journal of Corporate Finance journal homepage: SEO timing and liquidity risk Ji-Chai Lin a, YiLin Wu b, a Department of Finance, E.J. Ourso College of Business Administration, Louisiana State University, Baton Rouge, LA , USA b Department of Economics, College of Social Sciences, National Taiwan University, 21 Hsu-Chow, Road, Taipei City, Taiwan, ROC article info abstract Article history: Received 24 April 2012 Received in revised form 8 September 2012 Accepted 17 September 2012 Available online 25 September 2012 JEL classification: G14 G32 Keywords: Liquidity risk Seasoned equity offerings SEO timing Liquidity beta timing Cost of equity capital We extend the market timing literature to show that SEO timing can be characterized by the dynamics of liquidity risk. That is, firms tend to issue when liquidity risk declines to the point where investors have least concern of the risk. In the absence of liquidity risk, market risk rises right before and then gradually falls afterwards, consistent with the Q-theory (Carlson et al., 2010). However, once we include liquidity risk factor into the model for expected returns, issuing firms' market risk behaves like that of matched non-issuers, suggesting an omitted risk factor problem in SEO studies that does not take into account the effect of liquidity risk on stock returns. Furthermore, there is no evidence of post-issue long-run underperformance. Our results imply that, instead of timing alpha (i.e., exploiting overpricing, as behavioral finance has suggested), issuing firms time liquidity beta to minimize their cost of equity capital. The liquidity beta timing is especially evident in large offer size issuers Elsevier B.V. All rights reserved. 1. Introduction Acharya and Pedersen (2005), Liu (2006, 2010), and Pastor and Stambaugh (2003) among others, 1 have shown that market liquidity is time-varying and that a firm's liquidity risk, which captures the sensitivity of its stock returns to shocks to market liquidity, plays a very important role in determining its expected return. Given the importance of liquidity risk in asset pricing, would firms take this important factor into consideration when making seasoned equity offering (SEO) decisions? 2 If so, what role would liquidity risk play in SEO timing and post-issue long-run stock returns? In this paper, we address these questions to help resolve the SEO timing debate. We thank Robin Chou, Gary Sanger, Lingling Wang, and seminar participants at Louisiana State University, National Taiwan University, National Central University, Lappeenranta University of Technology, Peking University, University of Vaasa, University of Electronic Science and Technology of China, and Yuan Ze University for helpful comments and Weimin Liu for making his mimicking liquidity factor available to us. We would like to offer special thanks to Managing Editor (Jeffry Netter) and an anonymous referee for their insightful comments. YiLin Wu is grateful to the National Science Council of Taiwan, R.O.C. for funding this project. Corresponding author. Tel.: ; fax: addresses: filin@lsu.edu (J.-C. Lin), yilinwu@ntu.edu.tw (Y. Wu). 1 See, for example, Amihud (2002), Bekaert et al. (2007), Eckbo and Norli (2005), Korajczyk and Sadka (2008), Sadka (2006), and Watanabe and Watanabe (2008). 2 In addition to academic literature, considerable financial press has emphasized the importance of liquidity in equity offering decision. For example, a senior managing director at Morgan Stanley Dean Witter commented that, following the Internet crash of mid-2000, there has been a flight to quality and a flight to liquidity in the NASDAQ equity market. The number of NASDAQ IPO was down by 30% and the number of NASDAQ was down by 40% from the second quarter of 1999 (Wall Street Journal; June 27, 2000) /$ see front matter 2012 Elsevier B.V. All rights reserved.

2 96 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) It is well known that firms tend to conduct when their stock prices are high, and that their post-issue long-run stock returns tend to be low. Eckbo et al. (2007) survey the literature and conclude that The debate about what causes the apparent ability of firms to time their equity issues to periods that are followed by low market returns is still inconclusive. Loughran and Ritter (1995), Baker and Wurgler (2002), and Baker and Stein (2004) argue that the stylized facts imply that firms exploit a window of opportunity by selling shares at when their shares are overvalued, which is followed by market corrections. As we will discuss in Section 2, many studies have challenged this behavioral explanation and proposed risk-based and investment-based explanations. In this study we add to this SEO timing debate by proposing an alternative, rational explanation in which liquidity risk plays a leading role for SEO timing. Specifically, we propose that firms tend to issue when liquidity risk declines to the point where investors have least concern of the risk, and that low liquidity risk is the main cause of low post-issue long-run stock returns. Furthermore, we argue that timing to periods that investors have low concerns of liquidity risk would also help issuing firms mitigate the adverse effect of SEO announcements in the secondary market, and reduce the offering price discount in the primary market. To test our hypothesis, we use calendar-time factor model portfolio regressions and employ Liu's (2006) liquidity-augmented CAPM (LCAPM), a parsimonious and seemingly powerful model, and show that it can better explain post-issue stock returns than the CAPM, Fama French three-factor model, and Lyandres et al. (2008) investment-augmented CAPM. Furthermore, we use Liu's LCAPM and apply Ibbotson's (1975) RATS (returns across time and securities) regression technique to capture time-varying systematic risks. 3 Our main findings are: (1) issuing firms' liquidity risk steadily declines to the lowest point prior to SEO filing and then rebounds in the filing month, suggesting that managers monitor the market for the lowest liquidity risk to occur, and when it does, they promptly file for ; (2) to a lesser extent, issuing firms' market risk also declines prior to SEO filing; (3) comparable non-issuers experience the same improvement in market risk and a lesser improvement in liquidity risk as issuing firms, suggesting that issuing firms are more likely to file for when market conditions have generally improved; (4) issuing firms' liquidity risk remains relatively low for two to three years, compared to their matched non-issuers; and (5) for issuing firms that exhibit more liquidity risk reduction, the markets react less negatively to their SEO announcements, and investors demand less offering price discounts at issuance. 4 We repeat our analysis across three SEO offer size groups (small, medium, and large) and find a particularly sharp and persistent decline in liquidity risk prior to SEO filing for the large offer size issuers. We also find that the reason that larger issuing firms show lower post-issue stock returns, as documented by Fama and French (2008) and Pontiff and Woodgate (2008), is largely because their investors face lower liquidity risk in the post-issue period. Based on the notion that firms with greater stock liquidity tend to have lower liquidity risk (see Acharya and Pedersen, 2005; Pastor and Stambaugh, 2003), for a robustness check, we examine pre- and post-seo stock liquidity to see whether changes in liquidity are consistent with our inferences from liquidity risk. An advantage of drawing inference from examining changes in stock liquidity surrounding the is that it does not require a specific asset pricing model. As expected, we find that substantial liquidity improvement occurs prior to SEO filing and that stock liquidity remains high following the. While it is understandable that an SEO could lead to more share turnover and higher stock liquidity since it increases the number of outstanding shares and enlarges investor base, our hypothesis highlights the importance of stock liquidity improvement that leads to SEO filing decisions. Indeed, we estimate probit regressions of SEO filing decisions and show that the likelihood of filing for an SEO in a given quarter (or a given year) is significantly related to the stock liquidity level in the previous month and the improvement in stock liquidity over the previous six months. Our results that liquidity risk explains long-run stock performance of SEO issuers are consistent with Eckbo and Norli's (2005) findings of a liquidity risk effect in new-issue stocks, and complement Butler and Wan's (2010) findings that liquidity risk explains long-run stock performance of debt issuers and Ben-Rephael et al.'s (2012) findings that the liquidity is higher in actual stock repurchase months. The evidence implies that U.S. equity markets are efficient in pricing issuing firms' shares, and provides a strong contrast to the behavioral explanation that firms time to exploit overpricing. Regarding the risk dynamics of SEO firms, Carlson et al. (2010) present an alternative theory. They argue that committing to future investments raises the market risk of issuing firms, and that as the firms use SEO proceeds to put down investment commitments over time, the risk gradually drops. We test their commitment-to-invest theory using the market model (without considering liquidity risk), and find similar results to theirs. Indeed, issuing firms' market beta increases substantially right before the and then gradually decreases, but is still significantly higher than that of comparable non-issuers in the post-issue period. However, once we include Liu's (2006) liquidity factor to the market model, issuing firms' market beta behaves just like that of their matched non-issuers during both pre-filing and post-issue periods. Our analysis suggests that using the market model to capture market risk is subject to the omitted variable problem. Since shocks to market liquidity affect stock returns and 3 In Ibbotson's (1975) RATS approach, security excess returns are regressed on the excess market return and liquidity risk factor for each event month t relative to the month of SEO filing and relative to SEO issuance. The RATS methodology calculates a different market beta and liquidity beta for each month relative to the SEO event. Therefore, if risk systematically changes surrounding, then the coefficients on the risk factors are allowed to change month by month to reflect such changes in risk. Ibbotson's (1975) RATS technique has been employed by several studies to infer changes in systematic risks. For examples, Ikenberry and Lakonishok (1993) use it to estimate time-varying betas surrounding proxy contests; Peyer and Vermaelen (2009) use it to estimate time-varying betas surrounding share repurchases; and Lin et al. (2009) use it to infer changes in risks following stock splits. 4 Since information asymmetry and information quality are important elements of the liquidity environment, our finding is consistent with that of Lee and Masulis (2009), who show that as the quality of issuing firms' financial accounting improves, SEO underwriting spreads and the negative announcement return decrease.

