The Economic Role of Institutional Investors in Auction IPOs

Transcription

1 The Economic Role of Institutional Investors in Auction IPOs Yuechan Lu 1 Taufique Samdani 2 Abstract We examine the economic role of institutional investors in auction initial public offerings (IPOs) with and without a discretionary tranche of IPO shares pledged to institutional investors prior to public filing. We find that IPO underpricing in auction IPOs with a discretionary tranche is lower (higher) than IPO underpricing in auction IPOs without a discretionary tranche when institutional demand for IPO shares is high (low). The findings, which hold after controlling for potential endogeneity, reveal a cost-benefit tradeoff in auction IPOs that is sensitive to institutional demand, and explain why auction, albeit commonly used for debt instruments, is costly and rarely used for IPOs. JEL classification: Keywords: G15, G18, G24, G28, G38 Auction IPO, discretionary allocation, IPO underpricing, institutional investors. 1 College of Management, University of Massachusetts Boston, 100 William T Morrissey Blvd, Boston, MA yuechan.lu001@umb.edu. 2 College of Business and Innovation, University of Toledo, 2801 W. Bancroft, Toledo, OH tsamdani@gmail.com.

2 1. Introduction The standard methods for bringing initial public offerings (IPOs) to market are bookbuilding, fixed-price, and auction (Jagannathan et al., 2015). Book-building, the most popular of the three issue methods, is widely acknowledged for its superior information production and price discovery attributes (Jagannathan et al., 2015; Sherman, 2005; Sherman and Titman 2002; Benveniste and Spindt, 1989; Benveniste and Wilhlem, 1990; Hanley, 1993; Bradely and Jordan, 2002). Information production in book-building is superior to that in fixed-price because the underwriter in the former sets the offer price after marketing the issue to investors and recording their demand. The underwriter in fixed price, in contrast, sets the offer price prior to soliciting demand information from investors. Fixed-price, therefore, trades off information gathering against simplicity fixed-price requires less sophistication and less information gathering effort on the part of the underwriter. Information production in book-building is superior to that in auction because the underwriter in the former has substantial discretion over allocation of IPO shares, thereby allowing the underwriter to incentivize well-informed institutional investors with favorable allocations (Benveniste and Spindt, 1989; Benveniste and Wilhlem, 1990; Sherman and Titman, 2002). Auction, in contrast, gives little discretion to the underwriting bank and trades off information gathering and price discovery against reduced exposure to manipulation and abuse associated with discriminatory allocation of IPO shares (Ritter, 1998; Ritter and Welch, 2002). Given that each issue method uniquely trades off information gathering against simplicity and investor protection, a hybrid issue method that combines the information gathering and price-discovery benefits of book-building with the protection against manipulation and abuse benefit of auction appears to have a clear advantage over the standard issue methods (Jagannathan 2

3 et al., 2015). Despite the apparent advantage, auction, whether hybrid or standard, has had little success in equity markets worldwide new auction methods have failed to capture a meaningful market share and existing auction methods have lost a significant market share to book-building in almost every country where book-building was introduced (Sherman, 2005). In France, for example, book-building has all but replaced the auction mechanism (Degeorge et al., 2010). In Japan, auction IPOs disappeared as soon as book-building was introduced (Kutsuna and Smith, 2004). In the U.S., auction IPOs, albeit not non-existent, are rare and have an insignificant market share compared to book-building IPOs. 3 In emerging market countries, such as China and India, the regulatory authorities continue to experiment with hybrid auction methods searching for an optimal tradeoff between information gathering and investor protection. China, for instance, switched from a pro-rata allocation policy to a lottery system in November 2010, whereas India switched from a discretionary allocation policy to a non-discretionary pro-rata allocation policy in September 2005, and from this non-discretionary allocation policy to a hybrid discretionary non-discretionary allocation policy in July 2009 (Jagannathan et al., 2015). Motivated by the renewed academic and practitioner interest in auction IPOs as evident from recent experiments of hybrid issue methods in emerging markets, such as India and China (Jagannathan et al., 2015) and the successful use of auction in credit default swaps and government debt instruments in the US (Chernov et al., 2011), this paper explores a relatively less examined area of auction, namely, the impact of institutional investors demand for IPO shares on IPO proceeds in auction IPOs with a discretionary tranche of IPO shares pledged to institutional investors relative to auction IPOs auctions without a discretionary tranche of IPO shares. Specifically, the paper exploits the July 2009 amendment to the 2000 Disclosure and 3 The auction IPO method was used by 25 US firms between 2004 and 2016 ( 3

