Firms’ Access to External Capital Markets
by
Michele Dathan
A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Department of Rotman School of Management University of Toronto
c Copyright by Michele Dathan 2021 Abstract
Firms’ Access to External Capital Markets
Michele Dathan Doctor of Philosophy Department of Rotman School of Management University of Toronto 2021
This thesis examines external forces that impact firms’ willingness and ability to raise external funds in the capital markets. Chapter 1 introduces the research topics that the thesis addresses and briefly outlines their findings. Chapter 2 examines a regulation that was designed to increase the number of private firms choosing to go public in the United States, but which actually resulted in fewer firms completing initial public offerings. Specifically, the Jumpstart Our Business Startups Act was enacted to reduce the costs of going public for small firms by reducing the amount of required disclosure and allowing these firms to test the waters before filing public documents. This Chapter argues that these changes exacerbated information asymmetry in favour of firms over investors, discouraging all but the best firms from going public. Chapter 3 examines the impact of passive investors on firms’ bond issuance decisions. This Chapter hypothesizes that firms take advantage of the presence of passive demand by issuing index-eligible bonds with features that favour firms. It shows empirically that firms care whether their bonds are index-eligible, and that as the level of passive demand increases, firms are more likely to issue bonds and the bonds they issue are larger and have lower spreads. By focusing on changes in index thresholds, the Chapter then shows that while bonds become larger, bond issuance decreases, consistent with some firms ‘reaching’ to be included in the index while other firms opt out of the bond market. Chapter 4 examines two forms of seasoned equity offerings, which differ in terms of risk taken by financial intermediaries. Specifically, the Chapter compares bought deals, where underwriters buy shares from an issuer at a fixed price and then find buyers or risk holding the shares as inventory, to marketed deals, where underwriters build a book of interested buyers but do not take on any price or inventory risk. This Chapter examines why underwriters choose one form of intermediation over the other, and how bought deals change the way surplus is divided among firms, investors and underwriters and can result in inefficient seasoned equity offerings.
ii Acknowledgements
It is a pleasure to gratefully acknowledge the help and support of the many people who have made this thesis possible. I am forever grateful to my supervisor, Sergei Davydenko, who has been as much a personal support as an academic advisor, and with whom one of the Chapters of this thesis is co-authored. His encouragement through the toughest times of the doctoral program is the main reason that I continued on the academic path. I will cherish our hours working in his office where we discussed everything from high-level research topics, to how to best word an introduction, to detailed coding advice. I will use the tools that Sergei imparted on me, especially his ability to elegantly communicate important ideas, in all my future work. I am also deeply indebted to my committee members, Alexander Dyck and Andrey Golubov, for their guidance and wisdom. Alexander always emphasized the importance of asking research questions that address first order issues and remembering economic primitives. Andrey has been an inspiration as a young scholar who has developed an incredible repertoire of knowledge in our field. I would like to thank Yan Xiong, my cohort-mate in the doctoral program and with whom one of the Chapters is co-authored. Yan became a friend first and foremost, but is also a fantastic theorist with whom I am lucky to have worked. I also acknowledge the support of many faculty members and classmates at the Rotman School of Management with whom I have had the pleasure of exploring research ideas, such as Pat Akey, Olivier Dessaint, Tetyana Balyuk, Susan Christoffersen, Redouane Elkamhi, Christoph Schiller, and Mikhail Simutin. Finally, I would like to acknowledge those people who, in addition to my committee members, have provided invaluable feedback that has improved the quality of the Chapters in this thesis. Chapter 2 has benefited from comments from Pat Akey, Ling Cen, Craig Doidge, David Goldreich, Nicolas Inostroza, Charles Martineau, Jay Ritter, and seminar participants at Hong Kong University of Science and Technology and the University of Toronto. Chapter 3 has been improved by comments from Pat Akey, Ian Appel (AFA discussant), Susan Christoffersen, Gianpaolo Parise (EFA discussant), Melina Papoutsi (SFS discussant), Stuart Turnbull, seminar participants at the University of Toronto, and participants at the 2019 American Finance Association and European Finance Association Meetings and 2020 SFS Cavalcade. Chapter 4 has received helpful comments and suggestions from Pat Akey, James Barltrop, Matt Billett, seminar participants at the University of Toronto, and participants at the 2017 Canadian Law and Economics Association Conference. All errors are my own. I gratefully acknowledge financial support from the Canadian Securities Institute (CSI) Research Foundation and the Ontario Graduate Scholarship.
iii This work is dedicated to my dad, John Dathan, who inspires me every day with his work ethic and capacity to take care of others, and to my husband, Andrew Akkawi, without whose love and support I would have not been able to complete this thesis. Finally, I dedicate this work to my two year old daughter, Ada Dathan Akkawi, who was really more of a hindrance than a help (but who will always be my favourite distraction).
iv Contents
Acknowledgements iii
Table of Contents v
1 Introduction 1
2 Too much information? Increasing firms’ information advantages in the IPO process 5 2.1 Introduction ...... 5 2.2 Institutional Background ...... 8 2.2.1 JOBS Act ...... 8 2.2.2 Testing the Waters (TTW) ...... 8 2.3 Theoretical Framework ...... 10 2.3.1 Implications of Testing the Waters ...... 10 2.3.2 Implications of Reduced Disclosure ...... 15 2.3.3 Discussion ...... 16 2.4 Empirical Results ...... 17 2.4.1 Data ...... 17 2.4.2 Empirical Design ...... 19 2.4.3 Examining Theoretical Implications ...... 21 2.5 Concluding Remarks ...... 27 2.6 Tables ...... 28 Appendix ...... 40 2.7 Appendices ...... 40 Appendix A Proofs ...... 40 Appendix B Description of Variables ...... 42 Appendix C Sample TTW and Written TTW Language ...... 45 Appendix D Further Examination of Parallel Trends Assumption ...... 46 Appendix E Cattaneo et al. Density Test for Manipulation of EGC Status ...... 48
3 Debt issuance in the era of passive investment 49 3.1 Introduction ...... 49 3.2 Hypothesis development ...... 52 3.2.1 Model setup and the demand for bonds ...... 52 3.2.2 Model predictions ...... 53 3.3 Data description ...... 56 3.3.1 Sample selection ...... 56
v 3.3.2 Measuring passive demand ...... 57 3.3.3 Descriptive statistics ...... 59 3.4 Passive investment, bond characteristics, and issuance ...... 60 3.5 Index eligibility thresholds ...... 62 3.5.1 Threshold clustering ...... 62 3.5.2 Threshold changes ...... 63 3.5.3 Difference-in-difference regression analysis ...... 64 3.5.4 Other financing decisions ...... 67 3.6 Conclusions ...... 67 3.7 Figures and Tables ...... 69 3.8 Appendices ...... 85 Appendix A Variable definitions ...... 85 Appendix B Tracked bond indices ...... 87 Appendix C Aggregate passive demand ...... 89
4 SEO underwriters: The choice between matchmaking and market making 90 4.1 Introduction ...... 90 4.1.1 Contribution to the Literature ...... 92 4.2 Forms of Offering ...... 93 4.2.1 Matchmakers: Traditional Marketed Offerings ...... 93 4.2.2 Matchmakers: Accelerated Marketed Offerings ...... 94 4.2.3 Market Makers: Bought Offerings ...... 95 4.2.4 Comparing Bought and Marketed Offerings ...... 95 4.3 Deal Origination Process and Hypothesis Development ...... 96 4.4 Data ...... 97 4.4.1 Sample of New Issues ...... 97 4.4.2 Key Input: Announcement Dates ...... 98 4.4.3 Variable of Interest: Offering Type ...... 98 4.4.4 Empirical Proxies for Hypotheses ...... 99 4.4.5 Variables of Interest: Returns Around Key Dates ...... 101 4.5 Empirical Findings ...... 102 4.5.1 Likelihood of Doing a Bought Deal ...... 102 4.5.2 Effect of Underwriter Market Conditions ...... 102 4.5.3 Inefficient Transactions: Withdrawn and Hung Deals ...... 103 4.5.4 Returns Across Offering Types ...... 104 4.5.5 Bought Deals Change the Split of Surplus in SEOs ...... 104 4.6 Conclusion ...... 105 4.7 Tables ...... 108 Appendix ...... 114 4.8 Appendix ...... 114 Appendix A Description of Variables ...... 114
Bibliography 117
vi Chapter 1
Introduction
When firms hold positive net present value investment opportunities but lack the resources to fund these investment, they must turn to external sources of financing. There is a long literature that examines the issues when firms sell claims on their assets given the information asymmetry between firms and potential investors. In many theoretical frameworks, the act of selling securities itself is considered a signal of management’s assessment of the value of assets in place (e.g. Akerlof (1970) and Myers and Majluf (1984)). Despite a decades-old literature examining firms’ equity and debt issuance decisions, there is still debate on important issues, such as the ideal design of the capital raising process. Furthermore, there are emerging market factors that can impact if and how firms raise external funds. This thesis explores several forces that can affect firms’ willingness and ability to access external capital markets for funding. Chapter 2 theoretically and empirically examines the impact of a regulatory change in the initial public offering (IPO) process that was designed to increase the number of private firms going public, showing that the change actually had the opposite effect. Chapter 3 examines the impact of passive investors in corporate bonds and shows that firms appear to be taking advantage of increasing presence of bond index trackers. Chapter 4 examines financial intermediaries in the seasoned equity offering (SEO) market and shows that the form of intermediation affects the way deal surplus is divided between firms, investors and underwriters and can result in inefficient transactions. A decrease in the number of public companies in the U.S. has attracted significant concern from lawmakers and regulators, who are concerned that firms choosing to stay private limits investment opportunities for ‘Main Street’ investors and reduces the availability of “transparent, information rich and fair” public prices (Securities and Exchange Commission, 2019a). This concern has prompted changes to the IPO process to make it more welcoming for young, high growth firms. Chapter 2, co-authored with Yan Xiong, examines one set of changes - the Jumpstart Our Business Startups Act (JOBS Act) of 2012 - which was designed to reduce the costs of going public. The JOBS Act made two main changes to the initial public offering (IPO) process: it allowed private firms the option to test the waters with investors before filing registration statements, and it reduced the amount of required disclosure. Both of these changes increase the information asymmetry between firms and prospective investors and should thus be expected to affect firms’ ability and willingness to raise external equity. This Chapter asks theoretically what impact these changes should have, and empirically examines important IPO market outcomes. The first important result that we document is a decrease rather than an increase in the number of
1 Chapter 1. Introduction 2
firms choosing to go public. In order to control for important latent market conditions that affect the decision to go public, we rely on the fact that the JOBS Act only applies to a subset of firms: U.S. IPO firms with less than $1 billion in revenue in the year before IPO. This allows us to use a difference-in- difference setting, comparing firms eligible to use the JOBS Act provisions to those who are not eligible, before and after the enactment of the Act. The main control group is U.S. IPO firms with more than $1 billion in revenue, but in order to alleviate concerns about firm size, we repeat some tests using U.K. IPOs with less than $1 billion in revenue. We show that private firms have become less willing to go public in three tests. First, the average small firm in the U.S. economy is less likely to be public in the post-Act period (measured using the listing propensity from Doidge et al. (2017)). Second, focusing on the sample of private firms who choose either IPO, raising private equity or being acquired, eligible private firms are relatively less likely to choose the IPO option in the post-JOBS Act period. Finally, exits by private equity investors are less likely to be done via IPO in the post-JOBS Act period. These results lead us to conclude that the JOBS Act did not induce private firms to go public as intended. This result is consistent with our theoretical framework that focuses on the option to test the waters with investors before deciding to file documents to go public. We start with a benchmark economy with no testing the waters, where firms are either ‘good’ or ‘bad’, but neither firms nor investors know which type each firm is. In this benchmark, there is a pooling equilibrium where all firms go public. When the option to test the waters is available, firms use this option to learn their type, while investors remain in the dark (this is the source of information asymmetry). Knowing their type, good firms use a costly-to-fake signal to separate themselves from bad firms, while bad firms opt out of the IPO process. This separating equilibrium results in fewer firms choosing to go public than in the benchmark economy. Chapter 2 then explores other implications of the JOBS Act and our theoretical framework. First, we show that there have been benefits for eligible firms in the post-JOBS period, including lower likelihood of IPO withdrawal and shorter bookbuilding periods. Second, we show evidence that eligible firms are using signals of quality as a separating mechanism: underpricing has significantly increased for eligible firms in the post-JOBS period, and there is some evidence that use of underwriter certification has increased for eligible firms. Third, we provide initial evidence that while there are fewer IPOs, the quality of IPO firm is higher in the post-JOBS period for the eligible firm sample. Finally, we examine several tests that disentangle our testing the waters story from the alternative reduced disclosure story. Overall, Chapter 2 uses a relatively recent setting to highlight a long-established result that infor- mation asymmetry is paramount in equity raising, and that policy makers designing the IPO process should consider the effects of the information environment on both firms and investors. While Chapter 2 focuses on the trend of decreasing number of public firms in the U.S., Chapter 3, co-authored with Sergei Davydenko, focuses on another important trend: the increasing importance of passive investors in the corporate bond sector. Exchange-traded funds (ETFs) and other passive funds track indices rather than engage in active security selection; fund flows into these passive vehicles provides predictable demand for new issue corporate bonds as long as they meet index eligibility criteria, such as issue size, and regardless of other bond characteristics, such as yield or covenants. This Chapter asks how demand from bond index trackers affects firms’ corporate bond issuance decisions. We hypothesize that firms will take advantage of the presence of passive investors by issuing index-eligible bonds with features that are attractive for the firm but not for investors. We first show that firms care if their bonds are index-eligible. The most widely followed corporate Chapter 1. Introduction 3 bond indices use total face value of a bond to determine eligibility; when we examine the distribution of bond face values, there is significant bunching at exactly the threshold to be included in the index, with almost no issuance immediately below the threshold. When the index threshold changes, the ‘spike’ in issuance moves precisely from the old threshold to the new threshold. We then show that as our proxy for passive demand increases, newly issued bonds are larger and haver lower spreads, as well as fewer covenants and longer time to maturity. In addition, firms’ propensity to issue bonds is higher. Both of these results (bunching at the index threshold and a positive relationship between passive demand and firm-friendly bond characteristics) are consistent with our theoretical framework where passive investors replace the marginal, most pessimistic investor in a new issue bond, as long as that bond is index eligible. We then use the previously discussed changes in index thresholds as an identification strategy to establish a causal link between passive demand and bond issuance. We use a difference-in-difference setting in these tests, using high yield issuers as a control sample for the investment grade index changes and investment grade issuers as a control sample for the high yield index changes. By zooming in on relatively short windows before and after these threshold changes, we find that while bonds do become larger, firms’ propensity to issue decreases. We interpret this as some firms ‘reaching’ to be included in the index, while other firms who are not comfortable issuing a larger bond, opting out of the bond market altogether. We bolster this result by looking at bond rollovers, and find that bonds that expire in the post-change periods are less likely to be rolled over than bonds that expire in the pre-change periods. Overall, Chapter 3 contributes to the literature that examines the impact that ‘nonfundamental investor demand’ has on firm decisions, including capital structure decisions. Firms appear to be taking advantage of passive funds and other bond index trackers, which we see as a change in the supply curve of capital. The final Chapter of this thesis turns back to the equity markets and examines the impact of financial intermediaries in SEOs. In underwriting SEOs, financial intermediaries can choose to act as market makers, where they buy shares from a seller on their own account without knowing who the ultimate buyers will be (a “bought deal”). In contrast, intermediaries can choose to act only as matchmakers, simply introducing sellers and buyers and helping determine a clearing price for the shares (a “marketed deal”). These two forms of offering differ based on which party bears the risk that a deal can be cleared at a price acceptable to the seller and buyer; in particular, a bought deal involves a shifting of this risk from the seller to the underwriters. Chapter 4 explores the determinants of the choice of form of intermediation and its impact on sellers (firms or selling shareholders), buyers (investors) and underwriters. This Chapter uses data on SEOs in both the U.S., where the majority of deals are marketed, and Canada, where the majority of deals are bought. The first set of results shows that the deal type is correlated with seller bargaining power, competition between underwriters, and dispersion in buyer valuations. The likelihood of a bought deal is increasing in measures of seller bargaining power over the underwriters; this is likely related to the fact that a bought deal involves a negotiation on price between the underwriters and the seller. In contrast, the likelihood of a bought deal is decreasing in measures of seller bargaining power over investors; in marketed deals, the negotiation takes place between sellers and buyers. Proxies for underwriter competition are positively related to a deal being bought; this could indicate that underwriters take on additional risk as a way of ‘buying’ underwriting mandates. Finally, measures of buyer valuation dispersion are negatively related Chapter 1. Introduction 4 to the probability that a deal is bought. The Chapter then examines if there is a relationship between market conditions faced by underwriters and the likelihood that a deal is bought. Despite the fact that bought deals involve additional risk for underwriters, I do not find conclusive evidence that underwriters’ risk-bearing capacity is related to deal type. Next, I discuss the form of intermediation and the occurrence of inefficient deals, defined as seller valuation being greater than buyer valuation (when trade should break down). There is evidence that at least some inefficient marketed deals are not completed, as withdrawn marketed deals are found in the data. On the other hand, there are instances of ‘hung’ bought deals, where shares are left unsold with the underwriter; these deals indicate that either the deal was inefficient, or the seller was able to extract rents (i.e., a higher price) from the underwriters that they would not have been able to achieve in a deal negotiated directly with investors. The final test attempts to measure if the form of offering affects the split of deal surplus between sellers, buyers and underwriters. The seller’s all-in-cost (including discount and underwriters’ spread) in significantly lower in a bought deal, controlling for factors related to deal type, indicating that sellers benefit from receiving a bought deal. The benefit to sellers is related to the fact that the underwriters’ commission is significantly lower in bought deals, despite the fact that underwriters are taking on higher risk. This result indicates that bought deals shift surplus from underwriters to sellers (relative to marketed deals). Overall, this Chapter shows that SEO offering procedure is related to important outcomes for firms and investors and is one of the many ways that financial intermediaries impact equity issuance markets. Chapter 2
Too much information? Increasing firms’ information advantages in the IPO process
With Yan Xiong
2.1 Introduction
Traditionally, high growth private firms in the U.S. have used the public equity markets as their primary source of external financing to fund innovation and expansion. For this reason, well-functioning capital markets have been instrumental to the U.S. economy, supporting job creation and economic growth. In recent decades, however, there has been a decreasing number of U.S. companies going and staying public (e.g., Gao et al., 2013; Doidge et al., 2017), attracting significant concern from regulators.1 In particular, the SEC is concerned that firms using private rather than public sources of financing limits investment opportunities for ‘Main Street’ investors and reduces the availability of “transparent, information rich and fair” public prices (Securities and Exchange Commission, 2019a). In light of firms shunning the traditional initial public offering process, regulators have made interventions to make public equity markets more welcoming and attractive relative to private markets, especially for young and high growth firms. In this paper, we examine an important set of changes to the U.S. IPO process, the Jumpstart Our Business Startups Act of 2012 (the JOBS Act), and ask if they were designed to achieve the regulators’ goals. Through our difference-in-difference framework, we examine the impact of the legislation on treated firms’ ability and willingness to go public, and the related effects on important IPO market outcomes such as the number and quality of public firms. The results are important to consider for other countries considering similar reforms, as well as recent reforms in the U.S. In particular, the SEC has extended some of the IPO process changes to more firms as recently as December 2019.
1In his first speech as Commissioner of the SEC, Jay Clayton said that “the reduction in the number of U.S.-listed public companies is a serious issue for our markets and the country more generally” (Securities and Exchange Commission, 2017a).
5 Chapter 2. Increasing firms’ information advantages in the IPO process 6
The JOBS Act has two main components for eligible emerging growth companies: the ability to interact with investors before filing for an IPO (so-called ‘testing the waters’) and the option to reduce financial disclosure in IPO filings. Regulators designed these changes to ‘de-risk’ and ‘de-burden’ the IPO process for firms, but in doing so they increased the information advantage of firms over investors - testing the waters allows firms to collect information they do not have (buyers’ valuations) while reduced disclosure allows firms to reveal less to investors. Information asymmetry is a friction long-known to have substantial impacts on the ability to trade (e.g. Akerlof (1970), Myers and Majluf (1984)), and these problems are particularly prominent for IPO firms, who are often young and derive substantial value from uncertain future growth opportunities. We show that these changes may have had results that were opposite to the stated goal of increasing the number of listed firms. We argue that by exacerbating information asymmetry between firms and investors, these regulatory changes further contributed to the decline in the number of firms going public, highlighting an important unintended consequence of the JOBS Act. In particular, we demonstrate, both theoretically and empirically, that such changes have made it more difficult for all but the highest quality firms to go public, resulting in fewer firms choosing public listing as a viable option. While somewhat surprising at first glance, our conclusion is in line with the long literature in financial economics that recognizes information asymmetry between buyers and sellers as a key friction in capital markets. Our theoretical framework examines a private firm considering an IPO in the pre- and post-JOBS Act information environments. After the JOBS Act, a firm that is eligible to use its provisions must first decide whether or not do so. We argue that testing the waters is inexpensive and unobservable to the broad universe of IPO investors, so firms will do so and learn about their type. ‘Good’ firms can then signal this information through underpricing or through underwriter certification, while ‘bad’ firms do not go through with the IPO process. This self-selection out of the IPO process results in fewer firms going public but the resulting quality of public firms is higher, and under most parameters welfare is improved by the elimination of overinvestment in negative NPV projects. We also consider theoretically the impact of reduced disclosure. Investors directly observe if a firm chooses to offer reduced disclosure before IPO and will adapt accordingly. While reduced disclosure generates some similar predictions (such as higher underpricing), it also predicts more firms going public and lower quality of public firms. Our empirical analysis can help disentangle these alternative avenues and determine which effect dominates. We test the implications of our theoretical framework by examining initial public offerings between 2007 and 2017, approximately five years before and after the implementation of the JOBS Act. Since many of our outcomes of interest are strongly driven by latent IPO market conditions, our identification strategy relies on a difference-in-difference framework between firms that are eligible to use the JOBS Act provisions and those that are not. The treatment group is U.S. firms that have ‘emerging growth company’ status, or less than $1 billion in revenue in the year before IPO. The main control group is U.S. firms with more than $1 billion in revenue (U.S. non-EGCs). We also show that several of our results are robust to using U.K. firms with less than $1 billion in revenue (U.K. EGCs). On balance, our results are consistent with increased information asymmetry due to the ability to test the waters with investors. We first show that the JOBS Act provisions have succeeded in de-risking the IPO process, reducing withdrawal rates and shortening bookbuilding periods. Second, consistent with increased information asymmetry, we find increased underpricing for treated firms post-JOBS. Third and most importantly for policy considerations, we find that regulators’ goals were not achieved and Chapter 2. Increasing firms’ information advantages in the IPO process 7 that treated firms in the post-JOBS period are less likely to go public, as shown by both lower listing propensity and lower likelihood of choosing IPO over private equity or M&A. This is in contrast to initial evidence that the JOBS Act increased the number of U.S. IPOs (Dambra et al., 2015), which may be attributed to positive market conditions. Finally, we present initial results that treated IPO firms in the post-JOBS period are higher quality. Our paper makes four main contributions. First, we contribute to the growing literature on the declining number of U.S. public companies (Doidge et al., 2017) and IPOs (Gao et al., 2013; Doidge et al., 2013), which has been observed since the turn of the century. There have been several reasons proposed for the decline in IPOs, including the deregulation of private equity markets easing private capital raising (Ewens and Farre-Mensa, 2020), the fact that firms have more intangible assets that are better suited to private ownership (Stulz, 2019), and that market competition has increased the relative benefit of selling to a larger firm rather than going public (Gao et al., 2013). Our paper provides an in-depth analysis of the regulators’ response to the problem of decreasing IPOs, which does not directly address any of these possible explanations.2 Furthermore, the changes allow us to examine a shock that increased information asymmetry between firms and investors in the capital raising process, a friction that can hinder trade (Akerlof, 1970; Myers and Majluf, 1984), and we show that such a shock reduced the number of firms able to go public. Second, our paper provides a framework and evidence that can inform policy makers interested in improving the design of the capital raising process. We join previous empirical work that examines outcomes of regulator intervention in public equity markets, such as the introduction of bookbuilding in Japanese IPOs (Kutsana and Smith, 2004) or the expansion of accelerated seasoned equity offerings in the U.S. (Gustafson and Iliev, 2017). Our analysis suggests regulators both in the U.S. and globally3 should expand their focus from de-risking and de-burdening the IPO process for entrepreneurs and consider the impacts of information asymmetry on both firms and potential investors. While we show that allowing firms to collect additional information may prevent overinvestment in negative NPV projects, it will not increase the number of IPOs, which we understand to be a primary goal for U.S. regulators. Third, the paper contributes to the IPO literature. We provide a cohesive explanation for several changes in the U.S. IPO market in recent years, such as lower IPO withdrawal rates, higher underpricing and diminished propensity to be a public company. Our paper is most closely related to two papers on the effects of the JOBS Act on the IPO market, though we reach different conclusions. Chaplinsky et al. (2017) examine the impact of the JOBS Act on the cost of going public and show that the JOBS Act increased underpricing for IPOs in the post-JOBS Act relative to pre-JOBS, which they attribute to reduced disclosure. We model an alternative mechanism causing the information asymmetry (testing the waters) that drives the underpricing, and explore several alternative results that rule out disclosure as the explanation. Dambra et al. (2015) provide a detailed description of the IPO on-ramp provided by the JOBS Act and estimate that it increased the number of U.S. IPOs in the two years following the Act’s passage. We present several tests that show IPO propensity actually decreased for treated firms, so we suspect the increase in raw number of IPOs was likely related to overall market conditions and not JOBS Act specific. Finally, our setting provides insight into the consequences of information collection activities by
2Our paper confirms the position in Doidge et al. (2018) that deregulation is not the solution: “U.S. financial devel- opment has evolved in such a way that some types of firms can be financed more efficiently through private sources than through public capital markets ... . No deregulatory action is likely to restore the public markets in this case.” 3Canadian regulators adopted similar regulation as the JOBS Act in August 2013. Chapter 2. Increasing firms’ information advantages in the IPO process 8
firms, which are not usually observable. While recent papers are able to identify investors’ information acquisition activities in a few situations,4 we believe that testing the waters is one of the first ways to identify information acquisition by firms. This allows us to examine a concrete setting where increasing information asymmetry in favor of the seller can limit the ability for trade due to exacerbated information asymmetry (Levin, 2001).
2.2 Institutional Background
2.2.1 JOBS Act
The Jumpstart Our Business Startups Act was proposed in late 2011 in response to the 2008 financial crisis and enacted on April 5, 2012. Title I of the Act (“Reopening American Capital Markets to Emerging Growth Companies”) was designed to encourage small private companies to access the public capital markets via initial public offering. The regulatory response focused on the perspective of private firms: changes were proposed to de-risk and de-burden the process for firms considering an IPO. In particular, for firms with annual sales less than $1 billion (emerging growth companies or EGCs), the Act included provisions to reduce the risks of IPO failure and provisions to reduce the regulatory burden at IPO and on an ongoing basis. In terms of de-risking provisions, firms are permitted to confidentially file registration statements with the Securities and Exchange Commission (SEC) in order to clear regulatory comments. This allows firms to answer regulator concerns about the prospectus without signaling to the market the intention of selling securities and without revealing information to competitors. Firms are also permitted to “test the waters” by meeting with qualified institutional buyers before the filing of any registration statement. We discuss this further in the following section. In terms of de-burdening provisions, the Act reduced ongoing accounting and disclosure requirements, including requiring only two years of audited financials (from three), reducing disclosure around executive compensation, extending the deadline for auditor attestation to five years after IPO (from two), allowing the firm from opting out of future FASB or PCAOB standards, and removing the need to conduct say- on-pay votes. These provisions reduced the ongoing cost of being a public firm, and reduced the amount of information available to investors.
2.2.2 Testing the Waters (TTW)
Section 5(c) of the Securities Act prohibits the discussion of a potential offering without the filing of a registration statement. As such, discussions with potential investors before a public filing would be considered “gun-jumping” and could result in fines or delays. The JOBS Act added section 5(d) to the Securities Act, which provided exemptions for EGCs to discuss a potential offering with qualified institutional buyers and institutional accredited investors (two categories of large institutional investors who are financially sophisticated and generally do not require investor protection). The purpose of these “testing the waters” discussions is to gauge investor interest in the firm’s IPO, assess investors’ valuation of the securities, and determine which IPO terms may
4Examples include access to FDA information through the Freedom of Information Act (Gargano et al., 2017), news diffusion through Bloomberg terminal clicks (Fedyk, 2021), and web traffic on the SEC’s EDGAR website (Chen et al., 2020; Baugness et al., 2018; Crane et al., 2019; Drake et al., 2016). Chapter 2. Increasing firms’ information advantages in the IPO process 9 be important to investors. After engaging in TTW meetings, the firm then decides whether or not to actually begin the IPO process and publicly file a registration statement. Figure 2.1 compares how the U.S. IPO process timeline has changed for EGCs in the post-JOBS Act period. The most notable difference is when the firm first publicly declares its intention to go public and reveals corporate information to the market (and its competitors). In the pre-JOBS period (as well as non-EGCs in the post-JOBS period), the public knows about a firm’s desire to go public very early in the process, and the registration statement is available while the firm clears regulatory comments and prepares for the roadshow. The firm is not allowed to engage in any marketing activities before making the decision whether to begin the IPO process. In the post-JOBS period, on the other hand, EGCs are able to engage with potential institutional buyers and decide whether or not to pursue an IPO based on early feedback. In addition, through the confidential filing system, the firm is able to clear regulatory comments before declaring their intention to go public. Both these timeline shifts allow a firm to gather more information and ensure a smoother regulatory experience before filing a public prospectus.
Firm decides Firm publicly SEC clears Firm publicly Bookbuilding Underwriters Firm decides to begin IPO files registration registration files begins with determine to IPO or process statement statement prospectus initial price price and withdraw range quantity
(a) IPO process in the pre-JOBS period
Firm tests the Firm decides Firm SEC clears Firm publicly Bookbuilding Underwriters Firm decides waters with to begin IPO confidentially registration files begins with determine to IPO or investors process files registration statement prospectus initial price price and withdraw statement range quantity
(b) IPO process available to EGCs in the post-JOBS period
This figure compares the IPO process timeline in pre- and post-JOBS Act periods. Italics indicates when firm’s desire to go public is known by public.
Figure 2.1: U.S. IPO process in pre- and post-JOBS Act periods.
It is important to distinguish between TTW meetings and discussions with investors during the bookbuilding phase of the IPO (road show meetings). In both types of meetings, the firm will be telling their story to investors in hopes of garnering interest; in addition, investors will be providing the firm with information via their interest and valuation. In a TTW meeting, however, the firm is not committed to actually selling any shares, and any investor interest is non-binding.5 In addition, a TTW meeting can discuss an investor’s view on IPO terms, such as offering size or percentage of shares from primary vs. secondary sources; at the bookbuilding stage, these terms are generally taken as filed.
5Legally, indications of interest during the bookbuilding phase are also not binding since the firm has not filed a final prospectus, but due to repeated interactions between underwriters and investors, such interest can be considered a true “order” for the firm’s shares. Chapter 2. Increasing firms’ information advantages in the IPO process 10
Having emphasized its difference with road show meetings, we believe there are truth-telling moti- vations for both parties in a TTW meeting. For firms, any misrepresentation will be quickly discovered when they file a registration statement; this would be harmful for their reputation and may result in a failed IPO or legal penalties. For investors, providing valuable information in a TTW meeting will influence the firm’s decision to actually file for an IPO and may affect the investor’s ultimate allocation of shares in the case the IPO goes forward. The findings of our analysis of the JOBS Act have current relevance. The SEC, based on discussions with lawmakers and industry participants, extended the ability to test the waters to all firms regardless of size as of December 3, 2019.6 Testing the waters is seen as “a cost-effective means for evaluating market interest before incurring the costs associated with [an] offering” (Securities and Exchange Commission, 2019b). This is particularly beneficial if the firm decides based on TTW meetings not to proceed with an IPO, since “information has been disclosed only to potential investors and not to the company’s competitors” (Department of the Treasury, 2017). Overall, regulators view testing the waters as a means to increase the chances that an offering is successful, which may encourage more private firms to complete registered offerings, a primary goal of the SEC and other market participants.
2.3 Theoretical Framework
In this section, we build the conceptual framework that formalizes our subsequent empirical analysis. This also enables us to generate richer testable implications. Testing the waters and the reduced disclo- sure are two provisions in the JOBS Act that significantly affect firms’ information environments in the IPO process. Our framework focuses on testing the waters; information acquisition by firms is relatively understudied compared to optimal disclosure. While the regulators’ response views the issues in the IPO process solely from the firm perspective, we consider the effect of testing the waters on both firms and prospective investors.
2.3.1 Implications of Testing the Waters
Model Setup and Benchmark Economy
A firm is owned by an entrepreneur, and the entrepreneur is seeking financing for a project. Her initial wealth is A. The total amount needed to complete the project is I > A. Therefore, the entrepreneur seeks to raise I −A from investors in a competitive financial market. The project will yield R if successful and 0 otherwise. The likelihood of success is pH ∈ (0, 1] for a high-type project and pL ∈ (0, pH ) for a low-type project. That is, a high-type project has a higher probability of succeeding than a low-type project. The prior probability of a project being high-type is Pr(H) = α ∈ (0, 1). We assume that the high-type project is worth funding; that is, the net-present-value (NPV) of the high-type project is pH R − I > 0. While the low-type project is not as attractive as the high-type project, we assume that it can still be worthy of investment. That is, the NPV of the low-type project can be positive
(pH R > pLR > I) or negative (pH R > I > pLR). Since the entrepreneur owns the firm and seeks to issue securities to fund the project, we use entrepreneur, firm, and issuer interchangeably.
6The ability to confidentially file registration statements was extended to all firms on July 10, 2017 in order to “provide companies with more flexibility to plan their offering” (Securities and Exchange Commission, 2017b). Chapter 2. Increasing firms’ information advantages in the IPO process 11
In order to study the effect of information acquisition prior to the firm’s IPO decision, we consider the possibility that the firm can choose whether or not to acquire information, i.e., test the waters, at a negligible cost. To be specific, if the firm is not allowed to test the waters, neither the entrepreneur nor the investors have any information about the project type beyond the prior distribution specified above.7 This serves as a benchmark for us to gauge the effect of the firm’s potential information acquisition. If, however, the firm is allowed to test the waters, it can choose whether or not to exercise this option. We model the outcome of testing the waters as the entrepreneur learning the type of her project, while investors still do not know it. This assumption can be mapped to some realistic settings. For example, each investor’s assessment about the market environment determines his evaluation of the securities, which are provided to the firm in the form of a demand schedule (similar to bookbuilding). Investors collectively know the market environment better than the firm (e.g., Bond et al., 2012). So, by collecting the demand schedule of several institutional investors and aggregating their information, the firm gains superior knowledge about its market valuation. In our model, the expected value of the project equals to the success rate (p) times the return upon success (R). Given R, knowing the project expected value is equivalent to knowing the project success rate. Therefore, after testing the waters, the entrepreneur learns the success rate of her project (or the project type). Finally, since each investor’s demand schedules are submitted to the firm like a sealed bid auction (again, similar to bookbuilding), investors are not able to aggregate their information and learn the firm’ type. Further, the information acquisition process is not observed by the investors. While the investors with whom the firm has met will know that the firm is testing the waters, firms meet with only a small number of investors8 and they do not share information with the broader set of investors. In addition, the investors with whom the firm does not meet learn whether the firm has tested the waters only at the time of filing the underwriting agreement, after the offering has been priced and after the investor has decided to purchase securities. In summary, we present the timeline of the model in Figure 2.2. There are three dates. At t = 0, the firm that is eligible to test the waters chooses whether or not to exercise this option. At t = 1, given its information set, the firm determines the way to raise the funds, which will be specified shortly. At t = 2, all uncertainty about the project is revealed and all agents including the entrepreneur and the investors consume.
t = 0 t = 1 t = 2
The type of the project The firm chooses whether The firm determines the is revealed and all agents or not to test the waters. way to raise the funds. consume.