3 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Liu's liquidity factor is inversely related to market excess returns (e.g., liquidity risk premium tends to be high in bad markets), the market model, which omits the liquidity factor, would lead to a biased estimate of market risk and a biased inference. The results suggest that our liquidity risk timing hypothesis is a better alternative for explaining the risk dynamics surrounding the than the commitment-to-invest theory. The remainder of the paper is organized as follows. Section 2 compares our SEO timing hypothesis to existing theories. Section 3 describes our selection of SEO sample firms and their matched firms and Section 4 discusses Liu's LCAPM. Using the calendar-time portfolio analysis, Section 5 shows that the LCAPM is a useful model to explain post-issue sock returns. Section 6 presents the dynamics of liquidity risk and links it to SEO timing. Section 7 addresses whether equity markets price the liquidity risk reduction in SEO announcement returns and in offering price discounts. Finally, Section 8 contains our concluding remarks. 2. Hypothesis and literature review Motivated by studies on asset pricing with liquidity risk, we hypothesize that issuing firms tend to file for when their liquidity risk declines to the point where investors have low concerns of the risk, and that low liquidity risk is the main cause for low post-issue stock returns. Our hypothesis emphasizes that managers monitor the markets to time to periods in which investors have low concerns of liquidity risk, not only at issuance but also for the long run. This SEO timing is related to the IPO timing proposed by Pastor and Veronesi (2005) that ties IPO waves to improvements in market conditions and declines in expected market returns. Since market liquidity is pro-cyclical (Brunnermeier and Pedersen, 2009), improvements in market conditions could lead to a market-wide decline in liquidity risk, suggesting that improvements in market conditions and liquidity environment also play an important role in our SEO timing hypothesis. Our hypothesis emphasizes that the pre-filing improvement in issuing firms' liquidity environment contributes to lowering their liquidity risk, which exerts significant influences on the SEO decisions. Based on examining both Amihud's (2002) price impact measure and Liu's (2006)non-trading illiquidity measure from three years before to three years after, we find that improvements in the issuing firms' liquidity largely occur before the SEO filing. Our findings are consistent with that of Tripathy and Rao (1992), who observe declines in bid ask spreads of NASDAQ SEO firms in the pre-announcement period. In comparison with studies by Denis and Kadlec (1994), Eckbo and Norli (2005), andlease et al. (1991), who show that lead to more share turnover, more trades per day, and lower bid ask spread, we highlight the role of liquidity improvement on the SEO decisions. Our hypothesis is complementary to the time-varying adverse selection theories of Bayless and Chaplinsky (1996), Choe et al. (1993), and Korajczyk et al. (1991). However, the focus of our study differs from theirs while they examine the relation between time-varying asymmetric information and SEO announcement effects, our analysis mainly focuses on the dynamics of liquidity risk before and after and shows that liquidity risk timing allows issuing firms to raise capital at a lower cost of capital. Our hypothesis is different from the leverage risk reduction explanation. Eckbo et al. (2000) argue that an equity offering lowers the firm's financial leverage, 5 which reduces its stock's sensitivities to leverage-related macroeconomic risks and reduces its stock expected return as well. If induce any changes in leverage, they should affect SEO firms' market beta. However, inconsistent with their leverage risk reduction hypothesis, our analyses in Section 6 reveal that issuing firms and their matched non-issuers have qualitatively the same market betas before and after. Carlson et al. (2006, 2010), and Lyandres et al. (2008) propose a Q-theory based explanation that the SEO proceeds are used to fund investments that convert growth options into assets in place. Since these new assets are less risky than the growth options they replace, they predict that the risks and expected returns of SEO firms must decrease in the post-issue period. They do not consider the role of liquidity risk. Hence, it becomes an empirical question: Which factor can better explain the SEO risk dynamics? We address this question in Section 5. Finally, our hypothesis is different from the behavioral explanation that firms time to exploit overpricing, as proposed by Baker and Stein (2004), Baker and Wurgler (2002) and Loughran and Ritter (1995). Instead, our SEO timing hypothesis argues that, in markets with time-varying risks, issuing firms time their to periods of low liquidity betas. In the sections that follow, we will explain how we reach this inference, starting from the data we use in analysis. 3. Data 3.1. Selection of sample and control firms We describe our sample selection process as follows. To be included in our initial SEO sample, a firm must: (1) have a firm commitment, non-shelf registered offering undertaken during the sample period from January 1982 to December 2006, and have a non-missing filing date in the Thomson SDC Platinum database, (2) have four-digit SIC codes outside the intervals (financial companies), 6 (3) be listed on the NYSE, NASDAQ, or Amex, and have data for at least 12 monthly returns available on 5 In contrast, Leary and Roberts (2005) show that the effect of equity issuances on firms' leverage is removed within two years by subsequent debt issuances and Hovakimian (2004) documents that equity issues have no significant effect on capital structure. 6 Similar to Brav et al. (2000), Eckbo et al. (2000), and Lyandres et al. (2008), we include utilities (SIC ) in our sample.

4 98 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) the CRSP files during the three-year period prior to SEO filing and at least 12 monthly returns during the three-year period after SEO issuance, (4) be incorporated in the U.S. according to Compustat, and (5) offer only pure primary or combinations of primary and secondary issuances, but pure secondary offerings are excluded. 7 This procedure creates an initial sample of 5374 over the period. To assess whether the risk dynamics and lower post-issue stock returns are unique to SEO firms, we impose the following requirements in identifying control firms. First, we restrict our selection to the subset of non-financial firms on the CRSP stock return and the Compustat annual industrial files, which have no initial public offerings, with annual net share issuance (repurchase), or annual net debt issuance (retirement) in the range of 5% to 5% during any of the three-year period before the SEO filing. We impose this restriction because Butler and Wan (2010) and Pontiff and Woodgate (2008) show that equity issuances and debt issuances tend to be followed by periods of abnormally low returns while Affleck-Graves and Miller (2003) and Ikenberry et al. (1995) find that share repurchases and debt prepayments tend to be followed by periods of abnormally high returns. 8 Second, following Brav et al. (2000) and Gibson et al. (2004) that yearly portfolio rankings may be stale that could lead to incorrect matching, we choose comparable non-issuing firms based on quarterly portfolio rankings in terms of firm size (equity market capitalization), return momentum, and book-to-market equity ratio. Starting from March 31, 1981 to December 31, 2006, we form 125 size/momentum/bm-sorted portfolios for each quarter. 9 Third, we also take stock liquidity into consideration in selecting a matched firm for each SEO firm. This is because Butler et al. (2005) show that the cost of raising equity capital is inversely related to stock liquidity. Amihud et al. (2005) suggest that stock illiquidity raises the required rate of returns. Specifically, we match a non-issuing firm with each SEO issuer from the same size/ momentum/bm portfolio as the SEO firm in the quarter prior to the SEO announcement. The matching criteria are that a non-issuing firm must: (1) have at least 12 monthly returns during the three years prior to the SEO filing and at least 12 monthly returns during the three years after the SEO issuance, (2) have not been selected as a matched firm for any other sample firm during the event period (to avoid possible bias in test statistics due to dependence from the same matched firm), (3) have a pre-seo liquidity measure, PreLM12, closest to that of the SEO firm. The PreLM12 is Liu's (2006) liquidity measure: the standardized turnover-adjusted number of days with zero trading volume over a 12-month period estimated over months 18 to By using this matching procedure, we are able to individually match 5312, or 98.8% of the in the initial sample, with a non-duplicate control firm. Finally, to avoid unduly influences of low-priced stocks on illiquidity and liquidity risk and to avoid other potential problems associated with low-priced stocks (Kothari and Warner (2007)), we eliminate 174 firms with an offer price of less than three dollars. 11 Thus, our final sample consists of 5138 over the period Sample firm characteristics Table 1 reports summary statistics on size, momentum, BM, and PreLM12 for our SEO sample and control sample prior to SEO filing. The average issuing firm has a firm size (market value of equity) of about $1.5 billion, a B/M (ratio of book equity to market equity) of 0.64, a pre-seo liquidity measure over months 18 to 7, PreLM12, of 7.95, and a stock price of $23.01 per share. On average, their firm size, B/M, stock liquidity, and stock price level are similar to those of their matched counterparts. Consistent with the SEO literature, the average issuing firm has a significant stock price run-up of 43.3% over a six month period prior to the SEO filing. This price momentum is significantly higher than that of the control sample. Thus, except for momentum, our matching procedure performs quite well in selecting non-issuers with firm characteristics comparable to those of issuing firms. Also reported in Table 1 are firm characteristics that may affect stock returns, as Cooper et al. (2008) and Fama and French (2008) show. On average (and on the median), our SEO sample is associated with higher asset growth rate, higher leverage ratio, greater net share issuance, and greater net debt issuance during each of the prior three-year period than the control sample. This is largely by construction, however, since our control sample excludes firms that have experienced annual net share issuance or annual net debt issuance that exceeds 5% during any of the three-year period before the SEO filing. 7 Here and throughout the paper, SEO proceeds refer to cash raised by the firm and not by stockholders who simultaneously sell shares. 8 We thank a referee for the suggestion on the construction of the matched sample. We note, though, that our results are not sensitive to the matching procedure that includes or excludes observations with IPOs, annual net share issuance, or annual net debt issuance that exceeds 5% or below 5% during any of the three-year period before the SEO filing. 9 Specifically, we first sort NYSE, Amex, and NASDAQ firms by their market equity into five size groups (the size breakpoints are the NYSE market equity quintiles in that quarter) excluding firms with IPOs, annual net share issuance, or annual net debt issuance that exceeds 5% or below 5% during any of the threeyear period before the SEO filing. Within each size group, we further sort firms by their preceding six-month returns into five groups (the return breakpoints are NYSE return quintiles in that size group), resulting in a total of 25 portfolios. We calculate the preceding six-month return through the end of the month before quarter end. For sorting on six-month return momentum, we require that each firm has at least three monthly returns (out of the preceding six) available on the CRSP. Then, within each size/momentum portfolio, firms are further sorted by their book-to-market equity ratio into five groups (again the breakpoints are NYSE BE/ME quintiles in that portfolio). Following Fama and French (1993), we compute BE/ME using ME at the end of the previous year and BE at the last fiscal year end in the previous year. This procedure forms 125 size/momentum/bm-sorted portfolios for each quarter. 10 We find that issuing firms exhibit significant liquidity improvements over months 6 to 1 relative to SEO filing, which is the reason for choosing the time frame over months 18 to The results of including low-priced stocks (not reported) are similar to the results presented in this paper. 12 There are 5138 conducted by 3046 industrial firms. Among them, 1943 firms (63.8%) have one SEO, 627 firms (20.6%) have two, and 476 firms (15.6%) have more than two over the period.

