Market Volatility and the Timing of IPO Filings

Transcription

1 Market Volatility and the Timing of IPO Filings Walid Busaba*, Daisy Li, and Guorong Yang Ivey School of Business, University of Western Ontario November 2009 Abstract We investigate how aggregate IPO filing volume responds to changes in stock market volatility. The filing volume consists of all non-financial firms that filed with the SEC between 1984 and Controlling for factors shown in the literature to impact primary market activity, notably stock market returns, we find filing volume positively related to changes in market volatility, and the relation is especially pronounced when stock market return is at normal levels, i.e. neither too high nor too low. The relation also holds at the industry level, in a pooled time-series crossindustry regression context. The relation is more pronounced for IPO filings in new industries (computers, software, electronic equipment, and telecommunications) relative to traditional industries. These results are consistent with our hypothesis that the ability to discover investor valuations before deciding to sell shares gives firms filing with the SEC an option on the uncertain offer price. This option has value not only in a strong stock market but also in a volatile market. Furthermore, option theory implies that the marginal effect of volatility on this option is highest in normal stock markets. *Corresponding author: wbusaba@ivey.uwo.ca; Tel ; Fax

2 Market Volatility and the Timing of IPO Filings Abstract We investigate how aggregate IPO filing volume responds to changes in stock market volatility. The filing volume consists of all non-financial firms that filed with the SEC between 1984 and Controlling for factors shown in the literature to impact primary market activity, notably stock market returns, we find filing volume positively related to changes in market volatility, and the relation is especially pronounced when stock market return is at normal levels, i.e. neither too high nor too low. The relation also holds at the industry level, in a pooled time-series crossindustry regression context. The relation is more pronounced for IPO filings in new industries (computers, software, electronic equipment, and telecommunications) relative to traditional industries. These results are consistent with our hypothesis that the ability to discover investor valuations before deciding to sell shares gives firms filing with the SEC an option on the uncertain offer price. This option has value not only in a strong stock market but also in a volatile market. Furthermore, option theory implies that the marginal effect of volatility on this option is highest in normal stock markets.

3 1. INTRODUCTION The Initial Public Offering (IPO) literature has documented dramatic fluctuations in IPO activity over time (e.g. Ibbotson and Jaffe 1975, Ritter 1984, and Lowry and Schwert 2002). A common explanation for the fluctuations is that firms time their offerings to take advantage of high market valuations, whether such valuations are rational or otherwise. 1 Would-be issuers wait on the sidelines until they are lured by the expectation of going public at rich valuations. We provide evidence of another type of window of opportunity, one characterized not necessarily by high expected valuations but by high uncertainty in these valuations. We find that the number of firms attempting to go public (filing with the SEC), both at the market and the industry levels, increases with measures of valuation uncertainty. Notably, and as we hypothesize, we find the effect of valuation uncertainty on IPO filing activity to be pronounced mainly when expected valuations are not high and not low. We develop our hypothesis in a simple framework of a firm timing its attempt to go public. A firm that decides to tap the public market for the first time faces significant uncertainty at the time of the decision regarding how investors will receive its offering. The firm files with the SEC and engages an underwriter to discover investor interest and determine the offer price. 2 The firm then sells shares and goes public only if the price-discovery effort yields an acceptable offer price. 3 The ability to test the waters or roll the dice and then conditionally sell shares is tantamount to the firm having a call option on the uncertain offer price. Filing for an IPO and 1 See among others Pagano, Panetta and Zingales (1998), Lowry and Schwert (2002), Pastor and Veronesi (2005), Loughran and Ritter (1995), and Baker and Wurgler (2000). 2 Fourty-four percent of U.S. IPOs between 1984 and 2004 are priced outside the range specified in the SEC filing. Those priced above the high end of the range exceed this end by 21% on average, and those priced below the low end are priced 18% on average below this end. Ritter and Welch (2002) find similarly that around 47% of IPOs between 1980 and 2001 are priced outside the filing range. 3 Between15% and 20% of filed offerings are later withdrawn (Dunbar and Foerster, 2008; Busaba, Benveniste and Guo, 2001). 1

4 engaging an underwriter creates this call option, while selling shares if price discovery yields a high offer price amounts to exercising the option (Busaba, 2006). As such, the decision to attempt an IPO is a decision to create a call option on the uncertain offer price. The attractiveness of creating such option depends on the issuer s assessment at the time of potential investor valuations, 1) how high (i.e., the moneyness) and 2) how variable. A strong stock market increases the chances of high IPO valuations, i.e., the potential for high offer prices, and drives many firms to attempt a public offering. This is in line with the existing hypotheses of IPO timing, and in line also with what we find. Our focus, however, is on how ex ante uncertainty about investor valuations can by itself drive IPO filing activity. Firms might attempt a public offering also in less of a strong, yet volatile market because of the distinct possibility of selling shares at high prices. Put differently, the option on the uncertain offer price increases in value, inducing firms to attempt a public offering, when the ex ante uncertainty surrounding this price increases. We test the hypothesis that, all else equal, higher market volatility increases the propensity of firms to attempt an IPO. We also test the more specific hypothesis, using the call option analogy, that the impact of market volatility is most pronounced when the stock market is not particularly strong or weak. 4 This is because in strong markets, the call option on the offer price is sufficiently in the money that issuers will attempt an IPO irrespective of the perceived variability in the outcomes. Symmetrically, in weak stock markets, in which would-be issuers face grim prospects, higher variability in potential offer prices will do little to induce such firms to bear the fixed costs of attempting an IPO (e.g., filing costs, printing, auditing, and legal fees, as well as management distraction). We test our hypothesis at both the aggregate market and industry levels, by analyzing the 4 An option s sensitivity to the volatility of the underlying, or Vega, is highest when the option is at the money. 2

