Product Market Advertising, Heterogeneous Beliefs, and the Long-Run Performance of Initial Public Offerings

Transcription

1 Product Market Advertising, Heterogeneous Beliefs, and the Long-Run Performance of Initial Public Offerings Thomas Chemmanur* and An Yan** Current Version: June 2016 * Professor, Finance Department, Fulton Hall 440, Carroll School of Management, Boston College, Chestnut Hill, MA 02467, Tel: (617) , Fax: (617) , chemmanu@bc.edu. ** Professor, Finance Area, Graduate School of Business Administration, Fordham University, 113 West 60 th Street, New York, NY 10023, Tel: (212) , Fax: (212) , ayan@fordham.edu.

2 Product Market Advertising, Heterogeneous Beliefs, and the Long-Run Performance of Initial Public Offerings Abstract We study the long-run effects of product market advertising on the equity of firms conducting initial public offerings (IPOs). We find that firms going public with a greater extent of advertising prior to its IPO are valued higher both in the IPO as well as in the immediate aftermarket, are associated with greater upward price revisions from the pre-ipo filing range means, and has lower long-run post-ipo stock returns. The above results hold even after controlling for the effects of investor attention, as proxied by the pre-ipo media coverage received by firms going public. We show, using a number of additional tests, that the above findings are consistent with the implications of the heterogeneous beliefs theories of Miller (1977), Harrison and Kreps (1978), and Morris (1996), along with an assumption that product market advertising increases the heterogeneity in outside investor beliefs about firms going public.

3 1. Introduction The role of product market advertising on financial markets has recently begun to interest financial economists. Anecdotal evidence indicate that at least some firm managers take into account the effect of product market advertising not only on their customers but also on investors in their financial assets, especially during the period of going public. For example, a article published in Wall Street Journal ( Advertising Blitz? There Must Be an IPO Involved, May 30, 2011) pointed out the link between product market advertising and a firm s initial public offering (IPO) in Hong Kong: Hong Kong is awash in advertising from companies seeking billions of dollars from local share listings. The examples quoted in the article include: Samsonite International SA, which advertised heavily before it upped the size of its IPO from US$1 billion to $1.51 billion; Prada SpA, which ran an advertising campaign around Hong Kong while waiting in the wings with an IPO of $2 billion; L Occitane International SA, whose IPO was preceded by a marketing blitz and ended with 160 times oversubscription; and even insurance giant AIA Group Ltd., which bought considerable advertising space around Hong Kong under the slogan The Power of We before its giant IPO despite already being a household name with a strong customer base in HK. 1 Similarly, our sample of U.S. IPOs from also suggests that many firms advertise heavily prior to their IPOs. For example, Warner Music Group, which went public in November 2005, increased its advertising expenditures from $147 million in 2004 to $195 million in Subsequent to its IPO, Warner s advertising expenditures decreased to $144 million in 2007 and 1 Some firm managers also take into account the effect of product market advertising on investors in the secondary markets. For example, in the spring of 2002, Ford Motor Corp. ran a series of advertisements featuring Chairman and CEO Bill Ford Jr. without mentioning any products. The only purpose of these ads was clearly to boost confidence among investors in the firm. Also consider the following quote from a Business Week article ( What Price Reputation? July 9, 2007) on the advertising campaign by United Technologies Corp. (UTC): the color schematic of UTC s Silorsky S-92 copter is embedded with messages aimed at Wall Street... The underlying theme: UTC is a great investment because it is a leader in innovation and eco-friendly technologies that help the bottom line. 1

4 2008. Coincidentally, many financial analysts (e.g., Richard Greenfield at Fulcrum Global Partners) suggested that Warner s offering price represented an undeserved premium over EMI Group, a bigger music publishing unit with less debt. This phenomenon also exists across different industries. For example, consider the Business Services industry, which is defined based on the Fama-French 48 industry classification and is the biggest industry in our sample (with 256 IPO firms). The average advertising intensity (advertising/sales) of the IPO firms in the Business Services industry in the years prior to their IPOs is from 1990 to 2007 and from 1997 to In comparison, according to the industry advertising data from Schonfeld & Associates ( the average advertising intensity of all the IPO and non-ipo firms in the Business Service industry is from 1997 to Also consider the Retail industry, which is the second largest industry in our sample (with 121 IPOs). The average advertising intensity of these 121 IPO firms in the years prior to their IPOs is from 1990 to 2007 and from 1997 to In comparison, according to Schonfeld & Associates, the average advertising intensity of all the IPO and non-ipo firms in the same industry is from 1997 to Clearly, the IPO firms outspend their publicly traded industry peers prior to their IPOs. The preliminary evidence from our IPO sample indicates that the high level of advertising expenditures prior to IPOs is not driven by product market considerations alone (i.e., not driven solely by the need to improve sales and profitability). In particular, if we compare the IPO firms with high and low advertising intensity in the years prior to their IPOs, the sales of low advertising intensity IPO firms are almost twice as large as the sales of high advertising intensity IPO firms. The profitability (EBIT/Assets) of low advertising intensity IPO firms is on average 9.9% while the profitability of high advertising intensity IPO firms is on average -6.7%. More 2