5 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Table 1 SEO firms' and their matched non-issuers' firm characteristics, This table reports the average and the median of firm characteristics for 5138 and their matched non-issuers during the period. The last two columns show the p-value for the t-test (Kruskal Wallis test) on the mean (median) of the differences between the issuers and their matched non-issuers. We restrict our selection of control firms to the subset of non-financial firms on the CRSP stock return and the Compustat annual industrial files, which have no initial public offerings, with annual net share issuance (repurchase) or annual net debt issuance (retirement) in the range of 5% to 5% during any of the three-year period before the SEO. We choose comparable non-issuing firms based on quarterly portfolio rankings in terms of firm size (equity market capitalization), return momentum, book-to-market equity ratio, and pre-seo liquidity. Size is the equity market capitalization measured in the quarter prior to the SEO filing, Momentum is the preceding six-month stock return calculated through the end of the month before filing quarter end, B/M uses market value of equity at December of t 1 and book value of equity for the fiscal year end in t 1, and PreLM12 is Liu's (2006) LM12: the standardized turnover-adjusted number of days with zero trading volume over a 12-month period estimated over months 18 to 7. In addition, this table reports Price, the closing price or bid/ask average from the CRSP one day prior to SEO filing, and Asset growth, the percentage change in total assets from fiscal year end in t 2 to fiscal year end in year t 1. To compute asset growth, a firm must have nonzero total assets in both t 1 and t 2. Leverage is debt over total assets, where debt is the book value of long-term debt plus short-term debt in year t 1. We identify annual net share issuance (repurchase) during any of the three-year period before the SEO filing. For example, Net EQ 1 is the natural log of the ratio of the split-adjusted outstanding shares at the fiscal year end in t 1 divided by the split adjusted outstanding shares at the fiscal year end in t 2. The split adjusted outstanding shares is Compustat shares outstanding (CSHO) times the Compustat adjustment factor (ADJEX_C). To compute this measure, a firm must have nonzero split adjusted outstanding shares in both t 1 and t 2. We identify annual net debt issuance (retirement) during any of the three-year period before the SEO filing. For example, Net DE 1 is the change in the book value of debt from fiscal year end in t 2 tofiscalyearendint 1 over total assets in t 2. Issuers Matched Issuers vs. matched Mean Median Mean Median Mean [p-value] Median [p-value] Size (million $) [0.656] [0.381] Momentum [0.000] [0.000] B/M [0.841] [0.357] PreLM [0.827] [0.224] Price [0.903] [0.021] Asset growth (%) [0.000] [0.000] Leverage [0.000] [0.000] Net EQ 1 (%) [0.000] [0.000] Net EQ 2 (%) [0.000] [0.000] Net EQ 3 (%) [0.000] [0.000] Net DE 1 (%) [0.000] [0.000] Net DE 2 (%) [0.000] [0.000] Net DE 3 (%) [0.000] [0.000] Table 2 Offering characteristics of, This table reports summary statistics of offering characteristics for the 5138 in our sample during the period. The SEO offering characteristics are collected from the SDC database; Gross proceeds/total asset 1 is the total amount raised from over total assets for the fiscal year end in t 1; Block size is defined as shares offered/(shares offered+outstanding shares), where outstanding shares are measured on the announcement date or, if unreported, at the end of the quarter prior to the announcement. P 1 /P 0 is offering price discount, where P 1 is the share price one day prior to the issuance date and P 0 is the offering price; offering price discount is Winsorized at 1% in either tail. Mean Median Std. deviation Gross proceeds (million $)] Gross proceeds/total asset 1 (%) Block size (%) Offer price Offer price/per share book value of asset Investment banking fees (%) P 1 /P Table 2 reports summary statistics of offering characteristics. In terms of offering size, the average amount of SEO gross proceeds is $111.9 million; on average, the gross proceeds represent about 42.0% of the book value of assets measured at the previous fiscal year end; and the average block size accounts for 14.2% of the outstanding shares. The mean offer price is $24.9, which is close to 2.3 times of per share book value of assets at the previous fiscal year end. The average investment banking fee is around 5.4% of the gross proceeds, and the average offering price discount is 3.5%, 13 relative to the closing price on the day before the offering. In our sample, the sum of mean (median) investment banking fees and the offering price discount is about 8.9% (7.4%). Thus, in line with previous empirical studies, our data also show that SEO flotation costs are non-trivial. We turn next to the asset pricing model we will use to capture systematic risks of SEO firms. 13 We Winsorize the offering price discounts of our sample at 1% and 99% to mitigate the problem associated with outliers. Without the Winsorization, the average price discount for an offering is 4.0%. See Altinkili and Hansen (2003) and Corwin (2003) for discussions on the SEO pricing discount. For example, Corwin (2003) reports that in the 1980s, the average SEO pricing discount is 1.15%, while in the , it averaged 2.92%. He observes a rise in the average discount of.

6 100 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Liu's (2006) liquidity-augmented CAPM We mainly use Liu's (2006) liquidity-augmented CAPM (LCAPM), which is parsimonious with only two risk factors market risk and liquidity risk. Liu (2006) first proposes a liquidity measure, LM12, the standardized turnover-adjusted number of days with zero trading volume over the prior 12 months. 14 As Lin et al. (2009) note, Liu's LM12 captures the notion that the greater incidence of no trading implies higher latent trading costs and that non-trading reflects illiquidity. Liu demonstrates that while LM12 is correlated with the conventional liquidity measures, LM12 can better capture the liquidity premium in asset pricing. To capture shocks to market liquidity, Liu constructs a liquidity factor, LIQ, as the return difference between a low-liquidity portfolio (containing stocks with high LM12) and a high-liquidity portfolio (containing stocks with low LM12). The construction of LIQ, a zero-cost portfolio, is similar to Fama and French's (1993) size factor (SMB), and B/M factor (HML). With LIQ, Liu (2006) develops the LCAPM in which the expected excess return on stock i at time t can be expressed as h E R i;t r f ;t ¼ β m;i E R m;t r f ;t i þ β liq;i ELIQ ð t Þ; where E(R m,t ) is the expected return of the market portfolio at time t; E(LIQ t ) is the expected value of the mimicking liquidity factor, LIQ; and β m,i and β liq,i are firm i's market and liquidity betas, respectively. Thus, only the two beta risks matter in determining stock i's expected excess return. In asset pricing tests, Liu shows that the LCAPM can explain the anomalies associated with firm size, book-to-market, cash-flow-to-price, earnings-to-price, dividend yield, and long-run price reversals. Given its explanatory power on these anomalies, we choose the LCAPM to explain the pricing of SEO firms' common stocks and compare its performance to that under the CAPM, Fama French three-factor model (FF3FM), and Lyandres et al.'s (2008) investment-augmented CAPM (INVCAPM). Lyandres et al. show that SEO firms experience significant underperformance under the CAPM and FF3FM, and that the underperformance disappears when they augment the CAPM with an investment factor (INVF). 15 The premise of their INVCAPM is that firms facing lower cost of capital tend to make more investments and that issuing firms tend to make more investments following the than their comparable non-issuers. The premise suggests that issuing firms' factor loadings on the investment factor should be negative and could help explain the expected returns on their common stocks. It is plausible that firms with lower liquidity risk face lower cost of capital and make more investments. Hence, the investment-based explanation and the liquidity risk explanation for asset pricing may not be mutually exclusive. In fact, the correlation between Liu's (2006) liquidity factor LIQ and Lyandres et al.'s (2008) investment factor INVF is very significant at over the period. However, as we will demonstrate below, one can compare the relative importance of LIQ and INVF in explaining issuing firms' stock performance and show which factor is more relevant in pricing SEO stocks. 5. Post-issue long-run stock performance Fama (1998) and Mitchell and Stafford (2000) strongly advocate a monthly calendar-time portfolio approach for measuring long-term abnormal performance. In this section we follow their suggestion to use calendar-time factor model portfolio regressions to examine how well Liu's (2006) LCAPM explains post-issue stock returns, relative to the CAPM, FF3FM, and INVCAPM. After showing that the LCAPM is a useful model to explain the pricing of SEO firms' common stocks, we then use it in the next section to illustrate that we can characterize the SEO timing by the dynamics of liquidity risk The calendar-time portfolio approach We first construct two calendar-time portfolios each month, one equal-weighted and the other value-weighted, which consist of firms that have conducted during the 36 months prior to the month of portfolio formation. The portfolio constructions 14 Liu (2006) argues that conventional liquidity measures, such as trading volume, share turnover, bid ask spread, and the price-impact measures of Amihud (2002) and Pastor and Stambaugh (2003), have certain drawbacks. Specifically, the conventional liquidity measures (1) fail to capture the multi-dimensional properties of liquidity, (2) do not reflect the illiquidity of non-trading, and (3) fail to take into consideration the endogeneity of the trading decision as a function of trading costs. Liu (2006) formulates LM12 as 1= 12 month turnover LM12 ¼ Number of zero daily volumes in prior 12 months þ ð Þ Def lator NoTD ; where 12-month turnover is the stock's turnover in the prior 12 months, calculated as the sum of daily turnover over the prior 12 months; daily turnover is the ratio of the number of shares traded on a day to the number of outstanding shares at the end of the day, NoTD is the total number of the trading days in the market 1= 12 month turnover over the prior 12 months, and Deflator is chosen such that 0b ð Þ b1 for all sample stocks (for example, Liu chooses a deflator of 11,000 in Def lator constructing his LM12). Liu (2006) notes that LMx uses the pure number of zero daily trading volumes over the prior x months to identify the least liquid stocks, but it relies on turnover to distinguish the most liquid among frequently traded stocks as classified by the pure number of zero trading volumes. 15 The way Lyandres et al. (2008) create the investment factor INVF is as follows: in June of each year, they sort all stocks in ascending order independently on size, book-to-market, and investment-to-assets, and classify them into three groups, the top 30%, the medium 40%, and the bottom 30%. They define the investment factor as the average returns of the nine low-investment portfolios minus the average returns of the nine high-investment portfolios.

7 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Table 3 Calendar-time factor model portfolio regressions of SEO firms. This table reports calendar-time portfolio regression results under the CAPM, Fama French three-factor model (FF3FM), Liu's (2006) liquidity-augmented CAPM (LCAPM), Lyandres et al. (2008) investment-augmented CAPM (INVCAPM), and CAPM with Liu's liquidity risk factor (LIQ) and Lyandres et al. (2008) investment factor (INVF). Event firms that have issued an SEO in the past 12 (24, 36) calendar months form the basis of the calendar-month portfolio. We form equal- and value-weighted portfolios, and rebalance them every month. We estimate the regressions using weighted least squares, where the weight of each month corresponds to the number of event firms having non-missing returns during that month and compute the t-statistics (in parentheses) using the White (1980) heteroscedasticity-consistent standard errors. In parentheses are t-values and the *, **, and *** indicate significance of the coefficients at the 10%, 5%, and 1% levels, respectively. Equally-weighted portfolio Value-weighted portfolio CAPM FF3FM LCAPM INVCAPM INVCAPM+LIQ CAPM FF3FM LCAPM INVCAPM INVCAPM+LIQ Panel A: months (+1, +36) for the entire SEO sample Alpha (%) 0.290** ( 1.98) 0.403*** ( 3.64) ( 0.32) ( 1.32) ( 0.15) 0.251* ( 1.94) 0.360*** ( 2.83) ( 0.69) ( 0.90) ( 0.20) MKT 1.184*** (35.05) 1.159*** (45.72) 1.032*** (20.87) 1.167*** (33.61) 1.034*** (20.88) 1.185*** (40.03) 1.225*** (38.91) 1.088*** (24.78) 1.160*** (38.54) 1.092*** (25.08) SMB 0.628*** (19.31) 0.154*** (3.82) HML 0.261*** (7.01) 0.213*** (4.60) LIQ 0.217*** ( 4.11) 0.204*** ( 3.71) 0.140*** ( 2.98) 0.105** ( 2.18) INVF 0.163* ( 1.94) ( 0.87) 0.241*** ( 3.31) 0.196** ( 2.59) Adj R Panel B: months (+1, +24) for the entire SEO sample Alpha (%) 0.369** ( 2.35) 0.441*** ( 3.86) ( 0.44) ( 1.61) ( 0.27) 0.387*** ( 2.84) 0.470*** ( 3.21) ( 1.25) ( 1.54) ( 0.67) MKT 1.198*** (32.48) 1.147*** (40.77) 1.025*** (18.91) *** (30.91) 1.027*** (18.92) 1.173*** (36.73) 1.193*** (33.91) 1.054*** (22.22) 1.139*** (35.16) 1.059*** (22.62) SMB 0.664*** (19.15) 0.187*** (4.32) HML 0.167*** (4.02) 0.148*** (2.85) LIQ 0.241*** ( 4.25) 0.226*** ( 3.81) 0.166*** ( 3.34) 0.120** ( 2.35) INVF 0.185** ( 2.02) ( 0.89) 0.306*** ( 3.93) 0.252*** ( 3.12) Adj R Panel C: months (+1, +12) for the entire SEO sample Alpha (%) 0.325* ( 1.80) 0.238** ( 2.00) (0.33) 0.368* ( 1.92) ( 0.04) 0.246* ( 1.85) 0.276** ( 2.10) ( 0.49) ( 1.18) ( 0.29) MKT 1.215*** (28.50) 1.100*** (35.35) 1.005*** (15.65) 1.223*** (27.57) 0.999*** (15.63) 1.131*** (35.95) 1.124*** (32.62) 1.038*** (21.47) 1.116*** (34.22) 1.040*** (21.50) SMB 0.758*** (19.95) 0.238*** (5.66) HML (0.63) 0.089* (1.78) LIQ 0.282*** ( 4.28) 0.324*** ( 4.72) 0.126** ( 2.53) 0.110** ( 2.11) INVF (0.68) 0.223** (2.06) 0.136* ( 1.73) ( 1.05) Adj R start one month after issuance. To facilitate comparison to the results of Lyandres et al. (2008), we closely follow their calendar-time factor regressions modeling using the weighted least squares method to regress the equal-weighted (and value-weighted) SEO portfolio excess returns on the pricing factors under the CAPM, FF3FM, LCAPM, or INVCAPM. As in Lyandres et al. (2008), the weight of each month corresponds to the number of event firms having non-missing returns during that month, and the t-statistics for the coefficient estimates are based on the White (1980) heteroscedasticity-consistent standard errors. We find that the results (not reported) are essentially the same if we use the ordinary least squares method. Panel A of Table 3 reports the factor regression results. Indeed, under the CAPM, issuing firms show significant underperformance of 0.29% (t= 1.98) and 0.25% (t= 1.94) per month in the equal-weighted and the value-weighted portfolios, respectively. We also find large negative abnormal returns under the FF3FM, which produces an equal-weighted alpha of 0.403% (t= 3.64) per month and value-weighted alpha of 0.36% (t= 2.83) per month. However, under the LCAPM, the equal-weighted alpha reduces to 0.049% (t= 0.32) and the value-weighted alpha reduces to 0.095% (t= 0.69) per month and become insignificantly different from zero. These magnitudes of insignificant post-issue