5 time series of monthly IPO filings with the SEC between January 1984 and September Our results can be summarized as follows. First, controlling for factors other studies find to affect IPO timing, notably the strength of the stock market, we find that IPO filing volume is positively correlated with measures of market/industry volatilities, lagged and contemporaneous, especially when the stock market conditions are normal. Second, nascent industries singularly tend to experience an elevated IPO filing activity during periods of increased market and, in separate specifications, industry volatilities. This result is consistent with the view that increases in market volatility are associated on average with larger increases in volatilities in nascent industries. It could also indicate that issuers in these industries respond more noticeably to windows of opportunity presented by increases in valuation uncertainty. Following the burst of the stock market bubble in the spring of 2000, the SEC amended Rule 477 to facilitate the withdrawal of registered offerings, granting automatic approval of the withdrawal application unless it objects within 15 days. The SEC simultaneously adopted the new public-to-private safe harbor Rule 155 in order to allow firms to promptly pursue a private offering after withdrawing a public offering. (See Busaba, 2006, for details.). The stated objective of the new regulations was to reduce the legal uncertainty and the financial cost to an issuer who withdraws a public offering but still needs financing quickly. We indeed find that the adoption of the new rules had a distinct positive influence on IPO filing activity, consistent with the view that the new rules enhanced the attractiveness of the option to test the waters by reducing the repercussions of not completing offering. Our paper complements and contributes to the existing literature on IPO timing. At the conceptual level, the paper focuses on an important yet oft-ignored element of the IPO timing decision. Because the offer price is unknown at the time a firm contemplates going public, the 3

6 firm has to strategically decide when to roll the dice to discover that price. Existing studies implicitly suppress this strategic decision, analyzing directly the relation between market conditions and the time series of completed IPOs (as if offerings were filed with the full knowledge of the offer price). In line with the emphasis on the timing of the decision to attempt an IPO, our paper distinctly studies the time series of IPO filings not completions. While both the timing of the filing for an IPO and the subsequent decision to complete the offering are impacted by the strength of stock valuations at the respective times, the filing decision is uniquely impacted by the uncertainty in stock valuations, which is the subject of our analysis. Observing a flurry of offerings completed in a strong stock market tells only a part of the story of why the respective companies would have decided several weeks earlier it was time to file for an offering. The companies might very well have betted on the strengthening of the stock market. Alternatively, they might have been induced by an elevated level of stock market volatility, which presented the would-be issuers with an option on the uncertain valuation by the market. Symmetrically, a period of high IPO filing volume, driven either by the expectation of future strong valuations or by uncertainty in these valuations, might happen to be followed by a weak stock market and a correspondingly cold completed IPO market. In this case also, the completed IPO volume would fail to reflect the market-timing component of the decision to attempt a public offering. By studying the time series of IPO filings, we are better positioned to shed light on the timing element of the decision to attempt an initial public offering. Empirically, the monthly series of completed IPOs is not an adequate proxy for the monthly filing series in the study of how firms time their attempt to go public. There are relatively large differences between the two series month to month. In our sample of slightly 4

7 longer than 20 years, the absolute monthly difference between the two series, divided by the corresponding number of competed IPOs, is on average 69% (median is 36%). The correlation between contemporaneous changes in the two series is merely 30%. These statistics notwithstanding, we run our regressions using the completed IPO series instead of the filing series as the dependent variable. Our main results on the impact of stock market volatility on IPO activity are mostly eliminated, and sometimes reversed, with this substitution. This experiment not only illustrates that our results are uniquely related to the filing decision, as we hypothesize, but also helps position these results relative to the existing literature. We illustrate this latter point in the next section. The remainder of the paper is organized as follows. Section 2 relates our paper to the literature. Section 3 discusses data and methodological issues, while Section 4 presents the results. Section 5 presents robustness checks and Section 6 concludes. 2. RELATION TO THE LITERATURE The impact of market volatility, or valuation uncertainty in general, on IPO cycles has secondary attention in the literature so far received. The general tone has been that uncertainty in market valuations can be a disincentive to tapping the public market. Theory suggests a positive correlation between ex ante valuation uncertainty and underpricing (e.g., Rock, 1986, and Benveniste and Spindt, 1989), and consistent empirical evidence abounds. However, this tone ignores the option value provided by the uncertain market valuations, for firms attempting to go public. Busaba (2006) shows theoretically that this option value increases faster than underpricing as valuation uncertainty increases, supporting the hypothesis we test here that the higher uncertainty in and of itself provides an incentive for firms to attempt a public offering. 5