Figure 2.2: Timeline of events
7We normalize the information asymmetry between the entrepreneur and investors to zero so that we can cleanly identify the effect of the new information acquired by the firm by testing the waters. 8This is consistent with Jenkinson et al. (2018), who show that European firms undergoing IPO engage in ‘pilot fishing’ (similar to testing the waters) with 4.2% of bidders, which translates to approximately 6 investors assuming the median number of bidders of 140. Chapter 2. Increasing firms’ information advantages in the IPO process 12
Benchmark Economy Consider the benchmark economy where the firm is not allowed to test the waters. Since the two types of projects are pooled together, the expected success rate for the project is p¯ = αpH + (1 − α)pL. In order to have the project financed, the entrepreneur will offer to the investors I−A the return R0 = p¯ in the case of success (or equivalently, the entrepreneur gives out R0 of the shares of her company to investors), where R0 leaves the investors break-even with zero profits:
pR¯ 0 − (I − A) = 0. (2.1)
After paying to the investors, the entrepreneur’s expected payoff is thusp ¯(R − R0) − A =pR ¯ − I, where the equality follows by substitutingp ¯. Therefore, only whenpR ¯ > I can we observe the firm raise capital in public markets and the securities are fairly priced, that is, there is no underpricing. The following proposition summarizes the equilibrium in the benchmark economy without testing the waters.
Proposition 1. Consider the benchmark economy in which the firm is not allowed to test the waters.
Let p¯ = αpH + (1 − α)pL. There always exists a unique equilibrium.
(1) If pR¯ > I, in equilibrium both the high- and low-type firms obtain financing and they offer a stake I−A p¯ to investors when the project succeeds.
(2) If pR¯ ≤ I, in equilibrium neither type of firm obtains financing; that is, the market breaks down.
Note that as stated in Proposition 1, ifpR ¯ < I, the capital market does not exist in the first place because the firm won’t accept negative profits. We thus focus on the case in whichpR ¯ > I hereafter.
The Economy with Testing the Waters Permitted
We now consider the economy in which testing the waters is permitted. By comparing this economy with the benchmark, we can gauge the effect of the option to test the waters during the process of capital raising. To begin with, more knowledge about project valuation is always beneficial from the perspective of the individual firm, without considering the potential response of the investors to the increased information asymmetry. Further, even when investors’ reactions are taken into account, because testing the waters is not observable to the investors at the securities-pricing stage, given investors’ belief, the firm cannot be worse off by exercising this option. Therefore, in equilibrium, the firm should test the waters, which generates the following implication in the context of JOBS Act.
Implication 1 (Use of testing the waters provision). Firms that are eligible to test the waters will exercise this option. After testing the waters, firms’ individual decision making should be enhanced.
With better knowledge about its own type, if a high-type firm can do nothing to separate itself from the low type, it is well known that the public capital market can break down due to information asymmetry (Akerlof, 1970). However, we consider the possibility that high-type firms can take actions to mitigate the infor- mational disadvantage of investors and hence restore the fund raising. We focus on two costly signals that high-type firms can use to separate themselves from low-type firms that are prominent in the IPO literature. First, the high-type firm can use underpricing as a signaling device to separate from the low-type. Second, high-type firms can hire (more) reputable underwriters to certify firm quality. Below, Chapter 2. Increasing firms’ information advantages in the IPO process 13 we explore theoretically how these actions can be used as a separation mechanism, and in the following section we examine each of these actions empirically.
Underpricing as a signaling device Being the more informed side, the high-type firm can convey its quality to prospective investors by leaving money on the table. Following the signaling literature (e.g., Ibbotson, 1975; Allen and Faulhaber, 1989; Grinblatt and Hwang, 1989; Welch, 1989), when testing the waters is permitted, we can characterize the equilibrium as follows.
Proposition 2. Consider the economy in which testing the waters is permitted. (1) If I < R ≤ I +A 1 − 1 , there exists a unique equilibrium which features a separating allocation p¯ p¯ pL p¯ in which the high-type firm is funded whereas the low-type firm is not. Further,
I I 1 1 I−A (1.1) if < R ≤ + A − , in equilibrium, the high-type firm offers a stake of RH = p¯ pH pL pH pH to investors when the project succeeds; (1.2) if I + A 1 − 1 < R ≤ I + A 1 − 1 , in equilibrium, the high-type firm offers a stake pH pL pH p¯ pL p¯ A of RH = R − to investors when the project succeeds. pL (2) If R > I + A 1 − 1 , there co-exists two equilibria: a separating equilibrium and a pooling p¯ pL p¯ A equilibrium. In the separating equilibrium, the high-type firm offers a stake of RH = R − to pL investors when the project succeeds whereas the low-type firm is not funded. In the pooling equi- I−A librium, both the high- and low-type firm obtain financing and they offer a stake p¯ to investors when the project succeeds.
All else equal, the parameter R captures the profitability of the potential project. Proposition 2 states that if the to-be-funded project is not that profitable (small R, see part (1) of the proposition), the high-type firm will separate from the low-type by offering a contractual term that is unappealing to the low-type firm, who would then prefer not to be funded, and allow the investors to at least break even. Simple calculation shows that the low-type firm does not have any incentive to mimic the high- type firm, which thereby sustains the equilibrium. Intuitively, underpricing serves as a device for the high-type firm to signal its own type to investors. In the separating equilibrium, a high-type firm’s profit is pH (R − RH ) − A > 0. The investors’ payoff becomes
pH RH − (I − A) ≥ 0. (2.2)
Note only when R is very small (see part (1.1) of the proposition) can the separation entails no distortion to the high-type firm; that is, the investors break even and the high-type firm does not leave money on the table. However, if R is relatively large (see part (1.2) of the proposition), the separation is associated with distortion for the high-type firm: it leaves money on the table for the investors and there is underpricing in the IPO. Nonetheless, if the project is likely to be very profitable (large R, see part (2) of the proposition), in addition to the separating equilibrium, there can coexist a pooling equilibrium in which both types of firms raise funds in the capital markets. This is because the project is so profitable that the low-type firm always has the tendency to undertake it. Meanwhile, the profit potential is high, so the high-type firm can afford to be pooled with the low-type. Further, it can be shown that such a pooling equilibrium is preferred by the high-type firm to the separating equilibrium. Chapter 2. Increasing firms’ information advantages in the IPO process 14
R Region 1 Region 2 Region 3 Region 4 I I 1 1 I 1 1 I Left bound p¯ p + A p − p p¯ + A p − p¯ p H L H L L Right bound I + A 1 − 1 I + A 1 − 1 I pH pL pH p¯ pL p¯ pL No TTW Pooling Pooling Pooling Pooling Separating TTW Separating Pooling & Separating Pooling & Separating (no distortion)
Table 2.1: Summary of equilibrium in the economy with and without testing the waters
Table 2.1 summarizes the equilibrium in the benchmark economy without testing the waters and that in the economy where testing the waters is permitted for different values of R. By comparing the latter with the former, we can evaluate the implications of testing the waters when the high-type firm uses underpricing as a signaling device to separate from the low-type. Specifically, we have the following implications regarding underpricing, listing propensity, and firm quality.
Implication 2 (Underpricing). After testing the waters is permitted, firms that are eligible to do so experience (weakly) higher underpricing.
One of the most puzzling empirical regularities of the IPO market is the underpricing phenomenon (Ljungqvist, 2007). Our model predicts that a firm’s information acquisition may actually lead to more money left on the table in IPOs. Specifically, as summarized in Table 2.1, in Region 1, i.e., I < R < I + A 1 − 1 , there is no change in underpricing before and after the testing the waters p¯ pH pL pH regulation. However, in Region 2 – 4, i.e., R > I + A 1 − 1 , investors earn zero profits in pH pL pH expectation when testing the waters is not allowed, whereas they may have positive expected return when the firm is allowed to test the waters. That is, the firms eligible to test the waters can experience more underpricing after the JOBS Act due to greater information asymmetries between the firm and the investors and the high-type firm’s intention to separate from the low-type. Overall, after testing the waters is permitted, the underpricing of the firm is weakly higher. This result suggests one unintended consequence of the testing the waters provision in the JOBS Act. It is worth noting that the increasing underpricing effect applies to the firms that are eligible to test the waters, not constrained to the firms that actually test the waters. This is because testing the waters is not observable to the investors, and in equilibrium, investors believe that all firms test the waters and firms indeed do so. Therefore, the whole firm group that is eligible to test the waters will be affected.
Underwriter certification as a signaling device With more information after testing the waters, the high-type firm may choose to hire a more reputable underwriter to convey its type to investors (Booth and Smith, 1986). We model the certification in reduced form as the purchase, at cost c > 0, of a signal that perfectly reveals the firm’s type. That is, at a cost c, the firm can have access to a reputable underwriter who then provides accurate evidence regarding the quality of the project, namely, investors will then know whether the probability of success is pH or pL. A low-type firm obviously has no incentive to pay the cost to reveal its type. By resorting to a certifier, I−A a high-type firm can offer only RH = to the investors (which leaves the investors break even) in pH the case of success. The good type prefers to resort to a certifier if and only if R − RH − c ≥ R − R0, Chapter 2. Increasing firms’ information advantages in the IPO process 15 or equivalently,
1 1 c ≤ I − A − , p¯ pH wherep ¯ is given by equation (2.1). In other words, after testing the waters is permitted and if the cost to hire reputable underwriters is not prohibitively high, high-type firms will pay for the certification to mitigate the investors’ information disadvantage.
Implication 3 (Underwriter certification). After testing the waters is permitted, if the certification cost is low, the high-type firm resorts to a certifier.
The use of underwriter certification is an alternative mechanism to underpricing that allows high- type firms separate from themselves from low-type firms. However, under both separating mechanisms, the result is the same: low-type firms are not willing to access the capital markets. This leads us to two other important implications.
Other implications
Implication 4 (Listing propensity). After testing the waters is permitted, the listing propensity of firms that are eligible to do so can (weakly) decrease.
Our theory predicts that after testing the waters is allowed, while firms’ listing propensity remains unchanged if pooling equilibrium arises, in the separating equilibrium the low-type firm will forgo the IPO, thereby reducing an average firm’s propensity to go public. This effect is opposite to the regulators’ stated goal of increasing the number of listed firms.
Implication 5 (Firm quality). After testing the waters is permitted, the average quality of the firm group that is eligible to do can increase.
While there may be some costs to the separating equilibrium (either in the form of higher underpricing or a cost for underwriter certification), the average quality of IPO firms in this group will improve. This is because testing the waters enables the firm to better learn its type, and the high-type firm is likely to introduce or accept distortions in the financing contracting so as to signal attributes that are attractive to the uninformed side of the market. In the consequent equilibrium, the high-type firm is able to separate itself from the low-type and the low-type has to forgo the IPO, which leads to an improvement in the overall quality of firms that complete an IPO process. Related to the implications for firm quality, interestingly, the welfare implications of testing the waters are more nuanced. If firms use underpricing as a separating mechanism, we can see that in Regions 1 – 3, after testing the waters is permitted, social welfare can improve due to elimination of overinvestment in negative-NPV projects (pLR < I). However, in Region 4, while the low-type firm is worth investing (pLR > I), with more information about its own type the high-type firm may still want to separate from the low-type, thereby leaving the positive-NPV low-type firm unfunded. In this case, testing the waters can reduce social welfare.
2.3.2 Implications of Reduced Disclosure
Because the JOBS Act also allows small firms to provide reduced disclosure, a plausible alternative framework assumes that firms know their type and the JOBS Act allows low quality firms to better Chapter 2. Increasing firms’ information advantages in the IPO process 16 hide this from investors. Reduced disclosure also increases information asymmetry between firms and investors, and together with the lower firm quality, firms that are eligible to take advantage of this option can experience increased underpricing (Rock, 1986) to compensate the uninformed investors. Given the sample that is eligible to test the waters is the same as the sample that is able to reduce disclosure, the two provisions have the same theoretical implication: underpricing should increase for EGCs relative to non-EGCs. On the other hand, the two provisions differ on two other implications. While the above testing-the- waters provision predicts that low-type firms will self-select out of the IPO process, reduced disclosure allows these low-type firms to go public. The result is that while public firm quality increases due to the testing-the-waters provision, it decreases because of the reduced-disclosure provision. This also speaks to the likelihood that a given firm will be public; while the testing-the-waters provision results in lower listing propensity, the reduced-disclosure provision increases listing propensity. Table 2.2 summarizes these differences.
Implication Testing the waters Reduced disclosure Underpricing for EGCs Increase Increase Firm quality for EGCs Increase Decrease Listing propensity for EGCs Decrease Increase
Table 2.2: Comparison of implications of testing the waters and reduced disclosure
It is also plausible that reduced disclosure benefits certain types of firms. If disclosure is costly, then smaller firms, for whom IPO costs are relatively higher, will benefit more from the ability to reduce disclosure and associated costs. It could also benefit firms in certain types of industries or more complex firms. In the following section, we explore the different implications of the two provisions empirically.
2.3.3 Discussion
Regulators’ objectives
The JOBS Act was designed to encourage small companies to access the public capital markets. If the regulators’ objective is to increase the number of listed firms, testing the waters may not fulfill this purpose. For example, as shown in Table 2.1, the number of listed firms may decrease as the low-type firms are driven out of the market. If, however, the regulators care about efficiency such that all positive-NPV projects can be funded, then while firms’ information acquisition through testing the waters can eliminate the over-investment problem (see Regions 1 – 3 in Table 2.1), it can also result in under-investment problem since the low-type firms with positive-NPV project may not be funded (see Region 4 in Table 2.1).
Information asymmetries among investors
When issuing new claims, the firm should be preoccupied with two types of informational asymmetries: between the issuer and the investors, and among investors. Our framework of testing the waters is built upon the first type of information asymmetry. While we argue that in the context of testing the waters most information asymmetries relate to the firm’s private knowledge about assets in place and prospects, it is easy to envision that asymmetries of information among investors also change after the water testing: the institutional investors that have contact with the firm should learn something during Chapter 2. Increasing firms’ information advantages in the IPO process 17 the communication process and possibly gain some information advantage over other investors. We now briefly discuss the second type of information asymmetry and its implications for financing decisions. As in the well-known paper by Rock (1986), in fixed price-offerings, underpricing is needed to com- pensate small, uninformed investors for the winner’s curse. One implication of this theory is that the more severe the information asymmetry among the investors, the more underpricing for the IPO firm. When the firms test the waters before IPOs, the institutional investors that talk to the firm can gain some information advantage over others, thereby leading to potentially more information asymmetry among the investors and more underpricing. This is consistent with the above implication about underpricing (see Implication 2). While this might be a potential driving force behind the observed underpricing re- sult, we build our framework based on the information asymmetry between the firm and investors. This is because in the context of testing the waters, firms are the entities that initiate the whole IPO process, and the relative information status among the investors is only a by-product of the firms’ activities.
2.4 Empirical Results
We now proceed to test these predictions using data on all IPOs in the United States, and for robustness purposes, IPOs in the United Kingdom.
2.4.1 Data
We include in our sample all initial public offerings in the United States and the United Kingdom with an issue date between January 1, 2007 and December 31, 2017 and an offering size of at least $5 million. We exclude financial firms (with an SIC code beginning with 6), shell companies, limited partnerships, unit offerings, OTC listed firms, best efforts and non-underwritten offerings, and foreign issuers (filings of F-1 rather than S-1). We use firm and offering data from both Thomson Reuters’ Securities Data Company and Bloomberg, and supplement with information hand collected from firms’ IPO prospectuses. Confidential filings are found on EDGAR, while use of the testing the waters provisions is hand collected from firms’ underwrit- ing agreements and in correspondence with the SEC. Use of the reduced disclosure provisions is hand collected from firms’ prospectuses. We calculate listing propensity using the number of public firms in Compustat and total number of U.S. firms from the Census Bureau’s Statistics of U.S. Businesses (SUSB); IPO propensity is calculated using data on private equity and merger and acquisition trans- actions from Capital IQ. We use CRSP and Compustat data for post-IPO outcomes, as well as several sources for financial misconduct. Appendix A describes all variables and their sources in detail. We document that a firm has tested the waters if they include standard language in their underwriting agreements that they or their underwriters may have engaged in testing the waters activity. We also track when IPO firms provide more detailed information that confirms that they actually tested the waters. For example, firms may list the materials shown to investors or list the time frame in which they met with investors. We code this as ‘written TTW.’ Appendix C provides samples of TTW language taken from IPO documents. Our main IPO outcome variables include withdrawal rates, time in IPO process, price range adjust- ments, and final IPO price relative to original range. Withdrawal is measured as a dummy equal to 1 if the IPO is withdrawn before pricing and 0 otherwise. Time in IPO process includes the number of days between public filing and launch of bookbuilding period and the number of days between launch Chapter 2. Increasing firms’ information advantages in the IPO process 18 and pricing of the IPO (only applicable for completed IPOs). Adjust price range up (down) is a dummy equal to 1 if the price range is revised up (down) from the original price range. Similarly, priced above (below) original range is a dummy equal to 1 if the final IPO price is above the original high (below the original low). In terms of potential firms’ responses to increased information asymmetry, we examine underpricing and measures of underwriter certification. Underpricing is calculated as the price at the end of the first day of trading divided by the IPO price minus 1. We use several measures of underwriter certification, including the number of bookrunners, the spread paid to underwriters, the allocation of shares to the top left and all bookrunners, and the Carter-Manaster rank of the top left and all bookrunners. We examine listing propensity for the average firm in the U.S. economy using number of public companies from Compustat divided by the total number of companies from the SUSB. IPO propensity is calculated as the number of IPOs divided by the total number of transactions by private firms (including IPOs, private equity and mergers and acquisitions). We also examine post-IPO outcomes in three main categories. First, we look at trading outcomes including market- and match-adjusted two year return, dummy variables if the stock is still trading one year after IPO and at the end of 2019, and a dummy if the firm has gone bankrupt.9 Second, we examine financial statement ratios one year after IPO relative to the median firm in the same industry-quarter. We measure relative accounting outcomes including: year-over-year sales growth, return on assets, return on equity, profitability and Tobin’s Q. Finally, we look at proxies for financial misconduct, including dummies for actions by the SEC’s Accounting and Auditing Enforcement Releases, restatements due to fraud or resulting in an SEC investigation from Audit Analytics, and non-dismissed class action lawsuits as per the Securities Class Action Clearinghouse. We classify all IPO firms with revenue less than $1 billion as an emerging growth company, though that classification and its privileges only exist for U.S. firms in the post-JOBS period. We describe how we assign EGC status for listing propensity and IPO propensity under the discussion of implication 3. Table 2.3 shows summary statistics for the main variables of interest for U.S. EGCs, U.S. non-EGCs and U.K. EGCs in both the pre- and post-JOBS periods. Our U.S. sample includes 1,261 IPOs, of which 1,007 were launched and 868 were completed. Our U.K. sample includes 333 IPOs, of which 269 were completed. Our U.S. and U.K. samples include 22,704 and 4,837 PE and M&A transactions, respectively.
[INSERT TABLE 2.3 HERE]
Figure 2.3 shows the number of IPOs filed per year over the sample period, excluding 2012, the year of the enactment of the JOBS Act. The brown bars represent the number of U.S. EGCs while the blue bars represent the number of U.S. non-EGC plus U.K. EGC IPOs (normalized by the relative proportion of IPOs in the category to the number of U.S. EGC IPOs). U.S. EGC IPOs are much higher in 2013-2014 relative to pre-2012 levels, but we would highlight two important points. First, it appears that the trend of increasing IPOs begins in the recovery of the Great Recession. The finding of increased IPOs in the post-JOBS period discussed in Dambra et al. (2015), whose sample ended in 2014, may have been a premature conclusion. Second, the surge of IPOs by U.S. EGCs in 2013-2014 is similar in pattern to the surge by U.S. non-EGCs and U.K. EGCs, who were not affected by the JOBS Act. We believe this points to the importance of underlying market conditions, which we keep in mind in undertaking our empirical analysis. This is discussed further in the next section.
9We estimate bankruptcy if the stock is delisted, the return to the delisting date is -90% or lower, and the final trading price is $0.50 or less. This criteria leads us to estimate 6.3% of the IPO sample goes bankrupt. Chapter 2. Increasing firms’ information advantages in the IPO process 19 150 100 IPOs by year 50 0 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Firms subject to JOBS Act Firms NOT subject to JOBS Act
This figure plots the number of IPOs filed per year for U.S. EGCs, U.S. non-EGCs, and U.K. EGCs. The number of U.S. non-EGCs and U.K. EGCs are normalized by dividing the total number of IPOs in the respective category by the total number of U.S. EGC IPOs.
Figure 2.3: Number of IPOs
2.4.2 Empirical Design
We perform the majority of our analysis in a difference-in-difference setting around the introduction of the Jumpstart Our Business Startups Act. Our pre-JOBS period include IPOs publicly filed before December 1, 2011 (when the JOBS Act was first proposed) and the post-JOBS period includes IPOs publicly filed after June 30, 2012. Though the JOBS Act was enacted on April 5, 2012, we allow a period of time for firms to actually take advantage of the testing the waters provision. Our treatment group includes firms with less than $1 billion in revenue in the year before IPO; in the post-JOBS period, these firms are deemed emerging growth companies and are eligible to use the provisions of the JOBS Act. We use two different control groups. The first is U.S. firms with at least $1 billion in revenue in the year before IPO (U.S. non-EGCs). As shown in Table 2.3, this is a much smaller sample of IPOs and these firms are larger by design, so to the extent that firm size influences any outcomes of interest, this is an imperfect control group. As an alternative, we also examine IPOs of U.K. firms10 with the equivalent of less than US$1 billion in revenue in the year before IPO; in other words, these firms would be considered EGCs if they were going public in the U.S. after the JOBS Act (U.K. EGCs). This control group has more observations relative to the U.S. non-EGCs, but there are several limitations, such as potential latent differences in U.S. vs U.K. firms and markets, and limited data availability for some outcomes of interest. We acknowledge that neither control group is perfect, but we hope to improve the testing of our theoretical predictions by looking at results using both control groups. It is important to note that in contrast to a typical difference-in-difference analysis with panel data,
10We selected U.K. EGCs as the second control group by first considering the five control countries used in Dambra et al. (2015): Australia, Canada, Hong Kong, Japan and U.K. We eliminated Canada as they adopted rules similar to the JOBS Act in 2013. Given the importance of legal origin to capital markets (La Porta et al., 1997), we eliminated Japan (German legal origin compared to English origin for the U.S.). Of the remaining three countries, U.K. is the largest economy, and has most similar statistics in terms of IPOs relative to public companies (see Table B1 in Dambra et al. (2015)). Chapter 2. Increasing firms’ information advantages in the IPO process 20 a firm completes only one IPO and thus we do not observe the same sample in the pre- and post-JOBS periods. For this reason, we are drawing inferences from the pooled cross-section of IPO firms (i.e., comparing the average EGC or non-EGC firm in the pre- or post-JOBS period). The key assumption for the validity of the use of the difference-in-difference framework is that absent the treatment, the difference in outcome variables between the treatment and control groups would remain constant. In other words, we want to observe parallel trends in outcome variables for the treated and control groups in the pre-treatment period. Panel (a) of Figure 2.4 shows the average annual underpricing residual (after controlling for firm size and market performance) in each sample year before and after 2012 for U.S. EGCs compared to U.S. non-EGCs and to U.K. EGCs. Panel (b) shows the same analysis for the propensity for a given private firm deal to be an IPO (instead of a private equity or M&A transaction). Pre-treatment patterns for the treatment group and control groups generally follow similar patterns, though there are differences in levels and volatility. We show exposition of parallel trends in other outcome variables in Appendix D. .4 16 17 .08 14 16 .3 .06 15 12 .2 .04 14 10 IPO_propensity Underpricing residual .1 13 .02 8 12 6 0 0 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Year of filing Year of filing Year of filing Year of filing
U.S. EGC U.S. EGC U.S. non-EGC U.S. non-EGC U.K. EGC U.K. EGC
(a) Underpricing (b) IPO propensity Panel (a) of this figure plots the average underpricing residual (controlling for natural logarithm of assets and market return) per year for IPOs for U.S. EGCs (less than $1 billion in revenue in the year before IPO) compared to U.S. non-EGCs (more than $1 billion in revenue) and U.K. EGCs. Panel (b) of this figure plots IPO propensity per year for U.S. EGCs compared to U.S. non-EGCs and U.K. EGCs. Both panels have a vertical line at 2012, the year of implementation of the JOBS Act.
Figure 2.4: Parallel trends
We use a difference-in-difference setting because we believe many of the outcome variables for IPO firms are heavily influenced by market conditions that do not have a sufficient empirical proxy, such as investor sentiment. U.S. non-EGCs face the same market conditions as EGCs, but do not benefit from the provisions of the JOBS Act, allowing their inclusion to control for these latent market conditions. For this to work, it is important that firms cannot influence their treatment status; in other words, firms do not manipulate along the running variable (revenue in the year before IPO) in order to receive EGC status.11 While we do not believe firms engage in fraudulent accounting manipulation, it is possible that firms time their IPO as they approach the $1 billion revenue cutoff in order to benefit from the testing the waters provisions (any disclosure benefits would be lost as soon as the firm loses EGC status; for firms so close to $1 billion in revenue, this would likely be in the first year as a public company). As shown in
11Our focus here is on U.S. firms only, as revenue levels affect treatment status; U.K. firms are not subject to different treatment above or below the $1 billion revenue mark. Chapter 2. Increasing firms’ information advantages in the IPO process 21
Appendix E, we acknowledge that there appears to be a change in distribution in the post-JOBS period, with a small jump in IPOs just below the $1 billion cutoff compared to the pre-JOBS period (though the jump is not statistically significant). However, there are only 15 IPOs launched (12 completed) in the post-JOBS period with revenue between $900 million and $1 billion, representing 2.6% of the post-JOBS EGC sample.12 As an alternative to EGC status for our treatment and control groups, we considered the use of the small reporting company (SRC) exemption (also used in Chaplinsky et al. (2017), Balogh et al. (2020)), which allows U.S. firms with less than $75 million in public float to reduce disclosure in ways similar to the JOBS Act (see Anthony L.G. PLLC (2018) for a discussion of provisions). In particular, firms eligible for SRC status can provide reduced disclosure in the pre-JOBS period but are not able to test the waters, while in the post-JOBS period SRC (and all EGCs) are able to do both. However, eligibility for SRC status is determined by the IPO size, which is decided by the firm. For this reason, SRC status is easily manipulated13 and thus we do not use it in our empirical analysis.
2.4.3 Examining Theoretical Implications
Implication 1: Eligible firms will test the waters to enhance individual firm decision making. We first examine U.S. EGCs’ use of the JOBS Act provisions in the post-JOBS period in Table 2.4.
[INSERT TABLE 2.4 HERE]
More than 90% of EGCs confidentially file their prospectus with the SEC and include language in their IPO documentation that they or their underwriters may have tested the waters. We find that 28% of firms with completed IPOs engaged in written testing the waters activities. Comparing the usage of JOBS Act provisions for the 4.5 year post-JOBS period in our sample to the usage in the 2 year post- JOBS period in Table 8 of Dambra et al. (2015), we can see that the uptake of all JOBS Act provisions has increased with the time since the Act’s passage. It is reasonable to expect that a firm that acquires more information should make more informed decisions. We therefore start by examining IPO outcomes that could be ameliorated for a firm that engages in testing the waters with potential investors, such as likelihood of withdrawal, time in the IPO process, and accuracy of the marketing range.
[INSERT TABLE 2.5 HERE]
Table 2.5 shows the results of the difference-in-difference regression. The first two columns show that likelihood of withdrawal is statistically significantly lower for EGCs in the post-JOBS period relative to either U.S. non-EGCs (column (1)) or U.K. EGCs (column (2)). The reduction in withdrawal of approx- imately 11 percentage points is economically significant as well, given the sample average withdrawal rate for U.S. EGCs is 30%. Withdrawn IPOs can be seen as a proxy for market breakdown, where sellers and buyers cannot come to an agreement on a satisfactory deal. Increased information asymmetry is associated with a higher likelihood of market breakdown (Akerlof, 1970), but we note that the ability to test the waters should move the breakdown out of the view of the econometrician; in particular, EGCs
12This small sample size prevents us from making reasonable statistical inferences using a regression discontinuity design around the $1 billion sales cutoff. 13Chaplinsky et al. (2017) show the results of the density test and reject the null hypothesis that the density is continuous at the $75 million deal size threshold in the pre-JOBS period. Chapter 2. Increasing firms’ information advantages in the IPO process 22 can test the waters and learn that a deal will not be possible before starting the public IPO process. As a result, the firms that begin the IPO process after testing the waters should do so with more conviction in their potential success. The remaining columns in Table 2.5 examine other IPO outcomes using U.S. only non-EGCs as the control group due to data quality. Column (4) shows that the time between launch and pricing (the bookbuilding period) is statistically significantly shorter by about 12 days relative to a sample average of 22 days. A shorter bookbuilding period reduces uncertainty for issuers and investors alike. Finally, it would be reasonable to expect that once firms are able to test the waters, they are able to provide a more “accurate” initial price range. Column (6) shows that this appears to be true on the downside; EGCs in the post-JOBS period are significantly less likely to revise their price range down. However, column (5) shows that they are significantly more likely to adjust their price range up. Columns (7) and (8) show that post-JOBS EGCs are not significantly more or less likely to price outside their original price range, which is in contrast with the speculation in Jenkinson et al. (2006) that pilot fishing is the reason that European IPOs rarely price outside of the range. Finally, in untabulated results, we examine the partial adjustment phenomenon and confirm that EGC IPOs priced above (below) the marketing range in the post-JOBS period have high (low) underpricing.
Implication 2: The ability to test the waters will increase underpricing. In our theoretical framework, we discuss two potential ways that firms may respond to their increased information advan- tage endowed by the provisions of the JOBS Act. We will now examine whether firms engage in these activities empirically. We first predict that the option to acquire more information about investors’ valuations (the option to TTW) will lead firms to learn their type and high-type firms will use increased underpricing as a separating mechanism. Empirically, this would manifest as our treatment group experiencing increased underpricing in the post-JOBS period relative to the control group.
[INSERT TABLE 2.6 HERE]
Table 2.6 shows the results of the difference-in-difference regression of underpricing on treatment status and pre- and post-JOBS. The first three columns use U.S. non-EGCs as the control group, while the second three columns use U.K. EGCs. In all specifications, we control for industry using three-digit SIC code fixed effects and double cluster standard errors at the year and industry level. Columns (1) and (4) are the baseline regressions which include the time trend, treatment variable and interaction. The interaction term of EGC and Post-JOBS is statistically significantly positive and between 9.3% and 9.7%. This is also economically significant, as the sample average underpricing for U.S. EGCs is 17.7%. The other columns include controls that have been shown to be related to underpricing, including a measure of firm size (the natural logarithm of total assets), a measure of market performance (the performance of the S&P 500 or FTSE 100 Index from public filing to two months after public filing, consistent with Bernstein (2015)), firm age, a dummy for VC backed status, the number of bookrunners included in the syndicate, and the percentage of the IPO amount from secondary rather than primary sources. The U.S. only sample also includes the Carter-Manaster rank of the top left bookrunner and the IPO firm’s debt-to-assets ratio. The interaction term of EGC and Post-JOBS remains statistically significantly positive across all specifications, between 6.7% and 9.7%. We discuss in our theoretical section that it is the option to test the waters that drives our results, and not whether a firm actually tests the waters. This is because investors should assume that all eligible Chapter 2. Increasing firms’ information advantages in the IPO process 23
firms take advantage of this option and thus require additional underpricing from all firms. We test this implication by examining if underpricing differs for EGCs based on whether they tested the waters or not.
[INSERT TABLE 2.7 HERE]
Table 2.7 shows the results of this analysis. While firms that engage in written TTW activities do experience lower underpricing than those that do not, the difference is not statistically significant. While this is an imperfect proxy for TTW activities, it is supportive evidence that it is the ability to test the waters that matters.
Implication 3: The ability to test the waters will increase the use of certification. The second way firms may respond after learning their type is for high type firms to employ external certification, for example through their underwriter(s). There is no perfect measure of the use of underwriter certification. In designing proxies for this certification, we consider that it is the bookrunners of a deal that do the majority of the work in marketing and certifying a firm, with a special emphasis on the top left bookrunner. We believe firms could increase their use of underwriter certification in three different ways: (1) hiring more bookrunners, (2) paying their bookrunners more, or (3) hiring more reputable bookrunners.
[INSERT TABLE 2.8 HERE]
Table 2.8 shows the results of these difference-in-difference regressions. Columns (1) and (2) look at the number of bookrunners hired using either U.S. non-EGCs or U.K. EGCs as the control group. Table 2.3 show that the number of bookrunners has increased for all groups in the post-JOBS period compared to the pre-JOBS period, but these results show that the relative increase for U.S. EGCs depends on the control group used. We next look at the amount paid to bookrunners, which may indicate a firm’s willingness to incen- tivize better certification. Column (3) shows that underwriting fees are not statistically different for EGCs in the post-JOBS period, but columns (4) and (5) show the percentage of shares allocated (and consequently, percentage of deal fees) to bookrunners has significantly increased. As shown in Table 2.3, this is primarily driven by the decrease in the percentage allocated by U.S. non-EGCs, as there is little change for U.S. EGCs. Finally, we look at the quality of underwriters hired by U.S. EGCs relative to U.S. non-EGCs. Column (6) shows that the Carter-Manaster rank of the top left bookrunner for EGCs is signficantly lower, and column (7) the average rank for all bookrunners is non-significantly lower. Combined, the results for underwriter certification are mixed at best; it could be argued that U.S. EGCs are willing to pay their bookrunners relatively more in the post-JOBS period to overcome increased information asymmetry, but they don’t seem to hire more reputable underwriters which would bolster the certification effects. Though we specifically considers underwriters as the certifier in our theoretical framework, we also ex- amined three other potential signals of quality: auditor certification, legal certification and entrepreneur commitment to the post-IPO firm.
[INSERT TABLE 2.9 HERE] Chapter 2. Increasing firms’ information advantages in the IPO process 24
In columns (1) and (2), we show that EGCs in the post-JOBS period are not more likely to hire more reputable auditors (as proxied by a dummy for engaging Deloitte, E&Y, KPMG, or PwC), or pay higher fees to their accountants as a percentage of deal size. Similarly, in columns (3) and (4), EGCs in the post-JOBS period are not more likely to hire more reputable lawyers (as proxied by a dummy for engaging one of the top 5 law fims in terms of number of IPOs), or pay higher legal fees. Finally, we consider two signals of seller commitment to the post-IPO firm. Column (5) shows that there is no significant difference in post-IPO lockup periods, which is the amount of time after the initial public offering that company insiders are prohibited from selling shares. Column (6) shows that there is no significant difference in the percentage of the company sold, which is inversely related to the entrepreneur’s willingness to invest in their own project (Leland and Pyle, 1977). Overall, unlike a clear increase in underpricing described above, it does not look like EGCs in the post-JOBS period are resorting to certification to generate separating equilibria.