5 interestingly, we also find that the negative operating profits of many IPO firms in our sample is driven primarily by their advertising expenditures. For example, United Auto Group, prior to its IPO in October 1996, spent $10.71 million in advertising in 1995 and its EBIT in the year ended at -$3.70 million. Similarly, Adams Golf Inc., prior to its IPO in July 1998, spent $8.65 million in advertising in 1997 and its EBIT in the year ended at -$3.67 million. Recently, Constant Contact Inc., prior to its IPO in October 2007, spent $9.78 million in advertising in 2006 and its EBIT in the year ended at -$6.10 million. Teoh, Welch, and Wang (2002) suggest that IPO firms have a tendency to manage earnings upward prior to their IPOs. Considering this tendency, it is puzzling why many IPO firms still sacrifice their earnings by spending heavily in their advertising prior to their IPOs. However, there is very little research analyzing the effects of product market advertising around IPOs on the IPO market. The objective of this paper is to fill this gap in the literature. The only paper in the existing literature analyzing the effects of product market advertising on IPOs is Chemmanur and Yan (2009). They suggest theoretically and empirically that product market advertising is related to a firm s going public decision and its IPO underpricing. In particular, they develop a signaling model where insiders with private information about their firm s future cash flows use product market advertising and IPO underpricing to jointly signal firm value to outsiders. They also document that firms reach a peak amount of advertising in their IPO years, relative to the years before and after their IPOs; further, the higher the amount of product market advertising undertaken by a firm in its IPO year, the smaller the magnitude of underpricing in the firm s IPO. However, Chemmanur and Yan (2009) do not study any of the long-run effects of product market advertising on IPOs, which is the primary focus in the current paper. In contrast, in this paper we study how product market advertising around a firm s IPO is 3

6 related to the short and long run valuations of its equity, IPO offer price revision during the book-building period, its long-run stock return subsequent to the IPO, and the composition of investors (retail versus institutional) in the firm s equity in the immediate aftermarket. Further, unlike Chemmanur and Yan (2009), here we rely on the implications of heterogeneous beliefs models (such as Morris (1996)) rather than on asymmetric information models to interpret our findings. This is because, in general, asymmetric information models with fully rational outsiders cannot explain the long-run IPO stock return underperformance or other long-run effects of product market advertising in the IPO aftermarket. We first study the relationship between product market advertising and post-ipo long-run stock returns. Our measures for long-run stock returns consist of cumulative annual returns (Ritter (1991)), buy-and-hold long-run stock returns (see, e.g., Loughran and Ritter (1995)), and abnormal returns based on Fama-French s (1993) three-factor model and Cahart s (1997) four-factor model. The statistics of our long-run return measures are consistent with the IPO literature, showing that IPO stocks on average experience negative post-ipo stock returns in the long run. Based on these long-run return measures, we find that firms going public with a greater extent of product market advertising prior to their IPOs experience more negative long-run stock returns subsequent to their IPOs. For example, consider the IPO firms with advertising intensities (advertising/sales) above the sample median in the years prior to their IPO years and those with advertising intensities below the sample median. We find that a one-standard deviation increase in the pre-ipo advertising intensity would increase the post-ipo buy-and-hold return by 4.9% in the two-year window and 9.1% in the three-year window. Next, we study the relationship between product market advertising and IPO valuations. We use the ratio of offer price to intrinsic value to measure the IPO valuation at the offer price and 4

7 the ratio of first day secondary market closing price to intrinsic value to measure the IPO valuation in the aftermarket. 2 Using the above measures, we first establish that IPO firms are overvalued relative to intrinsic value both in the IPO and in the immediate aftermarket. More interestingly, we find that the extent of overvaluation of an IPO firm is increasing in the extent of its product market advertising prior to the IPO: this is true for the overvaluation in the IPO as well as in the immediate aftermarket. In particular, our results show that an IPO firm could increase its offer price to gain a valuation premium of 0.94% to 1.68% relative to its intrinsic value if it increases its advertising intensity by 1% in the year prior to its IPO. We also find that an IPO firm with a greater extent of product market advertising prior to its IPO revises its final IPO offer price upward to a larger degree from the midpoint of its filing range during the book-building period. Total IPO proceeds are also greater for the firm with a greater extent of product market advertising prior to its IPO. What explains the above effect of product market advertising on IPO valuations, IPO price-revision at the book-building stage, and long-run post-ipo stock returns? The large literature on the relationship between the heterogeneity in investor beliefs and stock valuation suggests a possible explanation, which we test in this paper. As Miller (1977) has argued (an argument subsequently formalized by Harrison and Kreps (1978) and Morris (1996)), when rational investors who are subject to short-sale constraints have heterogeneous prior beliefs, share prices reflect the valuation of the most optimistic investors, so that the equity of firms sells at a premium over their fundamental value. We conjecture that product market advertising induces some investors, especially retail investors, to become more optimistic about the future prospects of the firm going public, thereby increasing the heterogeneity of investor beliefs about 2 We measure an IPO stock s intrinsic value based on its matching firm s price/sales or price/ebitda multiples. Matching firms are selected by industry, sales, and EBITDA margin. We discuss our intrinsic value measures in detail in Section