8 102 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) abnormal returns are comparable to previous empirical studies. 16 Also, Liu's (2006) LIQ is significantly negative with a coefficient of (t= 4.11) and 0.14 (t= 2.98) in the equal-weighted and the value weighted portfolios, respectively. Thus, adding the liquidity factor to the CAPM can largely explain away SEO firms' post-issue underperformance (we will provide more evidence below to support this claim). The results suggest that investors of SEO firms face relatively low liquidity risk and thus require relatively low returns in the post-issue period. 17 Lyandres et al.'s (2008) INVCAPM is also useful in explaining SEO firms' post-issue returns, which produces insignificant alphas, and their investment factor INVF is significantly negative in the value-weighted portfolio and marginally so in the equal-weighted portfolio. For the final model reported in Table 3, we add the investment factor INVF to Liu's LCAPM to directly compare the explanatory power of the two pricing factors, LIQ and INVF. While both factors are significant in the value-weighted portfolio, only LIQ is significant in the equal-weighted portfolio, implying that the liquidity factor has a slight edge over the investment factor. The results are very similar when the calendar-time portfolios are alternatively constructed to consist of firms that have conducted during the 24 months or the 12 months prior to the month of portfolio formation, as we report in Panels B and C of Table 3, respectively. The results suggest that the effect of liquidity risk on post-issue stock returns is robust from one to three years after the issuance The issuance size effect Fama and French (2008) show that small net stock issues produce positive abnormal returns while large net stock issues account for the strong negative abnormal returns in the SEO literature. Pontiff and Woodgate (2008) also report that firms with a larger share issuance are associated with lower future stock returns. Their studies suggest that offer size matters in post-issue stock performance. However, neither study controls for liquidity risk. If liquidity risk, as we hypothesize, can help explain post-issue performance, it should also be able to help explain the difference in post-seo performance between small and large issuers. To address this issue, we repeat the factor regression analysis on each of the three SEO offer size groups (small, medium, and large), sorted by the ratio of gross proceeds/lagged total asset, using the 30th percentile and the 70th percentile of the SEO offer size as the cutoff points. Since the results from the equal- and the value-weighted portfolios are similar, to save space, we report in Table 4 the results based on the equal-weighted portfolio that consists of firms that have conducted during the 36 months prior to the month of portfolio formation. The results (not reported) are similar when we construct the calendar-time portfolios using firms that have conducted during the 24 months or the 12 months prior to the month of portfolio formation. The results show that, consistent with Fama and French (2008) and Pontiff and Woodgate (2008), the post-issue underperformance under the CAPM and the FF3FM is greater for large offer size issuers than for small offer size ones. Lyandres et al.'s (2008) INVCAPM also yields significant post-issue underperformance for the subsample of large offer size issuers ( 0.575%, t-value of 2.10), suggesting that their investment factor cannot explain away the issuance size effect. Interestingly, we find no post-issue underperformance for small as well as large offer size issuers under Liu's (2006) LCAPM, providing another evidence that the liquidity factor is more relevant than the investment factor to the pricing of issuing firms' stocks. To link the issuance size effect to liquidity risk, we find that the liquidity risk of the large offer size group behaves very differently from that of the small offer size group in the post-issue period. As shown in Table 4, underthe LCAPM, the large offer size subsample's liquidity beta is (t = 8.75), which is significantly lower than the small offer size subsample's liquidity beta of (t = 4.57), while the medium offer size subsample's liquidity beta of (t= 1.76) is between the two betas. Thus, the evidence suggests that the issuance size effect documented by Fama and French (2008) and Pontiff and Woodgate (2008) is at least in part attributable to lower post-issue liquidity risk being associated with larger offer size issues. The results of the factor regressions on the calendar-time portfolios suggest that liquidity risk plays an important role in the post-issue stock returns, and that Liu's (2006) LCAPM can explain the post-issue stock returns as good as and in some cases better than the CAPM, FF3FM, and INVCAPM. After establishing the legitimacy of the LCAPM in pricing issuing firms' common stocks in the post-issue period, we next use the model to show the dynamics of liquidity risk before and after the and then characterize SEO timing in terms of liquidity risk. 16 Eckbo et al. (2000) use a conditional six macro-factors model and produce 0.05% average abnormal returns (p-value of 0.749, Table 10); Eckbo and Norli (2005) use a five-factor model (market, size, book-to-market, momentum, and liquidity) and produce 0.03% (p-value of 0.812, Table 8); Lyandres et al. (2008) use a four-factor model (market, size, book-to-market, and investment) and produce 0.08% (t-value of 0.72, Table 4). 17 Liu's (2006) liquidity factor LIQ, a zero-cost portfolio, takes a long position in a high LM12 (low-liquidity) portfolio and a short position in a low LM12 (highliquidity) portfolio. Liu shows that LIQ captures market liquidity premium, which tends to be high in a bad market and low in a good market and that the average LIQ is significantly positive at 0.625% over , suggesting that investors demand a significantly higher liquidity premium on the low-liquidity portfolio than the high-liquidity portfolio. Hence, a portfolio having a negative liquidity beta means that the portfolio behaves more like (or covariates stronger with) the high-liquidity portfolio and it needs to offer relatively low liquidity premium, compared to a portfolio with a zero or positive liquidity beta. Since we match SEO issuers with non-issuing firms that have the closest stock liquidity, the matched portfolio also behaves like a high-liquidity portfolio.

9 Table 4 Calendar-time factor model portfolio regressions of SEO firms across three SEO offer size groups. This table sorts issuers on SEO offer size for three groups (small, medium, and large) using the 30th percentile and the 70th percentile of our SEO offer size, defined as the ratio of gross proceeds/lagged total asset, as the cutoff points. This table reports calendar-time factor regressions for the equal-weighted SEO portfolio excess returns using the CAPM, Fama French three-factor model (FF3FM), Liu's (2006) liquidity-augmented CAPM (LCAPM), Lyandres et al.'s (2008) investment-augmented CAPM (INVCAPM), and CAPM with Liu's liquidity risk factor (LIQ) and Lyandres et al.'s (2008) investment factor (INVF). Event firms that have issued an SEO in the past 36 calendar months form the basis of the calendar-month portfolio. We estimate the regressions using weighted least squares, where the weight of each month corresponds to the number of event firms having non-missing returns during that month and compute the t-statistics (in parentheses) using the White (1980) heteroscedasticity-consistent standard errors. In parentheses are t-values and the *, **, and *** indicate significance of the coefficients at the 10%, 5%, and 1% levels, respectively. Small Alpha (%) (1.16) MKT 0.875*** (27.01) CAPM FF3FM LCAPM INVCAPM INVCAPM+LIQ Medium Large Small Medium Large Small ( 1.34) 0.704*** ( 2.71) 0.221** ( 2.20) 0.404*** ( 3.16) 0.505*** ( 3.03) ( 0.60) ( 0.59) (0.51) (0.61) ( 0.49) 0.575** ( 2.10) ( 0.75) ( 0.16) 1.151*** 1.525*** 1.020*** 1.174*** 1.286*** 1.034*** 1.077*** 0.995*** 0.889*** 1.124*** 1.500*** 1.033*** 1.084*** (30.34) (26.40) (41.13) (36.83) (31.49) (22.07) (19.17) (12.47) (26.90) (29.26) (24.83) (22.03) (19.46) SMB 0.284*** 0.588*** 0.909*** (8.60) (14.18) (18.20) HML 0.679*** 0.428*** 0.251*** (18.32) (9.05) ( 4.30) LIQ 0.235*** 0.107* 0.723*** 0.224*** (4.57) ( 1.76) ( 8.75) (4.23) ( 1.00) INVF 0.155* 0.281*** *** (1.90) ( 3.05) ( 1.43) (0.89) ( 2.66) Adj R [p-value] [0.000] [0.000] LIQ Small =LIQ Large [p-value] [0.100] [0.733] INVF Small =INVF Large Medium Large Small Medium Large Small Medium Large (0.27) 0.993*** (12.44) 0.751*** ( 8.67) (1.09) J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013)