Implication 4: The ability to test the waters will lower the propensity to be a public company. Our theoretical framework posits that low quality firms will opt out of the IPO process after testing the waters and learning their type. Since the option to TTW is only available for EGCs in the post-JOBS period, we expect that the propensity for these firms to be public will decrease relative to non-EGCs. As a first test of this hypothesis, we follow Doidge et al. (2017) and calculate the listing propensity in the U.S. economy by comparing the number of public firms to the total number of firms. Using annual listing propensities at the three-digit NAICS level between 2007 and 2017, we use a similar difference- in-difference design with modified details. In order to align with our IPO analysis, we exclude pipelines (NAICS codes beginning with 486), many of which are structured as limited partnerships, and financial firms (NAICS codes beginning with 52 and 53). The pre-JOBS period is 2007 to 2011, while the post- JOBS period is 2013 to 2017. For public firms, EGC status is based on annual revenue and uses $1 billion as the cutoff; for the full sample of U.S. firms, the EGC pool includes firms with between 20 and 500 employees while the non-EGC pool includes firms with more than 500 employees.14
[INSERT TABLE 2.10 HERE]
Table 2.10 shows the results of this analysis. Column (1) includes all industries while column (2) includes only industries that are ‘active’ in terms of public companies (industries that have at least one public company classified as an EGC and non-EGC in each of the pre- and post-JOBS periods). EGCs in the post-JOBS period have listing propensities that are between 0.6 and 0.7 percentage points lower. This is economically large, given the average listing propensity for EGCs is only 1.9%. While this result is supportive of our theory, a more suitable empirical test would be to examine all firms that want to raise external capital and observe whether or not they pursue an IPO; comparing the IPO decisions of EGCs vs non-EGCs in the pre- and post-JOBS Act periods would allow us to determine if the JOBS Act really encouraged small firms to go public. We attempt this by looking at the universe of private firms that engaged in a public equity raise (IPO), private equity raise (PE), or completed merger transaction (M&A) over our sample period. We believe this sample more closely approximates possible IPO candidates, so examining their choice will be a better test of our implication. 14Unfortunately, the SUSB data does not include firm-level revenue, so we have to classify firms in the broader economy according to number of employees. Cutoffs deteremined using the full sample of U.S. public firms in Compustat: Of firms with less than $1 billion in sales, the median number of employees is 180 (and the 75th percentile is 890). While there are likely some firms in the 500+ employee category that would be EGCs, we are limited by the SUSB categorization. Chapter 2. Increasing firms’ information advantages in the IPO process 25
To do this, we gather from Capital IQ the full list of PE and M&A transactions by private companies between 2007 and 2017. A subset of these transactions include pre-transaction revenue for the target; we predict revenue for the full sample by regressing revenue on transaction value (and round type, e.g. Series A, Venture, Growth, etc., for PE transactions) and assigning EGC status if predicted revenue is less than $1 billion.15 Our dependent variables include a private firm’s IPO propensity, calculated as the number of IPOs to total number of transactions in a industry-year, and an IPO dummy that takes on a value of 1 if the transaction is an IPO and 0 otherwise.
[INSERT TABLE 2.11 HERE]
Table 2.11 shows the results of this analysis. In almost all specifications the coefficient on the interaction term is negative, and significantly so if we focus on industries with positive IPO activity across our treatment and control samples in the pre- and post-JOBS periods. In other words, the likelihood of a U.S. EGC to choose IPO over PE or M&A decreased in the post-JOBS period relative to either control sample. For a subset of firms in the U.S. PE/M&A dataset that we can link to the IPO dataset via SEC’s Central Index Key, we also examine PE-funded firms who subsequently engaged in an IPO or M&A exit within the sample period; for these transactions, we define an IPO exit dummy.
[INSERT TABLE 2.12 HERE]
Table 2.12 shows that PE exits of EGCs have become significantly less likely to be via IPO in the post-JOBS period, especially when focusing on firms with PE investments pre-2012.16 Consistent with our analysis of the listing propensity in the entire U.S. economy, we believe these results point to a reduced desire for private firms to go public in the post-JOBS period. While initial evidence pointed to an increase in the raw number of initial public offerings after the JOBS Act (Dambra et al., 2015), this may have been driven by market conditions that also drove the increase in total number of firms and the number of private company transactions in the economy. While the JOBS Act was designed to encourage small private firms to go public, it does not appear that it succeeded in that objective.
Implication 5: The ability to test the waters will increase public firm quality. If low quality firms self-select out of the IPO process, then the firms that do go public after testing the waters is permitted should be higher quality on average. In order to test this, we use the same difference-in- difference design as in Table 2.6.
[INSERT TABLE 2.13 HERE]
Firm quality is extremely difficult to measure, but we attempt to examine it from three perspectives. We look at trading outcomes, financial statement ratios, and proxies for financial misconduct.17 As shown in Table 2.13, the interaction term is not statistically significant for the majority of the variables,
15For the subset of deals with revenue available, the prediction is correct as follows: 95.6% for U.S. M&A, 87.0% U.S. PE, 97.7% for U.K. M&A, and 90.5% for U.K. PE. We correct EGC status for those misclassified. 16We think it makes sense to exclude firms who received their first PE investment post-2012, since they are already choosing private funding in the post-JOBS era. 17As discussed in Karpoff et al. (2017), the result of empirical tests using these proxies can depend on the source of data; for this reason, we use three different sources. Chapter 2. Increasing firms’ information advantages in the IPO process 26 other than post-JOBS EGCs being more likely to be trading one year after IPO, having higher growth in sales and higher return on assets. The coefficient for other dependent variables are not statistically significant, but generally move in the correct direction: post-JOBS EGCs are less likely to go bankrupt, have higher profitability and Tobin’s Q, and are less likely to be subject to an SEC Accounting and Auditing Enforcement Release or restate their financials. A better test of our theory would look at the firm quality of firms that could have gone public compared to those that do go public in the pre- vs. post-JOBS periods. This requires measures of quality of a sample of private firms, which is very difficult without proprietary data (see Chemmanur et al. (2020) for an example using U.S. Census data). However, we believe these are positive indicators of quality in the EGC sample post-JOBS.
Reduced disclosure as an alternative framework. We discuss above an alternative theoretical framework related to the reduced disclosure provisions in the JOBS Act. Firms reducing the amount of information provided to investors will also increase information asymmetry, which could result in increased underpricing. The fact that listing propensity, IPO propensity and IPO exits decreased in the post-JOBS period (and to a lesser extent, that U.S. EGC IPO firm quality has increased) is consistent with our theory and inconsistent with the alternative, but we also want to attribute our underpricing results to testing the waters rather than reduced disclosure. In order to do this, we are going to look at two alternative angles. First, to the extent that the costs of disclosure are fixed, reduced disclosure costs will disproportionately benefit smaller firms, so we are going to repeat our underpricing regressions for large firms only (firms with at least $100 million in revenue in the year before IPO). Second, we are going to look at the relationship between EGCs’ reduced disclosure in the post-JOBS period and their subsequent underpricing. Table 2.14 shows the results of this analysis.
[INSERT TABLE 2.14 HERE]
Panel A shows that the interaction terms in our difference-in-difference regression is still positive (though not statistically significant when we use U.K. EGCs as the control sample), indicating that the underpricing results are not driven by smaller firms, for whom we would expect the benefits of reduced disclosure to be larger. In Panel B, we examine the actual use of the reduced disclosure provisions by EGCs in their IPO prospectus and its impact on underpricing. We develop a score for EGCs’ use of the IPO disclosure exemptions and of ongoing disclosure requirements (which we call deburdening exemptions in the table). A higher score means that the firm used more of the exemptions and provided investors with less information. There is no statistically significant result between the reduced disclosure scores and the level of underpricing. Since potential investors know about this reduced disclosure before orders are submitted and the price is determined, then we can interpret the lack of significance in this regression as reduced disclosure not driving our underpricing results. The lack of a significant relationship between disclosure and underpricing is consistent with Chaplinsky et al. (2017), although they warn of the endogeneity in firms’ chosen disclosure levels. Chapter 2. Increasing firms’ information advantages in the IPO process 27
2.5 Concluding Remarks
The issue of information asymmetry and its negative impacts on the ability for trade have long been understood. In this paper, we examine how the quest to solve a very real and important problem (the decrease in the number of companies raising capital publicly) seems to have ignored this primitive issue of information asymmetry, and led to unintended consequences. In particular, we examine the JOBS Act of 2012, which allowed firms to test the waters before deciding to pursue an IPO, as well as allowed them to reduce disclosure in their IPO filings. By focusing on reducing the direct costs of going public, U.S. regulators made changes that exacerbated the informational advantage that firms have over investors. We show that eligible firms take advantage of these provisions and incur benefits, such as lower IPO withdrawal rates and shorter bookbuilding periods. However, as postulated in our theoretical framework, firms have had to compensate investors for the increased information asymmetry, which it appears they have done by signaling their type through underpricing. More surprisingly, we show that the testing the waters provision of the JOBS Act should be and is associated with a decrease in the number of firms going public, which goes against the regulators’ stated goals of increasing the number of IPOs (though can be welfare-enhancing). Our paper highlights the importance of analyzing regulatory policy changes from multiple perspec- tives; in particular, changes to the IPO process must consider responses from both firms and prospective investors. Chapter 2. Increasing firms’ information advantages in the IPO process 28
2.6 Tables
Table 2.3: Summary statistics.
Pre-JOBS Post-JOBS US EGC US Non-EGC UK EGC US EGC US Non-EGC UK EGC Number of filed IPOs 503 84 137 579 95 196 Withdrawal % 43.54% 44.05% 21.17% 17.44% 26.32% 9.69% Number of launched IPOs 358 53 523 73 Outcomes Days b/w filing & launch 156.3 226.7 44.2 122.6 Adjust price range up 4.47% 0.00% 8.80% 1.37% Adjust price range down 15.64% 1.89% 10.52% 4.11% Number of completed IPOs 284 47 93 470 67 176 Controls Deal size $147.4 $922.6 $106.7 $141.1 $584.1 $125.9 Secondary percent 18.55% 26.67% 16.47% 5.89% 19.89% 36.28% Total revenue $154.9 $5,934.7 $131.5 $133.8 $3,430.2 $139.8 Total assets (ln) 4.56 8.00 3.05 4.15 7.95 3.49 Firm age 15.14 50.19 9.04 12.49 50.87 10.39 VC backed 39.79% 6.38% 10.75% 54.68% 2.99% 7.95% Index perf two mths post-filing -0.29% 1.08% -0.30% 2.14% 2.55% 0.67% Debt to assets 30.91% 61.75% 34.85% 58.21% Outcomes Days b/w launch & pricing 27.2 13.3 18.0 14.1 Priced above orig. range 25.00% 14.89% 23.83% 14.93% Priced below orig. range 39.08% 36.17% 33.62% 35.82% Firm responses Underpricing 14.40% 6.28% 8.44% 19.75% 11.45% 10.11% Number of bookrunners 2.18 4.28 1.30 2.91 6.31 1.46 Underwriting spread 6.83% 5.89% 6.80% 5.81% Left bookrunner allocation 42.48% 27.11% 41.64% 24.11% All bookrunner allocation 63.01% 42.01% 62.24% 33.06% Left bookrunner rank 8.15 8.87 7.80 8.71 Avg bookrunner rank 7.94 8.71 7.40 8.37 Firm quality Mkt adj 2 yr ret -8.35% 10.19% 0.96% 11.72% Match adj 2 yr ret -9.84% 0.79% 15.76% 14.04% Trade 1 yr 97.87% 100% 99.36% 98.51% Trade 2019 43.26% 57.45% 73.66% 76.12% Bankrupt 10.99% 6.38% 3.64% 1.49% Sales growth 137.37% 101.89% 174.80% 103.01% ROA 0.12 0.27 2.58 0.99 ROE -13.93 4.70 -18.41 0.92 Profitability -2.74 0.19 3.08 0.58 Tobin’s Q 1.95 0.96 11.89 0.94 AAER 0.70% 2.13% 0% 1.49% Restate 2.11% 0% 0.85% 1.49% Class action 22.18% 17.02% 14.68% 14.93% Number of transactions (IPO, PE, M&A) 10,429 271 2,630 12,488 384 2,476 Listing propensity IPO dummy 2.72% 17.34% 3.54% 3.76% 17.45% 7.07% IPO propensity 2.68% 18.30% 4.32% 2.71% 19.11% 5.67% Chapter 2. Increasing firms’ information advantages in the IPO process 29
Table 2.4: Use of JOBS Act provisions by U.S. EGCs filing for IPO between July 1, 2012 and December 31, 2017.
All IPOs Completed IPOs De-risking provisions Confidential filing 91.9% 93.8% Testing the waters 91.9% 92.6% Written testing the waters 26.3% 28.3% De-burdening provisions Reduced statement disclosure 67.7% 66.6% Reduced compensation disclosure 75.0% 73.0% Auditor attestation opt-out 50.4% 50.2% Future accounting standards opt-out 14.5% 14.7% PCAOB rulings opt-out 24.2% 25.5% Executive compensation vote opt-out 46.6% 46.4% Chapter 2. Increasing firms’ information advantages in the IPO process 30 Days b/w filing Days b/w launch Adjust price Adjust price Priced above Priced below Table 2.5: Difference-in-difference regression: IPO outcomes. (0.0466) (0.0503) (1) (2) (3) (4) (5) (6) (7) (8) (0.0128) (0.00861)(0.0375) (2.712)(0.0433) (0.0345)(0.0299) (0.0384) (1.292) (8.497) (34.53) (0.00621) (3.526) (0.00654) (10.50) (10.56) (0.00909) (0.0259) (0.0581) (0.0175) (6.854) (0.0320) (0.0137) (0.0402) (0.0260) (0.0909) (0.0213) (0.0257) (0.0717) (0.0379) (0.0479) (0.0619)(0.0503) (0.0181) (26.97) (4.847) (36.62) (0.0133) (8.157) (0.0465) (0.0393) (0.0804) (0.0644) (0.100) (0.0859) (0.0862) (0.0579) (35.79) (4.945) (0.0271) (0.0530) (0.0850) (0.0896) (0.000576) (0.000764) (0.313) (0.200) (0.000428) (0.000810) (0.000862) (0.000918) Withdrawn Withdrawn & launch & pricing range up range down orig. range orig. range mths post-filing (0.200) (0.192) (108.1) (18.99) (0.127) (0.256) (0.288) (0.330) Index perf twoFirm ageSecondary -0.218 percentDebt to assets -0.00222*** -0.106** -0.130Observations -0.00160*R-squaredSample -0.0423 0.0579*FE -319.6** -0.0365Cluster 933 -22.61 -39.19* 0.174 US 0.0603 EGC/non US/UK EGC 1,078 Yr&SIC -12.41 US 0.0507 EGC/non 0.183 -0.000621 SIC3 18.20 US EGC/non Yr&SIC 0.00137 -0.0822 928 0.0213 US EGC/non 0.396 SIC3 US Yr&SIC EGC/non 5.855 -0.00120 -0.0895*** US EGC/non 0.437 US EGC/non SIC3 814 0.000819 0.088 Yr&SIC 0.122 -0.0366 -0.203 Yr&SIC -0.0923 0.0297 SIC3 0.121 933 Yr&SIC -0.0789* 0.123 SIC3 933 Yr&SIC 0.0811 0.210 Yr&SIC SIC3 933 0.173 SIC3 933 SIC3 Total assets (ln) -0.0196VC-backed -0.0118 -0.0988** 5.862* -0.0954** -2.533* -3.973 -0.00280 -13.61*** -0.0142* 0.0349 0.0161 -0.0257 0.0188 0.0885* 0.0226 Post JOBS x US -0.107* US 0.209*** Post JOBSEGCPost JOBS x EGC -0.00180 -0.112* -0.00363 0.0275 -115.2*** 12.83 2.278 -77.02* -11.67** -0.0301** 0.0804** 0.0513 12.37 -0.121** 0.0103 -0.0311 0.0427 0.0251 -0.0548 0.152** 0.0969 0.126 This table presents thecolumn difference-in-difference (2) and regression U.S. comparingclustered non-EGCs IPO at in outcomes the all year for other and the columns) SIC in treatment level. the group pre- (U.S. and EGCs) post-JOBS to periods. the Three-digit control SIC groups code (U.K. fixed EGCs effects in are included and standard errors are Chapter 2. Increasing firms’ information advantages in the IPO process 31
Table 2.6: Difference-in-difference regression: IPO underpricing.
This table presents the difference-in-difference regression comparing IPO underpricing for the treatment group (U.S. EGCs) to the control groups (U.S. non-EGCs in columns (1)-(3) and U.K. EGCs in columns (4)-(6)) in the pre- and post-JOBS periods. Three-digit SIC code fixed effects are included and standard errors are clustered at the year and SIC level. (1) (2) (3) (4) (5) (6) Underpricing Underpricing Underpricing Underpricing Underpricing Underpricing
Post JOBS -1.599 -1.567 -1.470 -1.957 -1.595 -1.032 (3.726) (3.744) (3.874) (2.686) (2.871) (2.649) EGC 2.864 0.658 -5.282* (2.858) (2.801) (2.897) Post JOBS x EGC 9.280* 9.041* 8.796** (4.214) (4.231) (3.499) US 3.645 4.929 3.449 (3.033) (4.475) (4.123) Post JOBS x US 9.655*** 9.004*** 6.722*** (1.072) (1.672) (2.042) Total assets (ln) -0.839 -1.752 -0.594 -0.825 (1.337) (1.465) (0.938) (0.902) Index perf two 11.20 9.937 11.31 10.67 mths post-filing (18.98) (17.68) (14.85) (15.29) Firm age -0.105 -0.0617 (0.0809) (0.0828) VC-backed 5.171 5.645** (3.568) (2.502) Number of bookrunners 0.252 1.078 (1.106) (1.091) Secondary percent 2.847 -1.619 (5.096) (3.840) Left bookrunner rank 1.226 (0.692) Debt to assets -4.725* (2.607)
Observations 815 815 814 955 952 949 R-squared 0.162 0.164 0.187 0.164 0.165 0.175 Sample US EGC/non US EGC/non US EGC/non US/UK EGC US/UK EGC US/UK EGC FE SIC3 SIC3 SIC3 SIC3 SIC3 SIC3 Cluster Yr&SIC Yr&SIC Yr&SIC Yr&SIC Yr&SIC Yr&SIC Chapter 2. Increasing firms’ information advantages in the IPO process 32
Table 2.7: IPO underpricing and testing the waters activity.
This table presents regressions of IPO underpricing on dummies for testing the waters and written testing the waters by EGCs in the post-JOBS period. Three-digit SIC code fixed effects are included and standard errors are clustered at the year and SIC level. (1) (2) (3) (4) (5) (6) Underpricing Underpricing Underpricing Underpricing Underpricing Underpricing
Testing the waters -6.023 -6.505 -6.935 (6.911) (6.567) (7.181) Written testing the waters -3.786 -3.842 -4.413 (3.678) (3.726) (4.156) Total assets (ln) 0.709 -0.237 0.538 -0.433 (1.581) (2.501) (1.763) (2.187) Index perf two 34.41 40.20 34.16 39.96 mths post-filing (48.68) (53.33) (47.71) (52.28) Firm age -0.148 -0.165 (0.162) (0.0937) VC-backed 6.772 6.539 (5.862) (5.613) Number of bookrunners 1.551 1.502 (1.932) (1.937) Left bookrunner rank -0.299 -0.266 (1.211) (1.093) Debt to assets -4.928 -5.154 (3.745) (3.779) Secondary percent 3.820 4.949 (8.890) (8.502)
Observations 464 464 462 464 464 462 R-squared 0.120 0.122 0.137 0.119 0.121 0.137 Sample US EGC US EGC US EGC US EGC US EGC US EGC FE SIC3 SIC3 SIC3 SIC3 SIC3 SIC3 Cluster Yr&SIC Yr&SIC Yr&SIC Yr&SIC Yr&SIC Yr&SIC Chapter 2. Increasing firms’ information advantages in the IPO process 33 (0.203) (0.301) Table 2.8: Difference-in-difference regression: Underwriter certification. (1) (2) (3) (4) (5) (6) (7) (0.287)(0.272) (0.0893) (0.00220) (0.0204) (0.00245) (0.0175) (0.0236) (0.0301) (0.121) (0.0927) (0.150) (0.186) (0.337) (0.365) (0.00177) (0.0149) (0.0221) (0.163) (0.140) (0.291) (0.00215) (0.0188) (0.0401) (0.135) (0.149) Number of Number of Underwriting Percent allocated Percent allocated Left Average bookrunners bookrunners spread to left bookrunner to all bookrunners bookrunner rank bookrunner rank Post JOBSEGCPost JOBS x EGC 2.503*** -1.549*** 0.225** -1.427*** -0.00314 0.00339 -0.0452* 0.00686** 0.0360* -0.116*** 0.125*** 0.0935** 0.0971 0.170*** -0.354** -0.315* -0.482*** -0.113 -0.670*** US 1.018*** Post JOBS x USDeal sizeObservationsR-squaredSampleFECluster 0.611* 0.618* 815 0.537 2.440*** US EGC/non US/UK EGC -0.00554** 955 Yr&SIC US EGC/non 0.469 SIC3 US EGC/non Yr&SIC -0.0228 815 0.376 SIC3 US EGC/non Yr&SIC -0.0331 US EGC/non SIC3 0.295 Yr&SIC 813 US EGC/non 0.196 SIC3 Yr&SIC 0.446 812 0.169 Yr&SIC SIC3 0.181 815 Yr&SIC SIC3 0.196 815 SIC3 This table presents thecolumn difference-in-difference (2) regressions and comparing U.S.effects IPO non-EGCs are outcomes in included all for and other the standard columns) treatment errors in group are the (U.S. clustered pre- EGCs) at and the to post-JOBS year the periods. and control SIC groups The level. (U.K. sample EGCs includes completed in IPOs only. Three-digit SIC code fixed Chapter 2. Increasing firms’ information advantages in the IPO process 34 (1) (2) (3) (4) (5) (6) (0.0751)(0.0691) (0.00263) (0.00272)(0.0274) (0.143) (0.0982) (0.00408) (0.00313) (2.400) (2.525) (0.0249) (0.0348) (0.0301) (0.468) (0.00570) (0.0904) (0.00214) (0.131) (0.00258) (2.806) (0.0311) Table 2.9: Difference-in-difference regression: Other certification. Big 4 auditor Accounting fees Big 5 lawyer Legal fees Lockup Percent of Post JOBSEGCPost JOBS x EGCDeal size -0.0288 0.0158ObservationsR-squared -0.170** 0.00199 0.00161SampleFECluster 0.0402 0.00533* 0.0536 0.0712 815 0.161 0.0144 US 0.00456 0.000690 EGC/non US EGC/non 815 Yr&SIC 0.00650* 2.002 -3.641 US EGC/non 0.188 SIC3 US EGC/non US -4.993* -0.0371 Yr&SIC EGC/non 0.0366 -0.0523 US EGC/non 815 0.214 SIC3 Yr&SIC 0.0230 0.202 815 Yr&SIC SIC3 -7.288*** Yr&SIC 0.213 SIC3 790 0.00417 Yr&SIC 0.242 SIC3 815 SIC3 VARIABLES dummy to proceeds dummy to proceeds period firm sold This table presents thein difference-in-difference the regressions pre- comparing and IPOat post-JOBS outcomes the periods. for year and the The SIC treatment sample group level. includes (U.S. completed EGCs) IPOs to only. the Three-digit control SIC group code (U.S. fixed non-EGCs) effects are included and standard errors are clustered Chapter 2. Increasing firms’ information advantages in the IPO process 35
Table 2.10: Difference-in-difference regression: Listing propensity.
This table presents difference-in-difference regression comparing listing propensity for EGCs to non-EGCs in the pre- and post-JOBS periods. Listing propensity calculated at the three-digit NAICS level based on number of public firms in Compustat divided by total number of firms in the U.S. EGCs include firms with less than $1 billion in annual revenue and between 20 and 500 employees. Non-EGCs include firms with more than $1 billion in annual revenue and more than 500 employees. The first column includes all industries while the second column includes only industries with at least one public EGC and non-EGC in each of the pre- and post-JOBS periods. Three-digit NAICS code fixed effects are included and standard errors are clustered at the year and NAICS3 level.
(1) (2) VARIABLES Listing propensity Listing propensity
Post JOBS 0.305 0.357 (0.210) (0.258) EGC (estimate) -3.652*** -4.329*** (0.673) (0.773) Post JOBS x EGC (estimate) -0.601*** -0.704*** (0.125) (0.136)
Observations 1,542 1,300 R-squared 0.740 0.741 Sample All industries Active industries FE NAICS3 NAICS3 Cluster Yr&NAICS3 Yr&NAICS3 Chapter 2. Increasing firms’ information advantages in the IPO process 36 (0.0102) (0.00792) (0.0226) (0.0121) (0.0160) (0.0103) (0.0310) (0.0122) -0.0249** -0.0148* -0.0377 -0.0130 Table 2.11: Difference-in-difference regression: IPO propensity. (1) (2) (3) (4) (5) (6) (7) (8) IPO IPO IPO IPO IPO IPO IPO IPO (0.0332) (0.0356)(0.0161) (0.0533) (0.0171) (0.0449) (0.0357) (0.0155) (0.0281) (0.0124) (0.0217) (0.0150) (0.0310) (0.0321) (0.0340) (0.0383) industries industries industries industries industries industries industries industries VARIABLES propensity dummy propensity dummy propensity dummy propensity dummy Post JOBSEGC (estimate)Post JOBS x EGC (estimate)US -0.0303Post JOBS 0.00593 x US -0.191*** -0.148*** 0.0290 -0.0710* Observations -0.145***R-squared 0.00242Sample -0.108** 0.217*** -0.114*** FECluster 0.0837* 0.0274 2,808 0.0345** 0.0967*** 0.235 23,537 0.0575*** All 0.052 443 Yr&SIC3 SIC3 0.283 Yr&SIC All 13,443 Yr&SIC3 SIC3 0.043 Yr&SIC Active 3,733 Yr&SIC3 0.189 SIC3 Active 27,988 -0.0267 Yr&SIC -0.0266** Yr&SIC3 0.041 SIC3 All -0.0676* 634 -0.0440*** Yr&SIC SIC3 0.225 All 16,472 SIC3 0.035 Active SIC3 Active SIC3 This table presents theEGCs) to difference-in-difference the regression control comparingprivate groups companies the (U.S. who non-EGCs propensity undertook in forvariable a columns in private public (1)-(4) companies the equity and odd-numbered raise to U.K.takes (IPO), columns EGCs do on private is in an a equity columns the value IPO, raise (5)-(8))with percentage of (PE) comparing in at of 1 or the the least transactions if merger pre- treatment one in the orclustered and group control the acquisition transaction at post-JOBS and (U.S. industry-year (M&A) was the periods. treatment that between an year IPO The 2007 were and IPO. in sample and IPOs, SIC Columns each includes 2017. level. while (1)-(2) of all in The and the dependent even-numbered (5)-(6) pre- columns include and is all post-JOBS a industries, periods. dummy while that columns Three-digit (3)-(4) SIC and code (7)-(8) fixed include effects only are industries included and standard errors are Chapter 2. Increasing firms’ information advantages in the IPO process 37
Table 2.12: Difference-in-difference regression: PE exit as IPO propensity.
This table presents the difference-in-difference regression comparing the propensity for private companies to do an IPO, comparing U.S. EGCs to U.S. non-EGCs in the pre- and post-JOBS periods. The sample includes only PE-funded firms that resulted in an M&A or IPO exit within the sample time frame. The dependent variable is a dummy that takes on a value of 1 if the exit was an IPO. Column (1) includes all transactions, while column (2) includes only firms with an initial PE investment between 2007 and 2012. Three-digit SIC code fixed effects are included and standard errors are clustered at the year and SIC level. (1) (2) VARIABLES IPO exit dummy IPO exit dummy
Post JOBS 0.493*** 0.562** (0.127) (0.232) EGC (estimate) 0.0335 -0.0521 (0.174) (0.0807) Post JOBS x EGC (estimate) -0.320*** -0.457** (0.0815) (0.197) Time since initial PE inv -0.0379** -0.0286 (0.0149) (0.0172) Amount of initial PE inv 2.010** 1.376** (0.666) (0.499) Exit amount -0.272** -0.443*** (0.118) (0.130) Year of initial PE inv -0.0733** -0.0200 (0.0235) (0.0335)
Observations 582 381 R-squared 0.323 0.292 Sample US PE exits US PE exits of inv before 2012 FE SIC3 SIC3 Cluster Yr&SIC Yr&SIC Chapter 2. Increasing firms’ information advantages in the IPO process 38 Table 2.13: Difference-in-difference regression: Firm quality. Trading outcomes Financial statement ratios Financial misconduct (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (0.118)(0.112) (0.184) (0.179) (0.0232) (0.0166) (0.139) (0.114) (0.0470) (0.0535) (0.229) (0.0710) (1.857) (0.844) (12.83) (3.341) (3.425) (1.435) (23.61) (2.913) (0.0232) (0.0219) (0.0182) (0.0192) (0.0724) (0.0901) (0.175) (0.183) (0.0179) (0.104) (0.0503) (0.103) (1.329) (20.24) (5.598) (15.56) (0.0224) (0.0307) (0.0966) Mkt adj Match adj Trade Trade Sales Tobin’s Class (0.0348) (0.0385) (0.00660) (0.0147) (0.00568) (0.109) (0.697) (1.522) (0.899) (8.966) (0.00214) (0.00363) (0.00894) (0.00109) (0.00134) (0.000334) (0.00130) (0.000730) (0.00171) (0.0493) (0.121) (0.225) (0.112) (0.000219) (0.000282) (0.000761) (0.00374) (0.0153) (0.000367) (0.00297) (0.000975) (0.00655) (0.0555) (0.173) (0.155) (0.504) (0.00125) (0.000211) (0.00287) ObservationsR-squaredSampleFECluster 862 0.005 All 810 Yr&SIC 0.009 None Yr&SIC All 811 0.186 Yr&SIC None Yr&SIC 0.211 811 All SIC3 Yr&SIC 0.133 811 SIC3 All Yr&SIC SIC3 Yr&SIC 0.018 All Yr&SIC 695 0.006 Yr&SIC None 0.001 810 Yr&SIC All None 810 0.003 Yr&SIC All None Yr&SIC 0.010 810 All None Yr&SIC 0.293 776 None All 0.294 SIC3 All 814 0.130 SIC3 814 All SIC3 814 All All Firm age 0.000925 0.00189 5.41e-05 0.00144 0.000277 -0.00385** 0.0213 0.0282 0.207 -0.102 8.85e-05 -0.000179 -0.00193** RevenueTotal assets (ln) 0.0146 -0.00334 -0.00248 0.0348** -0.000723* 0.0105 0.00467 0.0174 -0.000748 -0.00906 0.0129* -0.0556 -0.225* -0.0162 1.067 -0.115 0.911 -0.525 -0.265 0.00839*** 9.805 -9.73e-05 0.00677** 0.000759 0.00341 0.00713 VARIABLESEGCPost JOBS 2 yrPost ret JOBS x EGC 2 yr 0.0935 ret 0.00694 -0.122 0.124 1 yr 0.135 0.0747 0.0364* -0.0309* 2019 -0.000970 0.146 0.00689 0.109 Bankrupt -0.0216 0.0124 -0.0493 growth 0.271** ROA -0.406 2.428* 0.0607 -0.211 ROE 3.815* 0.542 Profitability -14.73 5.993 -3.855 Q 15.76 2.583 0.337 27.51 -2.164 -0.0266 AAER -0.0212 Restate 0.0416 0.0243 0.00116 action 0.0102 0.00824 -0.0577 0.0630 This table presentsColumns difference-in-difference (1)-(5) regression examine comparingthree-digit trading firm SIC outcomes quality code based measures industrytrading on in for outcomes CRSP Compustat. U.S. and data. financial Columns EGCs misconduct (11)-(13) Columns to proxy examine (6)-(10) U.S. regressions. proxies examine non-EGCs Standard for financial errors financial in statement are misconduct. the clustered ratios Three-digit pre- at relative SIC the and to code year post-JOBS the fixed and periods. median effects SIC are level. firm included in in a the given Chapter 2. Increasing firms’ information advantages in the IPO process 39
Table 2.14: Alternative framework: Underpricing and reduced disclosure.
This table examines IPO underpricing in the context of the reduced disclosure afforded to U.S. EGCs in the post-JOBS period. Panel A presents difference-in-difference regression comparing underpric- ing for the treatment group (U.S. EGCs) to the control groups (U.S. non-EGCs in column (1) and U.K. EGCs in column (2)) in the pre- and post-JOBS periods. Panel A includes only larger firms with at least $100 million in revenue in the year before IPO. Panel B examines the impact of dis- closure/deburdening exemptions on IPO underpricing for EGCs in the post-JOBS period. Three- digit SIC code fixed effects are included. Standard errors are clustered at the year and SIC level. Panel A: Subsample (firms with $100m revenue+) Panel B: Disclosure and underpricing (1) (2) (1) (2) VARIABLES Underpricing Underpricing VARIABLES Underpricing Underpricing
Post JOBS -1.665 -0.0306 Use of disclosure 0.731 (4.108) (3.910) exemptions (3.024) EGC -7.420 Use of deburdening -0.643 (5.988) exemptions (1.644) Post JOBS x EGC 8.382** (3.661) US 9.221* (4.440) Post JOBS x US 4.623 (6.602) Total assets (ln) -3.938 -2.414 Total assets (ln) -0.214 -0.232 (2.471) (1.689) (2.386) (2.301) Index perf two 17.27 28.24 Index perf two 9.382 10.78 mths post filing (24.10) (27.82) mths post filing (64.55) (65.24) Firm age -0.0660 -0.0256 Firm age -0.216* -0.226* (0.103) (0.152) (0.0848) (0.0997) VC-backed 5.199 4.690 VC-backed 6.809 6.732 (5.596) (4.012) (6.040) (5.892) Number of bookrunners 0.695 1.487 Number of bookrunners 0.367 0.442 (1.111) (1.490) (1.773) (1.768) Left bookrunner rank 2.176* Left bookrunner rank 0.0144 -0.0566 (0.985) (1.382) (1.401) Debt to assets -9.935** Debt to assets -5.486 -5.395 (3.623) (4.833) (4.812) Secondary percent 1.459 -0.378 Secondary percent 10.14 9.969 (5.965) (4.627) (9.744) (10.18)
Observations 336 339 Observations 435 435 R-squared 0.349 0.340 R-squared 0.147 0.147 Sample US EGC/non US EGC/non Sample US EGC US EGC FE SIC3 SIC3 FE SIC3 SIC3 Cluster Yr&SIC Yr&SIC Cluster Yr&SIC Yr&SIC Chapter 2. Increasing firms’ information advantages in the IPO process 40
2.7 Appendices
Appendix A Proofs
Proof of Proposition 1
See the main text.
Proof of Proposition 2
After testing the waters is permitted, there exists information asymmetry between the firm and the investors. We first focus on the pooling equilibrium and then discuss the separating equilibrium.