4 Investor Protection Act in India to examine the impact of (non-)discretionary allocation of IPO shares in conjunction with institutional investors demand for IPO shares proxied by institutional oversubscription level on IPO underpricing proxied by first-day return. The July 2009 amendment to the Disclosure and Investor Protection guidelines essentially grants the underwriters of auction IPOs the right to arbitrarily allocate a fixed quota (15%) of IPO shares to anchor institutional investors at their discretion prior to public filing, as if it were bookbuilding. The allocation of IPO shares to institutional investors and retail investors after public filing, in contrast, is pro rata and non-discretionary, as if it were an auction. The results, using a data sample of 312 auctions IPOs ( ) in India, reveal that auction IPOs with a discretionary tranche of IPO shares pledged to anchor institutional investors generate higher proceeds (lower first-day return) when the demand for IPO shares from institutional investors is high, or when the IPO is heavily oversubscribed by institutional investors. Conversely, auction IPOs without a discretionary tranche of IPO shares pledged to anchor institutional investors generate higher proceeds when the demand for IPO shares from institutional investors is low, or when the IPO is not heavily oversubscribed and there is less competition for IPO shares among institutional investors. Thus, the findings reveal a cost-benefit tradeoff in auction IPOs that is sensitive to institutional investors demand for IPO shares. The findings also reveal the complexity in pricing auction IPOs as a consequence of the auction IPO proceeds sensitivity to institutional investors demand and explain why auction, albeit commonly used for debt instruments, is costly and therefore rarely used for new equity issues. Jagannathan et al. (2015) examine auction IPOs in detail in countries around the world and provide an explanation for why so many countries have tried and then abandoned the auction method in favor of other IPO methods. They argue that the failure of auction IPOs in countries 4

5 worldwide is neither due to issuers lack of familiarity with the auction method nor due to the differences in the underwriting fees. The lack of popularity of auction IPOs is instead due to the high level of bidding sophistication demanded of investors to address the winner s curse inherent in auction IPOs. Specifically, they argue that it is difficult even for well-informed institutional investors to assess the bidding strategies of less-informed retail investors who are likely to exacerbate the winner s curse problem with their bidding mistakes in auctions. Jagannathan et al. (2015) posit that a competitive auction tranche which limits participation to institutional investors and a non-competitive tranche that is open to all investors reduces the impact of bidding mistakes made by naïve investors and at the same time increases information production, which in turn reduces IPO underpricing, or the abnormal first-day return. They further posit that the impact of bidding mistakes on IPO underpricing is likely to be more pronounced in auction IPOs when the number of bidders bidding for IPO shares is high than when it is low. Acknowledging that 1) IPO underpricing in auction IPOs without information production is at least partially attributed to the winner s curse problem facing less-informed retail investors (Jagannathan et al., 2015), and 2) IPO underpricing in book-building IPOs with information production is at least partially attributed to the quid pro quo for obtaining market demand information from well-informed institutional investors (Aggarwal et al., 2002), we conjecture that IPO underpricing in auction IPOs with anchor investment is higher than IPO underpricing in auction IPOs without anchor investment when the demand for IPO shares from institutional investors is high, or when the impact of the winner s curse on IPO price is more pronounced. Further acknowledging that institutional investors are incentivized to report information when the concern for the winner s curse is low and the probability of favorable allocation to institutional investors in the primary market is high (Busaba and Chang, 2010; Akron and 5

6 Samdani, 2017), we conjecture that IPO underpricing is lower in auction IPOs without anchor investment than with anchor investment when the demand for IPO shares from institutional investors is low, or when the competition for IPO shares among institutional investors and the concern for the winner s curse are both low. Conjectured differently, auction IPOs exposed to the winner s curse generate higher IPO proceeds (low IPO underpricing) than auction IPOs exposed to the quid pro quo when the demand for IPO shares from institutional investors is low, whereas auction IPOs exposed to the quid pro quo generate higher IPO proceeds than auction IPOs exposed to the winner s curse when the demand for IPO shares from institutional investors is high. These conjectures are borne out by the findings in this paper. The findings contribute to the IPO literature in general and to the auctions literature in particular by demonstrating a cost-benefit tradeoff in hybrid auction IPOs under a policy that combines the price-discovery benefits of book-building with the protection against manipulation and abuse benefits of auction, and that is sensitive to institutional investors demand for IPO shares. Jagannathan et al. (2015) examine auction IPOs in equity markets worldwide and find that when given a choice, issuers tend not to use the auction method once they have become familiar with the method. They posit that the auction method is less attractive to issuers because sophisticated institutional investors find it difficult to accurately shave for the winner s curse, implying that institutional investors find it difficult to assess the impact of naive investors bids on IPO share price in sealed-bid auction IPOs when the number of potential bidders is high. They conclude that a hybrid auction IPO mechanism which includes a competitive tranche of IPO shares pledged to institutional investors curbs the impact of naïve investors bidding mistakes on the IPO share price and at the same time provides the information gathering benefits of the U.S.- type book-building IPO method. Given the scarce data on auction IPOs (Degeorge et al., 2010; 6

7 Jagannathan et al., 2015), 4 there exists almost no empirical evidence, to the best of our knowledge, of the impact of institutional investors demand for IPO shares on IPO proceeds in auction IPOs with a discretionary tranche of IPO shares pledged to institutional investors prior to public filing relative to auction IPOs without a discretionary tranche, thereby making ours the first study to provide empirical evidence of the impact of institutional investors demand for IPO shares on IPO underpricing in two types of hybrid auction IPOs. In so doing, our study also adds to the existing literature on anchor investment (Samdani, 2017; Bubna and Prabhala, 2014). The findings also contribute to the area of research examining the effects of cornerstone investment in Europe, and the Jumpstart Our Business Startup (JOBS) Act in the U.S. designed to reduce the IPO risk and encourage firms to go public by allowing them to test the waters prior to public filing (Doidge et al., 2013). Even though cornerstone investment in Europe, unlike anchor investment in India and cornerstone investment in Hong Kong, is not regulated, the underwriters and the cornerstone/anchor investors in Europe, as in India and in Hong Kong, are committed to allocate and purchase IPO shares, respectively, prior to public filing (Samdani, 2017). The findings suggest that a hybrid auction IPO mechanism which includes a discretionary tranche pledged to anchor/cornerstone institutional investors prior to public filing provides the information production benefits of book-building. However, anchor investment in auction IPOs also reduces IPO proceeds, or increases IPO underpricing when the demand for IPO shares from institutional investors is low. Thus, the findings demonstrate a cost-benefit tradeoff in cornerstone investment in Europe and the JOBS Act in the U.S. that is sensitive to institutional investors demand for IPO shares in auction IPOs. 4 Twelve firms, including Google Inc. and Freddie Mac, used the open auction method of going public in the U.S. between 2004 and 2016 ( 7