8 A couple of empirical papers examine the impact of market volatility on aggregate IPO volume while testing hypotheses unrelated to ours. Schill (2004) argues that a higher stock market volatility increases a firm s exposure to offer price risk, share distribution risk and aftermarket performance risk, and thus is expected to have a negative impact on firms financing behavior. He studies the monthly series of completed IPOs and finds that high market volatility dampens IPO activity. Periods with above normal market volatility are associated with a significant 13% decline in the frequency of completed IPO and a 21% decline in the dollar amount raised. Our analysis differs from Schill s in many respects. First, as we noted above, we study the timing of the decision to attempt an IPO and, for that reason, analyze the monthly volume of SEC filings. When we estimate a regression specification like his (Table 3; column (1); page 670), also using the series of completed IPOs, we are able to replicate his finding. When we then use the series of IPO filings instead, the negative effect of market volatility becomes insignificant. Furthermore, the format of our hypothesis is different from his, and the specifications we estimate reflect this difference. We hypothesize and provide evidence that market volatility induces IPO filing especially when the stock market is in a normal state (not hot and not cold). The two studies, therefore, are complementary. Pastor and Veronesi (2005) develop a model in which the optimal timing of a firm s IPO is driven by the firm s market value, which is negatively related to the discount rate and positively related to the value of the firm s future growth options (expected future cash flows and ex ante variability in those cash flows). They find that the completed IPO volume is negatively related to recent market return volatility, which they use to proxy for the higher discount rate. The effect of market volatility on IPO volume is only through the effect on market returns, 6

9 however. So the two variables are included one at a time, not together, in a regression of IPO volume (Table VI, p. 1741). Our analysis differs from Pastor and Veronesi s in several ways. We are able to generate their result on the (negative) effect of market volatility when we use the time series of IPO completions in a regression specification like theirs. However, this result largely disappears when we then use the series of IPO filings. Furthermore, our specifications include market volatility together with market returns, both as standalone variables and as interactive terms. Hence, the positive effect we document of market volatility on IPO filing volume holds after controlling for the effect of market returns. Again, our paper and theirs are complementary. Finally, in a cross-sectional analysis of a sample of private German firms, Boehmer and Ljunqvist (2004) find a positive but generally insignificant relationship between IPO timing and valuation uncertainty, which is measured by the volatility of stock returns in the sample firm s industry. The literature has proposed and focused on other factors as potential drivers of the time variation in IPO activity. We control for these factors in our study. One such factor is time variation in real investment opportunities and technological innovation. Ritter (1984) finds the hot IPO market of 1980 to be associated almost exclusively with natural resource companies. Helwege and Liang (2005) find, on the other hand, that IPO firms in hot and cold markets between 1975 and 2000 do not appear to differ in their growth prospects or future operating performance, thus suggesting a weak link between investment opportunities and IPO volume. Both the hot and cold market IPOs are drawn largely from the same narrow set of industries which have accounted for most of the issuance in the sample period. Information spillover is also considered as a driver of hot markets. Lowry and Schwert 7

10 (2002) document strong autocorrelation of both IPO volume and initial returns. They find that more companies tend to file for IPOs and fewer companies withdraw their offerings following periods of high initial returns, because high initial returns are associated with positive information revealed during the registration period of the offerings. Benveniste, Busaba and Wilhelm (2002) argue that information produced during the marketing of an IPO is relevant to private firms in related industries. Given sufficient market power, investment banks can enforce more uniform sharing of the costs of information production by bundling together IPOs in industries. Benveniste et al predict a high degree of clustering by industry and a negative correlation between average initial returns and IPO volume among firms subject to a common valuation factor. Benveniste, Ljungqvist, Wilhelm and Yu (2003) find evidence that, although initial returns and IPO volume are positively correlated at the aggregate level, the correlation is negative among contemporaneous offerings subject to a common valuation factor. Another explanation of the fluctuation in IPO volume is time variation in the level of information asymmetry and adverse selection costs (Choe, Masulis and Nanda, 1993). Lowry (2003) tests the relative power of several potential determinants of IPO volume the demand for capital (i.e. investment opportunities), information asymmetry and investor sentiment hypothesis. Her empirical evidence suggests that both firms demand for capital and investor sentiment are important determinants of IPO volume in statistical and economic terms, while adverse-selection costs have a statistically significant but economically insignificant effect. 3. DATA 3.1 IPO Filing Volume We study the monthly series of IPO filing volume with the SEC from January 1984 until 8