Pooling equilibrium Following Tirole (2010), we can focus on the financial contracts that give the investors a compensation R1 in the case of success and 0 in the case of failure. In the pooling equilibrium, investors must at least break even, i.e., R1p¯ ≥ I −A. Since the firm’s profits are decreasing in R1, setting the inequality to zero yields
I − A R = . (2.3) 1 p¯
Further, the high- and low-type firms must make non-negative profits:
pH (R − R1) ≥ A, (2.4)
pL(R − R1) ≥ A. (2.5)
If (2.5) holds then (2.4) must hold as well. Now with (2.3) and (2.5) we know that the pooling equilibrium exists only when
I 1 1 R ≥ + A − . (2.6) p¯ pL p¯
Separating equilibrium We first prove by contradiction that in the separating equilibrium it cannot be the case in which both types of firms are funded: the high-type (low-type) firm offers RH (RL) in the case of success and 0 in the case of failure. Suppose not. Then for the low type not to mimic the high type, we need pL(R − RH ) ≤ pL(R − RL) and similarly for the high type not to mimic the low type we need pH (R − RL) ≤ pH (R − RH ). Thus the two inequalities imply that RH = RL, which contradicts with the separating equilibrium. Therefore, in the separating equilibrium, it must be the case in which the high-type firm is funded whereas the low-type firm is not funded. Suppose that the high-type offers a stake RH in the case of success and 0 otherwise. To have this equilibrium sustained, we need the following program:
max(R − RH )pH − A RH
s.t. pH RH ≥ I − A (IR)
pL(R − RH ) ≤ A (IC) Chapter 2. Increasing firms’ information advantages in the IPO process 41
The high-type firm’s profit function is monotonically decreasing in RH . Thus, the lower RH , the higher the high-type firm’s profits. Together with the individual rationality constraint (IR) and incentive compatibility constraint (IC) we obtain the optimal RH as follows:
I − A A + RH = ,R − (2.7) pH pL I−A I 1 1 p if R < p + A p − p , = H H L H R − A if R > I + A 1 − 1 . pL pH pL pH
I−A Further, when RH = , the high type’s profits are pH (R − RH ) − A = pH R − I > 0 and there is no pH underpricing in this case. In other words, this separating equilibrium is distortionless and we refer to it A as “separating equilibrium (no distortion).” However, when RH = R − p , the high type’s profits are L pH pH (R − RH ) − A = A − 1 > 0 and there is underpricing. pL With (2.6) and (2.7), the equilibrium in the economy with and without testing the waters permitted can be summarized as follows:
I (1) Assume R ≤ p¯. In the benchmark economy without TTW, the market breaks down. In the economy with TTW, the separating equilibrium (no distortion) is the unique equilibrium. (2) Assume that I < R ≤ I + A 1 − 1 . In the benchmark economy without TTW, the pooling p¯ pH pL pH equilibrium is the unique equilibrium. In the economy with TTW, the separating equilibrium (no distortion) is the unique equilibrium. (3) Assume that I +A 1 − 1 < R ≤ I +A 1 − 1 . In the benchmark economy without TTW, pH pL pH p¯ pL p¯ the pooling equilibrium is the unique equilibrium. In the economy with TTW, the separating equilibrium is the unique equilibrium. (4) Assume that R > I + A 1 − 1 . In the benchmark economy without TTW, the pooling equi- p¯ pL p¯ librium is the unique equilibrium. In the economy with TTW, there coexist two equilibria: the separating equilibrium and the pooling equilibrium.
Table 2.1 summarizes the equilibrium. QED. Chapter 2. Increasing firms’ information advantages in the IPO process 42
Appendix B Description of Variables
Variables of interest Variable Definition Source Post JOBS Dummy variable that takes on a value of 0 for IPOs filed before De- SDC cember 1, 2011 and a value of 1 for IPOs file after June 30, 2012. For listing and IPO propensity, post takes on a value of 0 in years 2011 and before, and 1 in years 2013 and after. EGC Dummy variable that takes on a value of 1 for U.S. firms with revenue Bloomberg, SDC and of less than $1 billion in the last fiscal year before IPO and 0 otherwise. firm filings US Dummy variable that takes on a value of 1 for EGC IPOs in the SDC United States and 0 for EGC IPOs in the United Kingdom. Confidential filing Dummy that takes on a value of 1 if the firm filed its IPO prospectus EDGAR confidentially. Testing the waters Dummy that takes on a value of 1 if the firm indicates that they or Underwriting agree- their underwriters may have engaged in testing the waters activity. ments and correspon- dence with SEC Written testing the waters Dummy that takes on a value of 1 if the firm has provided specific Underwriting agree- details about testing the waters activities undertaken. ments and correspon- dence with SEC Use of disclosure exemp- Score that sums the use of reduced disclosure in the IPO prospec- IPO prospectuses tions tus. Score includes reduced financial statement disclosure (1 if less than three years of balance sheet and income statement provided) and executive compensation disclosure (1 if three or fewer executives discussed). Score ranges from 0 to 2. Use of deburdening exemp- Score that sums the use of reduced disclosure as well as other ongoing IPO prospectuses tions deburdening provisions. Score includes components from use of dis- closure exemption, extension of auditor attestation (1 if extended five years years), opt out of accounting standard changes, opt out of fu- ture PCAOB changes, and opt out of executive compensation voting. Score ranges from 0 to 6.
IPO outcomes Variable Definition Source Withdrawn Dummy that takes on a value of 1 if the IPO was not completed. Deals Bloomberg and SDC in registration for longer than two years are considered withdrawn. Days b/w filing & launch Number of days between the date of the first public filing and the Bloomberg and IPO launch of marketing, which is the date of the announcement of the prospectuses. initial price range. Days b/w launch & pricing Number of days between the date of the launch of marketing, which Bloomberg and IPO is the date of the announcement of the initial price range, and the prospectuses. pricing date of the IPO (also known as the bookbuilding period). Adjust price range up Dummy that takes on a value of 1 if the firm adjusted the marketing Bloomberg and IPO price range up at least once. prospectuses Adjust price range down Dummy that takes on a value of 1 if the firm adjusted the marketing Bloomberg and IPO price range down at least once. prospectuses Priced above original range Dummy that takes on a value of 1 if the final IPO price is above the Bloomberg and IPO high end of the initial marketing range. prospectuses Priced below original range Dummy that takes on a value of 1 if the final IPO price is below the Bloomberg and IPO low end of the initial marketing range. prospectuses
Firm responses to information asymmetry Variable Definition Source Chapter 2. Increasing firms’ information advantages in the IPO process 43
Underpricing Price at the end of the first day of trading divided by the IPO offer Bloomberg and SDC price minus 1. Number of bookrunners Number of underwriters identified as bookrunners. Bloomberg and SDC Underwriting spread Underwriting commission divided by offer price. Bloomberg and SDC Percent allocated to left Number of shares sold by left bookrunner divided by total number of SDC bookrunner shares sold. Percent allocated to all Number of shares sold by all bookrunners divided by total number of SDC bookrunners shares sold. Left bookrunner rank Underwriter rank of first listed bookrunner. Rankings taken from Jay Ritter database and Ritter’s website (Loughran and Ritter, 2004). SDC Average bookrunner rank Average underwriter rank of all bookrunners. Rankings taken from Ritter website and Jay Ritter’s website. SDC Big 4 auditor dummy Dummy that takes on a value of 1 if the firm employs one of the Bloomberg and SDC following auditing firms: Deloitte, Ernst & Young, KPMG, or Price- waterhouseCoopers. Accounting fees to pro- IPO fees paid to accountants divided by the final IPO proceeds. Bloomberg and SDC ceeds Big 5 lawyer dummy Dummy that takes on a value of 1 if the firm employs one of the Bloomberg and SDC following law firms: Cooley, Davis Polk & Wardell, Goodwin Procter, Latham & Watkins, or Wilson Sonsini Goodrich & Rosati. Legal fees to proceeds IPO fees paid to law firms divided by the final IPO proceeds. Bloomberg and SDC Lockup period Number of days for first lockup period. Bloomberg Percent of firm sold Final IPO proceeds divided by post-IPO market capitalization. Bloomberg and SDC
Post-IPO outcomes Variable Definition Source Listing propensity Number of public firms in a three-digit NAICS industry divided by Compustat (public total number of firms with at least 20 employees in the same three- firms) and Census digit NAICS industry. Ratios calculated separately for EGCs and non- Bureau’s Statistics for EGCs using cutoffs of $1 billion in annual revenue and 500 employees. U.S. Businesses (all firms) IPO dummy Dummy that takes on a value of 1 if the private company transaction Bloomberg and SDC is an initial public offering and 0 otherwise. (IPOs) and Capital IQ (PE and M&A) IPO propensity Number of IPOs in a three-digit SIC industry divided by the total Bloomberg and SDC number of private company transactions (including IPOs, private eq- (IPOs) and Capital IQ uity and mergers and acquisitions) in the same three-digit SIC in- (PE and M&A) dustry. Ratios calcualted separately for EGCs and non-EGCs using private company revenue if available or predicted revenue if not. IPO exit dummy Dummy that takes on a value of 1 if the private company transaction Bloomberg and SDC is an initial public offering and 0 otherwise. Includes only firms with (IPOs) and Capital IQ a previous private equity transaction. (PE and M&A) Market adjusted 2 year re- IPO firm’s two year buy-and-hold return less the same two year buy- CRSP turn and-hold return of the CRSP value-weighted index. Match adjusted 2 year re- IPO firm’s two year buy-and-hold return less the same two year buy- CRSP turn and-hold return of the closest matched firm in the same three-digit SIC industry (matching based on market capitalization and book-to- market). Trade 1 yr Dummy that takes on a value of 1 if the shares of the IPO firm are CRSP still trading one year after IPO. Trade 2019 Dummy that takes on a value of 1 if the shares of the IPO firm are CRSP still trading as of December 31, 2019. Chapter 2. Increasing firms’ information advantages in the IPO process 44
Bankrupt Dummy that takes on a 1 value of 1 if the shares have been delisted, CRSP the buy and hold return to delisting is -90% or lower, and the delisting price is less than $0.50 per share. Sales growth Firm’s growth in sales in year after IPO over previous year, divided by Compustat same ratio of median firm in same industry as measured by three-digit SIC code. ROA Firm’s net income in year after IPO to total assets at the end of year, Compustat divided by same ratio of median firm in same industry as measured by three-digit SIC code. ROE Firm’s net income in year after IPO to shareholders’ equity at the Compustat end of year, divided by same ratio of median firm in same industry as measured by three-digit SIC code. Profitability Firm’s net income before extraordinary items in year after IPO to Compustat total assets at the end of year, divided by same ratio of median firm in same industry as measured by three-digit SIC code. Tobin’s Q Firm’s market value of equity plus book value of debt divided by book Compustat value of assets. AAER Dummy that takes on a value of 1 if the firm has been subject to an SEC Accounting and Auditing Enforcement Release. Restate Dummy that takes on a value of 1 if the firm has restated their fi- Audit Analytics. nancial statements due to fraud (RES FRAUD) or for another reason that resulted in an SEC investigation (RES SEC INVEST). Class action Dummy that takes on a value of 1 if the firm has been subject to a Securities Class Action class action lawsuit that has not been dismissed (case status updated Clearinghouse July 1, 2020).
Control variables Variable Definition Source Total assets (ln) Natural logarithm of total assets at the end of the last fiscal year Bloomberg, SDC and before IPO. IPO prospectuses; Compustat Index performance two Return of the S&P 500 Index (for U.S. firms) or the FTSE 100 (for Bloomberg months post-filing U.K. firms) between the date of public filing of the IPO prospectus and two months after the date of public filing. Firm age Where applicable, IPO year less founding year taken the Field- Field-Ritter dataset Ritter dataset of company founding dates (Field and Karpoff (2002), and IPO prospectuses; Loughran and Ritter (2004)). Otherwise, IPO year minus founding SDC; U.K. Companies year taken from prospectus or incorporation year taken from U.K. House Companies house. VC-backed Dummy that takes on a value of 1 if the firm is backed by a venture Bloomberg and SDC capital company at the time of IPO. Secondary percent Percentage of shares offered in IPO from selling shareholders. Bloomberg and SDC Debt to assets Total debt at the end of the last fiscal before IPO divided by total Bloomberg and IPO assets at the end of the last fiscal year before IPO. Winsorized at the prospectuses 5% and 95% level. Deal size Final IPO price multiplied by number of shares offered. Bloomberg, SDC and IPO prospectuses Chapter 2. Increasing firms’ information advantages in the IPO process 45
Appendix C Sample TTW and Written TTW Language
Testing the waters is covered in Section 5(d) of the Securities Act of 1933, so it is often referred to as Section 5(d) communications. The following excerpt is from the underwriting agreement for the initial public offering of Blue Apron Holdings, Inc., filed on June 28, 2017, which indicates that the company and underwriters were permitted to engage in testing the waters activities with qualified institutional buyers:
6. (b) The Company represents and agrees that (i) it has not engaged in, or authorized any other person to engage in, any Section 5(d) Communications, other than Section 5(d) Communications with the prior consent of the Representatives with entities that are qualified institutional buyers as defined in Rule 144A under the Act or institutions that are accredited investors as defined in Rule 501(a) under the Act; and (ii) it has not distributed, or authorized any other person to distribute, any Section 5(d) Writings, other than those distributed with the prior consent of the Representatives that are listed on Schedule II(b) hereto; and the Company reconfirms that the Underwriters have been authorized to act on its behalf in engaging in Section 5(d) Communications; ... (d) Each Underwriter represents and agrees that (i) any Section 5(d) Communications un- dertaken by it were with entities that are qualified institutional buyers as defined in Rule 144A under the Act or institutions that are accredited investors as defined in Rule 501(a) under the Act and (ii) it will not distribute, or authorize any other person to distribute, any Section 5(d) Writing, other than those distributed with the prior consent of the Company;
In correspondence with the SEC dated June 6, 2017, Blue Apron Holdings, Inc. justified the initial public offering price range with the following bullet point (among others): “In recent “testing-the-waters” meetings, the Company received positive feedback from potential investors in this offering.” For this reason, Blue Apron Holdings, Inc. was coded as engaging in written testing the waters activity. The following is an excerpt from the underwriting agreement for the initial public offering of Warrior Met Coal LLC, filed on April 10, 2017, which states that neither the company nor its underwriters engaged in testing the waters activities:
SECTION 1. Representations and Warranties. (a) Representations and Warranties by the Company. The Company represents and warrants to each of the Underwriters as of the date hereof, the Applicable Time, the Closing Time (as defined below) and any Date of Delivery (as defined below), and agrees with each Underwriter, as follows: ... (xii) Testing-the-Waters Communication. The Company (a) has not alone engaged in any Testing-the-Waters Communication and (b) has not authorized anyone to engage in Testing- the-Waters Communications. The Company has not distributed any Written Testing-the- Waters Communication. Chapter 2. Increasing firms’ information advantages in the IPO process 46
Appendix D Further Examination of Parallel Trends Assumption
We examine pre- and post-treatment trends in the outcomes examined in our difference-and-difference regressions. For all but listing propensity, we plot the average residual by year and group, controlling for the natural logarithm of firm assets in the year before IPO and market performance in the two months after public filing. 5 250 .35 .26 .055 0 .05 200 .045 -5 150 .24 .04 -10 .3 100 .035 Launch_time residual Adjust_price_up residual Bookbuilding_time residual -15 50 .03 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 .22 Year of filing Year of filing Year of filing .2 Withdrawal residual .12 .25 .25 .1 .2 .19 .245 .18 .08 .24 .17 .06 .235 .2 .16 .18 .04
2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 .23 .15 .02 Adjust_price_down residual Price_below_range residual Year of filing Year of filing Price_above_range residual 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Year of filing Year of filing Year of filing U.S. EGC U.S. non-EGC U.S. EGC U.K. EGC U.S. non-EGC
(a) Withdrawal rates (b) Other IPO outcomes Panel (a) examines the trends in withdrawal rates for U.S. EGCs, U.S. non-EGCs and U.K. EGCs. Panel (b) examines the other IPO outcomes examined in 2.5, comparing U.S. EGCs to U.S. non-EGCs, including: days between filing and launch, days between launch and pricing, dummy for adjusting the price range up, dummy for adjusting the price range down, dummy for pricing above the original range and a dummy for pricing below the original range.
Figure A1: Parallel trends in IPO outcomes .45 .65 2.8 .068 4 .6 .4 .066 .55 .064 2.6 .35 .5 .062 3.5 .3 .45 .06 2.4 Allocation_to_allbr residual Allocation_to_leftbr residual .4 Underwriter_spread residual .25 .058 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017
3 Year of filing Year of filing Year of filing 2.2 9 9.5 Number_of_BR residual 2 9 2.5 8.5 8 8.5 1.8 8 2 7.5 All_br_rank residual 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Left_br_rank residual 7
Year of filing Year of filing 7.5 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Year of filing Year of filing U.S. EGC U.S. non-EGC U.S. EGC U.K. EGC U.S. non-EGC
(a) Number of bookrunners (b) Other measures of underwriter certification Panel (a) examines the trends in the number of bookrunners for U.S. EGCs, U.S. non-EGCs and U.K. EGCs. Panel (b) examines the other measures of underwriter certification examined in 2.8, comparing U.S. EGCs to U.S. non-EGCs, including: underwriter spread, percent allocated to the left bookrunner, percent allocated to all bookrunners, the rank of the left bookrunner, and the average rank of all bookrunners.
Figure A2: Parallel trends in measures of underwriter certification Chapter 2. Increasing firms’ information advantages in the IPO process 47 2 -5 3 .1 .06 1.005 .04 1 2.5 .05 1.5 .02 2 0 -10 .995 0 1 1.5 .99 -.05 Rel_ROA residual Rel_ROE residual -.02 Trading_1yr residual Mkt_adj_perf residual Match_adj_perf residual 1 Rel_SalesGrowth residual -15 .5 -.1 -.04 .985 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Year of filing Year of filing Year of filing Year of filing Year of filing Year of filing 2 .1 25 .64 20 .63 .08 1.5 15 .62 1 .06 10 .61 .5 5 .04 .6 Bankrupt residual Rel_TobinsQ residual Trading_2019 residual Rel_Profitability residual 0 0 .59 .02 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Year of filing Year of filing Year of filing Year of filing
U.S. EGC U.S. EGC U.S. non-EGC U.S. non-EGC
(a) Trading outcomes (b) Financial statement ratios .015 .018 .016 .01 .014 .012 .005 AAER residual .01 Restatement residual 0 .008 2007 2009 2011 2013 2015 2017 2007 2009 2011 2013 2015 2017 Year of filing Year of filing .15 .14 .13 Class_action residual .12 2007 2009 2011 2013 2015 2017 Year of filing
U.S. EGC U.S. non-EGC
(c) Financial misconduct Panel (a) examines the trends trading outcomes: market adjusted 2-year return, match-adjusted 2 year return, dummy for trading 1 year after IPO, dummy for trading at the end of 2019, and dummy for bankruptcy. Panel (b) examines trends in financial statement ratios relative to industry-year average: sales growth, return on assets, return on equity, profitability, and Tobin’s Q. Panel (c) examines trends in financial misconduct proxies: dummy for being subject to an Accounting and Auditing Enforcement Release, dummy for financial restatement, and dummy for a non-dismissed class action lawsuit.
Figure A3: Parallel trends in measures of firm quality Chapter 2. Increasing firms’ information advantages in the IPO process 48
Appendix E Cattaneo et al. Density Test for Manipulation of EGC Status
We implement the Cattaneo et al. density test (Cattaneo et al., 2020) to try and determine if U.S. firms manipulate their pre-IPO revenue to receive EGC status in the post-JOBS period. The top two graphs include all firms with pre-IPO revenue less than $2 billion, while the bottom two graphs include only firms with pre-IPO revenue between $750 million and $1.25 billion. The left graphs include IPOs in the pre-JOBS period (where EGC status wasn’t applicable) while the right graphs include IPOs in the post-JOBS period. .01 .008 .008 .006 .006 .004 .004 .002 .002 0 0 -1000 -500 0 500 -1000 -500 0 500 .004 .006 .002 .004 0 .002 0 -.002 -300 -200 -100 0 100 200 -200 -100 0 100 200
This figure plots the Cattaneo et al. density test for manipulation along the running variable, revenue in the year before IPO. The x-axes represent the distance from $1 billion in revenue. Top row: includes U.S. IPOs with pre-IPO revenue of less than $2 billion. Bottom row: includes U.S. IPOs with pre-IPO revenue between $750 million and $1.25 billion. Left column: U.S. IPOs filed in pre-JOBS period. Right column: U.S. IPOs filed in post-JOBS period.
Figure A4: Cattaneo et al. density test for manipulation of EGC status Chapter 3
Debt issuance in the era of passive investment
With Sergei Davydenko
3.1 Introduction
Recent decades have seen a dramatic shift towards low-cost passive investment strategies, with passive funds now managing approximately 20% of aggregate investment assets (Sushko and Turner, 2018). Exchange-traded funds (ETFs) and other passive investment vehicles typically track various market indices and do not attempt to identify potentially mispriced securities. The trend towards passive index tracking has affected not only equity markets but also the corporate bond market, where bond ETFs and passive bond funds have attracted significant interest from investors. The inflow of money into such funds now provides predictable demand for newly issued corporate bonds as long as they meet certain criteria, such as the minimum issue size, that make them eligible for automatic inclusion in popular corporate bond indices. This passive demand from index trackers may be insensitive to bond yields, covenant protection, and other bond characteristics unrelated to index eligibility.1 This paper looks at how passive investors’ demand affects corporate bond issuance. We first document that firms take advantage of passive demand by issuing index-eligible bonds with favorable characteristics. Specifically, we show that higher demand from passive bond index trackers increases firms’ propensity to issue bonds, and is associated with larger bond issues, lower yield spreads, fewer covenants, and longer maturities. Thus, bond features undesirable for investors but irrelevant for index inclusion become more prevalent as passive demand for index-eligible bonds increases. These findings are consistent with a model in which, in addition to active investors who make lending decisions based on their evaluation of expected default losses, there is also a number of passive investors who buy all bonds eligible for index inclusion. In the model, higher passive demand results in larger issue sizes, lower spreads, and higher investment. The effect is predicted to be particularly pronounced
1While we focus on the demand from passive funds, our conversations with bond market investors suggest that being included in a major bond index is likely to result in substantially higher demand from active bond funds as well. Many institutional bond investors, such as insurance companies and pension funds, have the majority of their capital invested in a benchmark bond index, with the remainder used to actively select bonds. This implies that by default an index-eligible bond is likely to be bought by many active investors, unless they specifically decide to underinvest in the bond.
49 Chapter 3. Debt issuance in the era of passive investment 50 for firms that would normally choose to issue bonds with a face value somewhat below the threshold size required for indexing. In the presence of passive demand, such firms may decide to increase the bond issue size just enough to meet the index threshold requirement. By doing so, they can ensure the participation of passive investors, which allows them to issue bonds on better terms and to reduce their cost of capital. Consistent with this prediction, we find that firms issue a disproportionate number of bonds that just meet the criteria for inclusion into popular indices tracked by passive fixed income funds. This effect is illustrated in Figure 3.1, which shows the distribution of observed investment grade (IG) bond sizes during the time period when the size threshold for the main IG index was $250 million. The graph also plots the ‘target’ bond sizes predicted by a simple empirical model based on observed issuance over the same time period. While the distribution of estimated bond sizes is smooth, actual bond issuance experiences a sharp discontinuity at the index threshold, with very few bonds being slightly smaller than the threshold. Firms appear to ‘reach’ into the index by issuing larger bonds than otherwise desired. To establish a causal link between passive demand and bond issuance, we look at the effects of changes in index eligibility requirements. We find that when index providers raise the minimum bond size required for index inclusion, firms respond by issuing larger bonds and clustering at the new, higher threshold. The effect is more pronounced for those firms which, absent the change, would be likely to issue a bond slightly below the new threshold. At the same time, we show that an increase in the index threshold temporarily reduces firms’ propensity to issue bonds. Thus, rather than issuing a bond large enough to be eligible for indexing under the new rules, or selling a smaller bond that cannot take advantage of passive demand, some firms react to the tightening of index rules by abstaining from bond issuance altogether. As a falsification test, we repeat these analyses around dates that do not correspond to index rule changes, and find no such effects. 0.25 0.20 0.15 Fraction 0.10 0.05 0.00 0 200 400 600 800 Issue size ($ millions)
Predicted Actual
Figure 3.1: The distribution of investment grade corporate bond issue sizes between June 2005 and January 2017, when the threshold for the inclusion in the Bloomberg Barclays U.S. Corporate Index was $250 million (shown by the vertical line). Predicted bond sizes are estimated in-sample using log assets, bond rating, and quarter and industry fixed effects.
We reinforce these conclusions by examining refinancing decisions of firms whose bonds mature close Chapter 3. Debt issuance in the era of passive investment 51 to an index rule change. The need to roll over a maturing bond provides an instrument for the firm’s propensity to issue a new bond of a certain size, which is independent of investors’ demand. Based on this idea, we compare bond issuance decisions by firms whose bonds mature shortly before and shortly after an index rule change. We find that firms are less likely to roll over maturing bonds shortly after an index threshold increases. Overall, our findings suggest that passive demand for corporate bonds affects firms’ debt financing decisions, bond contract terms, and the cost of capital. It incentivizes firms to issue larger bonds than they would otherwise, while simultaneously allowing them to pay lower spreads and secure more favorable bond terms along the dimensions irrelevant for index inclusion. In our model, firms take advantage of the availability of cheap bond financing to increase investment, and hence passive bond demand has real consequences. In practice, there are reasons to expect looser bond terms to result in a relaxation of constraints on corporate investment. For example, covenants have been shown to constrain investment (Chava and Roberts, 2008). Thus, by allowing firms to issue bonds with fewer covenants, passive bond demand is likely to have real effects. To the best of our knowledge, our paper is the first to explore corporate financing implications of the rise of passive investment strategies in the corporate bond market. Two recent papers complement our findings about the effects of ETFs on firm decisions about bond issuance and leverage. Similar to our findings for US firms, Calomiris et al. (2019) document clustering of corporate bond issue sizes in emerging market at a face value of US$500 million, the threshold to be included in a popular international bond index. The authors attribute this firm behavior as catering to institutional investors in order to benefit from reduced yields. Gibbons (2019) finds that higher firm-level passive debt ownership exacerbates shareholder-debtholder conflicts as firms take advantage of passive demand for bonds by increasing leverage and payouts to shareholders. Our paper documents the effect of passive demand on firms’ willingness to issue bonds and on the resulting bond characteristics. A number of studies explore the effects of “nonfundamental investor demand” on firms’ capital structure decisions.2 Recently, Kashyap et al. (2018) theoretically explore the distortions to a firm’s investment and equity issuance decisions when the firm’s equity is included in asset managers’ per- formance benchmark. We contribute to this literature by showing how the presence of passive bond investors and benchmark bond indices affects firms’ debt issuance activity. A sizable literature studies the implications of the secular shift toward passive investment strategies, indexing, and ETFs, including its effect on market efficiency (Stambaugh (2014), Wurgler (2011), and Israeli et al. (2017)), market fragility and volatility (Ben-David et al., 2018), underlying correlations (Da and Shive, 2017), trading liquidity (Hamm, 2014), corporate governance (Appel et al., 2016), and corporate investment (including Massa et al. (2005) and Li et al. (2018)). These studies have focused exclusively on passive demand in public equity markets, whereas we look at the demand for corporate bonds. What makes bond issuance decisions particularly interesting in this context is that, compared with equities, firms can more precisely adjust different bond features to meet index eligibility criteria. In addition, firms may also adjust bond characteristics that are important for the firm (such as spread) but not for the passive bond funds’ decision to invest. There is a small but growing literature that examines the effects of ETFs and other index trackers on corporate bonds. Dannhauser (2017) examines the effects of ETF ownership on bond yields by looking at
2See Graham and Leary (2011) and Baker (2009) for reviews of the recent capital structure literature. Note that the existing literature sometimes refers to the effects of this ‘nonfundamental investor demand’ as ‘supply-side effects’ in capital structure. In this paper, we refer to investors’ demand for bonds and firms’ supply of bonds. Chapter 3. Debt issuance in the era of passive investment 52 changes to Markit iBoxx bond indices rules. Dick-Nielsen and Rossi (2019) estimate the cost of liquidity in corporate bonds using index exclusions as a laboratory. Chen et al. (2014) look at a change in rating calculations by the Lehman Brothers bond indices (now the Bloomberg Barclays indices) and the effect on bond yields. Several recent papers show that ETFs improve the liquidity of their underlying corporate bond securities (e.g., Holden and Nam (2019), Ye (2019), and Marta (2019)). All of these papers look at outstanding bonds and how they are affected by index rule changes or fund flows. already outstanding when index rules change. By contrast, we focus on the corporate finance implications of passive demand, investigating its effect on firms’ decisions to issue bonds and on the resulting bond characteristics in the new issuance market.
3.2 Hypothesis development
In this section we show how the structure of investors’ demand for bonds may affect firms’ issuance decisions. We distinguish between ‘active’ bond investors, who decide whether or not to buy a bond based on their evaluation of expected default losses, and ‘passive’ investors, who invest in each bond regardless of its characteristics as long as the bond’s face value is large enough for it to be included in a tracked index. We construct a simple model to illustrate the intuition and guide our empirical tests.
3.2.1 Model setup and the demand for bonds
A firm is considering raising funds in the corporate bond market for one period, and needs to choose the size of the bond to issue. The risk-free interest rate is zero, so that the bond interest rate coincides with the bond spread. For reasons that will be described shortly, in order to sell a bigger bond the firm needs to pay a higher spread; in other words, it faces an upward-sloping demand for bonds in the bond size/spread space. All bond characteristics other than size and spread are treated as given. Denote by s(D) the inverse bond demand function, i.e., the spread at which investors will agree to purchase a bond with a face value D, given the firm’s credit risk and other relevant parameters. The firm invests all funds raised by selling the bond under the production function f(D), and pays s(D)D in interest. To sell the bond, the firm must pay c in transaction costs. The firm chooses the face value D to maximize f(D) − (1 + s(D))D − c. If the above expression is never positive (for example, because transactions costs c are higher than potential profits), the firm eschews the bond market and does not invest. The upward-sloping bond demand schedule arises as follows. There is a continuum of active bond investors, each of which can invest one dollar in the firm’s bond.3 The investors differ in their assessment of the (risk-neutral) expected loss from default; given the firm’s and bond’s characteristics, an active investor invests in the bond if his or her estimate of the loss from default does not exceed the offered bond spread. Denote by F (s) the number of active investors whose estimate of the risk-neutral expected loss is below s, and assume that F is monotonically increasing. Thus, if the firm wants to issue a bond with the face value D and there are only active bond investors present, the lowest spread that will allow it to sell the whole issue can be found from F (s) = D, giving rise to an increasing inverse demand function s(D) = F −1(D). Intuitively, when faced with active investors with heterogeneous beliefs, in
3The assumption that each investor can only invest one dollar ensures that in order to sell a large bond the firm must attract a sufficient number of investors with diverging views of its prospects. It can be justified by introducing investors’ borrowing constraints or portfolio diversification requirements. Chapter 3. Debt issuance in the era of passive investment 53 order to sell a larger bond the firm needs to attract progressively more pessimistic investors, who require higher bond spreads to offset their default expectations. In the presence of passive bond funds, in addition to the active investors there are also P passive investors who will buy the firm’s bond regardless of the spread provided that the issue is of sufficient size to make it eligible for an index inclusion. Denote the size threshold for index eligibility as D¯. Thus, passive investors invest P dollars in the bond at any spread if D ≥ D¯, and do not invest otherwise. The resulting demand schedule is illustrated in Figure 3.2. Small issues (D < D¯) must be sold to active investors only at the spread given by s = F −1(D). By contrast, for large, index-eligible issues (D ≥ D¯) passive investors automatically contribute P dollars, leaving only D − P to be financed by active investors. This allows the firm to sell the bond at the reduced spread given by s(D) = F −1(D−P ).
Index 퐹−1(퐷) threshold
Active
demand 퐹−1(퐷 − 푃) Spread, s Spread,
Active and passive demand
푆∗
퐷ഥ − 푃 퐷ഥ Face Value, D
Figure 3.2: The bond demand curve. D¯ represents the index threshold amount, and P represents the amount of the bond purchased by passive investors.
3.2.2 Model predictions
It is convenient to distinguish between cases in which the firm’s optimal issue size in the absence of passive demand would be much smaller than D¯, ‘slightly’ smaller than D¯, or larger than D¯. We refer to these cases as small, medium, and large target issue size cases. For small target issues (D D¯), passive demand is irrelevant for the firm’s decision, and only active investors’ demand needs to be considered. We will now show that, by contrast, for medium and large issues the presence of passive investors will in general result in larger issues and lower spreads. Figure 3.3a illustrates the firm’s decision when the target issue size is large, with each dashed line representing bond size/spread combinations that result in a particular level of profits for the firm net of debt costs.4 In the absence of passive demand the firm’s optimal choice would correspond to point √ 4The firm’s indifference curves are plotted assuming a square-root production function (f = D) and linear demand from active investors. Chapter 3. Debt issuance in the era of passive investment 54
A, which results in the highest profit attainable given the active-only demand curve for bonds. The presence of passive investors shifts the optimal choice to point B, at which the size of the issue is larger while the yield spread is lower. It is straightforward to show formally that for large issue sizes, under reasonable parameterizations of the relevant functions, the optimal issue size is increasing in the amount of passive demand, P , and the equilibrium spread is decreasing in it. Moreover, passive demand also increases the firm’s propensity to issue bonds. Indeed, profits are strictly higher at point B than at A. Thus, there exists a range of parameters under which the cost of accessing the bond market, c, would deter bond issuance at point A, but not at point B. Figure 3.3b demonstrates that for firms with a medium target bond size (i.e., those which in the absence of passive investors would choose the bond size slightly below D¯), the introduction of passive investors may result in a corner solution, with the optimal bond size exactly equal to D¯. In the graph, without passive investors the firm would choose point A below the index threshold. However, in the presence of passive demand the firm’s profits are maximized at point B, i.e., at the issue size D¯ just sufficient to make the bond eligible for indexing. Thus, for a range of target issue sizes slightly below the index threshold, passive demand may induce firms to bump up their bond size to coincide with the threshold. While this bond size may be somewhat larger than otherwise desired for these firms, the ability to take advantage of passive demand makes this choice worthwhile. Thus, the model predicts that a disproportional number of bonds should be issued exactly at the index threshold, and issues just
below the threshold should be disproportionately infrequent.
Spread Spread
A
B A
B
퐷 Face Value 퐷 Face Value
(a) (b)
Figure 3.3: The effect of passive demand on firm’s financing choice. Point A represents the firm’s optimal issuance level in a world with only active demand, while point B represents the optimal issuance level in a world with both active and passive demand.
Now suppose that the index provider decides to increase the threshold size for index inclusion from D¯ to D¯ 0. This situation is depicted in Figure 3.4. Upon the threshold increase, a range of previously feasible size-spread combinations becomes infeasible. Specifically, firms that under the old threshold would optimally choose bond sizes between D¯ and D¯ 0 will no longer be able to do that while paying the same spreads as before, because bonds of this size are no longer bought by passive investors. As a result, these firms may decide to increase the bond size, decrease it, or abstain from bond issuance. Chapter 3. Debt issuance in the era of passive investment 55
Consider the case depicted in Figure 3.4a. With point A no longer attainable, the firm chooses to switch to point B (the new corner solution) and issue the bond at the new threshold size, D¯ 0. Thus, following the threshold increase, firms with target issue sizes slightly below D¯ 0 may cluster their issuance at the new threshold. However, because point B corresponds to lower expected profits,5 for high enough costs of bond issuance the firm may decide not to issue the bond at all and forgo the investment. Thus, firms’ propensity to issue bonds may decrease. It is also possible that, rather than inflating the bond size to meet the new threshold, the new optimal choice for the firm would be to issue a smaller bond, relying only on active investors’ demand. Consider the case illustrated in Figure 3.4b. As point A becomes unavailable, the firm’s profits are maximized at point C, where it sells the bond to active investors only instead of inflating its size to meet the new
indexing threshold.
Spread Spread
B
C B C
A A
퐷 퐷 퐷′ 퐷′ Face Value Face Value
(a) (b)
Figure 3.4: The effect of an increase in the index threshold. Point A represents the firm’s optimal issuance level before the index threshold increases, while points B and C represent feasible issuance levels after the index threshold increases. Figure 3.4a represents a small increase in the index threshold, while figure 3.4b represents a larger increase.