8 The rest of the paper is organized as follows. In section 2 we discuss the institutional context in India that is relevant to the analysis and develop the testable hypotheses. In section 3 we present the data. In section 4 we present our research methodology, and we report the results in section 5. In section 6 we check for robustness and we conclude in section Institutional background, related literature, and hypotheses development 2.1 Institutional background The hybrid auction book-building IPO method, introduced to the Indian IPO market in 1999 and used for the first time by issuers of new equity in January 2000, entails the issuing firm selecting an underwriting bank and going on a road show to gauge the demand for the issue. Following the information gathering period, the underwriting bank sets an initial price range and invites bids from investors for a period of four days prior to setting the offer price. In this regard, the hybrid auction book-building IPO method in India is similar to the standard book-building IPO method in the U.S. The two methods are different in that bids for hybrid auction IPO shares in India are binding, i.e., bidders cannot change or cancel their bids after the offer price is set, whereas bidders in the U.S. can retract their bids any time. The two issue methods are also different in that the offer price at which the IPO shares are allocated to investors in the primary market in India has to be between the lower and upper bounds of the initial price range, whereas the offer price of book-building IPO shares in the U.S. may deviate by up to 20% from the initial price range, thereby allowing the underwriter to set the offer price closer to the price revealed in investors bids. IPO shares in India are listed and allocated to investors twenty days after the underwriter sets the offer price. 8

9 A unique feature of the hybrid auction-book-building method in India is that IPO shares are allocated to institutional investors and retail investors based on pre-defined allocation quotas determined by the regulatory authority, namely, the Securities and Exchange Board of India (SEBI). Specifically, the SEBI mandates the allocation of 50% of IPO shares to retail investors and 50% to institutional investors. In the case that the IPO is oversubscribed, shares are allocated to investors in their respective categories on a pro-rata basis (Brooks et al., 2014). Bidders with the highest bids are therefore allocated the largest number of shares in oversubscribed IPOs (none of the book-building IPOs in the period in India are undersubscribed), whereas bidders with the lowest bids may not get any shares in oversubscribed IPOs. The SEBI contends that the fixed allocation quotas to investor types in India, initially introduced to ensure institutional participation in the otherwise retail investor dominated equity market (Bubna and Prabhala, 2011), curb underwriter bias for investor type, as documented for discretionary book-building IPOs in the United States (Aggarwal et al., 2002), and for non-discretionary auction IPOs in Taiwan (Chiang et al., 2010). Since its introduction to the Indian IPO market in 1999, the hybrid auction book-building method has undergone two major changes with regard to allocation of IPO shares to investors in the primary market. The first change, introduced by the regulatory authority SEBI in September 2005, took away the power of discretionary allocation of IPO shares to institutional investors from the underwriter. 5 The second change, introduced by the SEBI in July 2009, reinstated some of the discretionary power taken away from the underwriter by the September 2005 policy. Specifically, the July 2009 regulatory amendment to the SEBI 2000 Disclosure and Investor Protection Act grants the underwriter the power to allocate 15% of the IPO (30% of the 5 See the circular SEBI/CFD/DIL/DIP/14/ 2005/25/1 dated January 25, 2005 at 9

10 institutional investor quota) to institutional investors, also known as anchor investors, prior to the bidding period at the underwriter s discretion. The price at which the shares are allocated to anchor investors is the higher of the anchor price and the offer price. Another distinct feature of the auction IPO method in India that is relevant for this study is that the bidding phase in India is transparent bidders in auction IPOs in India can electronically observe bids from all bidders, at half-hour intervals, whereas bidders in bookbuilding IPOs in the U.S. cannot observe the bids of other bidders or learn about the allocation of shares to investors until after the allocation process is closed (Khurshed et al., 2014). 2.2 Related literature and hypotheses development A growing literature relates IPO underpricing to the winner s curse problem pointed out by Rock (1986). 6 For example, Chowdhry and Sherman (1996) relate IPO underpricing to the winner s curse by demonstrating that issuers in most countries, including in India, reduce the winner s curse and thereby the required level of IPO underpricing by favoring small investors over large investors. Likewise, Sherman (2005) relates IPO underpricing to the winner s curse by demonstrating that there are advantages, i.e., reduced winner s curse and thereby reduced IPO underpricing, to by invitation only auctions, whereas Biais and Faugeron-Crouzet (2002) relate IPO underpricing to the winner s curse by demonstrating that IPO underpricing encourages bidders to reveal endowed information. More recently, Jagannathan et al. (2015) relate IPO underpricing to the winner s curse by demonstrating that an equilibrium bid of winning bids and losing bids is a function of the strength of an investor s signal the strength of a negative signal grows as the number of bids grows. In 6 IPO underpricing in Rock s (1989) model addresses the uninformed investors concern for acquiring overpriced issues in a setting in which informed investors acquire only underpriced shares. 10