11 September The series is constructed from the completed IPO database, the withdrawn issues database (available starting 1984), and the issues-in-registration database of the Securities Data Company s (SDC) Global New Issues. We then verified the filing date with that in SEC EDGAR. After excluding ADRs, GDRs, Beneficial Interests, Capital shares, Income Depositary, Limited Liability/ Partnership Interest, Trust Receipts, Units, Master Limited Partnership, SPERs, REIT, financial firms, the final sample includes a total of 8,681 filed offerings, of which 6790 were completed, 1750 withdrawn, and 141 still in registration at the sample period s end. 5 The sample is characterized by computer and biotech companies in the late 1980s and internet companies in the late 1990s and early 2000s. Figure 1 illustrates the fluctuations in IPO filing volume during the sample period. We scale the number of filed offerings during month t by the total number of public firms in existence at the beginning of the month, and use the result, SFILG t, as the dependent variable in the time-series regression analysis of IPO filing activity. Scaling the IPO filing series is intended, as in Lowry (2003) and Pastor and Veronesi (2005), to avoid potential non-stationarity. The scaled series (SFILG) does exhibit auto-correlation the first-order autocorrelation of SFILG between 1984 and 2004 is 0.79 yet the Dickey-Fuller test does not show evidence of non-stationarity. 3.2 Valuation Uncertainty Our focus is on whether and how perceived valuation uncertainty impacts IPO filing volume. Issuers assessment of valuation uncertainty is naturally a function of what they expect 5 To ensure we have identified all issues that were filed during the sample period, we examined the filing date of all IPOs completed up until April 2005 and all issues still in registration up until June Eight completed IPOs in our sample have a missing filing date in EDGAR. Four of those were found in the CRSP database and assumed to be filed two months before their listing (one assumed filing date fell before January 1984 and thus the issue was excluded). The remaining four issues were deleted from the sample. 9

12 aggregate stock market volatility to be over the ensuing price-discovery period. Furthermore, in a survey of chief financial officers (CFOs), Brau and Fawcett (2006) find that the CFOs identify overall market conditions as the single most important determinant of the timing of IPOs. Industry conditions rank second while current IPO market conditions, such as the number of other good firms going public and first day stock performance of recent IPOs, are relatively less important. An obvious proxy for market-wide valuation uncertainty is implied volatility from options on broad stock indices. The implied volatility reflects the views of market participants and represents an objective, consensus forward-looking forecast of volatility. Furthermore, implied volatility indices can be tracked daily (even intraday) by firms contemplating filing for an IPO. We consider three CBOE implied volatility indices: 1) VXO, based on prices of S&P 100 options, which for convenience we denote by IVOL SP100 ; 2) VIX, based on prices of S&P 500 options, denoted as IVOL SP500 ; and 3) VXN, based on prices of NASDAQ-100 options, denoted as IVOL NQ All three implied volatility indices represent a 30-day forward-looking estimate of volatility of the underlying indices, a period that coincides with the road shows and IPO price discovery. IVOL SP100 has a price history dating back to 1986, IVOL SP500 to 1990, and IVOL NQ100 to Our data on IVOL NQ100 start from Therefore, IVOL SP100 is our primary proxy as it provides the largest number of observations. We employ IVOL SP500 as well in the analysis of the sub-sample 1995:1-2004:9 in Section 5. As shown in Figure 2, IVOL SP500 and IVOL SP100 closely track each other, while the IVOL NQ100 series exhibit higher values than both IVOL SP500 and 6 Prior to September 2003, VXO was actually named VIX. Since CBOE adopted a new formula for calculating VIX in September 22, 2003, the original index has been disseminated under the new ticker symbol VXO. The calculation of VXO includes only near-the-money options, while the calculation of VIX takes into account a broader range of strike prices. Starting September 22, 2003, VXN was calculated with the same new formula and methodology as VIX. 10

13 IVOL SP100. As an alternative to implied volatility, we consider standard deviations of historical stock market returns. We calculate monthly standard deviations of daily CRSP equally-weighted NYSE/AMEX/NASDAQ returns, denoted henceforth by HIST_VOL ALL, as well as of daily NASDAQ equally-weighted returns. However, we find that both measures are Granger (1969)- caused by filing volume, specifically during the last ten years of the sample (1995:1-2004:9), though the coefficients on the IPO volume terms are small. 7 Hence, we report (Section 5) results of alternative regression specifications employing HIST_VOL ALL in lieu of IVOL SP100, estimated only over the earlier sub-period 1984:1-1994: Other Explanatory Variables Factors that the literature suggests have an impact on firms decision to go public and thus the aggregate IPO filing volume include stock market returns, investment opportunities, technological innovations, the extent of information asymmetry, macroeconomic conditions, and the cost of debt capital. To rule out biases associated with omitted variables or other model misspecifications, we estimate regression equations with various subsets of these factors. We use past, current, and future CRSP monthly NASDAQ equally-weighted returns, MRT, to capture the market timing effect or the rational IPO waves in Pastor and Veronesi (2005). If firms tap the public market to take advantage of temporary high stock valuations, IPO filing waves should be preceded by strong stock returns and followed by weak returns. A similar pattern would emerge also if, as Pastor and Veronesi propose, firms go public when the cost of 7 The null hypothesis of the Granger-causality test is that forecasts of market volatility cannot be improved by using past observations of IPO volume in addition to past observations of volatility. Three lags are chosen in our Granger tests because Schwarz Information Criterion does not support longer lags. We found no evidence that, the filing volume Granger-causes the implied volatility measures. 11