8 the IPO firm. As a result, the increased heterogeneity of beliefs that is induced by advertising causes higher levels of valuations both in the IPO market and in the immediate aftermarket. Further, given our earlier findings that IPO valuations are greater for firms undertaking a larger extent of advertising, it is not surprising that the long run performance of such firms is poorer. As the effects of product market advertising around the IPO wear off over time and the information about the operating performance of the IPO firm becomes available to investors, the heterogeneity of investor beliefs about the firm s prospects is reduced, causing the firm s equity valuation to converge to its fundamental value. This convergence implies that the firms undertaking a greater amount of product market advertising will have lower post-ipo long-run stock returns. The heterogeneous beliefs theory can also explain our findings that product market advertising is associated with a larger offer price revision (from the mid-point of the IPO filing range to the final offer price) during the book-building period and a higher magnitude of the IPO offering proceeds. As some retail investors become more optimistic about an IPO firm s future prospects, institutional investors will also revise upward their valuation of the IPO firm s equity, taking into account the possibility that they can flip their equity to these optimistic retail investors at a higher valuation within a short period after the IPO (see, e.g., Aggarwal (2003) and Chemmanur, Hu, and Huang (2010), who show that institutional investors flipping within a month after the IPO is quite significant). Once IPO underwriters become aware through the book-building process that retail as well institutional investors have a higher valuation for the shares of the IPO firm associated with a greater amount of product market advertising, IPO underwriters will revise the final IPO offer price upward as well, yielding a larger offer price revision for such a firm. 6

9 To directly test the above heterogeneous beliefs theory as a possible explanation of the above findings, we develop the following testable hypotheses (discussed in more detail in section 7). We first hypothesize that an increase in product market advertising prior to an IPO increases the heterogeneity of investor beliefs regarding the IPO firm s stock value. Second, when the pre-ipo product market advertising increases the heterogeneity of investor beliefs regarding the IPO firm s stock value, IPO valuations increase both in the IPO market and in the immediate aftermarket, so that long-run post-ipo stock returns will be reduced. Thus, we expect the relationship between pre-ipo product market advertising, IPO valuations, and long-run post-ipo stock returns to be more pronounced when the degree of the heterogeneity in investor belief characterizing the IPO is greater. Third, retail investors are likely to be more susceptible to the effects of product market advertising than institutional investors. Thus, we expect that the relationship between pre-ipo product market advertising, IPO valuations, and long-run post-ipo stock returns to be more pronounced in the IPO firms with smaller institutional investor holdings. We use trading turnover and the dispersion in analysts earnings forecasts to proxy for the degree of the heterogeneity of investor beliefs. An increase in the heterogeneity of investor beliefs is associated with an increase in trading turnover (see, e.g., Harris and Raviv (1993), Hong and Stein (2003), and Scheinkman and Xiong (2003)) and a higher dispersion in analyst forecasts (see, e.g., Diether, Malloy, and Scherbina (2002)). We also control for investor attention in our tests of the heterogeneous belief hypotheses. One alternative explanation is that advertising could affect IPO valuations and long-run stock returns by increasing investor attention. To control for this possibility of increased investor attention, we use media coverage 7

10 around the IPO, i.e., the number of articles immediately prior to IPOs, to proxy for media mention (following Liu, Sherman, and Zhang, 2014). Our findings are consistent with the above hypotheses. First, we find that an IPO firm with a higher level of product market advertising prior to its IPO faces an increase in the trading turnover in the IPO aftermarket and an increase in the dispersion in analysts earnings forecasts. Second, product market advertising prior to an IPO increases firm valuations (both in the IPO and in the aftermarket) to a larger degree if the IPO firm experiences larger increases in its trading turnover and its analyst dispersion. For such an IPO firm (with larger increases in its trading turnover and its analyst dispersion), product market advertising prior to an IPO is also associated with lower post-ipo long-run stock returns. Finally, we find that the impacts of pre-ipo product market advertising on IPO valuations and long-run stock returns are greater when the IPO firm has a smaller fraction of its equity held by institutional investors in the IPO aftermarket. All the above results hold after we control for possible effects of advertising on investor attention, as proxied by the pre-ipo media coverage of firms going public. Overall, our findings support the heterogeneous beliefs theory, suggesting that the mechanism through which advertising affects IPO valuations and long-run post-ipo stock returns is by increasing the heterogeneity of investor beliefs about the IPO firm s stock value. The rest of this paper is organized as follows. Section 2 relates this paper to the existing literature. Section 3 discusses sample selection and variable construction. Section 4 studies the relation between product market advertising and long-run stock returns. Section 5 studies the relation between product market advertising and IPO valuations. Section 6 studies how product market advertising affects IPO price revision during the book-building period and the size of IPO 8