11 their sealed-bid auction IPO pricing framework, the offer price is sensitive to mistakes made by less-informed retail bidders. The impact of bidding mistakes on IPO price, which is more pronounced in auction IPOs with many bidders, makes it difficult for even well-informed institutional bidders to accurately shave for the winner s curse. To curb the impact of the winner s curse in auction IPOs in India, the regulatory authority SEBI requires the bidding phase in auction IPOs India to be transparent bidders in India can observe the book being built at half-hour intervals. Khurshed et al. (2009) and Neupane and Poshakwale (2013) argue that transparency in book-building allows retail investors in India to learn from institutional investors bids, which diminishes the winner s curse in Indian IPOs. We add to this argument and conjecture that, in auction IPOs without a discretionary allocation policy, institutional investors are less likely to report information when the competition for IPO shares among institutional investors is high and the probability of favorable allocation to institutional investors is low, implying that institutional bids are less likely to have meaningful content regarding investor demand when the number of bids from institutional investors is high. In this type of an auction IPO framework in which the underwriter is not permitted to commit to institutional investors and vice versa, institutional investors find it difficult to accurately shave for the winner s curse when the demand, or the number of bids, and thereby competition for IPO shares among institutional investors is high. Thus, we conjecture that retail investors are less likely to learn from institutional investors bids in auction IPOs under a policy that does not allow the underwriter to pledge a tranche of IPO shares to institutional investors when the demand for IPO shares from institutional investors is high, even if the bidding process is transparent. This implies that IPO underpricing in auction IPOs without anchor investment is likely to be higher when the number of bids is high and the winner s curse problem is more pronounced, than when the number of bids is low and the 11

12 winner s curse problem is less pronounced. Thus, we hypothesize: H1: First-day return in auction IPOs without anchor investment is high when institutional oversubscription is high and low when institutional oversubscription is low. With regard to IPO underpricing under a policy that allows underwriters to allocate IPO shares to investors at their discretion, the literature shows that IPO underpricing in IPOs with allocation discretion is a quid pro quo for obtaining market demand information from institutional investors in the primary market. For example, Aggarwal et al., (2002) examine the impact of institutional investors demand on IPO underpricing in book-building IPOs in the U.S. in the period and find that IPO underpricing in book-building IPOs is higher when institutional demand for IPO shares is high and lower when institutional demand for IPO shares is low, suggesting that underwriters reward institutional investors with underpriced IPO shares. Aggarwal et al., (2002), therefore, conclude that IPO underpricing in book-building IPOs is a quid pro quo for obtaining market demand information from institutional investors. In a related study, Rocholl (2009) examines the impact of institutional investors role in book-building IPOs and finds that institutional investors support in book-building IPOs with weak market demand is a quid pro quo for underpriced IPO shares. In a study of institutional investors role in auction IPOs, Degeorge et al. (2010) examine 19 auction IPOs completed by WR Hembrecht in the U.S., between 1997 and 2007, and find that institutional investors are rewarded with a large share of the deal in underpriced IPOs. Given that allocation of IPO shares to anchor investors in India is at the discretion of the underwriter, we conjecture that IPO underpricing in auction IPOs with anchor investment is a quid pro quo for obtaining market demand information from anchor 12

13 investors, regardless of the level of institutional investors demand for IPO shares. Thus, we hypothesize: H2: First-day return is higher, if not the same, in auction IPOs with anchor investment and with high institutional oversubscription than in auction IPOs with anchor investment and with low institutional oversubscription. Acknowledging the important information production role played by institutional investors in book-building IPOs worldwide, Jagannathan et al., (2015) propose a hybrid auction IPO mechanism that includes a competitive auction tranche pledged to institutional investors and a non-competitive tranche open to all investors. They argue that this type of a hybrid auction IPO mechanism provides the necessary incentive for information gathering and at the same time reduces the impact of the bidding mistakes on IPO price made by less-informed investors which, in turn, reduces the bidding and pricing complexity inherent in sealed-bid auction IPOs. IPO underpricing in their hybrid auction IPO framework, as in the book-building IPO framework in the U.S., is a quid pro quo for obtaining market demand information from institutional investors. Consistent with this line of reasoning, we argue that IPO underpricing in an auction framework that does not include a competitive tranche pledged to institutional investors is due the winner s curse which is increasing with increasing institutional demand and competition for IPO shares among institutional investors. Thus, we conjecture that IPO underpricing in auction IPOs under a policy that allows anchor investment in India is, at least partially, a quid pro quo for obtaining market demand information from institutional investors, whereas IPO underpricing in auction IPOs under a policy that does not allow anchor investment in India is, at least partially, a discount 13