14 equity (or expected return) drops. A drop in expected returns leads to an immediate increase in stock prices and, hence, in current period realized returns. It also leads to a drop in realized returns in future periods. Under either hypothesis, therefore, current month IPO filing volume is expected to be positively correlated with market returns in previous months and negatively correlated with returns in subsequent months. A monthly series of market-wide market-to-book ratio, MB, is calculated as a proxy for investment opportunities and/or investor sentiment. The variable MB in a certain month is the equally-weighted average of the market-to-book ratios for that month of all NYSE/AMEX/NASDAQ listed firms, excluding firms that went public during the past three years (to reduce potential endogeneity) and, as in Pastor and Veronesi (2005), firms whose market value or book value are smaller than one million dollars. Firms with a market-to-book ratio larger than 100 or smaller than 0.01 are also excluded from the calculation. The market value of equity of an individual firm in a certain month is the closing share price multiplied by the common shares outstanding at the end of the month, both obtained from CRSP. 8 The book value of equity for that firm is calculated as in Fama and French (1992). For any month from July of calendar year j to June of calendar year j+1, book value is measured at fiscal year ending anytime in calendar year j-1. 9 Following Lowry (2003) and Pastor and Veronesi (2005), we calculate the book value of a firm s equity as the book value of stockholder s equity plus deferred taxes and investment tax credits, minus the value of preferred shares, which is equal to redemption, liquidation or carrying value (in this order). Pastor and Veronesi (2005) find that the volume of completed IPOs increases in prior 8 We match the individual firms of Compustat and CRSP by the 6-digit CUSIP numbers. 9 This ensures that there is at least 6 months from the fiscal year end to the month over which IPO filing volume is being observed. The idea is that managers calculate market-to-book ratios based on available accounting data, which are filed typically with several month delay after fiscal year end. 12

15 uncertainty about the average future profitability of issuing firms, and declines in the discount rate. (The first variable is thought to increase the market value of young and growing firms, which is mostly growth options, and the second to reduce these firms market value.) We calculate a proxy for prior uncertainty as described in Pastor and Veronesi (2005), new firm excess volatility, VOL EXC, by subtracting the monthly standard deviation of market returns from the median monthly standard deviation of all new firms returns. New firms are defined as those which first appeared in the CRSP daily file in the previous month. As Pastor and Veronesi (2005), we require at least three completed IPOs in a month for the calculation of a valid new-firm standard deviation. We also use the risk-free rate of return, RF, measured by the yield on 90-day Treasury bills, as a proxy for the discount rate in particular and macroeconomic conditions in general. Lowry and Schwert (2002) find that higher initial returns are followed by a larger number of IPOs, because high initial returns indicate positive information. To control for this effect, we use the monthly average initial returns of IPOs, IR, from Jay Ritter s website. Following Lowry (2003), we include among the control variables the dummy EXPAN which equals 1 if a month is in the expansionary phase of a business cycle, as defined by NBER, and zero otherwise. Moreover, we use the real monthly continuously-compounded growth rate in industrial production, Grwth_IP, to capture the potential effect of technological innovations (fueling this growth) on the filing volume. Like Schill (2004), we control for expected economic growth by including the average of months t-2, t-1, and t growth rates in Conference Board s composite index of leading indicators, Lead. The slope of the term structure, Term, measured as the difference between the 10-year Treasury bond yield and the 90-day Treasury bill yield, is included to control for the relative cost of long term financing. We include the difference 13

16 between Moody s seasoned BAA corporate bond yield and AAA corporate bond yield, BAA-AAA, to measure the effect of the relative cost of corporate debt financing. It should be noted here that when considering explanatory variables, we are mindful of two potential problems. One concerns whether the variables capture the intended factors. The other is that a control variable might reflect more than one underlying economic factor. For example, the market-to-book (MB) ratio measures the general stock market condition in Lowry (2003), investment opportunities in Pastor and Veronesi (2005), and investor sentiment in Rajan and Servaes (2003). To keep focus on our hypothesis the impact of valuation uncertainty on the decision to attempt an IPO we control for variables that are found significant in prior studies but do not restrict our interpretation to a certain factor. 3.4 Industry-level Variables We conduct our analysis of IPO filing activity also at the industry level. We form 12 industries based on Fama and French (1997) industry specification, as specified in Table A.1 in the appendix. We exclude from the analysis the finance industry (SIC ), leaving 11 industries in total. Table A.2 in the appendix provides a summary of total filing volume and industry compositions for the period from January 1984 through September Industry-wide MB ratios, Ind_MB, are calculated in a similar fashion to the market-wide MB ratios. We use in the industry-level analysis the monthly equally-weighted industry stock returns, Ind_RT, and the monthly standard deviations of daily industry returns, VOL IND. 10 The calculation of industry-level MB ratios and stock returns excludes the stocks of firms that went public during the preceding three years. Other variables used in the market-level regressions remain the same in the 10 Stock returns at the market level are NASDAQ returns, while industry-level stock returns are based on the NYSE/AMEX/NASDAQ firms that constitute the various industries. 14