11 proceeds. Section 7 tests various hypotheses regarding the heterogeneous beliefs theory on the effect of product market advertising on IPOs. Section 8 concludes. 2. Relation to the Existing Literature Apart from the theoretical literature discussed earlier on heterogeneous beliefs and equity valuation (e.g., Miller (1977), Mayshar (1983), Harrison and Kreps (1978), and Morris (1996)), our paper is most closely related to the theoretical and empirical literature on the interactions between the product and financial markets (see, e.g., Dasgupta and Titman (1998)) and on the impact of the product markets on a firm s going public decision (see, e.g., Spiegel and Tookes (2010) or Chemmanur, He, and Nandy (2010)). 3 It is also related to the broader literature on the heterogeneity of beliefs between insiders and outsiders, and its consequences on various corporate financing decisions. 4 Three examples of this literature are Allen and Gale (1999), who examine how heterogeneous priors among investors affect the source of financing (banks versus equity) of new projects, and Bayar, Chemmanur, and Liu (2011), who develop a theoretical model of equity carve-outs under heterogeneous beliefs, and Bayar, Chemmanur, and Liu (2015), who develop a model of capital structure, price impact, and long-run stock returns following equity issues in an environment of heterogeneous beliefs. The relationship between the empirical results in the current paper and those in the theoretical and empirical analysis of Chemmanur and Yan (2009) is worth noting. As discussed earlier, Chemmanur and Yan (2009) focus on the relationship between product market advertising and 3 The theoretical literature on the interaction between the product markets and a firm s going public decision includes Battacharya and Ritter (1983), Maksimovic and Pichler (2001), and Stoughton, Wong, and Zechner (2001). 4 To the extent that we use some of the proxies for heterogeneous beliefs from this literature here, our paper is also related to the empirical asset-pricing literature on the relationship between heterogeneous beliefs among investors, short sale constraints on equity, and subsequent stock returns (see, e.g., Chen, Hong, and Stein (2002), Diether, Malloy, and Scherbina (2002), and Asquith, Pathak, and Ritter (2005)). 9

12 IPO underpricing. They show that the levels of advertising expenditures and underpricing function as substitutes for IPO firms to signal their true value to uninformed investors in the equity market. They, however, study neither theoretically nor empirically the relationship between product market advertising and long-run stock returns; neither do they study how product market advertising is related to IPO valuations or IPO offer price revision. Further, in the current paper, we show that long-run post-ipo stock returns decrease in response to an increase in the pre-ipo product market advertising, even after we control for IPO initial return. 5 From a theoretical point of view, information asymmetry, by itself, cannot explain long-run post-ipo stock returns. This is because fully rational investors will learn instantaneously from the actions of insiders with private information, so that the information flow from insiders to outsiders due to signaling will be fully reflected in the short-run effects (such as IPO underpricing or initial returns); asymmetric information models therefore cannot generate a systematic bias in the long-run stock prices (either upward or downward) subsequent to any corporate event. Thus, the impact of the pre-ipo advertising on post-ipo long-run stock returns and IPO valuations documented in our paper is clearly not driven by information asymmetry. In summary, in this paper, we are able to show empirically that, while considerations of asymmetric information may be important in the pricing of IPOs, the heterogeneity of investor beliefs characterizing the IPO market is also important in determining the valuations of IPO stocks, IPO offer price revisions, and long-run post-ipo stock returns. 6 5 There are two other papers in the broader literature on the linkages between product market advertising and the financial markets. Grullon, Kanatas, and Weston (2004) document that product market advertising helps improve the liquidity of trading in the stock market for firms that are already public. Demers and Lewellen (2003) document that, for a sample of internet firms, IPO underpricing is associated with a post-ipo increase in their website traffic. 6 Note, however, that our empirical results are quite consistent with those of Chemmanur and Yan (2009), who document that product market advertising is negatively related to IPO initial return. In our paper, we show that, while the first day secondary market closing valuation of the IPO firm s stock increases with product market advertising, the increase in the firm s IPO offer price with product market advertising is even greater, so that IPO underpricing (initial return) is decreasing in product market advertising (consistent with Chemmanur and Yan 10

13 Our paper is also related to the broader theoretical and empirical literature on the pricing and performance of IPOs and the behavior of investors in the IPO market: see Ritter and Welch (2002) for a review. For example, the model of Welch (1992) demonstrates that later investors in a firm s IPO will optimally ignore their private information and buy shares in the IPO after becoming more optimistic about the IPO firm s prospects if they observe strong demand from earlier investors for these shares. 7 To the extent that this literature also studies IPO offer price revisions during the book-building process, our paper is also indirectly related to the empirical literature on the partial adjustment phenomenon of IPO offer prices (see, e.g., Hanley (1993)) Sample Selection and Variable Construction 3.1. IPO Sample We obtain our initial sample of 5,120 IPOs from the Thompson Financial s Securities Data Corporation (SDC) new issues database. It covers the period from 1990 to We extract financial statement information from Standard & Poor s Compustat files, stock prices from the Center for Research in Securities Prices (CRSP), analyst coverage data from the Institutional Brokers Estimate System (IBES), and institutional investor holding data from CDA/Spectrum. We then exclude from our initial IPO sample spin-offs, ADRs, unit offerings, LBOs, limited partnerships, financial firms (SIC codes 6000 through 6999), and utilities (SIC codes 4900 (2009)). Further, the implications of our analysis here are also consistent with the implications of the theoretical analysis of Chemmanur and Yan (2009). Our analysis indicates that, while in the short run, advertising may play a signaling role in conveying firm insiders private information to new investors in the IPO market, in the long run a greater extent of advertising may also increase the heterogeneity in retail investor beliefs about the future prospects of the firm going public. This, in turn, leads to higher IPO and immediate secondary market valuations and lower post-ipo long-run stock returns (as the heterogeneity of investor beliefs about the firm at the time of IPO is dissipated over time with the arrival of hard information about the post-ipo operating performance of the firm). 7 Our paper is also related to the broader theoretical literature on the going public decisions of firms, see, e.g., Chemmanur and Fulghieri (1999) and Boot, Gopalan, and Thakor (2006), as well as the theoretical literature on the pricing of IPOs, see, e.g., Allen and Faulhaber (1989) and Welch (1989). 8 Two papers testing some of the implications of the heterogeneous beliefs models in the IPO context are Houge, Loughran, and Suchanek (2001) and Chemmanur and Krishnan (2012). But both papers are unrelated to product market advertising and the various hypotheses we test here. 11