14 for the winner s curse problem. We further conjecture that IPO underpricing in auction IPOs is lower under a policy that allows anchor investment than under a policy that does not when the demand for IPO shares from institutional investors is high, and the winner s curse in auction IPOs that does not allow anchor investment is more pronounced. This intuition is summarized in the following hypotheses: H3: First-day return is higher in auction IPOs with anchor investment than without when the level of institutional oversubscription is low than when it is high. H3 : First-day return is lower in auction IPOs with anchor investment than without when the level of institutional oversubscription is high than when it is low. The underlying premise for the hypotheses above is that IPO underpricing in auction IPOs with anchor investment is, for the large part, a quid pro quo for obtaining market demand information from anchor institutional investors, whereas IPO underpricing in auction IPOs without anchor investment is, for the large part, a discount for the winner s curse. In this type of an auction framework, IPO underpricing attributed to the winner s curse is higher in auction IPOs without anchor investment than with anchor investment when the number of potential bidders is high, and lower in auction IPOs with anchor investment than without anchor investment when the number of potential bidders is low. Figure 1 illustrates the relative impacts of institutional investors demand for IPO shares on IPO underpricing as a quid pro quo for obtaining market demand information from institutional investors, and as a price adjustment for the winner s curse. 14

15 [Insert Figure 1 about here] The steep sloping line in Figure 1 illustrates the impact of institutional investors demand for IPO shares on IPO underpricing in auction IPOs without anchor investment, which is mostly a price adjustment for the winner s curse, whereas the almost flat line in Figure 1 illustrates the impact of institutional investors demand for IPO shares on underpricing in auction IPOs with anchor investment, which is mostly a quid pro quo between the underwriter and the institutional investors. IPO underpricing as a quid pro quo for information gathering is mostly flat because anchor institutional investors who provide information to the underwriter of the IPO are guaranteed allocation regardless of market demand for IPO shares. The figure, therefore, illustrates that, consistent with hypothesis H1, the degree of underpricing in auction IPOs without anchor investment is higher when the demand from institutional investors is high, and lower when the demand from institutional investors is low. The figure also illustrates that, consistent with hypotheses H2, H3, and H3, IPO underpricing is lower in auction IPOs with anchor investment than without when the demand from institutional investors is high, whereas IPO underpricing is lower without anchor investment than with anchor investment when the demand from institutional investors is low. The explanation for the latter outcome, i.e., lower underpricing in auction IPOs without anchor investment when the demand from institutional investors is low, is that auction IPOs without anchor investment are less exposed to the winner s curse problem when the demand from institutional investors is low and when there is less competition among institutional investors for IPO shares. In this setting, i.e., in a setting in which IPO underpricing is not a quid pro quo for information reporting, reduced competition for IPO shares incentivizes institutional investors to report information, which diminishes the concern for the winner s curse 15

16 among bidders and leads to a relatively low IPO underpricing in auction IPOs without a quid pro quo mechanism for information gathering than in auction IPOs with a quid pro quo mechanism (Busaba and Chang, 2010; Akron and Samdani, 2017). 3. Data and sample characteristics The data for the analysis is collected from Prime Database, Chittograph, the Bombay Stock Exchange (BSE), and the National Stock Exchange (NSE). 7 The data sample consists of 312 auction IPOs listed on the BSE and the NSE between September 2005 and December IPOs of nine penny-stock firms, or firms with face-values of registered IPO shares below INR10 are excluded from the data sample. IPO listings on the smaller stock exchanges in India are also excluded from the data sample. The data on market index returns is collected form Money Control. 8 Daily market returns are computed using the BSE SENSEX index which is the Indian equivalent of the NYSE in the U.S. In addition to controlling for IPO size, firm size, IPO price update, retail oversubscription, and price-to-earnings (P/E) ratio, the regressions also control for auditor reputation, underwriter reputation, venture capitalist ownership, and IPO grading by a SEBI approved credit rating agency a grade of 1 signifies a low quality IPO firm and a grade of 5 signifies a high quality IPO firm, for example, relative to the industry. The variables are defined in detail in the Appendix. Table 1 provides the descriptive statistics for the data sample. [Insert Table 1 about here]

17 Panel A in Table 1 shows the statistics for 184 auction IPOs in the period September 2005 July Shares of auction IPOs in this period are allocated to investors on a pro-rata basis. Panel B and Panel C in Table 1 show the statistics for IPOs in the period July 2009 December 2013 in which anchor investors are allocated 15% of the IPO shares at the discretion of the underwriter. Panel B and Panel C show 50 auction IPOs with anchor investments and 78 auction IPOs without anchor investment, respectively, whereas Panel D shows the number of IPOs and the mean values IPO characteristics by year. The mean first-day return and the mean price update shown are highest (0.27% and 0.04%, respectively) for auction IPOs without anchor investment in the pre-july 2009 period (Panel A) and lowest (0.07% and 0.01%, respectively) for auction IPOs with anchor investment in the post-july 2009 period (Panel C), suggesting superior information production in auction IPOs with anchor investment. The mean oversubscription levels of both institutional investors and retail investors are highest in the pre-july 2009 period during which auction IPOs did not include a tranche of IPO shares pledged to anchor investors. Overall, the results in Table 1 show that it is important to control for year fixed effects in the analysis of IPO underpricing in India. Table 2 presents the Pearson correlations between variables used in the regressions. As documented in the extant literature, Table 2 shows that first-day return in India is positively related to both institutional oversubscription and retail oversubscription. The positive relations suggest that although both categories of investors tend to make similar choices with regard to IPO subscription, indicating a generally homogenous market reaction to an IPO, institutional oversubscription is higher in larger firms and larger issues, whereas retail oversubscription is higher in smaller firms and smaller issues. This suggests that control for firm size and issue size 17