17 industry-level analysis. This is particularly true for implied volatilities, which are unavailable at the industry level. Table 1 lists the variables used in the regression analysis, the variables definition, and sources of the data for each. 4. REGRESSION ANALYSIS 4.1 Market-Level Regressions We estimate a multivariate time-series regression model in which the dependent variable is the scaled level of the monthly IPO filing volume, SFILG t. To account for serial correlation in the data, we follow Pastor and Veronesi (2005) and regress the dependent variable against its first lag, SFILG t-1, and against the first difference,, in the other series of independent variables (except for stock market returns). Pastor and Veronesi (2005) argue that regressing levels on changes is appropriate because IPO volume is driven by changes in market conditions. Boehmer and Ljungqvist (2004) also support this regression methodology, as they find that recent returns matter more than the level of MB ratios. Furthermore, spurious correlations are less likely to occur with variables in differences than with variables in levels. Table 2 presents summary statistics and pair-wise correlations of selected variables. Since the pricing of an IPO happens on average a couple of months after filing the registration statement with the SEC, a company that is considering filing with the SEC makes the decision based on what it expects valuation uncertainty to be in the weeks that follow. Assuming that firms form expectations based on current and past market conditions, or that there is a lead time between when firms observe the favorable market conditions and then when they get to file with the SEC, we include as explanatory variables both current and lagged volatility measures. In 15

18 this respect, we consider three lags of the change in the volatility variable. 11 The regression analysis is done in steps. First, we estimate specifications that include the volatility terms as stand-alone variables. Then we introduce interactive terms between the volatility terms and the stock market return, to test our hypothesis that the effect of volatility on filing volume should be more pronounced in moderate-strength stock markets Baseline Specification We first estimate the following regression equation SFILG t 0 3 p 0 VOL 1, p t p γx t The dependent variable SFILG t is the scaled number of offerings filed with the SEC during month t. The independent variable VOL t-p is the change in the volatility measure, IVOL SP100 in this case, in month t-p, (p = 0,1, 2, and 3). We use the generic notation VOL instead of IVOL SP100 to allow for the inclusion of alternative volatility measures later in the paper. The vector X contains the other control variables discussed in Section 3 earlier. It is worth clarifying here that the control variable VOL EXC, monthly change in new firm excess volatility, is included in three months lag terms as well (consistent with Pastor and Veronesi, 2005). Estimation results are reported in Table 3, Column (1). The coefficients on the current 11 Given the high time-persistence of stock market volatility, GARCH has been popular in forecasting volatility. We have tried GARCH(1,1) and EGARCH(1,1) estimates of future volatilities in our regressions, using monthly equally-weighted NASDAQ returns. However, most of the coefficients on GARCH estimates are statistically insignificant, probably because the monthly data ignore important information contained in daily stock prices. This finding is consistent with Schill s (2004) finding that the GARCH(1,1) estimates are uncorrelated with financing transactions. 16

19 and lag terms of ΔVOL are positive, in line with our hypothesis, and those on both the 2-month (ΔVOL t-2 ) and the 3-month lags (ΔVOL t-3 ) are statistically significant, at the 5% and 1% levels respectively. The F-test rejects the joint hypothesis that all coefficients of the volatility terms (current and lags) are zero. It is important to note that the relation we find between filing volume and valuation uncertainty holds in the presence of control variables suggested in the literature. And the effect of these control variables on filing volume is consistent with that (on completed IPO volume) documented in the literature. The estimated coefficients for current-month and future stock returns are negative in general and, for past-months stock returns positive, consistent with the market timing hypothesis as well as with Pastor and Veronesi s (2005) rational IPO waves. The current-month change in new firm excess volatilities have a positive and statistically significant effect on IPO filing volume, again consistent with evidence on completed IPO volume in Pastor and Veronesi. Monthly changes in market-to-book ratios, MB t, are highly significant and positively related to the monthly IPO volume. Consistent with Lowry and Schwert (2002), IPO filing volume is positively related to monthly changes in average IPO initial returns, IR t. The business cycle dummy (EXPAN t ) as well has a positive and statistically significant coefficient. The one-month lagged filing volume has a significant effect on current IPO filing volume, and this serial correlation could reflect information spill-over effects in Benveniste et al (2003). On the other hand, the coefficients on the risk-free interest rate, default premium, slope of the yield curve, growth rate of industrial production, and average growth rate of composite index of leading indicators are all statistically insignificant. This is consistent with Helwege and Liang (2005), who provide evidence that technological innovations are not the primary determinants of the frequent and dramatic swings in the volume of IPOs over time. Boehmer and Ljungqvist 17

20 (2004) also find that macroeconomic conditions and the cost of debt capital have little effect on IPO timing. Lowry (2003) finds the extent of information asymmetry to not be a significant determinant of IPO cycles. We drop these control variables from the baseline specification, the specifications in the next subsection, and henceforth. Estimating the baseline specification without these variables leaves the results literally unchanged, as demonstrated by Column (2) of Table Main Specification If higher valuation uncertainty tempts firms to roll the dice and attempt an IPO (creating an option on the uncertain offer price), option theory would suggest that this effect is most pronounced when expected valuations are neither strong nor weak. We test this hypothesis by first identifying with a dummy variable the months in which the stock market return can be considered at a normal level. This variable, denoted as Mdm.MRT t, has a value of 1 when the market return MRT t is between the 33 rd and 67 th percentiles of the monthly returns for the entire sample, and zero otherwise. 12 We then include in the regression specification an interactive term between this dummy variable and each of the contemporaneous and lagged changes in the volatility measures, VOL. The specification we test is therefore SFILG t 3 l 0 0 2, l 3 p 0 ( VOL VOL p ) EXC t l t p q 0 Mdm. MRT VOL t q t t t q t k 3 1, k MRT MB IR EXPAN SFILG t 1 t k t 12 We attempt alternative definitions of the normal market return dummy, including when MRT t is between the 33 rd and 67 th percentiles of the monthly returns over the 2 years surrounding month t (one year before and one year after); and over the 4, 5, 10 and 11 years surrounding t; We also attempt a 25 th -75 th percentile band. Our main results and conclusions do not change. Detailed results are available upon request. 18