14 through 4999). This step of sample selection results in a sample of 3,948 IPOs. We also exclude from our sample those firms that are not covered by Compustat and CRSP, and especially those firms with missing data on advertising expenditures in the IPO year, where advertising expenditures are the cost of advertising, media, and promotional expenses from Compustat item #45. Our sample has 928 IPOs after this screen. Finally, we exclude those firms with sales revenue less than $1 million in the fiscal year prior to the IPO year, those with the book value of assets less than $1 million in the prior year, and those with IPO prices less than one dollar per share. Our final sample of IPOs consists of 875 firms. In the following sections, we may be constrained to use only part of the sample in some empirical tests, either due to the incomplete information on lagged values or due to the incomplete information in the SDC, IBES, or CDA/Spectrum to construct certain variables. We report the annual breakdown of our final sample in panel A, table 1, as well as the annual means and medians for the amount of equity raised from IPOs, the first day return, and advertising expenditures in IPO years. In general, the time series pattern of the average first day return in our sample is very similar to that in Ritter and Welch (2002), although we have a smaller sample. For example, the average first day returns are 13.2% between in our sample, 19% between , and 61.2% between The corresponding first day returns for these three periods in Ritter and Welch (2002) are 11.2%, 18.1%, and 65%, respectively. However, the average amount of equity raised from IPOs in our sample is larger than that in Ritter and Welch (2002). Panel A also shows that the standard deviation of advertising intensity (advertising/sales) in the IPO year is while the mean advertising intensity is The large standard deviation is due to the large variation in advertising expenditures in our sample firms. For example, the 12

15 largest advertising expenditure in the IPO year is $296.9 million by Vonage Holdings Corp in 2006, while there are six IPO firms with zero advertising in their IPO years. In panel B of table 1, we report the industry breakdown of our sample. Industries are defined based on Fama-French 48 industry specifications. As can be seen, there is also a large variation in advertising intensities among various industries both in the IPO year and in the prior year Matching Sample In the paper, we construct two matching samples for IPO firms. For the first matching sample, we match each IPO firm with a Compustat listed firm on the basis of industry, sales, and EBITDA margin. EBITDA margin is the earnings before interests, taxes, depreciation, and amortization divided by sales revenue. The matching algorithm consists of the following steps. First, from all Compustat firms in the fiscal year prior to the IPO year, we select those firms incorporated in the U.S., those that did not go public in the past three years, those that are identified by CRSP share type codes of 10 and 11, and those with stock prices more than one dollar as of the end of the prior fiscal year. Second, for the firms surviving the first step, we group them into 48 industries using the industry classification method in Fama and French (1997). Third, we group firms in each industry into three portfolios based on sales revenue in the fiscal year prior to the IPO year and then each sales portfolio into three portfolios based on EBITDA margin in the prior year. If there are not enough firms in an industry so that the above disaggregation yields a portfolio with less than four firms, we relax the sales classification or the EBITDA margin classification and construct only two sales portfolios or two EBITDA margin portfolios. Thus, we have a maximum of nine portfolios in each industry based on a 3 by 3 classification and a minimum of four portfolios on a 2 by 2 classification. Finally, we match each IPO firm to the industry-sales-ebitda margin portfolio to which the IPO firm belongs. From 13

16 this portfolio, we find a matching firm with the closest sales revenue to the IPO firm s sales revenue in the year prior to the IPO year. In this way, we assign each IPO firm a unique matching firm. We will use this matching firm as the benchmark firm in the studies of long-run stock performance and IPO valuations. In the study of IPO valuations, we also use the median firm in the industry-sales-ebitda margin portfolio (based on the medians of certain price multiples discussed later) as the benchmark to compute the intrinsic value for each IPO firm. 9 Our second matching sample is based on industry and firm size, similar to that used in Loughran and Ritter (1997). We select Compustat firms and assign them to 48 Fama-French industries, following the same steps as in constructing the first matching sample. Then, from each industry, we find a matching firm with the closest market capitalization to the IPO firm at the closing of the IPO date. We use the matching firm constructed based on industry and size as the benchmark firm in the study of long run stock performance Construction of the Advertising Variables We follow the industrial organization (IO) literature and use advertising intensity to measure the level of a firm s advertising. We calculate the advertising intensity in the IPO year as the advertising expenditures in the IPO year t (Adv t ) scaled by the sales revenue in the same year (Sales t ), i.e., Adi t = Adv t Sales t. Similarly, we calculate the advertising intensity in the year prior to the IPO year as Adi t-1 = Adv t 1 Sales t 1. We calculate both Adi t and Adi t-1 to study the advertising effect on IPO pricing (in the IPO and in the aftermarket), since both variables have their advantages and disadvantages. On the one hand, Adi t could capture the level of advertising related to the firm s going public decision better than Adi t-1. But Adi t could also capture the level of advertising after 9 We have also used the industry median firm as the benchmark to calculate abnormal returns in the study of long run stock performance. We do not present these results in the paper due to space constraints. The results based on the median firm to calculate abnormal returns are similar to those based on a single matching firm. 14