Which of these two outcomes will prevail depends on how big the threshold increase is relative to passive investment P . By continuity, when point A in Figure 3.4a is sufficiently close to point B, reaching for the new threshold will be optimal. But in general there may or may not be a region in which the bond will be sold to active investors only, as in Figure 3.4b. Nonetheless, in both cases profits at points B and C are unambiguously lower than at point A before the change, implying a lower propensity to issue bonds. To summarize, the model predicts that firms’ propensity to issue bonds is positively related to passive demand. Conditional on issuance, the bond size is increasing and the spread is decreasing in the amount of passive demand. Bond issuance by firms with medium target bond sizes clusters at the index threshold, with few issues just below it. When the threshold increases, firms with target bond sizes slightly below the new threshold issue larger bonds to meet the new requirement for inclusion, and the propensity to issue bonds decreases.
5To see that point B yields lower profits than point A, notice that under the old threshold point A was preferred even though point B was also feasible. Chapter 3. Debt issuance in the era of passive investment 56
An important limitation of the above analysis is that all bond characteristics other than the bond size and spread are assumed to be fixed. In reality, all bond characteristics are determined jointly, including covenants, maturity, seniority, etc. An increase in the passive demand expands the set of feasible contracts in the firm’s favor, because passive investors buy eligible bonds regardless of their creditor-unfriendly features. The resulting equilibrium thus corresponds to higher profits for the firm. This, however, does not imply that all bond features are necessarily adjusted in the firm’s favor. For example, in the presence of passive demand a firm that finds covenants very undesirable may choose a new bond contract with less covenant protection, even if such a bond can only be sold at a higher spread. Theoretically, when the choice of multiple bond characteristics is modelled jointly, the resulting equilibrium will depend on the relationship between these characteristics and expected default losses, as well as on how costly the unfavorable bond terms (such as strict covenants or short bond maturities) are for the firm. Thus, with multiple choice variables the equilibrium effect is an empirical question, one that we turn to below.
3.3 Data description
In this section, we describe the data and key variables used in our empirical analysis. More details regarding the construction of the variables can be found in Appendix A. Our tests employ both bond-level and firm-level data. Our first sample consists of corporate bonds included in Mergent’s Fixed Income Securities Database (FISD). The second sample, which we use to study firms’ propensity to issue bonds, consists of a broad set of Compustat firms not limited to bond issuers. We use various data sources, including CRSP Mutual Funds and index providers’ and funds’ websites, to construct our key independent variable summarizing the passive demand for corporate bonds. In particular, we identify passive investment funds and the amount they invest in corporate bonds, as well as the composition and size of popular tracked bond indices and changes to eligibility criteria for those indices.
3.3.1 Sample selection
Corporate bond sample
In order to study corporate bond characteristics, we use a sample of newly issued bonds included in FISD. We eliminate issuance by government, financial and utility issuers; issuers not domiciled in the United States; bonds not denominated in USD; duplicate bonds;6 convertible debentures, floating rate bonds, preferred shares and bonds issued as a part of a unit deal; and issues smaller than $25 million. The final sample includes 16,691 bonds issued by 4,063 unique issuers between January 1990 and September 2017. Our key variables of interest include the size (total face value) of the bond, the credit spread at issuance, the level of covenant protection, and the initial maturity, all taken from FISD. We estimate the strength of the covenant protection using a version of Moody’s Covenant Quality Index, which is based on the number of covenants in the bond contract related to restricted payments, risky investment, debt incurrence, lien subordination, and change of control. Our covenant score is a weighted average of
6For example, a registered bond that was originally issued as a Rule 144A bond and later exchanged would include two entries in FISD; we keep the original 144A bond in the sample but eliminate the bond into which it exchanges. Chapter 3. Debt issuance in the era of passive investment 57 covenants in these categories, and is normalized to be between 0 (no covenants) and 100 (all possible covenants tracked by FISD). To control for issuers’ financial characteristics, we match FISD to Compustat and CRSP using issuer-level CUSIPs where possible, and by firm name otherwise. This procedure allows us to match ap- proximately three quarters of our sample to Compustat. Where available, we include firm-level financial data from Compustat (from the quarter before issuance), and annualized daily volatility from CRSP. We winsorize all ratios at the 1% and 99% levels. We also include as controls macroeconomic variables such as the 10-year Treasury bond rate, the term slope between the 2-year and 10-year Treasury rates, and the spread between Baa and Aaa bonds, which we obtain from the Federal Reserve Bank of St. Louis.
Firm sample
We study the effects of passive demand on the propensity to access the bond market using a sample of firms included in the U.S. Compustat database. After dropping financial and utility firms, there are 16,620 unique firms and 654,190 firm-quarters between September 1989 and June 2017 (lagged one quarter from the issuance sample). We merge these data with the FISD database of bond issuance, and construct a dummy variable issuer, which takes on a value of 1 in a firm-quarter if in the next quarter the firm issues a bond meeting the criteria detailed in the previous section. We winsorize all ratios at the 1% and 99% levels. In order to determine whether a bond from a potential issuer would be included in a particular investment grade (IG) or high yield (HY) index, we need to determine whether the bond would be rated IG or HY. For firm-quarters with at least one rated bond outstanding, we use the current average rating of the outstanding bonds in FISD to determine the likely rating for a hypothetical new issue. This variable, however, is not observed for firms without outstanding bonds. In these cases, we use the following approach to estimate whether the issue would likely be rated IG or HY: Using the sample of Compustat firm-quarters with observed ratings, we regress the IG dummy on the five individual variables that comprise Altman (1968)’s z-score, and calculate the linear prediction from the fitted model. We then identify a cut-off prediction value that correctly classifies the highest number of firms as investment grade or not, and apply it to fitted values for non-rated firms to predict their IG/HY designation.
3.3.2 Measuring passive demand
We calculate our main measure of passive demand for bonds as the proportion of the index-eligible U.S. corporate bond universe that is held by passive funds. To this end, we apply index eligibility criteria to the universe of bonds in the FISD database, which allows us to estimate the fraction of the index attributable to corporate bonds. Using fund holdings data, we calculate the total assets in passive funds tracking each index, and then compute the amount of corporate bonds held by the funds. This subsection describes our approach in detail.
The bond indices
We use the CRSP Survivor-Bias-Free U.S. Mutual Fund database from January 1990 to September 2017 to measure passive investment in corporate bonds. We select funds that (i) invest at least part of their assets in U.S. corporate bonds and (ii) passively follow an index. In order to do this, we identify all funds with “Bond” and either “ETF”, “Exchange Traded”, “Exchange-Traded”, or “Index” in the name. Chapter 3. Debt issuance in the era of passive investment 58
We then identify which index each of these funds follows using information from ETFdb.com and funds’ online profiles. After removing all bond funds that do not hold U.S. corporate bonds (i.e, those investing in government, municipal, or foreign bonds only) as well as actively managed funds, we are left with 277 passive bond funds tracking 104 bond indices. Many different indices are administered by a few major providers, as illustrated in Figure 3.5 sep- arately for investment-grade (IG) and high-yield (HY) bond indices. Figure 3.5a shows that more than 80% of passive funds invested in investment grade corporate bonds follow indices provided by Bloomberg.7 For high yield funds, Bloomberg indices are followed by net assets representing approxi- mately 40% of high-yield bond funds, while about 50% follow Markit iBoxx indices (see Figure 3.5b). Based on these statistics, we limit our attention to Bloomberg and iBoxx indices only. Specifically, we use the Bloomberg Barclays U.S. Corporate Index (IGCI) and the Bloomberg Barclays U.S. Corporate High Yield Index (HYCI) to identify the set of bonds eligible for indexing.8 Most tracked indices are not exclusively focused on corporate bonds, and additionally include government, agency, and other bonds. For example, the popular Bloomberg Barclays U.S. Aggregate Index includes all bonds from IGCI, as well as Treasury, agency, non-U.S. government, and securitized bonds.9 We collect information on the composition of the relevant indices and the eligibility criteria that determine which bonds qualify for inclusion in each index. We record the history of major changes in the eligibility criteria, and use these rule changes in our test as quasi-natural experiments to identify the causal effects of passive index investment. Particularly important for our purposes are changes to the minimum face value for inclusion. Our model shows that such changes can have a direct impact on a firm’s issuance decision and bond sizes. A detailed discussion of the major indices and changes in eligibility criteria can be found in Appendix B. Based on the offering date of a bond, we calculate the difference between the bond size and the relevant index threshold, dist to threshold. A negative value for dist to threshold means the issue is too small to qualify for the index, while a zero or positive value means the issue meets the size criteria.
Aggregate passive demand for corporate bonds
To compute the fraction of bond demand that is attributable to passive funds, we aggregate the funds’ assets under management that are invested in U.S. corporate bonds and divide this amount by the total value of index-eligible bonds outstanding. Because index inclusion rules are different for investment grade and high yield bonds, we assign each bond to either the investment grade or high yield pool based on its initial rating and the rating criteria at the time of issuance. We estimate the proportion of corporate bonds held by passive funds as described in Appendix C. Figure 3.6 shows the total net assets invested in U.S. corporate bonds.10 Between 2009 and 2017, passive investment in investment grade and high yield U.S. corporate bonds has increased at the com-
7Bloomberg bond indices were run by Lehman Brothers before November 2008, and subsequently by Barclays until August 2016. 8We use the Bloomberg index for high yield bonds due to a dramatic change to the index methodology used by iBoxx in 2009 (see Appendix B for more detail). 9It should be noted that unlike equity index funds, bond index funds do not necessarily replicate the index they track. For example, according to its prospectus, the Vanguard Total Bond Market Index Fund “invests by sampling the Index, meaning that is holds a broadly diversified collection of securities that, in the aggregate, approximates the full Index in terms of key risk factors and other characteristics. All of the Fund’s investments will be selected through the sampling process, and at least 80% of the Fund’s assets will be invested in bonds held in the Index” (Vanguard, 2017). 10As of the end of our sample period, there is an additional $517 billion invested in other bond types, such as government, agency and securitized bonds, and non-U.S. corporate bonds. Chapter 3. Debt issuance in the era of passive investment 59 pound annual growth rates of 26% and 34%, respectively. Figure 3.7 shows the proportion of outstanding corporate bonds held by passive funds. As of September 2017, we estimate that more than 5% of invest- ment grade bond face value and more than 3% of high yield bond face value is held by passive funds. This is consistent with Sushko and Turner (2018), who estimate that 4.5% of all U.S. bonds are held by passive investment vehicles. It is important to stress that the effect of index inclusion on demand is likely to extend far beyond the automatic buying by passive funds. Anecdotal evidence suggests that some active bond investors may be ‘mostly’ following their benchmark bond index, with only a relatively small fraction of their capital used for active security selection. If this is the case, then a bond included in an important bond index would by default be bought by such funds, unless they specifically decide against it because they deem the bond to be particularly unattractive. This has two implications for our study. First, index inclusion is likely to result in a jump in demand not only because of the presence of explicitly passive investors, but because some active investors may also be ‘mostly’ tracking the index. Second, because our independent variable is constructed using only explicitly passive funds, it is likely to understate the total incremental demand that stems from a bond being index-eligible. If the unobserved indexing-induced demand from active investors is correlated with our independent variables, our regression coefficients may be biased upward, and hence their magnitudes should be interpreted with caution, as they would capture some of this ‘somewhat active’ index demand in addition to the explicitly passive demand.
Bond-level passive demand
As described above, we construct two versions of the variable Passive demand, which summarize passive demand for IG and HY bonds. In addition to these aggregate indices that only vary over time, we also compute a bond-level measure, Bond-level passive demand. This variable estimates the percentage of the given bond’s total face value bought by passive funds, which allows us to exploit cross-sectional differences in passive demand arising due to bonds’ varying eligibility for the 104 individual indices tracked by the 277 funds. To construct this variable for a particular bond, we check if the bond meets the criteria to be included in the bond index tracked by each of the funds, and if it is, how much money the fund has invested in the bond, which we estimate based on the fund’s total assets under management and the value weight of the bond in the index. The variable Bond-level passive demand is the fraction of the bond face value bought by all passive funds, assuming the funds tracked their target indices exactly.
3.3.3 Descriptive statistics
Table 3.1 reports the descriptive statistics for our variables of interest: Panel A describes our sample of bond issuers, while Panel B shows the statistics for the full Compustat sample. All variables are defined in detail in Appendix A.
[INSERT TABLE 3.1 HERE]
The median new issue bond in our sample is $300 million in face value, and has a spread of 2.02%, an initial term of 10 years, and a covenant score of 9.17 (out of 100). The median bond issuer is investment grade rated (a value of 10 corresponds to BBB- or Baa3). In the Compustat sample, 1.3% of firm-quarters include at least one bond issuance, with a median offering size of $275 million. Chapter 3. Debt issuance in the era of passive investment 60
Comparing the median bond issuer (Panel A) to the median Compustat firm-quarter (Panel B), the bond issuer has higher leverage, is larger in terms of book assets, is older, has more tangible assets, is more profitable, and has a lower q-ratio.
3.4 Passive investment, bond characteristics, and issuance
Our model predicts that as passive demand increases, firms issue larger bonds, either because firms ‘reach’ to be included in the index as in Figure 3.3b, or simply to take advantage of additional demand as in Figure 3.3a. At the same time, credit spreads are predicted to decrease. We also hypothesize that other bond characteristics irrelevant for index eligibility should be adjusted in the firms’ favor as passive demand increases.11 To test these hypotheses, we use the FISD sample of bond issues, and estimate the following regression:
bondcharit = β ∗ passive demand perct + controlsit + it, where bondcharit for bond i at time t is either the bond size (Log issue size), the credit spread (Spread), the level of covenant protection (Covenant score) or the initial time to maturity (Initial maturity). We expect the coefficient β to be positive for Log issue size and Initial maturity and negative for Spread and Covenant score. Our main independent variable is Passive demand, which is the fraction of indexed corporate bonds that is held by passive funds. This variable is measured at monthly frequency, separately for investment grade and high yield bonds. In all specifications, we use year-quarter fixed effects to control for time trends in the variable of interest, as well as industry fixed effects.12 We control for the bond rating, log assets, leverage, and other variables suggested by Graham and Leary (2011), and also use additional controls for macroeconomic conditions.The results of the regressions are shown in Tables 3.2 and 3.3.
[INSERT TABLES 3.2 AND 3.3 HERE]
Consistent with the model’s predictions, for all specifications in Table 3.2 passive demand for bonds is positively correlated with the size of bonds being issued and negatively correlated with the bond spread. The results are strongly statistically significant across all specifications. In addition, Table 3.3 shows that bond maturity is longer and covenant protection is lower when passive demand is high, though the latter correlation is not statistically significant. The control variables generally work as expected. Larger firms issue larger bonds with lower spreads, fewer covenants and longer maturities; controlling for firm size, lower rated issuers are associated with larger offering sizes, higher spreads, more covenants and shorter maturities. We next look at firms’ propensity to issue bonds. Because bond contract terms are predicted to be more favorable in the presence of passive demand, we hypothesize that firms should be more willing to issue bonds when passive demand is high. To test this hypothesis, we use a sample of Compustat firms, identify which of them issue bonds and which do not, and look at whether the probability of becoming
11More precisely, the model predicts taht higher passive demand will result in higher profits for bond issuers. A special case is when all bond characteristics are weakly adjusted in the firm’s favor, but other combinations of characteristics that amount to higher profits for the firm are also possible. 12For issuers in the Compustat database, we use the 2-digit SIC code to control for industry fixed effects. For specifications that use only FISD bond data, we use the industry code provided by FISD, which is as coarse as 1-digit SIC. In untabulated tests, we rerun all specifications using 2-digit SIC on the sample for which it is available, and the results are similar. Chapter 3. Debt issuance in the era of passive investment 61 a bond issuer is positively correlated with passive demand by estimating the following linear probability model:
issuerit = β ∗ passive demand perct + controlsit + it, where issuerit is a dummy variable equal to 1 in quarter t that a firm i accesses the bond market, and 0 otherwise. We include quarter fixed effects across all specifications, and either 2-digit SIC or firm fixed effects. We also control for variables used in studies of firms’ financing decisions, including log assets, leverage, and a dummy for the investment grade status. We also run the regression with the controls that have been shown in the literature to influence the debt issuance behavior, namely, the components of Altman (1968)’s z-score, the significant variables identified by Erel et al. (2012) (referred to in the tables as the EJKW variables), and the variables used by Leary and Roberts (2005) (LR controls). The results are given in Table 3.4. The coefficient on Passive demand is positive and strongly significant, indicating that higher passive demand is positively correlated with firms’ propensity to access the bond market. Thus, higher demand from passive bond investors makes bond issuance more likely, and conditional on issuance, it is associated with larger bonds, lower spreads, longer maturities, and fewer covenants.
[INSERT TABLE 3.4 HERE]
We test the robustness of these findings using panel-type regressions. Our independent variable of interest in the previous tests, Passive demand, differs for IG vs. HY issuers, but otherwise varies only over time and not across firms. While we control for time variation in firms’ issuing activity and bond features by including quarterly fixed effects and a number of macroeconomic variables, there may be residual unobserved variation in economic conditions within each quarter that may be affecting our results. To address this issue, we construct a bond-specific measure of passive demand, Bond-level passive demand. This variable exploits the fact that different bonds can be eligible for inclusion in different subsets of the 104 bond indices tracked by the passive funds, and thus bought by different funds. We re-run the bond characteristic regression as follows:
bondcharit = β ∗ passive demand bond percit + controlsit + it, where passive demand bond percit now varies for bond i issued at time t. The results are shown in Table 3.5. The results for the issue size and spread are similar to those found previously, with higher passive demand corresponding to bigger bonds and lower spreads. This is once again consistent with the model’s predictions. In addition, we also look a the effect on covenants and bond maturity. The coefficients for these variables flip signs compared to Table 3.3, although the relationship is statistically significant only for initial maturity.
[INSERT TABLE 3.5 HERE]
As discussed previously, the model makes no predictions on how multiple bond characteristics will be adjusted under higher passive demand; we can only conclude that the overall bond package will be more attractive to the firm. The bond-level regression results concerning the effect on maturity suggest that not all bond features may be adjusted in the firm’s favor. However, this result may also be a reflection of the fact that there are passive bond funds that are specifically tailored to invest in bonds of a particular maturity (such as 1-5 years, 5-10 years, or 10+ years), as firms may strategically choose a maturity to Chapter 3. Debt issuance in the era of passive investment 62 influence the amount of passive demand to which they are exposed. In other words, Bond-level passive demand may be more endogenous than the time series variable examined in Tables 3.2 and 3.3, especially with respect to the bond’s initial maturity. Overall, the results on bond sizes and spreads are consistent with those of our main regressions, con- firming that passive demand is related to bonds with terms that are attractive to issuers. To investigate whether these relationships are causal, we focus on effect that the threshold for index eligibility has on issuers’ behavior.
3.5 Index eligibility thresholds
In this section, we show that bond issuance decisions are affected by the requirements that determine eligibility for inclusion in popular bond indices tracked by passive funds. Each index has a number of criteria for inclusion, such as credit grade (IG vs. HY), minimum face value, and specific bond features (e.g. the bond must have a fixed coupon and cannot be convertible). In our tests, we take the bond’s IG/HY designation as given and focus on the minimum size (face value) requirement, because it is a characteristic most easily adjusted by the issuer. The size threshold differs for IG and HY indices and varies over time, allowing us to exploit changes in thresholds for identification purposes.
3.5.1 Threshold clustering
Our model predicts a ‘corner solution’ (illustrated in Figure 3.3b) for firms which, in the absence of passive investors, would issue bonds somewhat below the index threshold size. These firms may take advantage of passive demand by bumping up the issue size just enough for the bond to be included in the index. As a result, we expect bond sizes to cluster at the index size threshold, with few issues just below it. As previously discussed, Figure 3.1 illustrates this effect. It shows the distributions of ‘target’ bond sizes and the actually observed size, for investment grade bonds during the period when the IGCI threshold was $250 million. The graph demonstrates that, in contrast to the smooth distribution of target bond sizes, the actual distribution experiences a pronounced jump at the index threshold, with few firms issuing bonds below it. To formally check whether the index thresholds are relevant, we employ the density test from McCrary (2008), which is often used in the regression discontinuity design (RDD) context to check for manipulation of the running variable. We expect the total bond face value to be manipulated around the size threshold, so that the density of the distribution would be higher immediately to the right of the threshold than the left. The null hypothesis is that issuance is randomly assigned on both sides of the threshold. Using the dist to threshold as the variable of interest, Figure 3.8 shows that there exists a sharp discontinuity to the left and the right of zero (i.e., at the threshold), with higher density to the right of the discontinuity than the left. This is the case for the full sample, the IG sample and the HY sample. These tests confirm that the distribution of issue sizes is discontinuous at the index threshold, consistent with our model’s prediction of a corner solution at the threshold for a range of firms which in the absence of passive demand would issue a smaller bond not eligible for indexing. Chapter 3. Debt issuance in the era of passive investment 63
3.5.2 Threshold changes
In this section, we document the effects that changes in the index threshold have on firm’s issuance decisions. Our tests focus on three major indices, changes to which would affect corporate bond holdings of over 95% of passive bond funds by value. These are the Bloomberg Barclays U.S. Corporate Index (IGCI) for investment grade bonds, and the Bloomberg U.S. Corporate HY Index (HYCI) and the Markit iBoxx Liquid High Yield Index (iBoxx) for junk-rated bonds.13 During our sample period, which starts in 1990, the threshold for IGCI was updated five times, and each of the high yield indices was updated once. In addition, we look at the effect of the introduction of the iBoxx index in 2007.
Figure 3.9 plots the distribution of bond issue sizes in the 12 months before the announcements of threshold changes to IGCI and 12 months after the effective dates of the changes. The effect does not appear strong for the first two updates, in 1994 and 1999 (panels 3.9a and 3.9b), which took place before passive bond investment became significant. By contrast, Panel 3.9c, corresponding to two index threshold increases in October 2003 and July 2004, illustrates a strong effect. In particular, each graph shows that issuance immediately to the left of the threshold is smaller than the issuance immediately to the right (with large spikes in the bottom two graphs). In addition, we do not see issuance concentration at cut-offs from other time points; for example, issuance in the $250 million bin is lower than the $200 million bin in all graphs except the bottom one, when the threshold increased to $250 million. The effect can also be clearly seen in panel 3.9d, which shows the April 2017 increase in threshold from $250 million to $300 million. The bottom panel shows that in the six months after the effective date of the change, there is very little investment grade issuance at the $250 million level, but substantial issuance at that level in the six months before the change was announced.
Figure 3.10 shows the distributions for high yield bonds. We examine one change in the HYCI (the increase from $100 million to $150 million in 2000) and two changes in the iBoxx (its introduction in 2007 with a $200 million threshold and an increase from $200 million to $400 million in 2009, along with other changes). Though the change to HYCI is early in the sample, there is a clear cluster immediately to the right of the threshold in both graphs.
The introduction of the iBoxx in Panel B appears to result in a decrease in bonds at the $150 million level relative to the $200 million level. However, the results are less clear when the index dramatically increased its threshold to $400 million, as there is some clustering at both the old and new thresholds before and after the change. The ambiguous effect of the doubling of iBoxx is in fact consistent with the predictions that our model makes for large index changes. Indeed, a model prediction illustrated in Figure 3.4b is that a large increase in the index threshold may result in smaller bond sizes. In particular, a firm that under the old rules may have issued a $200 million bond to take advantage of passive demand, may find it unprofitable to issue a $400 million bond in order to reach for the new index threshold, and settles for a smaller issue size sold to active investors only.14
13Many funds track larger indices that include IGCI as a subset. However, changes to IGCI eligibility criteria will be mirrored in all indices that use IGCI as the basis for selecting eligible corporate bonds (for example, the Bloomberg Barclays U.S. Aggregate Index, which consists of U.S. corporate bonds from the IGCI, as well as other type of bonds, such as government and agency). 14Note that such bonds would still be tracked by the funds following the HYCI, as long as the issue size is above $150 million. Chapter 3. Debt issuance in the era of passive investment 64
3.5.3 Difference-in-difference regression analysis
To formally establish the effect that index changes have on the bond size, we examine the index changes in a difference-in-difference framework. In particular, for changes in the investment grade index (IGCI), we compare bond issuance by IG (treated) versus HY (control) issuers before and after the index changes. Similarly, we examine changes in the HY indices using high yield and investment grade bonds as the treated and control groups, respectively. We are able to do this because none of the changes across both the IG and HY indices occur at the same time. We first examine whether the treatment and control groups exhibit parallel trends before the ‘shocks’ to the respective index. Figure 3.11 shows the average offering size and issuance propensity before and after each change, with the announcement dates marked by the vertical lines (the ‘pre’ period is to the left of the lines). For the investment grade index changes (the top two panels), there appears to be a divergence in average bond sizes only in the months immediately before the fourth change. The issuance propensities move almost entirely in sync, except for the second change, where there appears to be an increase in high yield issuance immediately before the change. This change occurred in the middle of the dot-com bubble, which may have contributed to the observed trends. For the high yield index changes depicted in the bottom two panels, there may have been some divergence in the average issuance size immediately before the second change, which occurred in April 2007. Overall, however, the parallel trends assumption appears to be satisfied.
Threshold changes and bond sizes
We first estimate the following difference-in-difference regression for the changes to the investment grade index:
log issue sizeitc = α ∗ post igitc + γ ∗ treated igitc + β ∗ post igitc ∗ treated igitc + µc + controlsitc + itc, where post igitc is a dummy variable equal to zero in the period before a change in the investment grade index is announced and to one in the period after the change is implemented (the period in between the announcement and effective date is ignored), and treated igitc is a dummy equal to one for investment grade issuers and zero for high yield issuers. Because we look at one-year time windows before and after each change and the changes are spread out in time, we are left with four mutually exclusive difference-in-difference cohorts, indexed by c.15 By including industry by cohort fixed effects
(µc), β measures the average treatment effect across the four changes in a stacked difference-in-difference regression, controlling for industry effects in bond sizes. As controls, we include ratings, log assets and leverage, the variables from Graham and Leary (2011), and macroeconomic proxies. We also estimate regressions of Equation (3.5.3) for changes in the high yield indices (ignoring the middle period for the iBoxx changes). In both sets of regressions, we restrict our sample to bonds with a face value less than $1 billion, for which the index threshold is more relevant and closer to ‘binding’ (this does not materially affect the results).
[INSERT TABLE 3.6 HERE]
15We exclude the ‘middle’ period between the two consecutive changes in October 2003 and July 2004, because they occurred too close to each other. The pre- and post-periods correspond to the top and bottom graphs in Figure 3.9c, ignoring the middle graph. Chapter 3. Debt issuance in the era of passive investment 65
Table 3.6 shows the results of these regressions. Columns (1)-(4) pool all investment grade bond changes together, and show that investment grade bonds issued immediately after an investment grade index change are significantly larger, compared to the change in high yield bond issue sizes over the same period. Columns (5)-(8) pool all high yield changes together. The coefficient on the interaction term is negative across all specifications, though not always statistically significant. The lack of significance for the high yield changes may be due to the fact that HY index changes occurred while passive investment in high yield bonds was still not as widespread, or because of the large increase in index threshold in 2009. As illustrated in the two panels of Figure 3.3, our model suggests that the effect of an index threshold increase may be different for firms based on their target issue size absent the threshold constraint. In particular, a firm that is not constrained by the index threshold (i.e., would optimally issue above the threshold) should increase the bond size by a smaller amount than a firm that would otherwise issue below the threshold. To examine this prediction, we estimate the ‘target’ bond issue size and test whether the threshold increase is more relevant for firms which would otherwise issue bonds below the new threshold. The results are reported in Table 3.7. Using only bonds in the pre-change periods, we first predict the offering size of a bond based on the controls above (log assets, rating, and change and industry fixed effects), and place the bond in one of two buckets: predicted bond size that is above the new threshold (columns 1 and 3), and predicted bond size is above the old threshold but below the new threshold (columns 2 and 4).16 While we expect a positive β coefficient across all specifications (since the direction of the arrows in Figure 3.3 are both to the right), we expect a larger β for bucket 2 relative to bucket 1 as these issuers ‘reach’ to be included in the index in the post-change period but did not have to reach in the pre-change period. This is precisely what we find: the coefficient for investment grade issuers in bucket 2 is more than four times that of those in bucket 1. The coefficient on the interaction term for HY bonds is now positive (and three times larger for bucket 2 than 1), though still not statistically significant.
[INSERT TABLE 3.7 HERE]
Threshold changes and bond issuance
We next look at the effect that an increase in an index threshold has on the firms’ propensity to issue bonds. We hypothesize that larger thresholds discourage bond issuance by putting passive demand out of reach for some firms that target medium-sized issues. These tests parallel those in Table 3.6, except that we replace Log issue size with Issuer dummy, and estimate the following specification:
issueritc = α ∗ postitc + γ ∗ treateditc + β ∗ postitc ∗ treateditc + µc + controlsitc + itc, where postitc and treateditc represent either IG or HY index changes. The results of this regression are shown in Table 3.8. We include change-by-industry fixed effects and the same controls as those included in Table 3.4. In contrast to the long-term positive relationship between passive demand and issuance propensity, in the short term around index threshold changes, an investment grade firm’s propensity to access the bond market is reduced following a threshold increase (though only statistically significantly
16There are too few bonds that fall into the third mutually exclusive category (predicted size falls below the old and new threshold) that inference is not possible. Chapter 3. Debt issuance in the era of passive investment 66 with the LR controls). The results for high yield issuers are not statistically significant, which is not surprising given that we do not find significant results for bond sizes for these firms.
[INSERT TABLE 3.8 HERE]
As a robustness check, we perform falsification tests by re-estimating the regressions of Equations (3.5.3) and (3.5.3) using dates that do not correspond to any index changes. In contrast to Tables 3.6 and 3.8, no interactions terms are significant in these regressions for either IG or HY bonds.17
Evidence from bond rollovers
One potential problem with the previous analysis about firms’ propensity to issue is that we are not able to observe or control for firms’ need for capital, which may influence the decision of whether or not to issue a bond. We address this issue by examining one scenario where firms are more likely to want to issue a bond: when another bond is reaching its maturity date. We expect that under normal circumstances, some firms would likely roll over (refinance) their maturing bonds, and we compare the propensity to issue bonds for firms whose bonds mature in the year before and after an index change. Since bond maturity is chosen at issuance, refinancing needs induced by bond expiration are independent of the index threshold changes, and any difference in the behavior of firms whose bond mature shortly before and shortly after the change is likely to be attributable to the change. For a given bond, we calculate a Rollover dummy that takes on a value of 1 if a firm issues any new bonds in the 6 months before the old bond’s maturity date, provided that the total face value of the new bonds must be at least as large as the old bond’s face value. We then run the following regression:
rolloveritc = α ∗ postitc + γ ∗ treateditc + β ∗ postitc ∗ treateditc + µc + controlsitc + itc, where postitc and treateditc represent either IG or HY index changes. These regressions are reported in Table 3.9. In columns (1) and (3) we determine the treatment status based on whether the bond was IG or HY at issuance, whereas columns (2) and (4) require the bond to be rated IG or HY six months before expiry. In all regressions we control for change-by-industry fixed effects. The results in this table are consistent with our findings reported previously: An investment grade issuer has a lower propensity to issue after a index threshold change, as demonstrated by the lower likelihood of rolling over a bond that expires in the year after a threshold increase. The effect for high yield issuers goes in the same direction but is not statistically insignificant.
[INSERT TABLE 3.9 HERE]
To summarize our findings on index threshold changes, an increase in the index eligibility threshold for investment grade firms reduces their propensity to issue bonds, but for those firms that do end up issuing it results in larger bond sizes. These findings suggest that while some firms ‘reach’ to be included in the index, other firms are discouraged from issuance altogether given the options of either issuing a small bond that cannot take advantage of passive demand or a bond that is larger than their target issuance amount.
17The details of these falsification tests and the regression results are not reported here in the interest of space, but are available upon request. Chapter 3. Debt issuance in the era of passive investment 67
3.5.4 Other financing decisions
In this section, we study the net effect of passive demand for corporate bonds on other capital structure activities. We examine the level of firm debt, the decision to issue equity, and the change in leverage ratios. While our simple model doesn’t speak to activity other than bond issuance, these firm choices may be indirectly influenced by passive demand in corporate bonds. We estimate the following regression:
capital structureit = β ∗ passive demand percit + controlsit + it, where capital structureit is the change in firm’s total debt level (Change in total debt), whether the firm issues equity in the quarter (Equity issuer dummy), or the change in the firm’s market leverage ratio (Change in market leverage).
[INSERT TABLE 3.10 HERE]
The results are reported in Table 3.10. Panel A shows that passive demand in corporate bonds is positively correlated with firms’ total debt levels. While not surprising given the positive effect on bond issuance propensity, this finding does show that firms are not simply substituting other forms of debt (such as bank debt) with corporate bonds. Panel B shows that passive demand in corporate bonds is negatively related to firms’ propensity to issue equity, indicating some form of substitution from equity issuance to bond issuance. This result is particularly noteworthy given that there is a similar temporal trend in passive demand in equity markets. As previously mentioned, however, it is easier for firms to take advantage of passive demand for bonds given the fact that any bond that meets the index criteria is included in a bond index; firms have much less influence on their inclusion in major equity indices such as S&P 500 or Russell 1000/2000. Finally, Panel C shows that market leverage ratios are negatively related to passive demand in corporate bonds. Given the increase in total debt seen in panel A, it appears that result is driven by the increase in the market value of assets (the denominator in the dependent variable).