18 are important in the analysis. Price update is greater in smaller firms and smaller issues, indicating a high degree of information asymmetry in smaller firms and smaller issues (Benveniste and Spindt, 1989). Also consistent with the findings in the literature (Bubna and Prabhala, 2014; Clarke et al., 2016), Table 2 shows that IPO grade is positively related to institutional oversubscription and not related to first-day return. [Insert Table 2 about here] Table 3 presents, for each individual subsample of auction IPOs in India, the difference in means of IPO characteristics between auction IPOs in the top tercile and auction IPOs in the middle and bottom terciles for institutional oversubscription. High level of institutional oversubscription appears to be prevalent in larger firms and larger issues that use the auction method of IPO without anchor investment both in the pre- and post-july 2009 periods. In auction IPOs with anchor investment, in contrast, institutional oversubscription does not vary significantly by firm size and by issue size. Market-adjusted first-day return and price updates are significantly higher for auction IPOs with a high level of institutional oversubscription in the pre-july 2009 subsample and the post-july 2009 subsample of auction IPOs with anchor investment. Interestingly, these two variables are not very different across the tercile groups for auction IPOs without anchor investment in the post-july 2009 period. It is important to note that the differences in means reported in Table 3 do not preclude endogeneity due to selection bias. In the next section, we present our research methodology for the analysis of the differential effects of institutional demand allocation of IPO shares to institutional investors on IPO initial returns. 18

19 [Insert Table 3 about here] 4. Research methodology 4.1 Comparison across quantiles Wilcox (2006, 2012) contend that the quantile regression approach is more informative than the linear regression approach when examining the difference in causal effects between two independent groups. Thus, to present the differential effects of low and high institutional investor demand on the degree of underpricing in auction IPOs in India, we employ a quantile regression approach in which we divide our data sample into top, middle, and bottom terciles for institutional oversubscription. We then compare the impact on the dependent variable of institutional oversubscription in the top tercile with that in the middle and bottom terciles. Our research approach follows from that of Cornelli and Goldreich (2001) who divide their data sample into quartiles based on the size of the bid to examine the effect of bid size on allocation of IPO shares. Our research also follows from that of Brav and Gompers (1997) who divide their data sample into terciles to compare the average book-to-market ratio of venture-backed and non-venturebacked IPO firms in the U.S., and that of Clarke et al. (2016) who divide their data sample of Indian IPOs into quartiles to examine the effect of subscription patterns on initial returns. In our approach, we divide the IPO data sample into terciles to examine the differential effects on the first-day return (proxy for IPO underpricing) of high institutional oversubscription level relative to low institutional oversubscription level in auction IPOs with anchor investments and without. 19

20 4.2 Instrumental variable estimation A major challenge facing researchers investigating the effects of investors decisions and managers decisions on firm value using non-experimental data is to address the concerns for endogeneity due to selection bias non-random decisions and due to omitted variable bias an omitted variable correlated with both the explanatory variable and the error term that could bias the ordinary least squares (OLS) results. For example, an omitted unobservable variable related to institutional investors demand for IPO shares and IPO underpricing could bias the OLS results, whereas oversubscribed IPOs and IPOs backed by anchor investment, if they are nonrandom outcomes, could exhibit a selection bias: underwriters may selectively choose to back IPOs with anchor investment; institutional investors may selectively seek to acquire more shares in underpriced auction IPOs; and/or underwriters may selectively underprice those IPOs for which the demand from institutional investors is high. Brooks et al. (2014) posit that the oversubscription rate in Indian IPOs is an endogenously determined explanatory variable of IPO underpricing in that several factors, including IPO grades, analyst recommendation, and whenissued quotes, that appear prior to the subscription period affect the oversubscription rate, which in turn affects IPO underpricing. Specifically, they demonstrate that higher IPO grades lead to higher subscription rates for institutional investors, which in turn leads to higher IPO underpricing. They also demonstrate, as do several other studies (see, e.g., Khurshed et al., 2009; Bubna and Prabhala, 2014; Clarke et al., 2016), that retail investors bidding behavior in India is influenced by institutional investors bidding behavior in that retail investors follow the bidding behavior of institutional investors in the transparent IPO bidding phase, which suggests that IPO underpricing in India is likely be higher in IPOs with high retail oversubscription. Overall, these findings imply that IPO grade, which is known to investors prior to the IPO bidding phase and 20

21 which affects IPO underpricing through its effect on institutional oversubscription and anchor investment is a good instrument for the latter two variables in India. Following this line of reasoning, we address endogeneity concerns due to omitted variable bias, selection bias, and simultaneity or reverse causality by employing an instrumental variable (IV) regression approach, as in Angrist and Imbens (2012), in which we choose IPO grade as an instrument for institutional oversubscription and anchor investment. A good instrument in the IV regression framework satisfies both the relevance condition and the exclusion condition (Angrist and Imbens, 2012). The partial correlation (non-zero) between IPO grade and the potentially endogenous institutional oversubscription variable implies that IPO grade satisfies the relevance condition. This condition is also empirically testable via OLS and the test of the null hypothesis that the coefficient for the instrument is zero against the hypothesis that it is not. The exclusion condition, in contrast, requires that the covariance between the instrumental variable and the error term is zero. This condition is difficult to test because the error term is not observable. Thus, to find a good instrument, researchers focus on understanding the economics of the question at hand. For example, Brooks et al. (2014) argue that IPO grade, analyst coverage, and when-issued quotes are good instruments for subscription rate because they are known to investors prior to the subscription period and they affect IPO underpricing through their effect on the subscription rate IPO grade, analyst coverage, and when-issued quotes are observable prior to the subscription period and because they are expected to have information content about the quality of the IPO and the demand for IPO shares, they are likely to affect the subscription rate which, in turn, is likely to affect IPO underpricing in India. Following this line of reasoning, and the findings that IPO underpricing in India is, for the large part, driven by institutional investors demand (Bubna and Prabhala, 2014; Brooks et al., 2014; Clarke et al., 21