21 Again, our focus is whether and how market-wide valuation uncertainty, as a standalone variable ( VOL t-p ) and, particularly, conditioned on the strength of the stock market (Mdm.MRT t VOL t-p ), drive IPO filing activity. The estimation results are presented in Column (3) of Table 3. The standalone 2-month lag ΔVOL t-2 loses its significance in this specification while the 3-month lag (ΔVOL t-3 ) continues to show up with a positive and highly significant coefficient. Notably, however, all four interactive terms turn up with a positive coefficient, with Mdm.MRT t VOL t being significant at the 1% level and Mdm.MRT t VOL t-1 at the 10% level. Results related to the control variables are all but unchanged. Most relevant among those are market returns, for which lead terms typically have negative coefficients, some significant, and lag terms positive coefficients, some significant, in line with the literature. Also relevant is the change in new firm excess volatility, which shows up with a positive and marginally significant coefficient. These results provide direct support to our hypothesis that, holding all else constant, higher market volatility drives more companies to attempt to go public, especially in periods when stock returns are normal, i.e., neither too high nor too low. In the wake of the burst of the internet bubble starting April 2000, the U.S. Securities and Exchange Commission (SEC) amended Rule 477, effective March 2001, to facilitate the withdrawal of registered offerings, granting automatic approval of the withdrawal application unless it objects within 15 days. The SEC simultaneously adopted the new public-to-private safe harbor Rule 155 in order to allow firms to promptly pursue a private offering after withdrawing a public offering. (See Busaba, 2006, for details.) The stated objective was to reduce the legal uncertainty and the financial cost to an issuer who withdraws a public offering but still needs 19

22 financing quickly. In the framework of our hypothesis, the new rules would enhance the attractiveness of the option to roll the dice by reducing the repercussions of not completing the offerings, hence encouraging more firms to attempt public offerings. One way to empirically measure the impact of the new rules on the decision of firms to file with the SEC is through a dummy variable which equals 1 for March 2001 and subsequent months, and 0 otherwise. However, the problem is that this period partially coincides with a bear stock market in which filing activity was low. We disentangle the (positive) effect of the new rules from the negative effect of the bear market in two steps. First, we introduce to the specification a dummy variable, BUST, to indicate the months during the down stock market, April 2000 until April We expect IPO volume to be lower during this period relative to the other months in the sample. Then, we introduce an interaction term of BUST and the dummy variable SEC_D01 which indicates the months March 2001 through April This interactive term would capture any marginal effect the SEC rules had on potential issuers during the down market. Relative to the bust months prior to the adoption of the new SEC rules, we expect the months after to have a higher filing volume if the SEC rule were affective. Column (4) of Table 3 presents the estimation results when BUST and BUST SEC_D01 are added to regression specification. The coefficient of the dummy variable BUST is negative as expected and significant at the 10% level. The coefficient of the interactive term BUST SEC_D01 is positive, also as expected, with a t-statistic of 1.47 though not significant at conventional statistical levels. There is evidence, albeit mild, that the adoption of the new SEC rules in March 2001 had the effect of stimulating IPO filing activity at the margin, in line with 13 It is unclear when this bear market can be considered to have officially ended, the spring of 2002, the summer of that year or, as Wikipedia states ( October Our choice of April 2002 for the dummy variable ensures that the chosen date is undoubtedly during the bear market but close enough to its end. We tried months later than April 2002 including, in particular, October 2002, and obtained literally the same results. Details of this analysis are available upon request. 20

23 our hypothesis that firms view filing with the SEC as creating an option to go public conditional on investor reception. We revisit this question later in the analysis of filing volume during the sub-period 1995: :09. The Granger-causality tests notwithstanding, we repeat the estimation of the baseline and main regression specifications but with VOL set equal to the monthly standard deviation of daily CRSP equally-weighted NYSE/AMEX/NASDAQ returns, HIST_VOL ALL, in lieu of IVOL SP100. The standalone VOL t-3 shows up with a positive and statistically significant coefficient, just like in the corresponding baseline specifications, Columns (1) and (2) of Table 3. In regression specifications comparable to Columns (3) and (4) of Table 3, VOL t-3 turns up with a positive and marginally significant coefficient. The interactive terms, Mdm.MRT t VOL t-p, p = 0,...,3, all turn up with positive coefficients, like in the corresponding specifications that use IVOL SP100, with those corresponding to lags p = 0 and 2 being significant at the 10% level and that to p = 1 at the 5% level. The results regarding the other control variables are literally identical to those in Table 3. Finally, we also attempted the standard deviation of daily NASDAQ returns in lieu of HIST_VOL ALL, and found the regression results of both baseline and main specifications qualitatively unchanged, albeit marginally less significant. It appears, therefore, that support for our hypothesis on the impact of valuation uncertainty on IPO filing activity is robust to the use of the standard deviation of historic market returns as a proxy of future valuation uncertainty (maintaining the caveat that this variable might not be exogenous relative to IPO filing volume). 4.2 Industry-Level Regressions To test our hypothesis at the industry level, we determine the monthly IPO filing volume 21