17 the IPO offer date. On the other hand, Adi t-1 does not capture any advertising activities after the IPO offer date, but the effect from the advertising one year prior to the IPO offer year could wear out at the IPO offer date. Due to the above considerations, we use both Adi t and Adi t-1 in our empirical studies to ensure the robustness of our results Long Run Abnormal Stock Returns We compute long-run abnormal stock returns for IPO firms using three different approaches: cumulative abnormal return (CAR), buy-and-hold abnormal return (BHAR), and factor models. All three approaches have their merits and no approach has emerged as the universally optimal methodology in terms of statistical properties and implementability by investors. CAR has the advantage of easier statistical tests. But CAR is positively biased and hard to interpret in a practically meaningful way. In comparison, BHAR represents the return that is obtainable by an implementable investment strategy. However, BHAR could magnify underperformance due to the nature of compounding single-period returns. Brav (2000) also argues that the test statistics on BHAR suffer from cross-sectional dependence in sample observations. Unlike the BHAR method, the factor models eliminate the problem of cross-sectional dependence among sample firms since their inference is derived from the time series of monthly calendar-time portfolios. However, the time-series factor model could suffer from a low power to reject the null of no abnormal returns (see Barber and Lyon (1997) and Loughran and Ritter (2000)). In the paper, we use all three approaches to ensure the robustness of our results. We compute CAR following Ritter (1991). Define R im as the simple return on an IPO firm i in month m and E(R im ) as the expected return (i.e., the benchmark return). CAR is computed as the difference between the simple return and the benchmark return across M months: CAR im = [R im E(R im )] M m=1. (1) 15

18 Here, M equals the end month up to which CAR is computed or the delisting month of the IPO firm i, whichever occurs sooner. Similar to the approach in Ritter (1991), we calculate the benchmark return E(R im ) as the NYSE/AMEX/NASDAQ value weighted return. We compute BHAR as the return on a buy-and-hold investment in the IPO firm less the return on a buy-and-hold investment in its benchmark stock (the benchmark return): M M m=1 m=1 (2) BHAR = (1 + R im ) [1 + E(R im )]. We use three benchmarks for E(R im ). The first benchmark is the same as that in the computation of CAR, i.e., the NYSE/AMEX/NASDAQ value weighted index. The second benchmark is the matching firm based on the industry, sales, and EBITDA margin matching algorithm as discussed in Section 3.2. The third benchmark is the matching firm based on the industry and size matching algorithm. We compute BHAR for both the two-year and the three-year holding periods. Matching firms are included in the computation for the full holding period or until the IPO firm is delisted. If a matching firm is delisted before the end of the holding period or before the IPO firm s delisting day, whichever is earlier, a new match is drawn from the original list of matching candidates selected in the IPO offer day. This replacement procedure is analogous to Loughran and Ritter (1995). We have also experimented with other replacement procedures such as rematching using information at the time of delisting or monthly updating of matching firms, etc. The overall impact of these alternative procedures on BHAR is small. In the next section, we also study the difference in BHARs between the IPO firms with different advertising intensities. One concern on the test statistics for the difference is that they could be misspecified in a small sample (see Barber and Lyon (1997), Fama (1998), etc.). We address this concern by using the Monte Carlo simulation technique to compute critical 16

19 t-statistics (See, e.g., Noreen (1989)). Specifically, to conduct t-statistics for the difference in BHARs, we first shuffle advertising intensities on a yearly basis so that advertising intensities are randomly assigned to IPO firms in our sample. We then disaggregate the randomized sample in each year based on the randomly assigned advertising intensities and calculate t-statistics for the difference in BHARs between the disaggregated subsamples. We repeat the above randomization procedure 5000 times to generate a sample distribution for the t-statistics. This empirical distribution will be used in statistical inferences when we test the difference in BHARs between high and low advertising intensity firms. For the third measure of long-run stock returns, we use Fama-French s (1993) three factor model and Cahart s (1997) four factor model: (R pm R fm ) = α p + β p (R km R fm ) + s p SMB m + h p HML m + u p UMD m + ε pm. (3) Here, the first factor, R km R fm is the excess return on the market portfolio in month m, calculated as the return on the NYSE/AMEX/NASDAQ value weighted index R km minus the one-month T-bill return R fm (risk-free return); the second factor, SMB m, is the return on the large firms minus the return on the small firms in month m; the third factor, HML m, is the return on the high book-to-market stocks minus the return on the low book-to-market stocks in month m, and UMD m is the return on the high momentum stocks minus the return on the low momentum stocks in month m. R pm is the monthly return on the portfolio of IPO firms. Specifically, we assign each IPO firm to the portfolio on a monthly basis and each IPO firm will be held in the portfolio in a holding period of years starting from the end of the sixth month after the IPO offer date (Purnanandam and Swaminathan (2004)). We avoid the first six months after the IPO offer date in order to better understand how advertising affects the long run underperformance of IPO firms. 17