10 104 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Evidence of liquidity risk timing 6.1. Ibbotson's (1975) RATS regression Our SEO timing hypothesis posits that issuing firms time to periods when liquidity risk declines to the point where investors have low concerns of the risk. To estimate the dynamics of liquidity risk before and after and test our hypothesis, we apply Ibbotson's (1975) RATS regression technique. Specifically, for each pre-seo event month t relative to the month of SEO filing (and for each post-seo event month relative to SEO issuance), we estimate the following cross-sectional model: r it r ft ¼ α t þ β m;t r mt;i r ft;i þ β l;t LIQ t;i þ ε it ; for i ¼ 1; 2; ::; n; ð1þ where r it r ft, r mt,i r ft,i, and LIQ t,i are firm i's excess return, and its corresponding excess return on the value-weight CRSP market portfolio, and market liquidity premium in event month t, respectively; and the parameters, α t, β m,t, and β l,t, respectively, reflect the common abnormal return, market risk, and liquidity risk for the n firms in event month t. We apply the RATS regression technique to estimate the abnormal return (α t ) and the systematic risks (β m,t, and β l,t ) for each event month from month 36 to month 1 prior to the SEO filing and from month +1 to month +36 after the SEO issuance. The error term, ε it, in Eq. (1) reflects the idiosyncratic effects on firm i's excess return in event month t. Notice that since different firms undertake an SEO at different calendar times, their corresponding market excess returns and market liquidity premiums are likely to be different across firms in event month t. Eq. (1) assumes that the n firms have a common sensitivity to changes in market excess returns in event month t, which can be captured by β m,t and a common sensitivity to changes in market liquidity premiums, which can be captured by β l,t. The model in Eq. (1) allows us to infer SEO firms' (and separately their matched non-issuers') common market risk and common liquidity risk in each event month t. By estimating the model month by month, we would be able to pinpoint the time and the time span of changes in the issuing firms' systematic risks surrounding the SEO event, and compare the changes to those of their matched non-issuers. In a similar vein, we directly compare the dynamics of SEO firms' systematic risks to those of their matched counterparts by estimating the following equation, r it r bit ¼ α t þ β m;t r mt;i r ft;i þ β l;t LIQ t;i þ ε it ; for i ¼ 1; 2; ::; n; ð2þ where r it is the return of firm i and r bit is the return of its matched firm in event month t. In Eq. (2), the parameters, α t, β m,t, and β l,t, respectively, capture issuing firms' abnormal return, excess common market risk, and excess common liquidity risk, relative to their counterparts, in event month t. We rely on the model in Eq. (2) to control for possible market-wide shifts in market risk and liquidity risk, which might not be related to. To see if SEO firms' performance differs from that of their matched firms in the post-issue period, we test whether the average of α t from month +1 to month +36 after the SEO issuance (month 0) is significantly different from zero. Note that, in our RATS regressions, we implement the weighting procedure proposed by Asparouhova et al. (2010) to correct for possible biases due to microstructure noises, such as bid ask spread bouncing, in asset pricing tests. Asparouhova et al. suggest that biases in estimated return premiums for illiquidity can be effectively eliminated by use of a simple weighting procedure where each observed return is weighted by one plus the observed return on the same security in the prior period. In all cases in our study, using ordinary least squares yields similar results (not reported), but with slightly larger liquidity betas The dynamics of systematic risks of SEO firms Fig. 1 depicts the month-by-month results from the estimation of Eq. (1) for the SEO firms and their matched firms in each event month during the pre-seo period from month 36 to the SEO filing month and during the post-seo period from the SEO issuance month to month +36. Table 5 presents statistical analyses to assess whether there are significant shifts in systematic risks surrounding the. The figure and our statistical tests indicate a number of results that can be related to SEO timing and post-issue stock performance. We use Chow tests to examine whether the common beta in any given event month during the pre-filing (post-offering) period is significantly different from that in the filing (offering) month. We also perform Chow tests to examine whether the beta estimate in the offering month is significantly different from that in the filing month. First, Fig. 1 shows a steady decline in market beta starting from month 6 for both SEO firms and their matched firms. Specifically, Table 5 shows that the SEO firms' market beta is in month 6, which then gradually falls to in month 1 and to in the filing month. The Chow tests show that issuing firms' common market beta of in the filing month is significantly lower than that in month 6 but is not significantly different from that in month 1. We also use the match-adjusted model in Eq. (2) to compare issuing firms' common market risk to that of their matched firms, and Table 5 shows that there are no significant differences between the two samples throughout the SEO process. This finding suggests that the pre-seo decline in issuing firms' common market betas is largely due to market-wide improvements in market risk (e.g., economic improvements could reduce business risk for SEO firms and for their matched non-issuers as well).

11 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) A: The plot of Market beta Matches is the filing month 0 is the offering month B: The plot of Liquidity beta Matches is the filing month 0 is the offering month C: The plot of 0.07 Alpha Matches is the filing month 0 is the offering month Fig. 1. Plots of issuing firms' and their matched non-issuers' common market betas, liquidity betas, and abnormal returns surrounding. We first estimate issuing firms' (and separately their matched firms') common market beta, liquidity beta, and abnormal return, i.e., β m,t, β l,t, and α t using Ibbotson's (1975) RATS (returns across time and securities) regression technique for Liu's (2006) LCAPM: r it r ft =α t +β m,t (r mt,i r ft,i )+β l,t LIQ t,i +ε it for i=1, 2,, n sample firms in each event month t over the pre-seo period from month 36 to the SEO filing month and over the post-seo period from the SEO issuance month to month +36. To correct for possible biases due to microstructure noises, we implement Asparouhova et al.'s (2010) weighting procedure where each observed return is weighted by one plus the observed return on the same security in the prior period. We plot the SEO samples' and their matched firms' estimates of β m,t, β l,t, and α t in panels A, B, and C, respectively. For each issuing firm, we choose a comparable non-issuer, matched by firm size, return momentum, B/M, and pre-seo liquidity.

12 106 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Table 5 Issuing firms' and their matched non-issuers' common market betas and liquidity betas surrounding SEO filing and issuance. We estimate issuing firms' common market beta, β m,t, and common liquidity beta, β l,t, for each event month t surrounding SEO filing and issuance from the cross-sectional regression model: r it r ft = α t +β m,t (r mt,i r ft,i )+β l,t LIQ t,i +ε it, for i=1, 2,, n. We perform Chow tests and use superscripts *, **, and *** to indicate whether the pre-filing (post-issue) beta estimate in a given month differs significantly from that in the filing month (offering month) at the 10%, 5%, and 1% levels, respectively. We also perform Chow tests and use superscripts *, **, and *** to indicate whether the beta estimate in the offering month is significantly different from that in the filing month. We run the same model and do the same tests for the matched non-issuers. To obtain issuing firms' excess common market beta and excess common liquidity beta in each event month t, we run the match-adjusted model: r it r bit =α t +β m,t (r mt,i r ft,i )+β l,t LIQ t,i +ε it, where r it is the return of issuing firm i and r bit is the return of its matched firm in event month t. Each p-value in the bracket indicates whether the excess beta estimate is significantly different from zero. Panel A reports common market beta for all firms, and Panel B reports common liquidity beta for all firms, respectively. Pre-filing month Filing month Offering month Post-offering month Panel A: the estimates of common market beta, β m,t, for all SEO issuers 1.084** 1.012** 1.107*** * Matches 1.036* 1.052** 1.093*** Excess beta [p-value] Excess beta=0 [0.324] [0.336] [0.734] [0.427] [0.595] [0.386] [0.972] [0.344] [0.108] [0.316] [0.266] [0.875] Panel B: the estimates of common liquidity beta, β l,t, for all SEO issuers 0.008*** *** *** * Matches * Excess *** ** ** beta [p-value] Excess beta=0 [0.336] [0.122] [0.189] [0.058] [0.000] [0.035] [0.034] [0.000] [0.000] [0.802] [0.023] [0.000] Second, SEO firms' common market beta of in the offering month is not significantly different from that in the filing month. After the offering, the common market betas of SEO firms and their matched non-issuers fluctuate around their mean of and 1.015, respectively, as reported in Panel A of Table 6. To see whether there is a difference in the average common market risk estimates between the pre- and the post-seo periods, we regress the month-by-month common market betas on a post-seo dummy variable which equals one for the post-offering months and zero for the pre-filing months. The result of insignificant post-seo dummy, as reported in Panel B of Table 6, shows that the average pre-seo common market risk is not significantly different from that of the post-seo period for both the SEO firms and their matched firms. The results suggest that whatever causes SEO firms to have lower expected stock returns in the post-issue period than in the pre-filing period, it is not market risk. Third, there is also evidence of a pre-filing decline in liquidity risk for both SEO firms and their matched non-issuers. The decline in liquidity risk starts around month 6, and SEO firms show a more significant decline than their matched firms. Panel B of Table 5 shows that the SEO firms' common liquidity beta is in month 6, which then sharply declines to in month 1, much lower than their counterparts' decline to in month 1 from in month 6. In the filing month, SEO firms' liquidity beta rebounds to 0.305, which remains significantly lower than their matched firms' liquidity beta of The pattern of liquidity risk seems to suggest that a sharp decline and then a rebound in liquidity risk prompts issuing firms to file for. Fourth, compared to their matched non-issuers, issuing firms' liquidity risk remains relatively low for at least three years following the. As reported in Panel A of Table 6, the average post-issue liquidity beta of issuing firms is 0.217, which is significantly lower than 0.095, the average post-issue liquidity beta of their counterparts. Also, Panel C of Table 6 shows that SEO firms' average liquidity risk is lower during the post-seo period, compared to the pre-seo period. In contrast, there is no significant difference in the average liquidity beta between the pre- and the post-seo period for the matched non-issuers. These results imply that of the two risks in Liu's (2006) LCAPM, liquidity risk is the one that causes issuing firms to have low post-issue expected stock returns. This finding confirms our earlier inference from the calendar-time portfolio analysis that liquidity risk plays an important role in post-issue stock returns. Fifth, according to Fig. 1, SEO firms exhibit an impressive increase in abnormal returns prior to filing, which is consistent with the pre-seo price run-up suggested in the literature. Because one of our matching criteria is pre-seo return momentum, the matched firms also show significant abnormal returns, but to a lesser extent, during the pre-filing months. Intriguingly, SEO firms' abnormal returns abruptly decline toward zero in the filing month and further down to zero in the offering month. Thereafter, the estimates of alphas for both SEO firms and their matched non-issuers level off around zero. In fact, Panel A of Table 6 reports that the average monthly abnormal return during the three years after the is 0.056% (t-value= 0.90) and 0.058% (t-value= 1.42) for SEO firms and their matched firms, respectively. Thus, the results from Ibbotson's (1975) RATS regression technique are consistent with those based on the calendar-time portfolio regression analysis. Both approaches reveal that controlling for only two risk factors market risk and liquidity risk is sufficient to show that SEO firms have no post-issue underperformance.