3.6 Conclusions
In this paper, we examine the effects of passive investment on firms’ activity in the primary bond market. Investment in passive bond mutual funds and ETFs has increased drastically in the last decade, and many investment vehicles track a small number of bond indices that have well-defined eligibility criteria. We show that, in order to be exposed to passive investment funds, firms issue bonds of sufficient size to be included in the index, with clustering at index thresholds. Higher passive demand increases firms’ propensity to issue bonds in general, and firms are able to take advantage of it by improving bond terms that are irrelevant to index inclusion, such as lower spreads, fewer covenants, and longer maturities. These results are consistent with a model in which passive investors automatically buy index-eligible bonds, leaving less to be financed by active investors who evaluate the bond’s investment attractiveness based on its pricing and credit risk. We establish a causal link between passive demand and bond issuance by focusing on time windows around changes to eligibility criteria for popular bond indices. After an increase in the index threshold, firms’ propensity to issue a bond is temporarily decreased, though firms that do access the market sell larger bonds that meet the new higher criteria. We also show that firms issue a disproportionate number Chapter 3. Debt issuance in the era of passive investment 68 of bonds precisely at the threshold, with very few bonds immediately below it, and cluster at the new threshold when the index eligibility criteria are revised. An interesting extension of this research would be to study the effect of passive bond demand on firm’s real activities, such as corporate investment. Recent work on equity ETFs explores how passive investment affects stock price informativeness and subsequent investment (Li et al., 2018). Our model and empirical results show that passive investment in bonds influences firms’ cost of capital and access to bond markets, which has potential to be an even more important driver for firms’ investment decisions. We speculate that increased passive investment facilitates access to financing and thus investments. The differential impact among firms and over time is a potentially interesting area of research. Chapter 3. Debt issuance in the era of passive investment 69
3.7 Figures and Tables
Investment Grade Index Providers (1990-2017) 1 .8 .6 .4 .2 0 Bloomberg All Others
(a)
High Yield Index Providers (2007-2017) 1 .8 .6 .4 .2 0 Bloomberg iBoxx All Others
(b)
Figure 3.5: Proportion of total net assets following indices by Bloomberg, iBoxx and all other index providers combined. Chapter 3. Debt issuance in the era of passive investment 70 300 200 100 Total Net Assets (billions of $) 0 1992 1996 2000 2004 2008 2012 2016
IG HY
Figure 3.6: Monthly total net assets (in billions of US$) of passive funds invested in U.S. corporate bonds. 5 4 3 2 1 Index Face Value % Owned by Passive Funds 0 1992 1996 2000 2004 2008 2012 2016
IG HY
Figure 3.7: Monthly total net assets of passive funds invested in U.S. corporate bonds divided by the total face value of all bonds included in the relevant index. The investment grade index is the Bloomberg Barclays U.S. Corporate Index and the high yield index it the Bloomberg Barclays U.S. Corporate High Yield Index. Chapter 3. Debt issuance in the era of passive investment 71 40 30 25 20 30 20 15 20 10 10 10 5 0 0 0 -.5 0 .5 -.4 -.2 0 .2 .4 .6 -.5 0 .5 1
(a) (b) (c)
Figure 3.8: Results of the density test of McCrary (2008). The variable of interest is the dist to threshold, measured in billions of dollars, with a vertical line at $0 (or issuance at the threshold level). We include bonds within $500 million of the threshold. (a) Includes the full sample of bonds. (b) Sub-sample that includes investment grade bonds only. (c) Sub-sample that includes high yield bonds only. Chapter 3. Debt issuance in the era of passive investment 72
Increase from $50 to $100 million (1994) Increase from $100 to $150 million (1999)
Year before announcement Year before announcement 6 6 4 4 2 2 0 0 0 .1 .2 .3 .4 .5 0 .5 1 1.5
Year after effective date Year after effective date 3 10 2 5 1 0 0 0 .1 .2 .3 .4 .5 0 .5 1 1.5
(a) (b)
Increase from $150 to $200 to $250 million (2003 and 2004) Increase from $250 to $300 million (2017)
Year before first announcement Six months before announcement 4 4 3 2 3 1 0 0 .5 1 1.5 2 1 Time between first effective date and second announcement 0
4 0 .5 1 1.5 3 2
1 Six months after effective date 0
0 .5 1 1.5 4
Year after second effective date 3 5 2 4 3 1 2 1 0 0 0 .5 1 1.5 0 .5 1 1.5
(c) (d)
Figure 3.9: Histograms of investment grade issuance in short time frames before and after index threshold changes to the Bloomberg Barclays U.S. Corporate Index (excluding the time between announcement and effective date of the changes); only issuance up to $1.5 billion in face value shown. Each graph includes a vertical line at the issuance threshold of the time. Bins represent $50 million increments. (a) Change in index threshold from $50 to $100 million effective January 1, 1994 (assumed to be announced October 1, 1993). (b) Change in index threshold from $100 to $150 million announced February 24, 1999 (effective July 1, 1999). (c) Changes in index threshold from $150 to $200 million announced June 17, 2003 (effective October 1, 2003) and from $200 to $250 million announced March 18, 2004 (effective July 1, 2004). (d) Change in index threshold from $250 to $300 million announced January 24, 2017 (effective April 1, 2017). Chapter 3. Debt issuance in the era of passive investment 73
Increase from $100 to $150 million (Bloomberg HY-2000) Increase from $150 to $200 to $400 million (iBoxx-2007 & 2009)
Year before announcement Year before first announcement 4 4 3 2 3 1 0
2 0 .2 .4 .6 .8 1 1 Time between first effective date and second announcement 0 0 .2 .4 .6 .8 1 3 2
Year after effective date 1 0 5 0 .2 .4 .6 .8 1 4
3 Year after second effective date 4 2 3 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1
(a) (b)
Figure 3.10: Histograms of high yield issuance in short time frames before and after index threshold changes (excluding the time between announcement and effective date of the changes). Issuance up to $1 billion in face value. Each graph includes a vertical line at the issuance threshold of the time. Bins represent $50 million increments. (a) Change in Bloomberg Barclays U.S. Corporate High Yield index threshold from $100 to $150 million effective July 1, 2000 (assumed to be announced April 1, 2000). (b) Introduction of Markit iBoxx Liquid HY index in April 2007 (assumed to be announced April 1, 2007) with $200 million threshold, and change in index threshold from $200 to $400 million effective July 1, 2009 (assumed to be announced April 1, 2009). Chapter 3. Debt issuance in the era of passive investment 74
Table 3.1: Descriptive statistics
Panel A: Characteristics of Bond Issuer Sample N Mean Std. Dev. Min p25 Median p75 Max Passive demand Passive demand (percent) 15,272 1.145 1.321 0.000 0.000 0.466 2.255 5.123 Bond level passive demand (perc.) 15,272 0.229 0.445 0.000 0.000 0.001 0.237 2.117 Bond characteristics Issue size 16,691 430,541 499,675 25,000 150,000 300,000 500,000 15,000,000 Spread 15,935 2.701 2.183 0.096 0.939 2.015 4.020 9.663 Covenant score 10,221 11.421 7.392 0.000 7.000 9.167 12.833 41.167 Initial maturity 16,690 11.169 9.047 0.835 6.995 9.988 10.053 100.424 Index thresholds Threshold amount 14,748 205,519 108,105 25,000 100,000 150,000 250,000 400,000 Firm characteristics Average rating 14,801 10.278 4.256 1.0 7.0 9.5 14.5 22.0 IG dummy 15,272 0.562 0.496 0 0 1 1 1 Rated dummy 16,691 0.887 0.317 0 1 1 1 1 Log assets 12,129 8.743 1.709 -2.112 7.582 8.826 9.960 13.543 Market leverage 10,157 0.304 0.204 0.000 0.147 0.257 0.423 0.948 Age 12,397 18.451 20.440 0.000 6.500 13.084 23.253 170.000 Tangibility 12,067 0.392 0.260 0.000 0.169 0.347 0.590 0.920 Profitability (net income) 12,113 0.009 0.043 -1.334 0.003 0.012 0.020 0.176 Q ratio 10,646 1.108 0.635 0.392 0.712 0.929 1.297 8.929 Macro variables 10-year yield 16,690 4.462 1.810 1.380 2.670 4.580 5.900 9.090 Term slope 16,690 1.241 0.882 -0.490 0.390 1.320 2.000 2.910 BAA AAA spread 16,690 0.920 0.371 0.500 0.680 0.850 1.030 3.500 NBER recession dummy 16,691 0.080 0.272 0 0 0 0 1
Panel B: Characteristics of Full Compustat Sample N Mean Std. Dev. Min p25 Median p75 Max Passive demand Passive demand (percent) 406,733 0.807 1.101 0.000 0.000 0.173 1.298 4.905 Issuer characteristics Issuer dummy 654,190 0.013 0.115 0.000 0.000 0.000 0.000 1.000 Average offering 8,787 381,689 367,670 25,000 150,000 275,000 500,000 7,500,000 Firm characteristics IG dummy (estimate) 411,873 0.448 0.497 0 0 0 1 1 Rated dummy 654,190 0.115 0.319 0 0 0 0 1 Log assets 598,451 4.444 2.735 -6.908 2.748 4.535 6.296 13.649 Market leverage 517,540 0.219 0.249 0.000 0.008 0.125 0.351 0.948 Age 654,190 13.528 16.925 0.000 3.754 8.999 17.000 175.000 Tangibility 596,586 0.259 0.240 0.000 0.070 0.180 0.381 0.920 Profitability (net income) 594,871 -0.108 0.484 -3.911 -0.043 0.003 0.018 0.175 Q ratio 535,421 2.019 3.446 0.392 0.706 1.046 1.849 27.264 Chapter 3. Debt issuance in the era of passive investment 75
IG Change 1 IG Change 2 IG Change 1 IG Change 2 .04 .05 250 350 .04 .03 300 200 .03 .02 250 150 .02 .01 200 .01 0 100 150 1991m7 1992m7 1993m7 1994m7 1995m7 1997m1 1998m1 1999m1 2000m1 2001m1 1992q1 1993q1 1994q1 1995q1 1996q1 1997q3 1998q3 1999q3 2000q3 2001q3
IG Change 3 IG Change 4 IG Change 3 IG Change 4 .04 600 400 .035 .03 350 .03 550 .025 300 .02 .02 500 250 .015 .01 .01 450 200 2001m1 2002m1 2003m1 2004m1 2005m1 2006m1 2015m1 2015m7 2016m1 2016m7 2017m1 2017m7 2001q3 2002q3 2003q3 2004q3 2005q3 2006q3 2015q3 2016q1 2016q3 2017q1 2017q3
IG IG HY HY
(a) (b)
HY Change 1 HY Change 2 HY Change 1 HY Change 2 .04 .04 400 500 450 .03 .03 350 400 300 .02 .02 350 250 .01 .01 300 200 250 0 0 1998m1 1999m1 2000m1 2001m1 2002m1 2005m12006m12007m12008m12009m12010m12011m1 1998q3 1999q3 2000q3 2001q3 2002q3 2005q3 2007q1 2008q3 2010q1 2011q3
IG IG HY HY
(c) (d)
Figure 3.11: Graphs of parallel trends in bond sizes (panels (a) and (c)) and propensity to issue (panels (b) and (d)). Panels (a) and (b) examine the four changes in the investment grade index while panels (c) and (d) examine the two changes in the high yield indices (described in detail in Figures 3.9 and 3.10). The graphs show the average offering size (in millions) and the average issuance propensity for the two years before and after each change; the vertical lines indicate the timing of the announcement of the change. Data are smoothed using the locally weighted scatterplot smoothing (LOWESS) with a bandwidth of 0.5 for the offering size (monthly data) and a bandwidth of 0.8 for the issuance propen- sity (quarterly data). Note that the definition of the post variable includes only one year before the announcement date and one year after the effective date, excluding any period of time in between the two dates. As such, these graphs are not directly comparable to the difference-in-difference regressions. Chapter 3. Debt issuance in the era of passive investment 76 ). Dependent variable in columns Passive demand (0.338) (0.336)(0.291) (0.291) (0.910) (0.914) (0.663) (0.658) (0.0717)(0.0112) (0.0713) (0.0111)(0.0186) (0.0186) (0.108) (0.0194) (0.105) (0.0192) (0.0253) (0.0253) (0.00234) (0.00233) (0.00591) (0.00591) (0.000455) (0.000457) (0.000968) (0.000959) Log Issue Size Spread (0.0139)(0.0555) (0.0133) (0.0685) (0.0134) (0.0686) (0.0161) (0.0178) (0.151) (0.0177) (0.160) (0.159) . All regressions include quarter fixed effects and industry fixed effects. Additional macro controls Spread Table 3.2: Passive demand, bond size, and credit spread (0.0330) (0.0342) (0.0369) (0.0368) (0.130) (0.118) (0.128) (0.128) (0.00481) (0.00649) (0.00693) (0.00696) (0.00955) (0.0110) (0.0132) (0.0131) and in (5)-(8) is Log issue size Market to bookZ-scoreEarnings volatilityAdditional macro controlsFISD Industry FESIC2 Industry FEQuarter FE NoObservationsR-squaredCluster Yes No 0.0638*** No 0.0639*** 0.292 No 14,801 No Yes 0.00563** Yes 0.451 0.00552** 9,147 0.294 Quarter No Yes Yes Yes 0.559 Quarter 6,088 Yes Quarter No 0.595 Yes 6,088 No Quarter 0.0619*** Yes 0.596 0.0600*** No Yes 14,195 Quarter No 0.0331*** Quarter 8,677 0.722 0.0334*** 2.347*** Yes No No Quarter Yes 2.383*** 0.743 5,748 Quarter Yes No Yes Yes 0.755 5,748 Yes No Yes 0.756 Yes Average ratingLog assets -0.0180*** 0.0421***Profitability (net income) 0.0426***Tangibility 0.0426*** 0.377*** 0.270*** 0.238*** 0.262*** 0.255***Tax 0.263*** rate 0.256*** 0.0158 0.0303 -0.112*** -0.220*** -0.117*** -0.220*** -0.117*** -0.0469** -3.107*** -0.0475** -3.142*** 0.355*** 0.349*** 0.0482* 0.0500* VARIABLESPassive demand (percent) 0.165*** 0.106*** (1) 0.104***Market leverage 0.104***Age (2) -0.684*** -0.623*** (3) -0.629*** -0.630*** -0.209*** (4) 0.00176 0.00158 (5) (6) 0.000130 1.576*** (7) 0.000133 1.928*** 1.924*** (8) 0.000489 0.000527 include the 10-year yield,the BAA 10%, AAA 5% spread, and and 1% term level, respectively. slope. Standard errors are clustered at the quarter level. *,**,*** indicate significance at (1)-(4) is Regressions of bond characteristics on percentage of bond index value held by passive funds ( Chapter 3. Debt issuance in the era of passive investment 77 ). Dependent variable in columns Passive demand (6.291)(0.750) (6.304)(0.191) (0.750)(0.144) (0.191)(4.578) (0.144)(0.138) (4.594) (0.137) (3.002) (0.903) (3.004) (0.224) (0.905) (0.0270) (0.225) (0.0270) (1.961) (0.207) (1.936) (0.208) (0.00649) (0.00648) (0.00609) (0.00610) . All regressions include quarter fixed effects and industry fixed effects. Additional Covenant Score Initial Maturity (0.712) (0.984) (0.988) (0.587) (0.685) (0.687) (0.0977) (0.107) (0.107) (0.115) (0.147) (0.147) Initial maturity Table 3.3: Passive demand and other bond characteristics (0.535) (0.466) (0.632) (0.633) (0.376) (0.396) (0.427) (0.425) (0.0823) (0.0731) (0.0832) (0.0831) (0.0616) (0.0602) (0.0599) (0.0596) and in (5)-(8) is Covenant score Age -0.00415 -0.00434 -0.00231 -0.00224 VARIABLESPassive demand (percent) -0.145Log assets (1) -0.721Market leverage -0.645 (2)Profitability -0.644 (net income)Tangibility (3) 1.006***Market to book 0.505Z-score (4) -0.492*** -0.804 -0.448***Earnings 0.921** volatility -0.448*** -1.524Tax 0.916** rate (5) -1.525 -1.744Additional macro controlsFISD Industry FE (6) -1.642SIC2 Industry FE NoQuarter FE 0.529***ObservationsR-squared 0.510*** (7)Cluster -1.009 No 0.511*** Yes -0.176 -0.785 No -1.000 (8) -7.203 -0.176 No -1.124 No 8,840 Yes Yes 0.170 -7.206 -1.129 0.361 6,118 No Yes Yes Quarter -0.434*** 0.171 Yes -0.839 0.374 -0.440*** Quarter 4,050 No Yes -0.773 Quarter Yes 0.405 No 4,050 Quarter Yes 0.405 0.208 No Yes Quarter 0.0658 14,801 No Quarter 0.208 0.103 0.0637 -1.673 9,147 Yes No No Quarter Yes -0.0248 -1.643 0.116 Quarter 6,088 -0.192 Yes No Yes -0.0255 Yes 0.116 -0.197 6,088 Yes No Yes 0.116 Yes Average rating 0.939*** 0.852*** 0.931*** 0.930*** -0.404*** -0.331*** -0.258*** -0.258*** (1)-(4) is macro controls include thesignificance 10-year at yield, the BAA 10%, AAA 5% spread, and and 1% term level, slope. respectively. Standard errors are clustered at the quarter level. *,**,*** indicate Regressions of bond characteristics on percentage of bond index value held by passive funds ( Chapter 3. Debt issuance in the era of passive investment 78 ). All regressions include quarter fixed Passive demand Table 3.4: The propensity to issue bonds (0.00102) (0.000788) (0.00100)(0.00145) (0.000944) (0.00147) (0.00111) (0.00162) (0.000906) (0.00190) (0.000424) (0.000409) (0.000311) (0.000462) on percentage of bond index value held by passive funds ( VARIABLESPassive demand (percent) 0.00857***Log assets 0.00449***Market leverage 0.00923*** (1) 0.00495***Z-Score controls 0.00405***EJKW controls 0.00408*** LR controls 0.00701*** (2) 0.0102***SIC2 FE -0.00353**Firm FE 0.00496***Quarter -0.00460*** FE 0.00555***Observations -0.00553*** YesR-squared (3) 0.00503*** Cluster No No Yes (4) No Yes 378,164 Yes No No No 0.046 377,853 (5) No Yes Yes Firm Yes 301,445 0.128 No No (6) 301,015 Firm Yes Yes Yes No 0.072 272,415 No No Firm No No Yes 0.133 272,027 Yes Yes Firm No 0.061 No Yes Yes No Firm No 0.125 No Yes Yes Firm Issuer dummy Regressions of effects; regressions in columnsZ-score (1), controls (3), include and working (5) capitalto include to total industry total assets. assets, fixed retained effectsgrowth, EJKW earnings while equity to controls return, regressions total include in term assets,and columns slope, firm EBIT amortization, (2), and age, to (4) tangibility, NBER assets, market and profitability marketmarket recession leverage, (6) equity (net leverage. dummy. q-ratio, include to income), LR Standard total firm tangibility, controls earnings liabilities, errors fixed cash include and effects. are volatility, flow, size, sales z-score, clustered cash, market selling at to inverse the expense, book, interest firm equity ratio, capital level. return, expenditure, rated *,**,*** book cash, dummy, indicate sales depreciation leverage, significance and at change the in 10%, 5% and 1% level, respectively. Chapter 3. Debt issuance in the era of passive investment 79 ). Dependent . All regressions Initial maturity Bond-level passive demand , and in (7)-(8) is Covenant score , in (5)-(6) is Spread (0.487)(0.176)(0.437) (0.815) (0.232) (4.448) (0.675) (1.733) (4.360) (4.971) (2.260) (3.214) (0.101) (0.236) (1.571) (1.542) (0.0179)(0.0181) (0.0316) (0.0314) (0.238) (0.158) (0.523) (0.246) (0.0297)(0.0110) (0.0450) (0.0158) (0.310) (0.0908) (0.309) (0.107) (0.00268) (0.00351)(0.00322) (0.0596) (0.00634) (0.0238) (0.169) (0.0417) Table 3.5: Passive demand at the bond level , in (3)-(4) is Log Issue Size Spread Covenant Score Initial Maturity (0.0369) (0.0406) (0.102) (0.0631) (0.272) (0.219) (0.419) (0.541) Log issue size TangibilityMarket to bookZ-scoreTax rateAdditional macro controlsFirm FEMonth FEObservationsR-squaredCluster -0.415** 0.0667*** No 0.0105*** Yes 0.673*** 0.0474 13,989 -0.0434** Yes Yes No 0.707 5,652 Quarter 0.0171*** Yes Yes 0.741 Yes Quarter -4.150** 13,344 -0.163 0.00506 Yes Quarter Yes 5,312 0.806 No Quarter 0.0905 0.870 Yes Yes 8,260 Yes Quarter -1.117 Quarter 3,696 -0.159 0.719 0.376 Yes Yes Quarter No 0.777 13,989 0.0514 Quarter Yes Yes Yes 5,652 0.334 -0.574** Yes Yes 0.349 Yes Yes Profitability (net income)Earnings volatility 0.206 -1.263 0.906** 3.765 1.438** -5.971 1.166 -2.585 Market leverageAverage ratingAge -0.0266 0.00195 0.00457* 2.329*** 0.263*** -2.336 -0.00266 0.539*** 0.0516 -3.203** -0.551*** 0.000681 VARIABLESBond level passive demand (percent)Log assets 0.164*** 0.0925** -0.880*** -0.674*** (1) 0.361 (2) 0.327 -8.134*** (3) -11.06*** 0.208*** (4) (5) -0.126*** (6) -0.625** (7) (8) -0.284 variable in columns (1)-(2) is Regressions of bond characteristics on percentage of face value of a given bond purchased by passive funds ( include month fixed effects anderrors firm are fixed clustered effects. at Additional the macro quarter controls include level. the *,**,*** 10-year yield, indicate BAA significance AAA at spread, the and term 10%, slope. 5% and Standard 1% level, respectively. Chapter 3. Debt issuance in the era of passive investment 80 Treated ) takes a value of 0 in the 12 Post HY ( Post (IG) -0.113 0.186 0.273 0.261 (0.119) (0.134) (0.179) (0.181) 0.223** 0.235** 0.179 0.215 (0.0719) (0.113)(0.0805) (0.112) (0.112) (0.145) (0.111) (0.106) (0.144) (0.151) (0.159) (0.151) (0.173) (0.158) around changes in the index threshold. (0.0300) (0.0378) (0.0377) (0.0313) (0.0334) (0.0329) Table 3.6: Index threshold changes and issue size Log issue size (0.0101) (0.0166) (0.0212) (0.0208) (0.0211) (0.0238) (0.0326) (0.0329) (0.0285)(0.0668) (0.0432)(0.0629) (0.0596) (0.0975) (0.0735) (0.0917) (0.0856) (0.0930) (0.0922) (0.0857) ) takes a value of 1 for investment grade (high yield) issuers and 0 for high yield (investment grade) issuers. Columns (1)-(4) Treated (HY) ( Graham&Leary controlsAdditional macro controlsChangexFISD Industry FEChangexSIC2 FEObservations NoR-squared NoCluster YesSample No No No No 3,476 Yes No No 0.312 FISD Yes Industry 2,028 Yes IG SIC2 Yes Chgs No Yes 0.452 1,266 SIC2 IG Chgs 0.534 No IG 1,265 Yes Chgs No Yes SIC2 IG 0.537 Chgs FISD Industry 2,012 No HY No Chgs No No SIC2 0.196 HY Chgs Yes 1,179 SIC2 No HY No Chgs Yes 0.298 HY Chgs 783 Yes SIC2 Yes 0.340 No Yes 783 0.348 Yes VARIABLESPost (IG)Treated (IG)Post (IG)*Treated (IG)Post (HY) (1)Treated (HY) 0.118*Post 0.0203 (HY)*Treated (HY) -0.0788 (2)Average 0.0974 rating 0.0530 -0.223**Log 0.241** assets (3) -0.140 0.0141 0.237*** Market leverage -0.142 0.0702 (4) -0.0186* 0.0366** (5) 0.0565** 0.0576*** -0.0164 (6) -0.335** 0.214*** 0.251*** 0.00557 -0.187 (7) 0.249*** -0.0115 -0.232 -0.0116 (8) -0.0804 -0.195* 0.171*** -0.157 0.156*** -0.151 -0.146 0.154*** -0.0778 -0.0211 Difference-in-difference regressions of months before an investment grade (high yield) index change is announced and a value of 1 in the 12 months after the change is effective. examine the effectregressions across include all change-by-industry investmentbook, fixed grade z-score, effects. earnings index volatility, Graham&Learyerrors changes and controls are tax and clustered include rate. columns at firm Additional the (5)-(8) age, macro industry examine controls profitability level. include the (net *,**,*** the effect income), indicate 10-year tangibility, significance across yield, market at BAA all to AAA the high spread, 10%, and yield 5% term and index slope. 1% changes. level, Standard respectively. All (IG) Chapter 3. Debt issuance in the era of passive investment 81
Table 3.7: Index threshold changes and issue size by predicted issue size bucket
Difference-in-difference regressions of Log issue size around changes in the index threshold. Post (IG) (Post HY ) takes a value of 0 in the 12 months before an investment grade (high yield) index change is announced and a value of 1 in the 12 months after the change is effective. Treated (IG) (Treated (HY)) takes a value of 1 for investment grade (high yield) issuers and 0 for high yield (investment grade) issuers. For this analysis, bond sizes are predicted based on log assets, rating, and industry and change fixed effects. Issuers are then classified in one of two buckets: predicted amount above the new threshold (‘large’ bonds - columns (1) and (3)), or predicted amount above the old threshold but below the new threshold (‘medium’ bonds - columns (2) and (4)). Standard errors are clustered at the industry level. Change and industry fixed effects included. *,**,*** indicate significance at the 10%, 5% and 1% level, respectively.
VARIABLES (1) (2) (3) (4)
Post (IG) 0.0473 0.163 (0.0339) (0.164) Treated (IG) -0.00814 -0.294 (0.0780) (0.297) Post (IG)*Treated (IG) 0.0903 0.402** (0.0904) (0.180) Post (HY) 0.0626 0.753*** (0.122) (0.168) Treated (HY) -0.179* -0.567** (0.0927) (0.213) Post (HY)*Treated (HY) 0.131 0.370 (0.111) (0.317)
Change FE Yes Yes Yes Yes SIC2 FE Yes Yes Yes Yes Observations 2,250 201 1,156 211 R-squared 0.335 0.417 0.200 0.448 Cluster SIC2 SIC2 SIC2 SIC2 Bucket Large Medium Large Medium Chapter 3. Debt issuance in the era of passive investment 82 ) takes a value Treated (HY) ( Treated (IG) ) takes a value of 0 in the 12 months before an Post HY ( (0.00134) (0.00205)(0.00155) (0.00174) (0.00168)(0.00183) (0.00161) (0.00217) (0.00223) Post (IG) around changes in the index threshold. (0.00180) (0.00179)(0.00165) (0.00193) (0.00195) (0.00199) (0.00216) (0.00242) (0.00235) (0.00273) (0.000998) (0.00135) (0.00116) (0.000488) (0.000368) (0.000462) (0.000370) Table 3.8: Index threshold changes and bond issuance Issuer dummy VARIABLESPost (IG)Treated (IG)Post (IG)*Treated (IG) (1)Post (HY) -0.00241 -0.00811***Treated -0.00842*** 0.0102*** (HY) (2) -0.00223 -0.00746*** 0.00874***Post (HY)*Treated (HY) -0.00393** 0.00613*** Log assets (3)Market leverageZ-Score controls (4)EJKW controlsLR 0.00952*** 0.0121*** controlsChangexSIC2 FE 0.00524***Observations 0.000246 (5)R-squaredCluster YesSample No Yes (6) No 0.00858*** No 96,541 0.00662*** -0.00111 0.00467*** Yes 0.00448*** -0.00573*** Yes 0.045 -0.00346 80,423 -0.00735*** 0.00134 0.00158 No No Firm IG -0.00138 Chgs No 0.00522*** 0.000654 0.074 68,944 Yes IG Chgs Yes Firm No 59,797 0.057 No IG Chgs Yes Firm No 49,686 HY Chgs 0.038 Yes No Yes HY Chgs Firm 44,286 0.060 No No HY Chgs Yes Yes Firm 0.056 No Firm investment grade (high yield) index change is announced and a value of 1 in the 12 months after the change is effective. Difference-in-difference regressions of of 1 forearnings investment grade to (high total yield)tangibility, assets, cash issuers EBIT flow, and to cash, 0market assets, inverse to for market interest book, high equity ratio, capital yield toreturn, rated expenditure, book (investment total cash, dummy, grade) sales leverage, depreciation liabilities, and and issuers. growth,respectively. and amortization, change equity sales tangibility, in Z-score return, profitability to market controls (net term leverage. total income), include slope, Standard assets. earnings working and errors volatility, capital z-score, NBER EJKW are selling to recession clustered controls expense, total dummy. at include equity the assets, LR firm firm retained controls age, level. include market *,**,*** size, leverage, indicate q-ratio, significance at the 10%, 5% and 1% level, Chapter 3. Debt issuance in the era of passive investment 83
Table 3.9: Index threshold changes and bond rollovers
Difference-in-difference regressions of Rollover dummy around changes in the index threshold. Post (IG) (Post HY ) takes a value of 0 in the 12 months before an investment grade (high yield) index change is announced and a value of 1 in the 12 months after the change is effective. Treated (IG) (Treated (HY)) takes a value of 1 for investment grade (high yield) issuers and 0 for high yield (investment grade) issuers. Treated (IG) (Treated (HY)) in columns (2) and (4) take a value of 1 for issuers rated investment grade (high yield) at the time of bond maturity and 0 for high yield (investment grade) issuers. Standard errors are clustered at the industry level. *,**,*** indicate significance at the 10%, 5% and 1% level, respectively.
VARIABLES (1) (2) (3) (4)
Post (IG) -0.00218 0.0128 (0.0295) (0.0418) Treated (IG) at issue 0.235*** (0.0526) Post (IG)*Treated (IG) at issue -0.0744* (0.0395) Treated (IG) at expiry 0.212*** (0.0560) Post (IG)*Treated (IG) at expiry -0.0961** (0.0474) Post (HY) 0.0243 0.0216 (0.0432) (0.0407) Treated (HY) at issue -0.0876** (0.0381) Post (HY)*Treated (HY) at issue 0.00376 (0.0501) Treated (HY) at expiry -0.0395 (0.0542) Post (HY)*Treated (HY) at expiry -0.0195 (0.0606)
ChangexSIC2 FE Yes Yes Yes Yes Observations 1,145 1,105 878 829 R-squared 0.306 0.288 0.223 0.213 Cluster SIC2 SIC2 SIC2 SIC2 Sample IG Chgs IG Chgs HY Chgs HY Chgs Chapter 3. Debt issuance in the era of passive investment 84
Table 3.10: Passive bond demand and other other financing decisions
Regressions of capital structure activities and leverage on percentage of bond index value held by passive funds (Passive demand). As dependent variables, panel A uses Change in total debt, panel B uses a Equity issuer dummy, and panel C uses Change in market leverage. All regressions include quarter fixed effects; regressions in columns (1), (3) and (5) include industry fixed effects while regressions in columns (2), (4) and (6) include firm fixed effects. See Table 3.4 for description of controls. Standard errors are clustered at the firm level. *,**,*** indicate significance at the 10%, 5% and 1% level, respectively.
Panel A: Change in Total Debt VARIABLES (1) (2) (3) (4) (5) (6) Passive demand (perc.) 12.94*** 11.17*** 25.79** 26.91** 9.948*** 13.61*** (3.182) (3.121) (12.95) (13.60) (2.973) (3.554) Log assets 7.966*** 4.273*** 12.87** 14.58* (0.823) (1.382) (5.215) (7.788) Market leverage -28.97*** -58.16*** 23.29 -40.93* (2.244) (4.615) (48.05) (22.05) Observations 365,569 365,199 291,987 291,542 272,198 271,807 R-squared 0.006 0.036 0.005 0.158 0.011 0.037 Panel B: Equity Issuance Dummy VARIABLES (1) (2) (3) (4) (5) (6) Passive demand (perc.) -0.00608*** -6.51e-05 -0.0135*** 0.000820 -0.00255 0.00290 (0.00197) (0.00194) (0.00202) (0.00199) (0.00199) (0.00201) Log assets -0.0166*** -0.0255*** -0.0122*** -0.0288*** (0.000497) (0.00133) (0.000545) (0.00149) Market leverage -0.00575 0.00343 0.000541 0.000921 (0.00395) (0.00486) (0.00427) (0.00535) Observations 365,800 365,424 292,214 291,761 272,282 271,892 R-squared 0.052 0.176 0.060 0.177 0.054 0.172 Panel C: Change in Market Leverage VARIABLES (1) (2) (3) (4) (5) (6) Passive demand (perc.) -0.413*** -0.472*** -0.428*** -0.457*** -0.373*** -0.424*** (0.0305) (0.0401) (0.0332) (0.0435) (0.0350) (0.0455) Log assets -0.0564*** 0.475*** -0.0505*** 0.418*** (0.00547) (0.0250) (0.00708) (0.0292) Market leverage -3.480*** -9.552*** -3.925*** -10.28*** (0.0646) (0.126) (0.0801) (0.152) Observations 363,063 362,685 289,892 289,463 271,182 270,797 R-squared 0.066 0.131 0.072 0.142 0.079 0.146 Z-Score controls Yes Yes No No No No EJKW controls No No Yes Yes No No LR controls No No No No Yes Yes SIC2 Industry FE Yes No Yes No No No Firm FE No Yes No Yes Yes Yes Quarter FE Yes Yes Yes Yes Yes Yes Cluster Firm Firm Firm Firm Firm Firm Chapter 3. Debt issuance in the era of passive investment 85
3.8 Appendices
Appendix A Variable definitions Dependent variables Variable Definition Source Log issue size Log of the total bond face value. FISD Spread Offering yield (or coupon directly if offering yield is not available) minus Offering yield: FISD; the applicable risk-free rate. The applicable risk-free rate is linearly inter- risk-free rate: Federal polated between Treasury bonds (using the 1, 2, 3, 5, 7, 10, 20 and 30 year Reserve Bank of St. bonds as available) based on the bond’s initial time to maturity. For bonds Louis with an initial maturity less than one year or greater than the longest avail- able government bond, we take the risk-free rate as the yield on the 1-year Treasury or on the longest Treasury bond, respectively. Negative spreads are removed. Winsorized at the 1% and 99% levels. Covenant score cov score = (0.25∗RP +0.1∗RI+0.25∗DEBT +0.2∗LIEN +0.1∗SS+0.1∗ Formula: adapted COC) ∗ 100, where where RP stands for Restricted Payments, RI stands from Moody’s for Risky Investments, DEBT represents Debt Incurrence, LIEN represents Covenant Quality Liens Subordination, SS stands for Structural Subordination and COC Index; covenants: stands for Change of Control. Covenants are classified into each bucket FISD and the bond receives a category score equal to the number of covenants included divided by the total number of covenants in the category. Ex- amples of each category: “restricted payments” for RP, “investments” for RI, “indebtedness” for DEBT, “negative pledge covenant” for LIEN, “fixed charge coverage” for SS, and “change control put provisions” for COC. Initial maturity Maturity date minus offering date, divided by 365.25. FISD Issuer dummy Dummy that takes on a value of 1 if the firm issued a bond in the following FISD financial quarter. Rollover dummy Dummy that takes on a value of 1 if a firm issues one or more bonds with FISD total face value at least as high as the face value of an expiring bond within six months of the bond’s expiry date. Change in total debt Change in quarterly total debt. Compustat Equity issuer dummy Dummy that takes on a value of 1 if a firm’s number of common shares Compustat outstanding increases by at least 5% over the previous quarter. Change in market leverage Change in quarterly market leverage ratio. Winsorized at the 1% and 99% Compustat levels.
Independent variables Variable Definition Source Passive demand (percent) Total value of net assets invested in passive corporate bonds divided by CRSP Survivor-Bias- total face value of bonds eligible to be included in the index. Separated Free U.S. Mutual Fund by investment grade (uses Bloomberg Barclays U.S. Corporate Index) and database (total net high yield (uses Bloomberg U.S. Corporate High Yield Index). assets) and FISD (face value of index) Bond level passive demand Percentage of face value of a bond purchased by passive funds. Sum across Index rules (dummy (percent) all passive funds of the product of (i) a dummy for inclusion in an index for inclusion), CRSP at the time of bond issuance and (ii) the fund’s total net assets divided by Survivor-Bias-Free index face value at the time of bond issuance. Winsorized at the 1% and U.S. Mutual Fund 99% level. database (total net assets) and FISD (face value of index) Log threshold Log of the rating-appropriate index threshold. Bloomberg (IG) and iBoxx (HY) Chapter 3. Debt issuance in the era of passive investment 86
Post (IG) Dummy that takes on a value of 0 for the period 15 months to 3 months Bloomberg IG index before an investment grade index threshold change and a value of 1 for the 12 months after the change. Post (HY) Dummy that takes on a value of 0 for the period 15 months to 3 months Bloomberg and iBoxx before a high yield index threshold change and a value of 1 for the 12 HY indices months after the change. Treated (IG) & Treated Dummy that takes on a value of 1 for investment grade (high yield) bonds Bloomberg and iBoxx (HY) with respect to investment grade (high yield) index changes, and 0 for high yield (investment grade) bonds.