20 At the end of each holding period of years, the IPO firm drops out of the portfolio. We then calculate R pm as the equally weighted returns for the portfolio in each calendar month. In the above factor model, the intercept α p is the monthly risk-adjusted abnormal return in percent. The slope coefficients β p, s p, h p, and u p are factor-loading, measuring the sensitivities of the return of the IPO portfolio with respect to various factors IPO Valuations We measure IPO valuations using price-to-value ratio. Our calculation follows Purnanandam and Swaminathan (2004). In particular, we first compute the intrinsic value V for each IPO firm based on its matching firm s Price/Sales or Price/EBITDA multiple: 10 Market Price Shares Outstanding Sales V sales = ( ) Sales Match ( t 1 ) t 1 Shares Outstanding IPO, and (4) Market Price Shares Outstanding EBITDA V EBITDA = ( ) EBITDA Match ( t 1 ) t 1 Shares Outstanding IPO. (5) Here and also in the following, we define year t as IPO year, year t-1 as one year prior to the IPO year, etc. In both computations, each IPO firm is matched to either a single firm based on industry, sales, and EBITDA margin or to the median firm in the portfolio matched by industry, sales, and EBITDA margin. Due to space constraints, we do not present the results based on other types of matching, such as the industry and size matching algorithm discussed earlier. The results based on the alternative matching algorithms are similar to those presented in the paper. Based on the calculated intrinsic value, we measure the level of valuation at the IPO offer price by the following price-to-value ratios: ( P 0 ) IPO Offer Price V sales = log ( ), and (6) V sales 10 We do not use the matching firm s Price/Net Profit as in Purnanandam and Swaminathan (2004) since doing so would substantially reduce our sample size given that many IPO firms in our sample have negative earnings before going public. 18

21 ( P 0 ) IPO Offer Price V EBITDA = log ( ). (7) V EBITDA P 0 stands for the IPO offer price. Both ( P 0 V ) sales and ( P 0 V ) EBITDA measure the premium or discount in the IPO offer price relative to intrinsic value. A higher ( P 0 V ) sales or ( P 0 V ) EBITDA indicates that an IPO firm s equity is valued higher in its IPO relative to its intrinsic value. Since many IPO firms in our sample have negative EBITDA and are left out in the calculation of ( P 0 V ) EBITDA, the sample size for ( P 0 V ) EBITDA is smaller than the sample size for ( P 0 V ) sales. We also measure the level of valuation of each IPO stock in the immediate aftermarket by the following ratios: ( P 1 ) First Day Closing Price V sales = log ( ), and (8) V sales ( P 1 ) First Day Closing Price V EBITDA = log ( ). (9) V EBITDA P 1 stands for the closing price of the IPO stock at the end of the first trading day in the secondary market. The difference between ( P 0 V ) sales and ( P 1 V ) sales (and also between ( P 0 V ) EBITDA and ( P 1 V ) EBITDA) is that the former ratio measures how an IPO firm prices its new equity issue and the latter ratio measures how investors in the aftermarket price the IPO stock. The IPO valuation at the first day closing price captures both the IPO first day return and the valuation at the IPO offer price. We winsorize all the price ratios, including ( P 0 V ) sales, ( P 0 V ) EBITDA, ( P 1 V ) sales, and ( P 1 V ) EBITDA, at the 1% level in both tails of the distribution to avoid the outlier effect. Table 2 reports the sample statistics for the above IPO valuation measures. Table 2 shows that the average IPO offer price P 0 is overvalued. For example, consider ( P 0 V ) sales, with intrinsic value V is calculated based on matching firms. The mean and median of 19

22 ( P 0 V ) sales are and 0.589, corresponding to pricing premiums of 77% (i.e., e ) and 80%, respectively. Table 2 further shows that the IPO stock is also overpriced relative to its intrinsic value by investors in the aftermarket. Consider ( P 1 V ) sales, in which V is calculated based on matching firms. Its mean and median are and 0.705, respectively. These statistics suggest that the first day closing price of an average (median) IPO firm is 107% (102%) higher than its intrinsic value. Thus, the premium carried in the first day closing price is higher than that in the IPO offer price. This difference between our measures of the IPO valuation in the first day closing price ( P 1 ) and the measures of the IPO valuation in the offer price V (P 0 ) can be viewed V as capturing the amount of money left on the table by the IPO firm. It is also worth noting that the statistical properties of the price-to-value ratios in our sample are similar to those in Purnanandam and Swaminathan (2004), although we have a smaller sample. Consider the ( P 0 V ) sales ratio that we discussed earlier. Purnanandam and Swaminathan (2004) find a median premium of 54% based on a sample covering only non-bubble years from year 1980 to The median ( P 0 V ) sales based on our sample is (as in table 2) or a premium of 80%. However, if we exclude the 111 IPOs that took place in the bubble period, i.e., years 1999 and 2000, then the median ( P 0 V ) sales is 0.462, which corresponds to a pricing premium of 58.7%. For the 111 IPOs in the bubble period, the median ( P 0 V ) sales is 1.223, which corresponds to a pricing premium of 239.7%, four times larger than the premium during the non-bubble period. Thus, the high price-to-value ratios reported in table 2 are mostly due to the inclusion of the bubble period in our sample. The magnitude of the pricing premium in our non-bubble sample period is similar to that in Purnanandam and Swaminathan (2004). 3.6 Controlling for Investor Attention 20