22 2016), we argue that IPO grade is a good instrument for institutional oversubscription and anchor investment in India. In addition to the economic rationale noted above, we examine the appropriateness of IPO grade as a plausible instrumental variable by conducted several post-estimation statistical tests, including a test to see whether institutional oversubscription and anchor investment are endogenous (Wooldridge, 1995). 5. Analysis and results In this section, we test the predictions of our hypotheses H1, H2, H3, and H3. We start by examining the factors that affect anchor investment and institutional oversubscription in auction IPOs in India using Probit regressions based on the following three regression models: ANCHOR_ INVESTMENT = b 1 IPO _ GRADE + b 2 PE _ RATIO + b 3 VC _ OWNERSHIP + b 4 AUDITOR_ REP + (1) b 5 UNDERWRITER_ REP + b 6 Ln(ISSUE _ SIZE) + e HIGH _ OVERSUBSCRIBED _ INST = b 1 IPO _ GRADE + b 2 PE _ RATIO + b 3 VC _ OWNERSHIP + b 4 AUDITOR_ REP + (2) b 5 UNDERWRITER_ REP + b 6 Ln(ISSUE _ SIZE) + e ANCHOR_ INVESTMENT * HIGH _ OVERSUBSCRIBED _ INST = b 1 IPO _ GRADE + b 2 PE _ RATIO + b 3 VC _ OWNERSHIP + b 4 AUDITOR_ REP + (3) b 5 UNDERWRITER_ REP + b 6 Ln(ISSUE _ SIZE) + e The left-hand side of Eqns. (1), (2), and (3) show the dependent binary anchor investment variable, the binary high institutional oversubscription variable, and the interaction term between the two, respectively. The right-hand side of the three equations show, along with the control 22

23 variables commonly used in IPO studies in India (see, e.g., Deb and Marisetty, 2010; Bubna and Prabhala, 2011; Neupane and Poshakwale, 2012; Akron and Samdani, 2017), and in the U.S. (see, e.g., Beatty, 1989; Brav and Gompers, 1997; Aggarwal et al., 2002), the important IPO grade variable that is used as an instrumental variable for the three dependent variables in the instrumental variable regressions. Tables 4, 5, and 6 report the results of the Probit regression models based on Eqns. (1), (2), and (3), respectively. The variable definitions are presented in the Appendix. [Insert Table 4 about here] [Insert Table 5 about here] [Insert Table 6 about here] Models 1 and 2 in Tables 4, 5, and 6 are baseline models that include the variable of interest, namely IPO_GRADE, whereas Models 3 and 4 are expanded models that, in addition to IPO_GRADE include control variables, namely, the natural logarithm of issue size, Ln(ISSUE_SIZE), the price-to-equity ratio, PE_RATIO, VC ownership, VC_OWNERSHIP, underwriter reputation, UNDERWRITER_REP, and auditor reputation, AUDITOR_REP. Given that IPO grading introduced to the Indian equity market in April 2006 was not mandatory until May 2007, not all IPOs in the September 2005 December 2013 period in India are graded (Brooks et al., 2014). To control for potential selection bias in the analysis, we exclude IPO listings prior to May 2007 from the data samples used in the regression models in Tables 4, 5, and 6. Further given that the anchor investment option was not available to firms in India until 23

24 after July 2009, we also exclude IPO listings prior to July 2009 from the data samples used in the regression models in Tables 4, 5, and 6. The only difference between Model 1 and Model 2, and between Model 3 and Model 4, in Tables 4, 5, and 6 is that Model 1 and Model 3 control for year fixed-effects and industry fixedeffects, whereas Model 2 and Model 4 do not. The statistically significant and positive coefficients, 1.02, 0.79, 0.86, and 0.57, for the IPO_GRADE variable in the four models in Table 4 support the argument that anchor backed IPOs have high ratings and therefore high quality (Bubna and Prabhala, 2014). The statistically significant and positive coefficients, 1.70 and 1.51, for the VC_OWNERSHIP variable in Model 3 and Model 4, respectively, and the statistically significant and positive coefficients, 1.55 and 1.03, for the AUDITOR_REP variable in Model 3 and Model 4, respectively, also support support the argument that the underwriter in India is likely to solicit anchor participation in IPOs that have VC ownership and that are audited by high quality auditors. The statistically significant and positive coefficients, 0.57, 0.64, 0.80, and 0.56, for the IPO_GRADE variable in the four models in Table 5, and the statistically significant and positive coefficients, 0.48, 0.84, and 0.40, for the IPO_GRADE variable in Model 2, Model 3, and Model 4, respectively, in Table 6 imply that institutional oversubscription in anchor-backed auction IPOs increases with increasing IPO grade. Overall, the results in Table 4, Table 5, and Table 6 show a strong relationship between the IPO_ GRADE variable and the three variables of interest, namely, the ANCHOR_INVESTMENT variable, the HIGH_OVERSUBSCRIBED_INST variable, and the ANCHOR_INVESTMENT* HIGH_OVERSUBSCRIBED_INST interaction term, in the analysis of the effect of institutional demand for IPO shares on IPO underpricing in auction IPOs with and without anchor investment. 24