Controls Variable Definition Source Average rating Simple average of the initial ratings assigned to the bond by Moody’s, Fitch FISD and Standard & Poor’s (converted to a common inverted numerical scale, i.e. AAA bonds receive a rating of 1, bonds in default receive a rating of 22). If a particular bond does not have an initial rating that meets this criterion, the initial rating is assumed to be the same as the issuer’s most recent rating. IG dummy Dummy that takes on a value of 1 if the avgrating is less than or equal to FISD 10 (corresponds to BBB-/Baa3). IG dummy (estimate) Dummy that takes on a value of 1 if the firm is estimated to be investment Compustat and FISD grade and 0 otherwise. Investment grade estimation is done by regressing investment grade status for rated firms on wc2ta, re2ta, ebit2ta, me2tl and s2ta, predicting the value for all firms and assigning a cutoff value that correctly classifies the highest percentage of rated observations. Rated dummy Dummy that takes on a value of 1 if the issuer has a bond rating. FISD Log assets Log of the firm’s total assets. Compustat Size Sales divided by the total sales of all firms in a quarter (variable multiplied Compustat by 1000). Winsorized at the 1% and 99% levels. Market leverage Total debt divided by total debt plus market value of common equity, Compustat measured as of the most recent quarter before the offering date. Winsorized at the 1% and 99% levels. Book leverage Total debt divided by book value of assets, measured as of the most recent Compustat quarter before the offering date. Winsorized at the 1% and 99% levels. Volatility Trailing 250 day stock price volatility. Winsorized at the 1% and 99% CRSP levels. Profitability (operating in- Trailing twelve month income before extraordinary items divided by total Compustat come) assets. Winsorized at the 1% and 99% levels. Tangibility Net property, plant and equipment divided by total assets. Winsorized at Compustat the 1% and 99% levels. Q ratio Total market value of common equity plus book value of liabilities divided Compustat by book value of assets and liabilities. Tax rate Trailing twelve month taxes payable dividend trailing twelve month pre-tax Compustat income, measured as of the most recent quarter before the offering date. Winsorized at the 1% and 99% levels. Age Where applicable, financial reporting year less founding year taken the Field-Ritter dataset, Field-Ritter dataset of company founding dates (Field and Karpoff (2002), Compustat Loughran and Ritter (2004)). Otherwise, financial reporting date minus first recorded reporting date, divided by 365.25. Market to book Book value of assets minus book value of equity plus market value of equity, Compustat divided by book value of assets. Winsorized at the 1% and 99% levels. Profitability (net income) Quarterly net income divided by total assets. Winsorized at the 1% and Compustat 99% levels. Chapter 3. Debt issuance in the era of passive investment 87
Earnings volatility Absolute change in quarterly net income, divided by total assets. Win- Compustat sorized at the 1% and 99% levels. Cash Cash and short-term investments divided by total assets. Winsorized at Compustat the 1% and 99% levels. Capex Capital expenditures. Compustat Depreciation & amortiza- Depreciation and amortization divided by total assets. Winsorized at the Compustat tion 1% and 99% levels. Equity return Cumulative four quarter stock price return. Winsorized at the 1% and 99% Compustat levels. Z score z score = 1.2 ∗ wc2ta + 1.4 ∗ re2ta + 3.3 ∗ ebit2ta + 0.6 ∗ me2tl + 0.999 ∗ s2ta. Compustat Calcualted using raw values; result winsorized at the 1% and 99% levels. Working capital to assets Current assets less current liability, divided by total assets. When used as Compustat a direct variable, winsorized at the 1% and 99% levels. Retained earnings to assets Retained earnings divided by total assets. When used as a direct variable, Compustat winsorized at the 1% and 99% levels. EBIT to assets EBITDA minus depreciation and amortization, divided by total assets. Compustat When used as a direct variable, winsorized at the 1% and 99% levels. Market equity to liabilities Market value of equity divided by liabilities. When used as a direct vari- Compustat able, winsorized at the 1% and 99% levels. Sales to assets Sales divided by total assets. When used as a direct variable, winsorized Compustat at the 1% and 99% levels. R&D to sales Research and development expenses divided by sales. Winsorized at the Compustat 1% and 99% levels. Selling expense Selling, general and administrative expenses divided by total assets. Win- Compustat sorized at the 1% and 99% levels. Cash flow Income before extraordinary items plus depreciation and amortization, di- Compustat vided by total assets. Winsorized at the 1% and 99% levels. Inverse interest ratio The natural logarithm of 1 plus interest paid divided by pretax income Compustat plus interest. Winsorized at the 1% and 99% levels. Sales growth Trailing twelve month sales divided by the previous twelve month sales. Compustat Winsorized at the 1% and 99% levels. 10-year yield 10-year Treasury bond rate. Federal Reserve Bank of St. Louis Term slope 10-year Treasury bond rate minus 2-year Treasury bond rate. Federal Reserve Bank of St. Louis BAA AAA spread Moody’s seasoned Baa corporate bond yield minus Moody’s seasoned Aaa Federal Reserve Bank corporate bond yield. of St. Louis NBER recession dummy Dummy that takes on a value of 1 in recession months. NBER
Appendix B Tracked bond indices
Investment grade bond indices
Given its dominance in the investment grade index market, we focus exclusively on the Bloomberg indices (formerly administered by Barclays and Lehman Brothers) for investment grade bonds. The single index with the largest passive bond following is the Bloomberg Barclays U.S. Aggregate Index, the flagship benchmark index that measures the investment grade, U.S. dollar-denominated, fixed-rate taxable bond market. It is comprised of the constituents of the U.S. Treasury Index, the U.S. MBS Index, the U.S. CMBS Index, and the U.S. Credit Index (comprised of the corporate bonds and government-related bonds, such as agencies, sovereigns, supranationals and local authorities). The pure Chapter 3. Debt issuance in the era of passive investment 88 corporate bond index is the Bloomberg Barclays U.S. Corporate Index. The indices are value weighted and rebalanced/reconstituted on a monthly basis. Though it varies over time, corporate bonds make up approximately 20% of the Aggregate Index. The investment grade indices have had changes in index inclusion rules, including changes to minimum size (face value) for inclusion, changes to bond types that are includable, and changes to the calculation of ratings. The minimum face value to be included in the index evolved as follows18:
• Until August 1, 1988: $1 million • Between August 1, 1988 and January 1, 1992: $25 million • Between January 1, 1992 and January 1, 1994: $50 million • Between January 1, 1994 and July 1, 1999: $100 million • Between July 1, 1999 and October 1, 2003: $150 million (announced February 24, 1999) • Between October 1, 2003 and July 1, 2004: $200 million (announced June 17, 2003) • Between July 1, 2004 and April 1, 2017: $250 million (announced March 18, 2004) • Since April 1, 2017: $300 million (announced January 24, 2017)
In terms of includable bonds, the following bond types became eligible for inclusion: covered bonds on January 1, 2011; fixed-to-floating perpetual bonds without a coupon step-up on the first call date on January 1, 2008; and 144A bonds with registration rights on July 1, 2000. In terms of rating methodology, until October 1, 2003, Moody’s was the only rating considered, with S&P if Moody’s not available. From October 1, 2003 to July 1, 2005, the lower of Moody’s and S&P was considered the rating. Finally, since July 1, 2005, the middle rating of Moody’s, S&P and Fitch is considered the rating (if only two are available, the lower of the two is taken).
High yield indices
There are two dominant players in the high yield index space - Markit iBoxx and Bloomberg. The Markit iBoxx Liquid High Yield Index measures the high yield, U.S. dollar-denominated, fixed rate corporate bond market. Though the index was adapted from the Goldman Sachs $HYTop Index which was established on December 31, 1998, the first fund following the iBoxx HY index was only established in March 2007; for this reason, we consider the impact of this index on bonds issued in or after April 2007. The iBoxx HY index went through a complete overhaul of rules in June 2009, moving from including only the 50 largest eligible bonds to including all eligible bonds (it also moved from equal weighting to market value weighted and moved from best rating to average rating for determination of high yield status). At the time of this index overhaul, it also revised the minimum face value for inclusion:
• Until July 1, 2009: $200 million • Since July 1, 2009: $400 million (and issuer must have at least $1 billion total face value)
In April 2012, the index removed the requirement that an eligible bond must be less than 5 years old and reduced the required time until maturity to one year from three, among other changes. Because of the drastic change in the Markit iBoxx Liquid High Yield Index in April 2009, we also look at the Bloomberg Barclays U.S. Corporate High Yield Index, which we use to calculate the total face value of eligible HY bonds (used as the denominator in our passive demand perc variable). This index tracks the high yield, U.S. dollar-denominated, fixed-rate corporate bond market; it is value weighted and
18For our analysis with index inclusion eligibility, we use announced dates where found; otherwise, we assume the change was announced three months prior to the effective date. Chapter 3. Debt issuance in the era of passive investment 89 rebalanced monthly. The index increased the threshold for inclusion from $100 million to $150 million in July 2000 (the only change in index threshold). Though there was no direct passive investment in high yield bonds at the time of the change, we include this change as we believe there were active funds using this index as a benchmark.
Appendix C Aggregate passive demand
In this section, we provide the details of the procedure used to compute the total demand for eligible corporate bonds from bond ETFs. For each fund, the CRSP Mutual Fund database reports the total value of assets under management. Unfortunately, for most of the sample period there is no information on what proportion of their portfolio is in corporate bonds. For each fund, we infer this proportion by aggregating all bonds in the FISD database that are eligible to be included in the index tracked by the fund, and dividing the value of corporate bonds in the index by the total index size. To this end, for each bond (corporate or otherwise) in FISD we track key bond characteristics (including amount outstanding, rating and coupon status, as well as non-time varying characteristics) throughout the life of the bond. Each month, we determine which bonds are eligible to be included in each relevant index based on their characteristics, and then aggregate their outstanding amounts to find the total size of the index.19 We then find the proportion of this value attributable to corporate bonds, and multiply the fund’s assets under management to estimate the total dollar amount invested in corporate bonds by the fund. We aggregate these amounts across all passive funds to find our estimate of dollar passive bond demand. Finally, we divide this quantity by our estimate of the size of IGCI (HYCI) for investment grade (high yield, respectively) bonds to find the proportion of eligible corporate bonds that are held by passive funds. Thus, we obtain our main independent variable, Passive demand.
19FISD does not track securitized bonds (such as asset-backed securities), which are included in a number of important bond indices. To correct for this omission, we adjust the corporate bond percentage down in the case of aggregate bond market indices. The monthly adjustment factor is based on relative total index face values of the Bank of America Merill Lynch Broad Market Index, which includes securitized bonds, and the U.S. Corporate and Government Index, which excludes them. This information can be obtained from a Bloomberg terminal (index tickers US00 and B0A0, respectively). Chapter 4
SEO underwriters: The choice between matchmaking and market making
4.1 Introduction
It has been documented that seasoned equity offerings (SEOs) are a rare occurrence for most firms (see for example Fama and French (2005)). Despite this, the SEO market results in significantly more deal volume than the initial public offering (IPO) on an annual basis,1 since the deals are much larger on average. Because the deals are large and infrequent, it is reasonable to believe that firms engage in careful consideration before undertaking an SEO and optimize the deal structure in as many ways as possible. Recent literature has highlighted the rise in accelerated SEOs as one example of optimizing deal structure (Bortolotti et al. (2008), Gao and Ritter (2010), Gustafson (2018)).2 For example, between 2008 and 2014, 65% of U.S. SEOs were issued on an overnight basis, compared to less than 30% between 2001 and 2007 (Gustafson, 2018). The focus of the literature to date has been on timing effects: while a traditional marketed offering gives underwriters time to market and flatten the demand curve for shares given increased supply (Gao and Ritter, 2010), the firm’s shares face price pressure in the open market trading between announcement and pricing (a pressure that does not exist in the case of accelerated offerings) (Gustafson, 2018). Another key difference in SEOs in the U.S. and Canada is the offering procedure.3 Within the category of accelerated offerings described above, most papers combine two deal types: bought deals and accelerated marketed offerings. In this paper, I argue that the observed offering procedure (either marketed or bought) is determined by the underwriter’s choice to be a matchmaker or a market maker
1During my sample period, SEO volume was approximately 3.5 times IPO volume. 2As noted in Bortolotti et al. (2008), there is no generally accepted definition for “accelerated” offerings. I define an accelerated offering as one that does not include any open trading hours between announcement and pricing, which is different than Bortolotti et al. (2008), who use a 48 hour window to identify accelerated offerings, but aligns with “overnight” offerings as discussed in Gustafson (2018). 3In this paper, I use the terms “offering procedure” and “form of offering” interchangeably.
90 Chapter 4. SEO underwriters: Matchmaking or market making 91 and the firm’s decision to proceed with the SEO or not reflects this underwriter’s choice.4 In a marketed offering, the underwriters contact potential buyers of shares and build a book of demand; the price is determined based on the orders as well as the amount of equity that the seller (either a firm in the case of primary offering or a shareholder in the case of a secondary offering) requires. In this sense, the underwriters are simply matchmakers: they help sellers of shares find potential buyers without bearing any risk in terms of the price or the deal completion. 5 On the other hand, in a bought offering, the underwriters and seller privately agree on a number of and price at which the underwriters will buy shares from the seller; the deal is then announced and the underwriters sell the shares from their own account. In undertaking a bought deal, the underwriters are acting as market makers, taking the risk that the shares will sell and incurring trading losses if they have to sell below the agreed upon deal price. It is important to highlight that in a bought deal, the underwriters are not permitted to pre-market the shares; they are buying the shares from the seller without knowing the valuations of potential buyers. Given this, there are clear additional risks for the underwriter in acting as a market maker over a matchmaker. In this paper, I show that there are certain factors that influence the observed offering procedure based on the choice of the underwriter to act as a matchmaker or market maker. I then show that this choice has impacts on the efficiency of deals completed as well as the split of deal surplus between sellers, buyers and the underwriters. I first show that the type of offering procedure observed is affected by seller bargaining power, competition between underwriters, and dispersion in buyer valuations. Given that a bought deal SEO involves first a negotiation on offering price between the underwriter and the seller, empirical proxies for seller bargaining power relative to the underwriters are positively related to the likelihood that a deal is bought (in other words, seller bargaining power increases the offering price to which the underwriters commit, which makes the seller more likely to accept such price and deal form). On the other hand, a marketed offering involves a negotiation between the seller and buyers on offering price; empirical proxies for seller bargaining power relative to buyers is negatively related to the likelihood that a deal is bought. Measures of the level of underwriter competition are positively related to a deal being bought: as an underwriter faces competition, it will offer a bought deal price that balances potential market making trading losses against potential lost business, increasing the offer price to the seller. Finally, empirical proxies for buyer valuation dispersion are negatively related to the probability that a deal is to be bought. In the next set of analyses, I examine if there is evidence that adverse market conditions faced by underwriters reduce their ability or willingness to act ask market makers over matchmakers (given a restricted capability to take on risk, an underwriter would only be able to do a marketed deal). I find that measures of such market conditions do not provide conclusive explanatory power to the form of intermediation observed in SEOs. However, I briefly examine two categories of offerings that indicate
4The literature exploring middlemen began with Rubenstein and Wolinsky (1987), and the closest related paper for this paper is Yavas (1992), who specifically discusses why some markets are observed with matchmakers while other markets are observed with market makers (for example, real estate brokerage vs. market makers on a stock exchange). To the best of my knowledge, no paper explores why an intermediary would switch between one form of intermediation to another depending on the transaction. 5I will note that in this paper I do not distinguish between “firm commitment” and “best efforts” marketed offerings; while the former includes a commitment by the underwriters to buy any unsold shares, this commitment is made after the shares are marketed and a book of demand is built (in other words, the underwriters are agreeing to take up any shares if investors who have submitted orders change their mind). See Eckbo et al. (2007) for a full description of the full commitment process. Section 2 in this paper describes the bought deal process in detail. Chapter 4. SEO underwriters: Matchmaking or market making 92 that inefficient deals are likely to happen only in bought and not marketed offerings, where inefficiency is defined by seller valuation being greater than buyer valuation. Withdrawn marketed offerings and hung bought deals (where shares are left unsold with the underwriter) indicate deals that should not be completed. Marketed deals that are withdrawn do not result in any shares changing hands, while hung bought deals are completed by construction (as the underwriter has bought the shares from the seller), resulting in losses to the underwriters for the benefit of the sellers. Finally, I examine whether the form of offering affects the split of surplus in an SEO between sellers, buyers and underwriters. While the surplus is inherently unobservable since it is based on seller and buyer valuations, I examine proxies for this surplus, including the discount, spread, all-in-cost, underpricing and overall SEO return. In both the full sample and accelerated subsample of deals, a seller’s all-in-cost (full discount plus underwriters’ spread) is significantly lower in a bought deal, controlling for the factors that were previously found to influence the form of offering, indicating that sellers benefit from a bought deal (though not included, this is likely partly due to the inefficient deals as described above). This effect is economically important; given an average deal size of $188 million and a 0.80% savings on all-in cost, a firm can save $1.51 million by choosing a bought deal over a marketed deal (this translates to savings worth approximately 13 bps of pre-offering market capitalization). The savings are driven by lower commissions paid to underwriters, indicating a shift of surplus from underwriters to sellers. A note of caution is warranted on this conclusion: though I have attempted to control for the factors that drive an underwriter’s and seller’s joint offering form decision, it is still an endogenous decision and a causal interpretation is difficult to make. Overall, this paper shows that the choice of offering procedure is statistically and economically important to firms that are engaging in SEOs.
4.1.1 Contribution to the Literature
This paper is related to the recent literature on the increasing popularity of accelerated SEOs. While Autore et al. (2011), Gao and Ritter (2010), Gustafson (2018) focus on U.S. deals only, Bortolotti et al. (2008) explores the fact that the method of SEO financing is converging internationally to the accelerated form of offering. While these papers mention bought deals, with Bortolotti et al. (2008) specifically discussing their popularity in Canada, none of these papers explicitly consider the choice of doing a bought deal and its impact. Bought deals are relatively rare in the U.S., which may be why they have not been in focus in this literature. In contrast, the majority of SEO proceeds raised in Canada are done via bought deal, so the few papers that have examined bought deals have focused exclusively on a Canadian sample. Examples include Pandes (2010) and Gunay and Ursel (2015) (the latter focuses on the method in which Canadian underwriters compete for bought deal mandates). In contrast to this literature, I examine the full sample of transactions from both U.S. and Canada in a single sample. This paper is also related to the literature on discounts in SEOs. Previously literature on discounts, such as Altinkilic and Hansen (2003), Corwin (2003), and Mola and Loughran (2004), have all focused purely on the discount between offer price and the price immediately before pricing. In this paper, I explicitly consider the share price impact during the marketing period, which gives a more complete estimate of the discount (what I call the “full discount”). In addition, I compare the discounts across the offering types. The rest of the paper is structured as follows. Section 2 describes the forms of offering in detail and Chapter 4. SEO underwriters: Matchmaking or market making 93
4:00pm 9:30am 4:00pm 9:30am 4:00pm Announce Price
Stock: A B C D Index: X Y Z (a) Traditional marketed offering.
4:00pm 9:30am 4:00pm Announce Price
Stock: A = B C D Index: X = Y Z (b) Accelerated marketed offering.
4:00pm 9:30am 4:00pm Announce with Price
Stock: A = B C D Index: X = Y Z (c) Bought offering.
Figure 4.1: Timeline of offering types. The red line indicates the period of time when the offering has been announced but the offering terms (size and price) are unknown. provides a comparison from the underwriters’ and firms’ point of view. Section 3 describes a hypothetical deal origination framework and develops hypotheses related to the choice of form of offering. Section 4 describes the data. Section 5 discusses the empirical findings. Section 6 concludes.
4.2 Forms of Offering
In this section, I describe the process by which marketed and bought deals are undertaken. In addition, announcement and pricing times are highlighted (shown in figure 4.1).
4.2.1 Matchmakers: Traditional Marketed Offerings
In a traditional marketed offering, the firm announces that it intends to sell a certain number of shares or a certain dollar amount and describes the proposed use of proceeds. The last trade before the deal is announced is A, while the equivalent level of the market index at this time is X. The shares are then traded in the market with the news that the firm is raising equity, but the terms of the offering are uncertain. During the period between announcement and pricing, the underwriters solicit investors and build a book with quantity and price demand. At the end of the marketing period, the underwriters and the firm agree to and announce final terms of the offering, including the number of shares and the price per Chapter 4. SEO underwriters: Matchmaking or market making 94 share. The last trade before pricing is finalized is B and the equivalent market index level is Y ; the offer price is C. The day following the pricing announcement, the shares now trade with all information, including the final terms of the offering. The closing price of the shares on the day following pricing is D while the equivalent market index level is Z. These steps are shown in figure 4.1a (note that the time between announcement and pricing may vary). A marketed offering can be done as either firm commitment or best efforts, where a firm commitment includes a guarantee by the underwriter to buy all shares at the agreed upon fixed price. The key distinction between a firm commitment marketed offering and a bought deal is that the underwriters’ commitment in the marketed offering is fixed at the signing of the underwriting agreement, which is done after the shares are marketed and the underwriters have built a book of demand (Eckbo et al., 2007). So while the underwriters are protecting against any unsold shares with a firm commitment, the underwriter knows who the buyers will be and their valuations, drastically diminishing the risk taken. In this paper, I do not distinguish between firm commitment and best efforts marketed offerings. There are several important distinctions between traditional marketed offerings and the two other deal types described below. First, the extended time period allows the underwriters to engage in more extensive marketing of the firm’s shares (in some cases, even including a formal road show), designed to flatten the demand curve for the shares (Gao and Ritter, 2010). Second, the longer marketing period allows investors additional time to analyze and value the information in the equity offering, such as the use of proceeds. Finally, in a traditional marketed offering, the shares continue to trade in the interim period between announcement of the offering and the finalization of the offering terms. While SEC Rule 105 prohibits investors from purchasing securities in an SEO when short sales are made within a certain period before pricing (Securities and Exchange Commission, 2013), current shareholders are not prohibited from selling their holdings and buying new issue shares on the offering, and natural buying of the shares in the open market may fall as investors plan to buy in the offering. In addition, Henry and Koski (2010) show that this rule constrains but does not eliminate manipulative trading. These combined effects often result in pricing pressure during the marketing period.
4.2.2 Matchmakers: Accelerated Marketed Offerings
In my definition of accelerated marketed offerings, the offering is announced and priced within a time period which does not include any open market trading of the shares (ignoring after- and pre-market trading). A common example includes the announcement of a deal shortly after market close on a given day, with final terms announced before market open the following trading day. Because there is no trading between the announcement and the pricing, the last trade before announcement is also the last trade before pricing (A = B and X = Y ). Figure 4.1b shows a typical accelerated marketed offering. Similar to the traditional marketed offering, in the period between announcement and pricing, the underwriters build a book of demand from investors including price and quantity. In contrast with a traditional marketed offering, the offering is much quicker, necessarily limiting the amount of marketing the underwriters are able to complete and the amount of time investors are able to ‘digest’ the offering details. In addition, there is less market pressure on the shares during the pricing period due to the absence of open trading hours. In bidding to lead an accelerated marketed offering, the underwriters submit to the firm a proposal that includes the underwriting fee and may also include a “backstop” (price at which the underwriter will guarantee the firm can raise money) and any associated incentive fee (for example, the total commission Chapter 4. SEO underwriters: Matchmaking or market making 95 to the underwriter may increase based on how far above the backstop the deal is priced). A backstop provides some certainty to the firm in terms of price, though not all deals include this backstop. 6 According to industry participants, any backstop price is generally below the price at which the firm could do a bought deal.
4.2.3 Market Makers: Bought Offerings
The key difference between bought offerings and the two forms of marketed offerings is the fact that in a bought deal, the underwriters and the firm negotiate the quantity and price of shares that the underwriter will buy from the firm.7 In Canada, bought deals are done on a “fixed price” basis, which means that the underwriters are required to sell all the shares at the announced price.8 In the U.S., even if bought deals are announced with a price, the underwriters are able to sell the deal at different prices (for example, an investor may receive a discount if has agreed to make a large purchase of the securities being offered (Shearman and Sterling LLP, 2010)). This practice will impact underwriter profits but not the proceeds to the firm.9 Figure 4.1c shows the corresponding timeline for a bought offering. While the timeline for bought offerings is similar to accelerated marketed offerings in that they are both announced during non-trading hours, bought deals have the additional feature that the announce- ment of the SEO includes the deal price which has been pre-negotiated. Similar to an accelerated marketed offering, the price before announcement is the same as the price before pricing (A = B and X = Y ). The book building process for a bought deal is different than that described above since investors only provide the number of shares they are interested in purchasing given the price.
4.2.4 Comparing Bought and Marketed Offerings
The descriptions above highlight the differences in timing among the deal types, with the accelerated marketed and bought offerings most similar. However, bought offerings are different than marketed offerings in terms of which party (issuer or underwriters10) bears the uncertainty. In particular, when the underwriter acts as a match maker in either the traditional or accelerated marketed offering, it provides a platform for sellers and buyers to negotiate deal terms. From a seller’s perspective, therefore, both types of marketed offering carry uncertainty in terms of whether the desired size of the offering can be achieved (deal risk) and at what price (price risk). Prospective investors may
6Unfortunately, there is no way to identify if a deal has a backstop clause included or if the price at which the deal prices is at the backstop price. 7Some papers refer to U.S. bought deals as “block trades” (Thomson SDC Platinum also uses this term). Given the ambiguous nature of this term, which can also mean a trade of a large block of shares on an exchange, I will opt to use only the term “bought deal” or “bought offering”. 8There are exceptions for deals that do not sell (“hung deals”), which are almost always sold below the announced price at a later date. 9In this paper, I assume all bought deals are completed on a fixed price basis. 10The reference to “underwriters” means the syndicate of underwriters, which often includes multiple investment dealers. The lead underwriter is referred to as the bookrunner (or left lead bookrunner if there are multiple bookrunners). The final terms of the deal are agreed upon by the issuer and bookrunner(s); in the case of a marketed offering, this is based on the book of demand while in the case of a bought offering it is solely based on the bookrunners’ views. In a bought offering, the deal is often structured “subject to syndication” to protect the bookrunner in case that no other underwriters agree to the term. The syndication process is very short (often less than one hour). Each underwriter in the syndicate is entitled to a portion of the commission from the offering and is responsible for a portion of any syndicate costs, such as legal fees, trading losses if any, and any future liabilities from the offering. The composition of the underwriting syndicate is at the discretion of the issuing firm. Chapter 4. SEO underwriters: Matchmaking or market making 96 not like the offering details or use of proceeds and may submit orders at very low prices or not submit orders at all, resulting in an order book with very low demand or prices.11 In a bought deal, on the other hand, the underwriters acts as a market maker and take on the price and deal risk because they have agreed to buy the shares at a negotiated price. Assuming that the offering is not unsold at close, the underwriters must sell the shares at the same price as they purchased them, so they have no upside potential on the sale price. The case where the underwriters cannot sell all of the shares purchased in a bought deal is referred to as a “hung deal”, because the underwriters are left holding unwanted shares.12 In these instances, the underwriters may either continue to maintain the syndicate post-closing and complete a clean-up trade of the unsold shares as a group, or the syndicate may break and each underwriter receives its portion of unsold shares (determined by their underwriting percentage). In either case, the underwriters subsequently sell the shares at a price that is almost always below of the original offering price (since a hung deal means that the original deal price was too high relative to the market’s valuation of the shares post-offering). The underwriters therefore incur trading losses on the portion of the deal that is left unsold at deal closing.
4.3 Deal Origination Process and Hypothesis Development
In this section, I describe a hypothetical SEO origination process and the factors that are likely to affect the type of deal that is subsequently observed. I consider an SEO to be an example of a transaction where it is difficult for sellers and buyers to meet without the help of an intermediary.13 Given this, if a firm would like to raise cash via equity financing or a large holder would like to sell (some of) its stake in a secondary offering, it must first approach a potential underwriter or a group of underwriters and describe the amount it would like to raise, desired price, and use of proceeds (not applicable in the case of secondary offering). The underwriter(s) will then decide if it is willing to act as a market maker for this deal (offer a bought deal) and if it is, at what price it is willing to buy the shares. This price per share can be considered a negotiation between the underwriter and the seller (firm or selling shareholder). If the underwriter is willing to be a market maker, the seller must then decide whether or not to accept the terms of the bought deal offered. In particular, the seller should accept if the bought deal price per share is above its reservation value, and it does not believe that it can receive a higher price in a marketed offering, where the seller will negotiate on a price per share with the investors directly. This leads to the first hypothesis: H1: Seller Bargaining Power. Bargaining power of the seller over the underwriter will increase the bought deal offer price, increasing the likelihood of a bought deal. Bargaining power of the seller over investors will increase the potential marketed deal offer price, decreasing the likelihood of a bought deal. How will the underwriter decide whether it is willing to be a market maker and at what price it should buy the shares? Assuming that the underwriter is risk neutral, it should offer a price that maximizes expected profits, incorporating the fact that in the case it offers a price too high, it may not be able to sell the shares to investors and be forced to realize trading losses. If the underwriter was a monopolist and earned comparable fixed compensation (commission) as a market maker or matchmaker, it should
11A backstop price in an accelerated marketed offering removes the deal risk and limits, but does not eliminate, the price risk. 12This is also known as being “long and wrong”. 13There are non-brokered offerings, but they are rare and very small in size. I exclude them from my sample. Chapter 4. SEO underwriters: Matchmaking or market making 97 offer a price that will never generate losses - in other words, the lower bound of the buyers’ valuation range. If the seller accepts the bought deal terms or instead asks the underwriter to act as a matchmaker (execute a marketed deal), the underwriter will receive the same commission payment and does not take any risk of trading losses in a bought deal. However, if there is competition among underwriters, offering too low of a bought deal price may result in an underwriter being excluded from an SEO. In this case, the underwriter must offer a bought deal price that maximizes expected profits, balancing potential lost business and expected losses: H2: Underwriter Competition. As underwriter competition increases, an underwriter will offer a higher bought deal offer price, increasing the likelihood of a bought deal. When there is competition between underwriters, the underwriter will offer a price that is above the buyers’ valuation lower bound, and that price will be based on the distribution of such valuation. Without specifying a functional form of that distribution, I assume that the underwriter considers the dispersion of valuations in offering a bought deal price: H3: Buyer Valuation Dispersion. As buyer valuation dispersion increases, an underwriter will offer a lower bought deal offer price, decreasing the likelihood of a bought deal. Previously I assumed that underwriters were risk neutral. It is reasonable, however, to examine if the market conditions faced by underwriters affects their willingness or ability to be a market maker rather than a matchmaker. For example, if a bank-affiliated underwriter is required to maintain a certain capital level and faces a non-SEO related hit to their capital ratios, it may be unwilling to take on additional risk of a bought deal. This leads to the last hypothesis: H4: Underwriter Market Conditions. As underwriters face adverse market conditions, they are less willing or able to offer a bought deal, decreasing the likelihood of a bought deal. In summary, in the data, we will observe a bought deal if the underwriter is willing to act as a market maker for a given SEO and the seller of shares believes this is the best price at which they are able to sell their shares.
4.4 Data
4.4.1 Sample of New Issues
In order to identify seasoned equity offerings, I analyzed deals from two data sources. The first data source, SDC Platinum, is widely used in the literature on corporate events, such as equity offerings and mergers and acquisitions. The second source, Bloomberg, is used to supplement the deals identified in SDC with additional deal information, including form of offering and pricing dates. It is also easier to extract date-specific data through Bloomberg after adjusting announcement dates. The original sample included SEOs of at least $10 million by Canadian and U.S. corporations and trusts in the years 2000-2015 found in both sources. The sample was then reduced by excluding non- brokered offerings (deals done directly with investors without underwriters), offerings of tax-advantaged flow-through shares, deals where no announcement could be found, at-the-market offerings, offerings of rights, and offerings of shares bundled with other securities such as notes or warrants. After removing these deals, the sample includes 9,855 deals (2,755 Canadian deals and 7,100 U.S. deals). In order to include expanded financial data for the firms in the new issue sample, deals were linked to the issuer’s previous quarterly financial data in Compustat using the firms’ six-digit CUSIPs or names. Chapter 4. SEO underwriters: Matchmaking or market making 98
Of the 9,855 deals in the new issue sample, 8,893 were matched to firm-quarters in the Compustat database (approximately 90%). The match rate is slightly higher for U.S. firms (6,460 out of 7,100 deals or 91%) than for Canadian firms (2,433 out of 2,755 deals or 88%).
4.4.2 Key Input: Announcement Dates
As discussed below, for return analysis, exact dates and times of announcements are required to identify correct prices. For more than 2,000 deals, the date and time of the announcements was manually checked using the news releases on Bloomberg, which are timestamped at release. The earliest available news release is taken to be the applicable announcement. For the remaining deals, a method similar to Altinkilic and Hansen (2003) was used. As discussed in Safieddine and Wilhelm (1996), there is a sharp increase in trading volume upon the announcement of an SEO, as well as upon the pricing of an SEO. When the dates for deal announcement and price announcement matched between Bloomberg and SDC, the trading volume for the 11 days centered around such dates was analyzed, with the applicable announcement assumed to come before market open on the day with the highest volume in these periods (this is equivalent to after market close on the previous day). When the chosen dates match the data source dates, this was assigned a ‘high’ confidence level; when only one of the two dates matches, this was assigned a ‘medium’ confidence level; and when neither date matched, this was assigned a ‘low’ confidence level. For the analysis that requires exact pricing, only deals that were manually checked or had a ‘high’ confidence level from the volume methodology were included. This includes 5,366 deals, more than half of the deal sample. For the remaining deals, the dates from the volume methodology were used when available, or the dates given by Bloomberg when not available.
4.4.3 Variable of Interest: Offering Type
Key to this analysis is the classification of a deal as a bought or marketed offering. For Canada, deals for which Bloomberg’s Offer Type or SDC’s Offering Technique includes a reference to “Bought Deal” were determined to be bought deals, while deals including a reference to “Best Efforts” were determined to be marketed offerings. The news releases for deals with neither or both references were manually checked. For the U.S., deals for which SDC’s Offering Technique included a reference to “Block Trade” were initially identified as bought deals, while all other deals were classified as marketed offerings. Upon closer inspection, however, many of these block trades had announcement dates different from their pricing dates, which is not consistent with a bought deal timeline.14 The list of U.S. bought deals was therefore limited to those with block trade classification where the given announcement and pricing date were the same. The announcement dates and times are used to determine the number of trading days between announcement and pricing. If the deal is classified as marketed (not bought) with no trading days between, it is classified as accelerated marketed; otherwise, it is traditional marketed. Table 4.1 shows the percentage of offering types between Canada and the U.S. While only approxi- mately 10% of deals in the U.S. are done by bought deal, almost 85% of deals in Canada are bought (the
14I suspect that this issue occurs in SDC due to the ambiguous use of “Block Trade”, which may also describe share sales from existing holders (secondary offerings) that are not necessarily bought. Chapter 4. SEO underwriters: Matchmaking or market making 99
Gross Proceeds Gross Proceeds 1 1 .8 .8 .6 .6 .4 .4 .2 .2 2000q1 2005q1 2010q1 2015q1 2000q1 2005q1 2010q1 2015q1 YrQtr YrQtr
Bought Accelerated Marketed Traditional Marketed
Figure 4.2: Percentage of deals and SEO proceeds raised by offering type by quarter.
figures are comparable in terms of gross proceeds raised: less than 9% in the U.S. compared to almost 85% in Canada).
[INSERT TABLE 4.1 HERE]
The previous table showed that accelerated marketed offerings represent 1.3% and 27% of SEOs in Canada and the U.S. over the entire sample, but Figure 4.2 highlights the increase in accelerated marketed offerings over time. This is consistent with findings in Bortolotti et al. (2008) and Gustafson (2018), among others, who discuss the increase in accelerated offerings over time. The left graph in figure 4.2 shows that the number of accelerated marketed deals is increasing at the expense of both traditional marketed and bought deals. The second figure, however, shows that amount of gross proceeds raised through bought deals is relatively constant, while the amount raised via traditional marketed offerings has decreased over the sample period.