23 In order to make sure that our results are not driven by the effects of investor attention to a firm s IPO, but are indeed due to the increase in the heterogeneity in investor beliefs due to the effect of product market advertising undertaken by a firm around the time of its IPO, we control for the extent of pre-ipo investor attention garnered by the IPO firm. To assess the degree of attention that investors pay to IPO firms, we follow Liu, Sherman, and Zhang (2014) and Bajo, Chemmanur, Simonyan, and Tehranian (2016) and use two measures of pre-ipo media coverage of firms going public as proxies for investor attention. Liu, Sherman, and Zhang (2014) argue that media sources compete to attract readers and advertising revenues and, consequently, editors expect their reporters to cover the firms that have already received investor attention or are expected to receive such attention in the future. Even though media coverage does not contain any new hard information about the IPO firm (such hard information must be disclosed in the IPO prospectus), the fact that the firm receives coverage indicates that reporters or their sources, or both, expect the firm to attract investor attention. According to Liu, Sherman, and Zhang (2014), when choosing a firm to cover, reporters use not only their own judgment but also talk to Wall Street professionals, so that media coverage of IPO firms is more than mere noise. While media coverage can include some firms due to short-term demand from retail investors (who are driven by sentiment), it also includes firms that sophisticated investors care about or that reporters expect to do well in the future. The pre-ipo media coverage of firms going public thus is a good proxy for the degree of attention investors pay to such firms. We construct two measures of pre-ipo media coverage of firms going public by hand searching all US English language media sources in Factiva for news articles covering such firms. Our first measure is LogArticles, which is the log of one plus the number of times English language publications in the US have mentioned the IPO firm s name in an entire article in the 21

24 two months prior to the IPO. Our second measure (used by Bajo, Chemmanur, Simonyan, and Tehranian (2016)) is LogHeadlines, which is the log of one plus the number of times English language publications in the US have mentioned the IPO firm s name in article headlines in the two months prior to the IPO. We use the first measure as our main proxy for investor attention, and use it as a control variable in tables 9 to 12. We use the second measure only in our robustness tests (due to space limitations, available only in additional tables in an internet appendix on the authors websites) Construction of Other Variables We follow the IPO literature and construct the following variables related to a firm s going public decision. We calculate the first-day return on the firm s equity, RET, as the percentage change from the IPO offer price to the first day closing price. We calculate underwriter rank, Rank, following Loughran and Ritter (2004) (see also Carter, Dark, and Singh (1998)), with higher ranks representing higher quality. The syndication dummy, Syndicate, equals one if an IPO is underwritten by a syndicate and zero otherwise. The exchange dummy, Exchange, equals one if a firm is listed in the NYSE or AMEX, and zero otherwise. The dummy of venture capitalist backing, Venture, equals one if the firm making the IPO is backed by venture capitalists and zero otherwise. The dummy of high technology firms, Tech, is calculated based on the firm s SIC code and follows the definition of technology firms in Loughran and Ritter (2004). Age is the difference between the IPO year and the founding year. Amount is the log of the amount of money raised from the IPO. Revision is the difference between the IPO offer price and the mid-point of the filing range, scaled by the IPO offer price. We also measure market conditions by Runup, calculated as the compounded return of an equally weighted market index 60 days prior to the IPO offer date. In the paper, we do not include the dummy for the bubble years, i.e., 22

25 years 1999 and 2000, since we will include year dummies in most regressions. The dummy of the bubble years would be perfectly correlated with our year dummies. We further follow the industrial organization literature and construct the following product market variables to capture the factors that may affect a firm s advertising decision. We calculate Assets t-1, the book value of assets in year t-1, and Sales, the change in sales revenue from year t-1 to year t. We also calculate 1 Sales t and 1 Sales t 1, the inverse of sales revenue at year t and t-1, as control variables to ensure that the variations of advertising intensity Adi t and Adi t-1 are not driven by the variations in their divisors Sales t and Sales t-1. We measure a firm s trading activities by adjusted trading turnover. Adjusted trading turnover is the log ratio of a firm s trading turnover to the market average trading turnover in the corresponding stock exchange, where trading turnover is trading volume in shares scaled by shares outstanding after the IPO. In the paper, we measure adjusted trading turnover on the first trading date after a firm goes public. In unreported tests, we have tried adjusted trading turnover in the first week and in the first month. Our results are similar using these two alternative measures. We also construct analyst dispersion based on financial analysts earnings forecasts reported in I/B/E/S in the first month subsequent to the IPO offer date. Analyst dispersion is the standard deviation of analysts one-year-ahead earnings forecasts scaled by the IPO offer price. Both the dispersion in financial analysts earnings forecasts and trading turnover are used to measure the degree of investors heterogeneous beliefs regarding a firm s future prospects. In addition, we calculate institutional investor holdings as the fraction of the shares held by institutional investors on the first reporting date reported by CDA/Spectrum subsequent to the IPO offer date. Table 2 reports the sample statistics of the above variables. 23