13 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Table 6 Cross-sectional estimates of issuers' and their matched non-issuers' pre- and post-seo common abnormal returns and betas. This table provides summary statistics for the month-by-month estimates of issuers' and their matched non-issuers' common alpha and betas, estimated from the cross-sectional regression model: r it r ft =α t +β m,t (r mt,i r ft,i )+β l,t LIQ t,i +ε it, for i=1, 2,, n. In Panel A, we test whether issuers' and their matched non-issuers' average common alpha and betas estimated over the pre-seo filing period from month 36 to month 1 are significantly different from zero, and do the same over the post-seo offering period from month +1 to month +36. In Panel B (C), we run a time-series regression of β m,t (β l,t ) on a dummy variable, D t, which equals one for the post-seo offering months and zero for the pre-seo filing months. In brackets are the t-values and the *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Statistics with p-values of.10 or lower are highlighted in bold. SEO firms Matched firms Match-adjusted SEO firms Matched firms Match-adjusted Pre-filing (months 36 to 1) Post-issuance (months +1 to +36) Panel A: time-series averages of cross-sectional regression coefficients Mean (α t (%)) 2.030*** [7.02] 0.915*** [6.95] 1.115*** [5.92] Mean (β m,t ) 1.045*** 1.031*** [59.40] [60.61] [1.56] Mean (β l,t ) *** 0.129*** [ 1.25] [4.40] [ 5.48] [ 0.90] 1.027*** [70.91] 0.217*** [ 11.64] [ 1.42] 1.015*** [76.82] 0.095*** [6.93] [0.02] [1.20] 0.312*** [ 17.17] SEO firms Matched firms Match-adjusted Panel B: time-series regression of common market beta, β m,t, on the post-issue dummy variable, D t Intercept 1.045*** [64.84] 1.031*** [67.69] D t [ 0.76] [ 0.74] [1.44] [ 0.11] Panel C: time-series regression of common liquidity beta, β l,t, on the post-issue dummy variable, D t Intercept [ 1.49] 0.092*** [5.20] D t 0.181*** [ 5.19] [0.13] 0.129*** [ 6.13] 0.184*** [ 6.19] More importantly, the dynamics of liquidity risk shown in Fig. 1 are consistent with our SEO timing hypothesis that managers time to periods when liquidity risk declines to the point where investors have low concerns of the risk, not only at but also for the long run, which leads to low expected stock returns in the post-issue period. Table 7 presents the dynamics of systematic risks across the three SEO offer size groups. Panel A of Table 7 shows that there are no apparent differences in market beta between the small and large offer size issuers. The liquidity risk timing is especially Table 7 The dynamics of market-betas and liquidity-betas across the three SEO offer size groups. This table sorts issuers on SEO offer size, defined as total amount raised from an SEO over lagged total asset, for three groups (small, medium, and large) using the 30th percentile and the 70th percentile of the SEO offer size as the cutoff points. We estimate issuers' common market beta and liquidity beta for each event month t surrounding SEO filing and issuance from the cross-sectional regression model: r it r ft =α t +β m,t (r mt,i r ft,i )+β l,t LIQ t,i +ε it, for i=1, 2,, n. We perform Chow tests and use superscripts *, **, and *** to indicate whether the pre-filing (post-issue) beta estimate in a given month differs significantly from that in the filing month (offering month) at the 10%, 5%, and 1% levels, respectively. We also perform Chow tests and use superscripts *, **, and *** to indicate whether the beta estimate in the offering month is significantly different from that in the filing month. Each p-value in the bracket indicates whether the beta estimates between the small offer size group and the large offer size group differ significantly. Pre-filing month Filing month Offering month Post-offering month Panel A: the estimates of common market beta across the three SEO offer size groups Small * N=1433 Medium 1.179* ** N=1909 Large * 1.105** N=1432 [p-value] Small=Large [0.533] [0.869] [0.537] [0.165] [0.265] [0.204] [0.003] [0.146] [0.068] [0.691] [0.601] [0.077] Panel B: the estimates of common liquidity beta across the three SEO offer size groups Small * ** Medium 0.094* ** Large 0.296* * *** [p-value] Small=large [0.108] [0.006] [0.068] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

14 108 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) evident in large offer size issuers. As we report in Panel B of Table 7, large offer size issuers' common liquidity risk shows a particularly sharp and persistent decline from in month 6 to in month 1, which then rebounds to in the filing month and to in the offer month. There is also some evidence of pre-seo declines in liquidity risk for the medium offer size subsample. But, we find no evidence of pre-seo liquidity risk declines for the small offer size subsample. Thus, Table 7 shows that large offer size issuers behave differently from small offer size issuers in the liquidity risk not only in the post-issue period but also in the pre-filing period. While we document the differences in the liquidity risk dynamics between large and small offer size issuers, what factors may cause the differences are not clear. It is possible that a larger decline in liquidity risk allows issuing firms to go for a bigger SEO. Or, some issuers might not be able to wait until a substantial improvement in their liquidity environment, and choose to issue smaller. We leave this issue for future research Additional evidence from time-series regression In addition to the models in Eqs. (1) and (2), which allow us to examine the dynamics of SEO firms' common systematic risks during the SEO process, we also construct the following time-series LCAPM model to examine changes in each individual firm i's systematic risks from before to after the : r it r ft ¼ α i;0 þ α i;1 D t β im;0 þ β im;1 D t r mt;i r ft;i þ β il;0 þ β il;1 D t LIQ t;i þ ε it ; ð3þ Table 8 Time-series estimates of issuers' and their matched non-issuers' pre- and post-seo abnormal returns and betas. Panels A and B use the following CAPM model to examine changes in each individual firm i's risks from before to after the : r it r ft =α i,0 +α i,1 D t +(β im,0 β im,1 D t )(r mt,i r ft,i )+ε it, where D t =1 if t is in the post-offering period, and D t =0 if t is in the pre-filing period; β im,0 and β im,1 are firm i's pre-seo market risk and the differences between its post- and pre-seo market risk, respectively; α i,0 is its pre-seo abnormal return; and α i,1 is the difference between its post- and pre-seo abnormal return. We run the regression for each SEO firm for t from month 36 to month 1 prior to SEO filing and from month +1 to month +36 after SEO issuance. Panels C and D use the following Liu's (2006) LCAPM model to examine changes in each individual firm i's risks from before to after the : r it r ft =α i,0 +α i,1 D t (β im,0 +β im,1 D t )(r mt,i r ft,i )+(β il,0 +β il,1 D t )LIQ t, i+ε it., where β il,0 and β il,1 are firm i's pre-seo liquidity risk and the differences between its post- and pre-seo liquidity risk, respectively. To control for market-wide movements in risks, we use the match-adjusted returns in regression: r it r bit =α i,0 +α i,1 D t (β im,0 +β im,1 D t )(r mt,i r ft,i )+(β il,0 +β il,1 D t )LIQ t,i +ε it. Panels B and D sort issuers on SEO offer size for three groups (small, medium, and large) using the 30th percentile and the 70th percentile of our SEO offer size as the cutoff points. Wereportthe cross-sectional average of each regression coefficient and test whether it is significantly different from zero. In brackets are the t-values and the *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. To see whether there is post-issue underperformance, we examine whether the average of α 0 +α 1 across each is significantly different from zero. α 0 (%) α 1 (%) β m,0 β m,1 α 0 +α 1 (%) Panel A: results under the CAPM for the entire SEO sample 1.952*** [55.08] Matches 0.970*** [37.56] Match adjusted 0.982*** [17.28] 2.227*** [ 43.53] 0.962*** [ 27.01] 1.265*** [ 15.93] 1.134*** [85.07] 1.064*** [83.06] 0.070* [1.83] 0.107*** [7.48] [ 0.86] 0.119*** [7.88] 0.275*** [ 8.17] [0.26] 0.283*** [ 9.01] Panel B: results under the CAPM across the three SEO offer size groups Small 0.933*** [21.93] 0.800*** [ 12.30] 0.771*** [42.72] 0.147*** [7.45] 0.133*** [2.94] Medium 1.829*** [34.79] 2.090*** [ 27.07] 1.131*** [52.88] 0.109*** [4.84] 0.262*** [ 4.98] Large 3.233*** [40.05] 3.866*** [ 33.37] 1.535*** [53.72] 0.066* [1.94] 0.634*** [ 7.92] [33.58] [ 27.25] [49.94] [5.72] [ 6.22] α 0 (%) α 1 (%) β m,0 β m,1 β l,0 β l,1 α 0 +α 1 (%) Panel C: results under Liu's (2006) LCAPM for the entire SEO sample 1.906*** [43.19] 1.887*** [ 32.42] 1.123*** [64.40] Matches 0.919*** 0.939*** 1.109*** [32.04] [ 22.55] [64.91] Match adjusted 0.987*** 0.949*** [12.95] [ 9.55] [0.89] 0.065*** [ 2.84] [ 1.28] [ 1.30] [0.60] 0.032* [1.91] [ 1.17] 0.262*** [ 10.24] [ 1.53] 0.226*** [ 9.32] [0.48] [ 0.56] [0.76] Panel D: results under Liu's (2006) LCAPM across the three SEO offer size groups Small 0.647*** [13.64] 0.613*** [ 8.44] 0.962*** [43.40] Medium 1.761*** 1.745*** 1.134*** [27.46] [ 19.83] [42.41] Large 3.513*** 3.371*** 1.274*** [33.51] [ 24.26] [29.27] [1.15] 0.061* [ 1.69] 0.177*** [ 3.17] 0.244*** [11.73] [1.11] 0.289*** [ 5.64] 0.170*** [ 5.74] 0.276*** [ 6.87] 0.353*** [ 5.52] [0.63] [0.27] [1.44]

15 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Market beta Matches Event Time Fig. 2. Market betas from the market model with daily data. To facilitate comparison to their Figure 4a, we closely follow Carlson et al.'s(2010)specification and use daily return data. Specifically, we obtain realized beta calculated by regressing daily log returns of the firm on a constant and daily log returns of the CRSP value-weighted index in each 21-day windows ( months ) prior to the announcement date and after issuance. The interval between announcement and issuance is considered as a single period (and designated as month 0), regardless of how long that interval is. All positive numbers on the horizontal axis are months after issuance; and all negative numbers are months prior to announcement. As in CFG, we aggregate returns across any days in which trading volume is zero, and require a minimum of fifteen valid daily returns for firm i in month t to produce a valid beta. The solid line in the plot shows the average across all issuing firms, while the dashed line represents average betas of their matched non-issuers. For each issuing firm, we choose a comparable non-issuer, matched by firm size, return momentum, B/M, and pre-seo liquidity. where D t =1 if t is in the post-seo period, and D t =0 if t is in the pre-seo period; β im,0 and β il,0 are firm i's pre-seo market risk and liquidity risk, respectively; β im,1 and β il,1 are the differences between its post- and pre-seo market and liquidity risks, respectively; α i,0 is its pre-seo abnormal return; and α i,1 is the difference between its post- and pre-seo abnormal return. We run the model in Eq. (3) for each firm for t from month 36 to month 1 prior to the SEO filing and from month +1 to month +36 after the SEO issuance. Like Eq. (2), to control for market-wide movements in risks, we use the match-adjusted returns in the time-series LCAPM model below: r it r bit ¼ α i;0 þ α i;1 D t β im;0 þ β im;1 D t r mt;i r ft;i þ β il;0 þ β il;1 D t LIQ t;i þ ε it : ð4þ To see if there is post-issue underperformance, we examine whether the average of α i,0 +α i,1 across the SEO sample firms is significantly different from zero. In sum, the time-series models in Eqs. (3) and (4) provide robustness checks on the estimation results from the cross-sectional models in Eqs. (1) and (2). 18 Table 8 reports the averages of time-series results of pre-filing and post-offering abnormal returns and systemic risks from the estimation of the model in Eq. (3) for each of the 5138 SEO sample firms under the CAPM and LCAPM. Table 8 also reports the averages of time-series results from the match-adjusted model in Eq. (4). We expect that the cross-sectional averages of the estimates from the time-series models in Eqs. (3) and (4) should be close to the time-series averages of the estimates from the cross-sectional models in Eqs. (1) and (2), respectively. Consistent with prior research, Panel A of Table 8 reports significant post-issue underperformance under the CAPM. Specifically, the average of α i,0 +α i,1 across the 5138 SEO sample firms is 0.275% (t-value= 8.17) from the model in Eq. (3) and 0.283% (t-value= 9.01) from the match-adjusted model in Eq. (4) under the CAPM (i.e., without the liquidity factor). Panel B shows that post-issue abnormal returns using the CAPM from the model in Eq. (3) for small, medium, and large offer size issuers are 0.133% (t-value=2.94), 0.262% (t-value= 4.98), and 0.634% (t-value= 7.92), respectively. The results suggest that the negative abnormal returns are driven predominately by large offer size firms and to a lesser extent by medium offer size firms. The evidence is consistent with Fama and French (2008) and Pontiff and Woodgate (2008). Once we incorporate Liu's LIQ factor into the CAPM, Panel C shows that the averages of α i,0 +α i,1 across the 5138 SEO sample firms are 0.019% (t-value=0.48) from the model in Eq. (3) and 0.039% (t-value=0.76) from the match-adjusted model in Eq. (4). Furthermore, Panel D shows that, in time-series analysis, Liu's LCAPM also has explanatory power on post-seo stock performance 18 For clarity, the cross-sectional models in Eqs. (1) and (2) calculate a different liquidity beta for each event month. The time-series models in Eqs. (3) and (4) calculate a different pre-seo (and post-seo) liquidity beta for each sample firm.