25 We next examine the effect of anchor investment and high institutional oversubscription on first-day return in auction IPOs in India using OLS regressions based on the following regression model: FIRST _ DAY _ RETURN = b 1 ANCHOR _ INVESTMENT + b 2 HIGH _ OVERSUBSCRIBED _ INST + b 3 ANCHOR_ INVESTMENT * HIGH _ OVERSUBSCRIBED _ INST + b 4 IPO _ GRADE + b 5 PE _ RATIO + b 6 VC _ OWNERSHIP + b 7 AUDITOR _ REP + (4) b 8 UNDERWRITER _ REP + b 9 Ln(ISSUE _ SIZE) + b 10 PRICE _UPDATE + b 11 OVERSUBSCRIBED _ RETAIL + e The left-hand side of Eqn. 4 shows the dependent variable, FIRST_DAY_RETURN, which is a proxy for IPO underpricing. The right-hand side of Eqn. 4 shows, in addition to the control variables commonly used in Indian IPO studies (Khurshed et al., 2009; Marisetty and Subrahmanyam, 2010; Neupane and Poshakwale, 2012; Bubna and Prabhala, 2011; Brooks et al., 2014), the binary anchor investment variable, ANCHOR_INVESTMENT, the binary high institutional oversubscription variable, HIGH_OVERSUBSCRIBED_INST, and the interaction term, ANCHOR_INVESTMENT*HIGH_OVERSUBSCRIBED_INST. These variables are defined in the Appendix. Table 7 reports the results of the regressions based on Eqn. 4. [Insert Table 7 about here] Model 1 in Table 7 shows the results of the baseline OLS regression which includes the independent variables of interest, namely, the anchor investment variable, ANCHOR_INVESTMENT, the high oversubscription by institutional investors variable, 25

26 HIGH_OVERSUBSCRIBED_INST, and the interaction term, ANCHOR_INVESTMENT* HIGH_OVERSUBSCRIBED_INST. Model 2, which is also a baseline model, includes the IPO GRADE variable. Models 3 and 4 in Table 7 are extended Models 1 and 2, respectively. Model 5 is identical to Model 3 in all aspects except for the number of observations in the data sample used in each. While the data samples used in Model 1 and Model 3 include all observations in the period September 2005 December 2013, the data samples used in Model 2, Model 4, and Model 5 include observations in the period May 2007 December 2013 in which IPO grading is mandatory in India. All five regression models shown in Table 7 control for industry-fixed effects and year-fixed effects. The statistically significant and positive coefficients, 0.42, 0.31, and 0.31, reported for the HIGH_OVERSUBSCRIBED_INST variable in the regression Model 1, Model 3, and Model 5, suggest that, ceteris paribus, IPO underpricing is positively related to institutional oversubscription rate. This result, which is consistent with the results reported by, among others, Brooks et al. (2014) who find that IPO underpricing is positively associated with institutional investor subscription rate in a setting in which pre-ipo when-issued trading also contributes to information production, supports the prediction of hypothesis H1. The statistically significant and negative coefficients, -0.23, -0.29, and -0.26, for the interaction term, ANCHOR_INVESTMENT*HIGH_OVERSUBSCRIBED_INST, in Table 7 imply that anchor investment reduces IPO underpricing in auction IPOs with a high level of institutional oversubscription. This result, together with the statistically significant and positive coefficients reported for the HIGH_OVERSUBSCRIBED_INST variable in the regression Model 1, Model 3, and Model 5, in Table 7, is consistent with the argument that IPO underpricing in auction IPOs is lower with anchor investment than without when the level of institutional 26

27 oversubscription high, whereas IPO underpricing in auction IPOs is higher with anchor investment than without when the level of institutional oversubscription is low. The findings also support the underlying premise for the arguments in H2, H3, and H3, that IPO underpricing in auction IPOs with anchor investment is a quid pro quo for information provided by anchor institutional investors in the primary market, and that IPO underpricing in auction IPOs without anchor investment is a discount for the winner s curse, which is more pronounced in auction IPOs with a high level of institutional oversubscription than in auction IPOs with a low level of institutional oversubscription. Finally, the statistically insignificant coefficients for the IPO_GRADE variable reported in Table 7 together with the statistically significant coefficients for the same variable reported in regression models in Table 4, Table 5, and Table 6 support the conjecture that IPO grade indirectly affects IPO underpricing through anchor investment and through the oversubscription rate (Bubna and Prabhala, 2014; Brooks et al., 2014; Clarke et al., 2016). The results, thus, confirm that the IPO_GRADE variable is a suitable exogenous proxy for the ANCHOR_INVESTMENT variable and the HIGH_OVERSUBSCRIBED_INST variable in the analysis of IPO underpricing in India. Although the results reported in Table 7 support the predictions of hypotheses H1, H2, H3, and H3, it is important to note that the allocation of auction IPO shares to anchor institutional investors subsequent to the July 2009 Disclosure and Investor Protection Act is not mandatory, and it is therefore likely that the decision to allocate shares to anchor investors is non-random, as demonstrated by Bubna and Prabhala (2014). Thus, it is important to address the concern for endogeneity in general and selection bias in particular in the OLS regression results. 27