24 for each of the 11 Fama-French industry classifications identified in the previous section, and then run a pooled cross-sectional, time-series analysis of filing volume. Categorizing the entire sample into 11 industries significantly decreases the number of filings per individual months. Some industries have a zero filing volume in a number of months, which means that the industry-level filing volume is truncated at zero. We therefore use a Tobit model instead of OLS. 14 The dependent variable, Ind_SFILG i,t,, is the observed number of filings in industry i in month t, scaled as in the market-level analysis by the total number of public companies in that industry at the beginning of month t. The independent variables used in the industry-level regressions are similarly defined as those used in the market-level regressions, except as described before in the Data section: The industry-level MB ratio (Ind_MB), equally weighted stock returns (Ind_RT), and standard deviations of the equally weighted stock returns, VOL IND, replace their market-level counterparts. Two volatility measures are considered in this regression, IVOL SP100 and VOL IND. The former is our main variable, being exogenous, forward-looking, and readily observable by market participants. The second is used to check robustness. It is also exogenous given how it is calculated, as described earlier in the Data section. We define a new-industry dummy variable, New.Ind, to capture the potential differential impact of market volatility on the filing decision of high technology firms. The new industry category comprises computer-related companies (computers, software, and electronic equipment) as well as telecommunication firms. Innovative activities are associated with higher idiosyncratic risks (e.g. Mazzucato 2002, and Shiller 2000). Given Campbell et al s (2001) finding that volatilities at the market, industry and firm levels are related, it is reasonable to 14 At the market level, only two months in the entire sample have a filing volume of zero. We run Tobit at the market level as a robustness check, and the results are very similar to those obtained from OLS. 22

25 assume that when the market-wide/industry-wide valuation uncertainty rises, high technology firms will be associated with an amplified level of valuation uncertainty. We hence conjecture that the filing activity of these firms will respond more to increases in market/industry valuation uncertainty than will the filing activity of traditional companies. To test this effect, we include among the independent variables an interactive term between the new-industry dummy and the month t change in the volatility measure. Also included among the independent variables individual industry dummy variables, INDi (10 in total, excluding Consumer Non-durables), to control for potentially missing withinindustry factors. We do not include a time dummy variable, on the other hand, since the time series is long (249 months) relative to the number of the cross sections. The final specification is as follows: Ind _ SFILG * i, t Ind _ MB t p 0 i, t VOL p t t i, t p q 0 BUST BUST SEC _ D Mdm. Ind _ RT q k 3 t 1, k i, t VOL i, t k IR EXPAN Ind _ SFILG Ind _ RT 6 3 l 0 i, t 1 i, t q 2, l IND 11 New. IND VOL ( VOL i EXC i, t l t, i ) i i, t where i, t Ind _ * i, t Ind _ SFILG SFILG if *, t Ind _ SFILG i > 0 Ind _ SFILG i, t = 0 if *, t Ind _ SFILG i < 0 Table 4 presents the results of Tobit estimation of this specification, organized in a way similar to Table 3 for ease of comparison. The influence of volatility on IPO filing volume is evident in two ways. First, we find that higher volatility significantly increase the IPO volume, especially when the industry-level returns are neither too high nor too low. In the specification without the interactive terms between volatility and industry returns (Column (1)), the standalone 23

26 2-month and 3-month lagged IVOL SP100 are both positive and statistically significant at the 5% and 1% respectively. This is consistent with the results we observe at the market level. In Columns (2) and (3), the dummy variable of normal market condition interacted with current and three lags of IVOL SP100 all have positive and statistically significant coefficients. All such coefficients except that of the 2-month lag are in fact significant at the 1% level. In the specifications which replace IVOL SP100 with the industry-level volatility VOL IND, the standalone 3-month lagged VOL IND turns up with a positive coefficient that is significant at the 10% level (Column (4)). This variable loses its significance in the presence of interaction terms with normal industry returns (Columns (5) and (6)), but the coefficients on the interaction terms associated with 1- and 2-month lagged VOL IND in Column (5) and 1-month lagged VOL IND in Column (6) turn up positive and significant. An F-test rejects the null hypothesis that all four interaction terms are simultaneously zero. Second, the interaction of the new industry dummy and the contemporaneous change in the volatility measure shows up with a positive and highly significant coefficient, irrespective of how valuation uncertainty is measured. Changes in market-wide volatility have a larger influence on IPO filings of high technology firms than on filings traditional firms. Inference about the effect of other variables on IPO volume remains the same as at the market level. The dummy for computer-related industry and telecom industry are positive and highly significant, indicating that these two industries have had higher monthly IPO filing volumes during the sample period than have other industries. In general, the industry-level analysis provides even stronger evidence than the marketlevel analysis in support of our hypothesis. This is probably because the industry-level regressions control more adequately for industry factors missed in the market-level regressions. 24