16 110 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) A: The plot of 1.4 Market beta Matches is the filing month 0 is the offering month Fig. 3. Estimates from the market model with monthly data. In each event month t around the SEO event, the monthly log returns of issuing firms (and separately for their matched non-issuers) are regressed on a constant and the corresponding monthly log returns of the value-weighted CRSP index, i.e., ln(1+r i,t )=α t +β m,t ln(1+r m,t )+ε i,t, where r i,t and r m,t are the holding period return and value-weighted CRSP-index return, respectively. We obtain a common market beta β m,t in each event month t,fort from month 36 (before SEO filing) to month +36 (after the SEO). For each issuing firm, wechooseacomparablenon-issuer,matched by firm size, return momentum, B/M,and pre-seo liquidity. across the three offer size groups. The findings corroborate our earlier findings of no post-seo abnormal return from calendar-time factor regressions and from Ibbotson's RATS regressions. Furthermore, Panel C shows that the averages of the time-series estimates of SEO firms' pre-filing excess market betas and their post-seo changes in excess market betas are (t-value=0.89) and (t-value= 1.30), respectively. The findings confirm that there are no significant differences in market betas between the SEO sample firms and their matched non-issuers throughout the SEO process. Also, given our findings from the cross-sectional regression analyses reported in Tables 5 and 6, we expect time-series regression analyses to yield a significant post-issue change in liquidity risk for issuing firms. As expected, Panel C shows that the average of the SEO firms' post-issue changes in excess liquidity beta is a significant The finding suggests that, compared to their matched non-issuers, SEO firms experience a significant reduction in liquidity risk in the post-issue period. In sum, the evidence validates our earlier inference that lower liquidity risk is the reason why SEO firms have lower post-issue stock returns Robustness checks The market model Our findings based on Liu's (2006) LCAPM show that market betas of issuing firms and their matched non-issuers are statistically not different from each other both before and after the. The results suggest that do not bring significant changes to market risk. The evidence is inconsistent with Carlson et al.'s (2006, 2010) investment-based explanation. In particular, without considering the liquidity factor, Carlson et al. (2010) use the market model to infer possible shifts in issuing firms' market beta during the SEO process, and find that, compared to matched non-issuers, issuing firms' market beta increases prior to SEO issuance and then decreases gradually thereafter. We investigate whether the differences are due to our sample selection or their analysis is subject to the omitted risk factor. In order to facilitate a comparison to their Figure 4a, we closely follow Carlson et al.'s (2010) specification and obtain realized beta calculated by regressing daily log returns of the firm on a constant and daily log returns of the CRSP value-weighted index in each 21-day window ( months ) prior to the announcement date and after issuance. 20 Fig. 2 clearly illustrates that issuing firms' market beta increases dramatically right before, and drops gradually thereafter, but remains significantly higher in the post-issue period than that of their matched non-issuers. The magnitudes and the pattern of our estimated market betas around are virtually the same as their Figure 4a, suggesting that our sample selection is not the problem. We also obtain similar market beta dynamics using the market model and monthly return data coupled with Ibbotson's (1975) RATS regression, as shown in Fig. 3. Our earlier evidence from the time-series regressions under the CAPM reported in Panel A of 19 Note that the results reported in this subsection also hold (not reported) if we apply Petersen's (2009) two-way-clustering panel data model (clustered by firm identifier and SEO year) to account for the cross-sectional dependence of standard errors. The results are available upon request. 20 The realized beta is defined as the ratio of the realized covariance of stock and market to the realized market variance (see Andersen et al., 2006). We also follow Scholes and Williams (1977) who take into account the infrequent trading problem and our realized beta estimate consists of the current market beta plus the lagged beta. Since adding the lagged market return does not change the results much, we do not report them here to save space. The results are available upon request.

17 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) A: The plot of Matches 1.2 Market beta is the filing month 0 is the offering month B: The plot of 0.4 Liquidity beta SEO Match 0 is the filing month 0 is the offering month Fig. 4. Estimates from the market model augmented with Liu's liquidity factor. In each event month t around the SEO event, the monthly log returns of issuing firms (and separately for their matched non-issuers) are regressed on a constant, the corresponding monthly log returns of the value-weighted CRSP index, and Liu's (2006) liquidity factor LIQ, i.e., ln(1+r i,t )=α t +β m,t ln(1+r m,t )+β l,t LIQ t ε i,t, where r i,t, r m,t, and LIQ are the holding period return, value-weighted CRSP-index return, and Liu's liquidity factor, respectively. Panel A plots the common market beta β m,t and panel B plots the common liquidity beta β l,t for each event month t from month 36 (before SEO filing) to month +36 (after the SEO). For each issuing firm, we choose a comparable non-issuer, matched by firm size, return momentum, B/M, and pre-seo liquidity. Table 8 corroborates our finding in Fig. 3, which shows significant post-issue increases in market risk for issuing firms. Panel A of Table 8 indicates that the averages of the time-series estimates of post-seo changes in market beta and changes in excess market betas are (t-value=7.48) and (t-value=7.88), respectively. However, given the importance of the liquidity factor in asset pricing and the significant negative correlation ( 0.713) between LIQ and value-weighted CRSP index return over the period, 21 omitting the liquidity factor could result in biases in estimated market beta. To illustrate, Fig. 4 displays the market beta estimates from the market model augmented by LIQ, using Carlson et al.'s (2010) log return specification. As expected, similar to Fig. 1, Fig. 4 shows that the market betas of issuing firms and their matched non-issuers behave similarly both before and after the, but liquidity betas of the two samples show obvious differences prior to and after the. Thus, instead of market risk, it is liquidity risk that allows us to differentiate issuing firms from comparable non-issuers in terms of SEO timing and the behavior of stock returns in the post-issue period. 21 The significant negative correlation suggests that large shocks to market illiquidity tend to be associated with large declines in the market index, which reflects the fact that liquidity could decline considerably or even disappear in bad markets. Consequently, investors require a higher liquidity risk premium in bad markets.

18 112 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) A: Liu s (2006) LM1 1.2 Liu s (2006) LM1 (Day) Matches is the filing month 0 is the offering month B: Amihud s (2002) IM 1.8 Amihud s (2002) IM (1X10 6 ) Matches is the filing month 0 is the offering month Fig. 5. Illiquidity measures for issuing firms and their matched non-issuers surrounding. This figure plots issuing firms' (and matched non-issuers') averages of Liu's (2006) LM1 (standardized turnover-adjusted number of zero daily trading volumes over one month) and Amihud's (2002) illiquidity measure (IM) in each event month t from month 36 (before SEO filing) to month +36 (after the SEO). For each issuing firm, we choose a comparable non-issuer, matched by firm size, return momentum, B/M, and pre-seo liquidity. IM in each event month is Winsorized at 1% and 99% to mitigate problems with outliers. Furthermore, to assess the robustness of our findings, we have experimented to add to Liu's LCAPM the lagged market return to account for the nonsynchronous trading problem (Scholes and Williams, 1977), and we have also tried to substitute Liu's (2006) liquidity factor with the one proposed by Pastor and Stambaugh (2003) or by Sadka (2006). Overall,these changes do not materially alter our conclusion that liquidity risk plays an important role in shaping issuing firms' post-issue stock returns Evidence of improvement in stock liquidity To provide another robustness check, we examine changes in stock liquidity surrounding the, which do not require an asset pricing model. In theory, firms face higher liquidity risk when their stock returns are more sensitive to shocks to market liquidity. It is conceivable that firms with greater stock liquidity, which could better facilitate informed trading and allow stock price to impound more information, would have lower liquidity risk (see Acharya and Pedersen, 2005; Pastor and Stambaugh, 2003). Hence, based on the inverse association between stock liquidity and liquidity risk, we examine pre- and post-seo stock liquidity to see whether changes in liquidity are consistent with our inferences from liquidity risk. Specifically, we expect that substantial liquidity improvement occurs prior to SEO filing and that stock liquidity remains high following the. Furthermore, we expect that the liquidity improvement is tied to SEO filing decisions. Indeed, Fig. 5 and Table 9 demonstrate a very steady and visible liquidity improvement prior to SEO filing. The demonstration is based on Liu's (2006) LM1, standardized turnover-adjusted number of zero daily trading volumes over one month, and Amihud's (2002) illiquidity measure, IM. 23 While LM1 emphasizes trading discontinuity, IM reflects the price impact of trades. Both measures have a close relation to the (latent) costs of trading. 22 To conserve space, the results of these experiments are not reported, but are available upon request. 23 Following Amihud (2002), IM in each event month is Winsorized at 1% and 99% to mitigate problems associated with outliers.

19 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Table 9 Stock liquidity of issuing firms and their matched firms surrounding SEO filing and issuance. This table reports issuing firms' (and their matched firms') averages for Liu's (2006) LM1 in Panel A, and Amihud's (2002) illiquidity measure (IM) in Panel B at pre-filing months 12, 9, 6, 3, and 1, SEO filing month, SEO issuance month, and post-issue months +1, +3, +6, +9, and +12. This table also reports the average excess (match-adjusted) LM1, and IM. The IM in each event month is Winsorized at 1% and 99% to mitigate problems with outliers. Superscripts *, **, and *** indicate whether the pre-filing (post-issue) average measure in a given month differs significantly from that in the filing month (offering month) at the 10%, 5%, and 1% levels, respectively. We also test whether the average measure in the offering month is significantly different from that in the filing month. Each p-value in the bracket indicates whether the match-adjusted estimate is significantly different from zero. Pre-filing months Filing month Offering month Post-offering months Panel A: Liu's (2006) LM *** 0.625*** 0.540*** 0.440*** *** 0.161** Matches 0.699*** 0.673*** 0.636** Excess LM *** 0.048*** 0.096* *** Excess LM1=0 [0.414] [0.046] [0.003] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] Panel B: Amihud's (2002) IM ( ) 0.840*** 0.655*** 0.515*** 0.351*** *** 0.122*** 0.151** *** 0.262*** Matches Excess IM * Excess IM=0 [0.094] [0.030] [0.001] [0.000] [0.019] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] As expected, we obtain similar results from LM1 and IM as liquidity measures. Both measures show that issuing firms start to have higher stock liquidity than their matched firms about one year prior to SEO filing, and that the liquidity improvement of SEO firms stays for at least three years after the. To further confirm the results, Table 9 shows that the average differences in LM1 (and IM) between the SEO sample and the matched sample are statistically significant from month 9 and afterwards. It is reasonable to expect that improve stock liquidity since they substantially increase the number of outstanding shares and expand issuing firms' investor base. However, as we emphasize, most of the liquidity improvement during the SEO process occurs prior to SEO filing. Such liquidity improvement could lower liquidity risk concern of stock investors, and lead them to require a lower liquidity risk premium after the Probit analysis of the decision to conduct an SEO To test whether the liquidity improvement and the resultant high liquidity prompt SEO decision, we run probit analysis on the decision to file an SEO and report the results in Table 10, which includes quarterly as well as annual results. For models (1) to (3) that contain quarterly results, the dependent variable is a dummy equal to 1 if the firm has an SEO filing in a given quarter t and zero for the rest of Compustat firms in that quarter. For models (4) to (6) that contain annual results, the dependent variable is similarly a dummy equal to 1 if the firm has any SEO filing in a given year t and zero for the rest of Compustat firms in that year. To capture the importance of prior liquidity improvement in SEO filing decision, we use ΔLn(LM1) 6, 1, the natural logarithm of Liu's (2006) LM1 in month 1 (i.e., one month before the filing month) minus that in month 6, and Ln(LM1) 1, the natural logarithm of LM1 in month 1. Furthermore, we include a set of variables to control for possible effects on SEO decisions from market-timing opportunities, near-term need for cash (Cash), and corporate investment opportunities (Growth). Similar to DeAngelo et al. (2010), we measure market-timing opportunities by market-to-book, and prior 12-month stock returns. We use asset growth to measure the demand for investments. Following Butler and Wan (2010), we also include Ln(size), the natural logarithms of equity market capitalization; Leverage, debt over total assets; and fixed assets ratio, the ratio of net property, plant, and equipment to total assets. Models (1), (2), (4), and (5) report our basic results from a pooled probit model of SEO probability, with standard errors obtained from the two-way (firm and time) clustering method proposed by Petersen (2009). All four models show that ΔLn(LM1) 6, 1 and Ln(LM1) 1 are highly significant, which imply that the likelihood of a firm to file for an SEO in a given time period increases with the prior liquidity improvement and the resultant high stock liquidity. Consistent with earlier studies, the results also show that smaller firms and firms with higher prior one-year return, higher market-to-book, more asset growth, higher leverage, lower cash balance, and higher fixed asset ratio, are more likely to file for an SEO. Thus, the probit analysis confirms that, along with these firm characteristics, liquidity conditions on the stock market also play a very important role in issuing firms' SEO timing decisions. One might still be concerned with possible endogeneity of stock liquidity and liquidity improvement. For instance, institutional investors show a strong and persistent demand for liquidity (Gompers and Metrick, 2001, Table 4) and increases in institutional ownership are associated with a higher likelihood of an SEO (Hovakimian and Hutton, 2010). To address this possible endogeneity, we run a two-stage least squares regression. We first estimate respective models of Ln(LM1) 1 and ΔLn(LM1) 6, 1 using exogenous

20 114 J.-C. Lin, Y. Wu / Journal of Corporate Finance 19 (2013) Table 10 Probit analysis of the decision to conduct an SEO. In models (1) to (3), the dependent variable is a dummy equal to 1 if the firm has any in quarter t and zero for the rest of Compustat firm-quarter. In models (4) to (6), the dependent variable is a dummy equal to 1 for any in year t and zero for the rest of Compustat firm-year. ΔLn(LM1) 6, 1 is the natural logarithm of Liu's (2006) LM1 in the pre-filing month 1 minus that in month 6; Ln(LM1) 1 is the natural logarithm of LM1 in the pre-filing month 1. For the two-stage-probit regression in models (3) and (6), PRLn(LM1) 1 and PRΔLn(LM1) 6, 1 are the fitted values from their respective first-stage regressions of Ln(LM1) 1 and ΔLn(LM1) 6, 1, respectively. Market-to-book is market value of asset at December of t 1 and book value of asset for the fiscal year end in t 1. Prior stock return is the preceding 12-month stock return calculated through the end of the month before filing quarter end. Ln(size) is the natural logarithms of equity market capitalization measured in the quarter prior to the SEO filing. In models (1) to (3) all Compustat variables (Growth, cash, leverage, and fixed asset ratio) are measured in quarter t 1 and in models (4) to (6) all Compustat variables are measured in year t 1. Growth is the percentage change in total assets. Cash is the ratio of cash and short-term investments to total assets. Leverage is debt over total assets. The fixed assets ratio is the ratio of net property, plant, and equipment to total assets. Reported z-statistics, in parentheses, are calculated using standard errors obtained from the two-way (firm and time) clustering method in Petersen (2009). *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Intercept ΔLn(LM1) 6, 1 Ln(LM1) 1 Quarterly regression Annual regression Pooled probit Pooled probit Two-stage probit Pooled probit Pooled probit Two-stage probit (1) (2) (3) (4) (5) (6) 3.125*** ( 72.57) 0.075*** ( 12.95) 1.041*** ( ) 3.054*** ( 69.28) 0.075*** ( 12.75) 1.061*** ( ) 4.680*** ( 32.68) 3.214*** ( ) 0.076*** ( 13.12) 1.045*** ( ) 3.218*** ( ) 0.076*** ( 13.07) 1.060*** ( ) 5.035*** ( 52.57) PRΔLn(LM1) 6, *** ( 4.30) 0.051*** ( 4.15) PRLn(LM1) *** ( 57.71) 2.547*** ( 57.61) Market-to-book 0.037*** (17.49) 0.030*** (10.70) 0.012*** (10.86) 0.012*** (10.93) Prior stock return 0.127*** (36.71) 0.126*** (36.64) 0.130*** (37.37) 0.130*** (37.30) Ln(size) 0.101*** ( 9.86) 0.134*** ( 19.41) 0.123*** ( 13.34) 0.000*** ( 11.72) 0.000*** ( 11.24) 0.000*** ( 6.07) Growth 1.603*** (27.98) 1.497*** (26.00) 1.496*** (25.96) 0.465*** (28.66) 0.443*** (27.04) 0.444*** (27.08) Cash 0.111*** ( 4.06) 0.114*** ( 4.18) 0.108*** ( 3.84) 0.100*** ( 3.79) 0.140*** ( 3.83) 1.044*** ( 3.64) Leverage 0.389*** (12.43) 0.423*** (13.60) 0.426*** (13.69) 0.420*** (14.30) 0.428*** (14.60) 0.431*** (14.69) Fixed asset ratio 0.203*** (9.31) 0.252*** (11.40) 0.256*** (11.57) 0.202*** (9.38) 0.225*** (10.39) 0.229*** (10.55) Pseudo R Observations/ 353,077/ ,077/ ,894/ ,066/ ,066/ ,182/3968 variables. We then estimate the probit model of SEO probability using predicted values, PRLn(LM1) 1 and PRΔLn(LM1) 6, 1,as explanatory variables. Models (3) and (6) report the second stage results. 24 Because we require the analyst coverage information from the I/B/E/S data and institutional holding information from CDA/ Spectrum Institutional 13(f) filings for the first-stage regression, the sample size is reduced from 353,077 firm-quarters to 103,894 firm-quarters. Nevertheless, the second stage regression results show that the variables PRLn(LM1) 1 and PRΔLn(LM1) 6, 1 remain highly significant. Overall, the results from the six models presented in Table 10 suggest that the importance of prior liquidity improvement in SEO decisions is robust and not sensitive to model specifications. 7. Mitigating the negative SEO announcement effect and the offering price discount When investors have low concerns of liquidity risk, issuing firms face relatively low cost of equity capital, permitting them to issue shares at relatively high price. Could low concerns of liquidity risk also (i) allow issuing firms to mitigate the negative SEO 24 To save space, Table 10 omits the results from the first-stage regression. We follow closely the specification for liquidity by Fang et al. (2009).Specifically, LnðLM1Þ 1 ¼ α þ β 1 LnðLM1Þ 2 þ β 2 Market to book þ β 3 IND LnðLM1Þ 1 þ β 4 Prior stock return þ β 5 Lnð1 þ InstÞ q 1 þ β 6 LnðInstownÞ q 1 þ β 7 Lnð1 þ AnalystsÞ 1 þ β 8 Stock return volatility 1 þ ε 1 ; and Δ LnðLM1Þ 6; 1 ¼ α þ β 1 Market to book þ β 2 ΔIND LnðLM1Þ 6; 1 þ β 3 Prior stock return þ β 4 Δ Lnð1 þ InstÞ q 3;q 1 þ β 5 Δ LnðInstownÞ q 3;q 1 þ β 6 Δ Lnð1 þ AnalystsÞ 6; 1 þ β 7 Stock return volatility 1 þ ε 6; 1 ; where IND Ln(LM1) 1 is the median Ln(LM1) 1 in firm i's three-digit industry. ΔIND LnðLM1Þ 6; 1 is the median Δ LnðLM1Þ 6; 1 in firm i's three-digit industry. Marketto-book is market value of asset at December of t 1 and book value of asset for the fiscal year end in t 1. Prior stock return is the preceding 12-month stock return calculated through the end of the month before filing quarter end. ΔLn(1+Inst) q 3,q 1 is the natural logarithms of one plus the number of 13f institutions in quarter 1 minus that in quarter 3; ΔLn(Instown) q 3,q 1 isthechangeinnaturallogarithmsofinstitutionalownershipoverthesameperiod.δln(1+#analysts) 6, 1 are the natural logarithms of one plus the number of analyst in the pre-filing month 1 minus thatin month 6. Stock return volatility is the variance of OLS regression residuals where excess monthly return of stock over month 1 is regressed on the market risk premium for the full set of industrial firms.