4.4.4 Empirical Proxies for Hypotheses
As discussed in the Deal Origination Process section, the likelihood of observing a bought deal is affected by seller bargaining power, underwriter competition, buyer valuation dispersion, and underwriter market conditions. I now describe the empirical proxies used for these concepts; all variables and their sources are described in greater detail in the Appendix. There are two forms of seller bargaining power to consider: bargaining power in relation to the under- writer (positively related to probability of bought deal) and bargaining power in relation to the buyers (negatively related to the probability of a bought deal). In terms of the former, I include Secondary, a Chapter 4. SEO underwriters: Matchmaking or market making 100 dummy variable that takes on a value of 1 if the SEO includes shares sold by an existing shareholder. Holders who are selling in a secondary offering are often large, influential participants in the market (consider a private equity firm selling some of its stake post-IPO, or a large institutional holder who is looking to move a large block of shares). Such an influential holder is likely to hold bargaining power over underwriters through the potential for future business. I also include InstOwnP erc, which measures the percentage of shares held by institutional investors. As the level of current holder sophistication increases, I expect that relative bargaining power over underwriters increase. In terms of relative bargaining power of the seller over buyers, I first consider IndexP erf and V IXIndex as measures of investor sentiment which are positively and negatively related to firm bargaining power, respectively. The final measures of seller bargaining power are related to the financial constraints faced by the firm; more financially constrained firms doing an SEO would hold lower relative bargaining power. Following Hadlock and Pierce (2010), I first use LogAssets and Age as proxies for financial constraints. IGCreditRatingSP is a dummy that takes on a value of 1 if the firm has an investment grade credit rating from S & P; this indicates that the firm has access to the credit markets if the equity market is not favourable. Finally, I include P rice, which is the offering price of the SEO, which is expected to be negatively related to financial distress. In terms of underwriter competition, I consider the following proxies, all of which are expected to be positively related to the probability that a deal is bought. NumUW is the number of underwriters included in the SEO syndicate. Since the syndicate members are mostly chosen by the seller, this variable represents a good proxy for the number of underwriter relationships that the seller has, all of whom could be expected to be in the running for the bookrunner role. NumUW Avail, on the other hand, is a more general measure: on a lagged quarterly basis, I calculate the number of underwriters with more than 1% market share in the equity league tables for a given country. Finally, Concurrent is dummy that takes on a value 1 if the use of proceeds includes a merger or acquisition or if the security issued is a subscription receipt (used in Canada almost exclusively for M& A). This variable is trying to capture the possibility that the underwriter has been paid additional fees outside of the SEO, which may affect how aggressive they are in offering a bought deal price. To measure buyer valuation dispersion, I consider the following proxies, all of which are expected to be negatively related to the probability that a deal is bought. The first measures relate to size, since larger deals are harder for an underwriter to place fully. LogOfferSize is the natural logarithm of the gross proceeds of the offering, while RelOfferSizeMC and RelOfferSizeSH are the gross proceeds as a percentage of pre-offering market capitalization and the number of shares issued as a percentage of pre-offering daily trading volume, respectively. I also consider RecentIss, a dummy that takes on a value of 1 if the issuer has done another SEO in the previous year; this would increase buyer recognition but would also potentially indicate that buyers are “full” on the name and do not want to buy additional shares. Finally, I include V ol30Day, the recent trading volatility of the shares before the offering. Finally, in terms of underwriter market conditions, I consider four proxies derived from financial data of underwriters found in Compustat. For the underwriring industry in each country, I include de-trended total assets (UW industryAssetsdt), total income (UW industryIncomedt), total revenues (UW industryRevenuesdt), and capitalization (UW industryCap, where capitalization is defined in Murfin (2012) as book equity divided by total assets; this variable is not de-trended). These are posi- tively related to market conditions, so I expect them to be positively related to the probability that a deal is bought. Chapter 4. SEO underwriters: Matchmaking or market making 101
In most specifications, I include industry-quarter fixed effects, where industry is measured at the 2-digit SIC code level. Because the underwriting industry measures are at the country-quarter level, only industry fixed effects are included in these regressions. Finally, in most specifications, I include a Canada dummy, which takes a value of 1 if the deal was done in Canada and 0 otherwise. This is to control for any country-specific factors that may affect the probability that a deal is bought. The following table summarizes the empirical proxies as well as their expected relationship with a deal being bought:
Seller Bargaining Underwriter Com- Buyer Valuation Underwriter Market Condi- Power petition Dispersion tions Secondary (+) NumUW (+) LogOfferSize (-) UWindustryAssetsdt (+) InstOwnPerc (+) NumUWAvail (+) RelOfferSizeMC (-) UWindustryIncomedt (+) IndexPerf (-) Concurrent (+) RelOfferSizeSH (-) UWindustryRevenuedt (+) VIXIndex (-) Vol30Day (-) UWindustryCap (+) LogAssets (-) RecentIss (-) Age (-) IGCreditRatingSP (-) Price (-)
Table 4.2 shows summary statistics of the empirical proxies across the three offering types.
[INSERT TABLE 4.2 HERE]
4.4.5 Variables of Interest: Returns Around Key Dates
I also examine returns around key SEO dates, such as deal announcement and pricing announcement. All returns are described in detail in the Appendix. For traditional marketed offerings, the P ricingReturn is the market-adjusted return from the price before announcement to the price before pricing. There is no concept of return during the pricing period for bought and accelerated marketed offerings since there is no trading between the two times. I also examine the discount that has been traditionally used in the literature, T radDiscount, which is the return from the price before announcement to the offer price (T radDiscountAfterF ees accounts for the underwriter fees). While both definitions of discount has been examined in previous literature focused on SEO discounts15, it ignores the return during the pricing period. For this reason, I examine the F ullDiscount, which is the market-adjusted return from the price before announcement to the offer price (AllInCost accounts for underwriter fees). For accelerated marketed offerings and bought offerings, the full discount is equal to the traditional discount, and the all-in cost is equal to the traditional discount after fees. The final key return analyzed is the ReturnSEO, which measures the return from before announce- ment to the close of trading on the day after pricing, adjusted for the market return over the same period.
15Altinkilic and Hansen (2003), Corwin (2003), and Mola and Loughran (2004) use a definition similar to T radDiscount while Bortolotti et al. (2008) define underwriting spread similar to T radDiscountAfterF ees. Chapter 4. SEO underwriters: Matchmaking or market making 102
4.5 Empirical Findings
4.5.1 Likelihood of Doing a Bought Deal
In this first analysis, I am going to test the first three hypotheses described in the Deal Origination Process section. In particular, I would like to determine whether the probability of deal being bought is affected by the empirical proxies for seller bargaining power, underwriter competition, and buyer valuation dispersion. Table 4.3 shows the results from a linear probability model where the dependent variable, BotDeal, is a dummy that takes on a value of 1 if the SEO is bought. All deals are included in this Table. Industry-quarter fixed effects are included in almost all specifications, and standard errors are clustered at the left bookrunner level.
[INSERT TABLE 4.3 HERE]
Column (7) is the full specification, and while some proxies are significant and affect the probability of a bought deal in the direction expected, many of the variables are insignificant and two (LogAssets and RecentIss) load significantly with the opposite sign. There may be differences between traditional marketed and the other forms of offering that are not properly accounted for in Table 4.3 (e.g. the deal is so complex that it requires additional time for investors to process), so the next analysis includes only bought and accelerated marketed offerings. Table 4.4 shows the results from linear probability model with the same dependent and independent variables as Table 4.3. Industry-quarter fixed effects are included in most specifications, and standard errors are clustered at the left bookrunner level.
[INSERT TABLE 4.4 HERE]
In the full specification, both measures of seller bargaining power over the underwriters load positively in the correct direction. The measures of seller bargaining power over investors are less convincing, though having an investment grade credit rating decreases the likelihood of a bought deal as expected. The number of underwriters available, a proxy for underwriter competition, loads significantly positively as expected. Finally, none of the measures for buyer dispersion load significantly (which means that RecentIss does not load in the opposite direction as in Table 4.3). Based on the results in both Tables 4.3 and 4.4, proxies for seller bargaining power, underwriter competition and buyer valuation dispersion do influence the likelihood that a deal is bought in the expected direction.
4.5.2 Effect of Underwriter Market Conditions
I now turn to hypothesis 4, which measures the effect of underwriter market conditions on the probability that a deal is bought. In this section, I would like to test if underwriters are less likely to offer bought deals (or offer bought deals at lower prices, leading sellers to reject the bought deals) if they face adverse market conditions, which may be expected if underwriters care about downside risk. Table 4.5 includes the full specification from Table 4.3 and adds the empirical proxies for underwriter market conditions in a linear probability model with BotDeal as the dependent variable. Since the underwriting industry measures are very different for U.S. and Canadian deals, Panel A includes U.S. deals while Panel B includes Canadian deals. Industry (2-digit SIC) fixed effects are included in all specifications, and standard errors are clustered at the left bookrunner level. Chapter 4. SEO underwriters: Matchmaking or market making 103
[INSERT TABLE 4.5 HERE]
In both the full sample (columns (1)-(4)) and the accelerated only sample (columns (5)-(8)), none of the measures load significantly in the same direction across the two country sub-samples. While de- trended industry income is positively correlated in the U.S., it is insignificant in the Canadian subsample. In addition, while industry capitalization is negatively related in the U.S. (which is not as expected), it is positively related in the full Canadian sample and insignificant for accelerated deals. It is difficult from this analysis to draw concrete conclusions about the effect of underwriter market conditions on the probability of bought deals.
4.5.3 Inefficient Transactions: Withdrawn and Hung Deals
Based on the previous section, I cannot claim that adverse market conditions are consistently related to the types of deals that are done, at least not in the way that would be expected. However, in the data I observe two categories of deals that can be considered inefficient, in that the seller’s valuation is above the buyer’s valuation (negative surplus) and where trade should not occur. The SEO offering procedure does affect whether the deal is completed or not: a marketed deal can be withdrawn whereas a bought deal will still be completed. I will first discuss withdrawn offerings. In the case of marketed offerings, the seller is not obliged to complete the financing based on negotiations with buyers; this can occur based on the price per share or the size of deal the seller is able to achieve. According to data collected from Bloomberg, there is between 286 and 637 withdrawn marketed transactions over the sample period or between 4% and 9% of the number of completed marketed deals.16 On the other hand, because the price and size of a bought deal is negotiated privately between the underwriter and the seller, there are very few cases where a bought deal is withdrawn after announcement; I have found evidence of 12 transactions over the 16 year sample.17 The second category is hung deals. As previously described, in the case that the underwriters in a bought deal cannot sell the shares, they are left holding the shares and often incur large losses when liquidating the shares off of their balance sheets. In the sample period, I have found news articles describing this occurring for 23 offerings.18 This is a lower bound for the number of deals that are hung, since it will include only deals that are discovered by journalists and are ’newsworthy’, which usually means that the losses to the underwriters are large. I argue that both categories of deals indicate a potentially inefficient transaction and the trade should not occur. In the case of a withdrawn deal, the inefficient transaction is not occurring. In the case of a hung deal, the offering should not be taking place at the bought deal price, but it still does. The losses incurred by underwriters are essentially a transfer of surplus from the underwriters to the seller.
16351 deals indicate a deal size of $0, likely due to the fact that the intended deal size was never revealed; given that some of these deals would have been less than $10 million and may be excluded for other reasons, the true number of withdrawn deals is likely between these two bounds. 17While they do not have a “market out” clause in a bought deal, underwriters have a “disaster out” clause in the case of catastrophic macroeconomic events that are not solely based on the state of the financial markets, and a “material adverse change” clause in case of changes that materially affect the issuer. 1813 of these transactions involved renegotiation between the underwriters and the sellers, resulting in a smaller deal size or lower price per share. While the seller is not contractually obligated to accept this renegotiation, perhaps they do given long-term relationships with underwriters involving other services, such as lending. Underwriters still incur losses in these re-negotiated deals, but the losses are reduced Chapter 4. SEO underwriters: Matchmaking or market making 104
4.5.4 Returns Across Offering Types
In this section, I examine the abnormal returns around key dates in SEOs across the three offering types.19 In this section and the following, the sample includes only deals that were manually checked or were matched based on volume with a high confidence level, as discussed under the Data section. The mean return during from before announcement to the day after pricing (ReturnSEO) is -2.8%, while the median is -3.2%. This is in line with previous studies.20 Based on means, accelerated marketed offerings have the lowest returns (-4.4%), bought deals have the best (-1.9%), and traditional marketed deals fall in between (-2.3%). Based on medians, accelerated marketed deals (-3.6%) do better than traditional marketed (-3.9%), but both are worse than bought deals (-2.7%). Figure 4.3 shows a visual representation of the price movement around the pricing date and over the whole SEO process. Figure 4.3a shows returns around the pricing date only. The pre-pricing price, offering price, net price after commission, and closing the day after pricing are all normalized by the pre-pricing price. This figure shows that traditional marketed offerings have a smaller T radDiscount and T radDiscountAfterF ees then accelerated marketed and bought offerings, resulting in higher net proceeds per share to the firm. This is true whether the measure used is the mean, median, 25th percentile, or 75th percentile. Figure 4.3b, where prices are normalized by the pre-announcement price, shows a different story for traditional marketed offerings. Consistent with the price pressure story in Gustafson (2018), traditional marketed offerings experience a mean -2.4% return (median -4.2%) between announcement and pre- pricing. After accounting for this pricing period return, traditional marketed offerings are the most expensive on an all-in cost basis (discount and underwriting fee) both on a mean (-9.3%) and median (-9.7%) basis. The method of calculating discounts in previous literature did not properly account for this pricing return. While accelerated marketed and bought deals have very similar all-in costs at the median (-7.5% and -7.7%, respectively), the mean bought deal has better all-in cost (-8.4% vs. -9.0%), as does the 25th percentile bought deal (-10.5% vs -10.9%). On the other hand, the 75th percentile accelerated marketed deal (-5.2%) has a lower all-in cost than the 75th percentile bought deal (-5.8%). In other words, while the average bought and accelerated marketed deals have similar cost to an issuer, there is a larger range of all-in cost for accelerated marketed deals. The firm conducting an accelerated marketed deal therefore has more upside but also faces more downside in terms of the net proceeds per share.
4.5.5 Bought Deals Change the Split of Surplus in SEOs
Given the above analysis of mean and median returns around SEOs, I now examine if the form of offering has impact on the returns after controlling for the factors that influence whether or not a bought deal is observed. Table 4.6 shows the results of a fixed effects regression that includes BotDeal as the variable of interest as well as controls for the deal type from the full specification of the previous linear probability
19Because of the relatively short period under study (the mean number of days between announcement and pricing is 4.0 trading days; the post-pricing return is always 1 trading day), I calculate abnormal returns using only the market return, as proxied by the S&P 500 Index for U.S. deals and the S&P TSX Composite Index for Canadian deals. As discussed in Kothari and Warner (2007), given such a short time period, the sensitivity of results on the definition of normal returns is low. 20Older studies include Asquith and Mullins (1986), who find average two-day excess returns of -2.7% around announce- ment dates, and Masulis and Korwar (1986), who find average returns of -3.25%. Ritter (2003) finds a two-day cumulative return of -2%. Chapter 4. SEO underwriters: Matchmaking or market making 105 models. The dependent variables examine include: (1) F ullDiscount, which measures the return from the pre-announcement price to the offering price, as an indication of the new issue price relative to the pre-announcement price; (2) Spread, which is the commission paid to the underwriters; (3) AllInCost, which measures the return from pre-announcement price to the net proceeds per share received by the issuer (offering price less fees paid to underwriters), as an assessment of the true cost of raising equity; (4) Underpricing, which measures the return from the offering price to the closing price the day after pricing, as a measure of how cheap the new shares are relative to a proxy of their true value; and (5) ReturnSEO, which measures the return from pre-announcement to the closing price the day after pricing, which measures the immediate impact of financing once the full deal news is impounded into the share price.
[INSERT TABLE 4.6 HERE]
Panel A shows the results for the full sample of deals, while Panel B shows the results for accelerated deals only.21 In both the full sample and the sub-sample, Spread loads significantly negatively, indicating that the commission charged in bought deals is less, controlling for other factors. In the accelerated subsample, the AllinCost loads significantly positively on the BotDeal dummy, indicating that the cost for the sellers (including both full discount plus underwriter’s commission) is less expensive. The benefit accruing to sellers is economically significant. In the full sample, the average deal size is $188.7 million, so the 0.80% savings on AllInCost results in savings to the seller of $1.51 million by choosing a bought deal over a marketed deal (this translates to savings worth approximately 13 bps of pre-offering market capitalization). As previously indicated, the spread paid by sellers in consistently lower in bought deals, indicating a shift in surplus from underwriters to the sellers. It is difficult to make a causal interpretation in this scenario, however. While I have attempted to control for the factors that drive an underwriter’s and seller’s joint form of offering decision, it is likely that the deal type is decided in conjunction with the price and other deal factors (such as offering size and commission charged).
4.6 Conclusion
In this paper, I have explored the effect of offering procedure on outcomes associated with seasoned equity offerings, such as costs and returns. I have shown that whether underwriters act as market makers or matchmakers depends on factors related to seller bargaining power, underwriter competition and buyer valuation dispersion. Controlling for these factors, it appears that bought deals are less expensive and have less negative share price impacts than marketed offerings, whether accelerated or traditional. Given this, why are bought deals not more prevalent in the U.S. (e.g. similar to the levels of adoption in Canada)? While underwriter competition offers a partial explanation, whereby U.S. underwriters are often seen as viable competition in Canada but not vice versa, there are other possible structural differences between Canada and the U.S. including simple market acceptance of bought deals. It is possible that these factors may evolve over time and change the U.S. landscape toward more bought deals.
21Note that in the dataset, F ullDiscount and AllInCost are calculated as returns from pre-announcement price, so they are almost always negative, whereas Spread is documented as a positive commission paid. Chapter 4. SEO underwriters: Matchmaking or market making 106
Future work could include the expansion of analysis to alternative offering forms, such as rights offerings, which are widely used in non-North American countries. Such analysis would require more extensive discussion on the factors that lead to different equilibria in different countries. Chapter 4. SEO underwriters: Matchmaking or market making 107
Normalized by price before pricing 110 100 90 80 Traditional Mktd Accelerated Mktd Bought
1 2 3 4
excludes outside values
(a) Returns normalized by pre-pricing price. Normalized by price before announcement 120 110 100 90 80 70
Traditional Mktd Accelerated Mktd Bought
0 1 2 3 4
excludes outside values
(b) Returns normalized by pre-announcement price.
Figure 4.3: Abnormal returns of firms announcing and pricing SEOs. 0=Price before announcement; 1=Price before pricing; 2=Offering price; 3=Net price after commission (proceeds to issuer); 4=Closing price on day after pricing. Raw returns are normalized by index performance over the relevant period (S&P 500 Index for U.S. firms and S&P/TSX Composite Index for Canadian firms). Chapter 4. SEO underwriters: Matchmaking or market making 108
4.7 Tables
Table 4.1: Offering types
This table shows the breakdown of offering types in Canada and the U.S. Panel A uses number of deals while Panel B uses gross proceeds raised. Traditional Marketed Accelerated Marketed Bought Deal
Panel A: Number of Deals U.S. 62.3% 27.2% 10.5% Canada 13.8% 1.4% 84.8%
Panel B: Gross Proceeds U.S. 65.4% 25.7% 8.9% Canada 14.0% 1.3% 84.7% Chapter 4. SEO underwriters: Matchmaking or market making 109
Table 4.2: Summary statistics
Traditional Marketed Accelerated Marketed Bought Number of Observations 4,806 1,967 3,082 Secondary 36.2% 18.9% 15.3% InstOwnPerc 34.3% 17.3% 16.7% IndexPerf 1.1% 1.9% 1.3% VIXIndex 19.0 18.5 19.2 LogAssets 6.45 6.97 6.40 Age 10.33 11.32 8.17 IGCreditRatingSP 12.1% 18.0% 8.1% Price $25.38 $23.52 $16.08 NumUW 6.23 5.37 5.67 NumUWAvail 13.23 13.12 14.77 Concurrent 39.4% 36.7% 41.6% LogOfferSize 4.81392 4.708375 4.168996 RelOfferSizeMC 22.7% 15.7% 16.2% RelOfferSizeSH 92.5 37.7 88.4 Vol30Day 53.75 43.81 45.22 RecentIss 19.8% 45.4% 35.5% US UWindustryAssetsdt -0.101 -0.105 -0.142 UWindustryIncomedt -1.112 -1.766 1.409 UWindustryRevenuedt -4.909 -7.186 -1.602 UWindustryCap 16.4% 18.8% 14.9% Canada UWindustryAssetsdt -0.014 -0.017 -0.010 UWindustryIncomedt -0.013 0.046 -0.041 UWindustryRevenuedt -0.404 -0.313 -0.222 UWindustryCap 7.3% 7.5% 7.4% Chapter 4. SEO underwriters: Matchmaking or market making 110
Table 4.3: Likelihood of bought deal
Linear probability model where the dependent variable, BotDeal, is a dummy that takes on a value of 1 if the offering was done via bought deal. All deals are included in the sample. In- dependent variables are defined in the appendix. Industry-quarter fixed effects, where industry is measured at the 2-digit SIC level, are included in all specifications but (6), where indus- try and quarter fixed effects are included separately. Standard errors are clustered at the left bookrunner level. ***, **, * indicate significance at the 0.01, 0.05 and 0.10 level, respectively. (1) (2) (3) (4) (5) (6) (7) VARIABLES BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal
Secondary 0.000 0.0188 0.0144 (0.0121) (0.0122) (0.0124) InstOwnPerc 0.0421** 0.0228 0.0330* (0.0177) (0.0153) (0.0173) IndexPerf -0.0607 0.0355 -0.00145 (0.122) (0.100) (0.122) VIXIndex -0.00279** -0.00135 -0.00277** (0.00121) (0.000886) (0.00121) LogAssets 0.0303*** 0.0456*** 0.0468*** (0.00302) (0.00530) (0.00512) Age -0.0460 -0.0976** -0.0582 (0.0377) (0.0388) (0.0382) IGCreditRatingSP -0.0711*** -0.0496** -0.0611*** (0.0192) (0.0206) (0.0188) Price -0.0321 -0.00208 0.000970 (0.0225) (0.0219) (0.0239) NumUW -0.00923*** (0.00180) NumUWAvail 0.0136*** 0.0189*** 0.0156*** (0.00345) (0.00251) (0.00306) Concurrent -0.00934 -0.0151 -0.00612 0.00391 (0.00919) (0.00953) (0.00779) (0.00923) LogOfferSize -0.0122*** -0.0146*** -0.0574*** -0.0564*** (0.00390) (0.00406) (0.00825) (0.00816) RelOfferSizeMC -0.231*** -0.163*** -0.139*** (0.0239) (0.0291) (0.0262) RelOfferSizeSH -0.0179*** (0.00290) Vol30Day -0.0309** -0.0364*** -0.0176 -0.0114 (0.0127) (0.00910) (0.0122) (0.0114) RecentIss 0.0592*** 0.0659*** 0.0615*** 0.0492*** (0.00813) (0.00847) (0.00824) (0.00769) Canada 0.764*** 0.732*** 0.706*** 0.728*** 0.736*** 0.698*** 0.713*** (0.0162) (0.0175) (0.0169) (0.0162) (0.0157) (0.0174) (0.0170)
Observations 8,439 8,962 8,962 8,624 8,653 9,054 8,192 R-squared 0.631 0.624 0.620 0.641 0.637 0.578 0.649 Fixed Effects Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Ind/Qtr Ind*Qtr Chapter 4. SEO underwriters: Matchmaking or market making 111
Table 4.4: Likelihood of bought deal (accelerated transactions only)
Linear probability model where the dependent variable, BotDeal, is a dummy that takes on a value of 1 if the offering was done via bought deal. This sample includes only accelerated transactions. Independent variables are defined in the appendix. Industry-quarter fixed effects, where industry is measured at the 2-digit SIC level, are included in all specifications but (6), where industry and quarter fixed effects are included separately. Standard errors are clustered at the left bookrunner level. ***, **, * indicate significance at the 0.01, 0.05 and 0.10 level, respectively. (1) (2) (3) (4) (5) (6) (7) VARIABLES BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal
Secondary 0.0951*** 0.0937*** 0.0950*** (0.0231) (0.0263) (0.0241) InstOwnPerc 0.121*** 0.111*** 0.111*** (0.0324) (0.0290) (0.0315) IndexPerf -0.0755 -0.212 -0.0214 (0.155) (0.141) (0.150) VIXIndex 0.000602 -0.000254 -0.000147 (0.00126) (0.00117) (0.00127) LogAssets 0.0154*** 0.0309*** 0.0284*** (0.00444) (0.00541) (0.00734) Age 0.0312 -0.108* 0.0195 (0.0882) (0.0641) (0.0899) IGCreditRatingSP -0.0721** -0.0596** -0.0714** (0.0302) (0.0286) (0.0285) Price 0.00446 0.0120 0.0156 (0.0451) (0.0273) (0.0468) NumUW -0.0121*** (0.00231) NumUWAvail 0.0222*** 0.0266*** 0.0192*** (0.00453) (0.00344) (0.00434) Concurrent 0.00270 -0.00527 -0.00161 0.00795 (0.0116) (0.0114) (0.0102) (0.0129) LogOfferSize 0.00723 0.00744 -0.0231*** -0.0234** (0.00676) (0.00663) (0.00746) (0.0102) RelOfferSizeMC -0.0573* -0.0274 -0.0118 (0.0325) (0.0326) (0.0385) RelOfferSizeSH 0.00263 (0.00348) Vol30Day -0.00624 -0.0146 -0.000434 0.0137 (0.0156) (0.0150) (0.0154) (0.0204) RecentIss 0.00161 0.00399 0.00679 -0.00505 (0.0116) (0.0113) (0.00760) (0.0110) Canada 0.741*** 0.717*** 0.653*** 0.701*** 0.696*** 0.662*** 0.695*** (0.0227) (0.0236) (0.0240) (0.0255) (0.0255) (0.0213) (0.0246)
Observations 4,014 4,341 4,341 4,192 4,207 4,599 3,924 R-squared 0.718 0.718 0.716 0.712 0.711 0.640 0.723 Fixed Effects Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Ind/Qtr Ind*Qtr Chapter 4. SEO underwriters: Matchmaking or market making 112
Table 4.5: Likelihood of bought deal (underwriter conditions)
Linear probability model where the dependent variable, BotDeal, is a dummy that takes on a value of 1 if the offering was done via bought deal. Panel A includes only deals by U.S. companies and Panel B includes only deals by Canadian companies. All specifications include the variables included in tables 3 and 4 as controls other than the Canada dummy. Columns (1)-(4) include all deals, while columns (5)-(8) include accelerated transactions only. Industry (2-digit SIC) fixed effects are included in all specifications. Standard errors are clustered at the left bookrunner level. ***, **, * indicate significance at the 0.01, 0.05 and 0.10 level, respectively. Panel A: U.S. (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal
UWindustryAssetsdt -0.561 -1.366 (0.391) (0.854) UWindustryIncomedt 0.0436*** 0.105*** (0.0154) (0.0336) UWindustryRevenuedt 0.0106 0.0273 (0.0104) (0.0279) UWindustryCap -1.245*** -4.167*** (0.112) (0.258)
Observations 6,740 6,740 6,740 6,740 2,561 2,561 2,561 2,561 R-squared 0.152 0.153 0.152 0.174 0.262 0.263 0.262 0.349 Sample All deals All deals All deals All deals Accel. deals Accel. deals Accel. deals Accel. deals Controls Yes Yes Yes Yes Yes Yes Yes Yes Fixed Effects Ind Ind Ind Ind Ind Ind Ind Ind
Panel B: Canada (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal BotDeal
UWindustryAssetsdt 0.0266 -0.00256 (0.0547) (0.0341) UWindustryIncomedt -0.00103 -0.00208 (0.00677) (0.00327) UWindustryRevenuedt 0.0107*** 0.000328 (0.00336) (0.00149) UWindustryCap 2.173** -0.0129 (0.958) (0.228)
Observations 2,308 2,308 2,308 2,308 2,026 2,026 2,026 2,026 R-squared 0.135 0.134 0.139 0.138 0.058 0.058 0.058 0.058 Sample All deals All deals All deals All deals Accel. deals Accel. deals Accel. deals Accel. deals Controls Yes Yes Yes Yes Yes Yes Yes Yes Fixed Effects Ind Ind Ind Ind Ind Ind Ind Ind Chapter 4. SEO underwriters: Matchmaking or market making 113
Table 4.6: Impact of deal type on stock returns
OLS regressions where the dependent variable is listed at the top of the column (see the Appendix for definitions). Panel A inludes all deals while Panel B includes only accelerated transactions. The explanatory variable of interest is BotDeal, a dummy that takes on a value of 1 if the offering was done via bought deal. All specifications include the variables included in tables 3 and 4 as controls. Standard errors are clustered at the left bookrunner level. ***, **, * indicate significance at the 0.01, 0.05 and 0.10 level, respectively. Panel A: All deals (1) (2) (3) (4) (5) VARIABLES FullDiscount Spread AllInCost Underpricing ReturnSEO
BotDeal -0.0512 -0.00796*** -0.0400 0.00911 -0.0479 (0.0568) (0.00145) (0.0532) (0.0121) (0.0608)
Observations 4,342 4,347 4,342 4,345 4,340 R-squared 0.169 0.623 0.178 0.199 0.188 Controls Yes Yes Yes Yes Yes Fixed Effects Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr
Panel B: Accelerated Deals (1) (2) (3) (4) (5) VARIABLES FullDiscount Spread AllInCost Underpricing ReturnSEO
BotDeal -0.00446 -0.00383*** 0.00802*** 0.0157 0.0188 (0.00310) (0.00137) (0.00287) (0.0140) (0.0125)
Observations 2,868 2,871 2,868 2,869 2,866 R-squared 0.543 0.650 0.631 0.258 0.256 Controls Yes Yes Yes Yes Yes Fixed Effects Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Ind*Qtr Chapter 4. SEO underwriters: Matchmaking or market making 114
4.8 Appendix
Appendix A Description of Variables
Dependent Variable: Of- Definition Source fering Type BotDeal Dummy that takes on a value of 1 if the offering is done Bloomberg, SDC and via bought deal. Assigned Type=2 manual AccelDeal Dummy that takes on a value of 1 if the offering is mar- Bloomberg, SDC and keted with no trading days between deal announcement manual and pricing announcement. Assigned Type=1 TradDeal Dummy that takes on a value of 1 if the offering is Bloomberg, SDC and marketed with more than zero trading days between manual deal announcement and pricing announcement. Assigned Type=0
Independent Variable Definition Source Secondary Dummy that takes on a value of 1 if deal involves shares sold by Bloomberg and SDC current shareholder InstOwnPerc Number of shares owned by institutions in the preceding calendar Bloomberg and quarter divided by pre-offering number of shares outstanding Thomson Reuters IndexPerf Index return (S&P 500 for U.S. deals and S&P TSX Composite for Bloomberg Canadian deals) from 21 trading days before the offering announce- ment to the day before the announcement VIXIndex Level of the VIX Index Bloomberg LogAssets Natural logarithm of total assets in the preceding calendar (variable Compustat winsorized at the 1 and 99 percent level within the entire 2000-2015 Compustat database) Age Number of years since the IPO date where available, otherwise num- Bloomberg and Com- ber of years included in the Compustat sample as of year of the SEO pustat IGCreditRatingSP Dummy that takes on a value of 1 if the firm has an investment Compustat grade credit rating from S&P in the quarter preceding the offering Price Offering price Bloomberg NumUW Number of underwriters included in the offering Bloomberg NumUWAvail Number of underwriters with at least a 1% market share in the Bloomberg previous calendar quarter’s equity league table for the applicable country Concurrent Dummy that takes on a value of 1 if the use of proceeds includes Bloomberg M&A or the firm issues subscription receipts LogOfferSize Natural logarithm of total gross proceeds, in millions of dollars (in- Bloomberg cludes over-allotment option if exercised) RelOfferSizeMC Total deal size divided by pre-offer market capitalization, calculated Bloomberg five days before the deal announcement (variable is winsorized at the 1 and 99 percent level) RelOfferSizeSH Total deal size, in number of shares, divided by average daily trading Bloomberg volume in the month ending five days before the deal announcement (variable winsorized at the 1 and 99 percent level) Chapter 4. SEO underwriters: Matchmaking or market making 115
Vol30Day Share price volatility five days before the offering announcement, Bloomberg calculated as the standard deviation of the share price over the prior 30 calendar days RecentIss Dummy that takes on a value of 1 if the issuer has completed another Deal database SEO in the prior year UWindustryAssetsdt Detrended total assets of underwriters found in Compustat database Compustat UWindustryIncomedt Detrended total profits of underwriters found in Compustat Compustat database UWindustryRevenuedt Detrended total revenues of underwriters found in Compustat Compustat database UWindustryCapdt Quarterly average capitalization of underwriters found in Compu- Compustat stat, where capitalization is calculated as book equity divided by total assets
The following table references the prices in figure 4.1, as follows: A = P riceBeforeAnnt, B = P riceBeforeP ricing, C = OfferP rice, D = P riceDayAfterP ricing, X = IndexBeforeAnnt, Y = IndexBeforeP ricing, Z = IndexDayAfterP ricing.
Dependent Variable: Definition Formula Returns UnadjPricingReturn The return from the price before announce- ment to the price before pricing B − 1 A
PricingReturn The unadjusted pricing return adjusted for the market performance over the same period. For Y UnadjP ricingReturn − ( − 1) accelerated marketed offerings and bought of- X ferings, B = A and Y = X since there is no trading between announcement and trading. For these deals, the pricing return is not ap- plicable TradDiscount Traditionally defined discount or premium; the return from the price before announcement C − 1 to the offer price B
TradDiscountAfterFees Traditional discount accounting for the under- writing spread, c C ∗ (1 − c) − 1 B
UnadjFullDiscount Price change from before the announcement to the offer price C − 1 A Chapter 4. SEO underwriters: Matchmaking or market making 116
FullDiscount Unadjusted full discount adjusted for the mar- ket performance over the same period. For Y UnadjF ullDiscount − ( − 1) accelerated marketed offerings and bought of- X ferings, the full discount is equal to the tradi- tional discount UnadjAllInCost The return from the pre-announcement price to the net proceeds of the offering price after C ∗ (1 − c) − 1 the underwriter’s commission, c A
AllInCost Unadjusted all-in cost adjusted for the market performance over the same period Y UnadjAllInCost − ( − 1) X
UnadjUnderpricing Return from the offer price to the closing price on the trading day after the offer price is an- D − 1 nounced C
Underpricing Unadjusted under or overpricing adjusted for the market performance Z UnadjUnderpricing − ( − 1) Y
UnadjReturnSEO The return from before announcement to the close of trading on the day after pricing D − 1 A
ReturnSEO Unadjusted SEO return adjusted for the mar- ket performance Z UnadjReturnSEO − ( − 1) X Bibliography
Akerlof, G. A. (1970). Quality uncertainty and the market mechanism. Quarterly Journal of Eco- nomics 84 (3), 488–500.
Allen, F. and G. R. Faulhaber (1989). Signalling by underpricing in the IPO market. Journal of Financial Economics 23 (2), 303–323.
Altinkilic, O. and R. S. Hansen (2003). Discounting and underpricing in seasoned equity offerings. Journal of Financial Economics 69, 285–323.
Altman, E. I. (1968). Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. Journal of Finance 23 (4), 589–609.
Anthony L.G. PLLC (2018). Smaller reporting companies (SRCs) and emerging growth companies (EGCs). Client Publication.
Appel, I. R., T. A. Gormley, and D. B. Keim (2016). Passive investors, not passive owners. Journal of Financial Economics 121, 111–141.
Asquith, P. and D. W. Mullins, Jr. (1986). Equity issues and offering dilution. Journal of Financial Economics 15, 61–89.
Autore, D. M., I. Hutton, and T. Kovacs (2011). Accelerated equity offerings and firm quality. European Financial Management 17 (5), 835–859.
Baker, M. (2009). Capital market-driven corporate finance. Annual Review of Financial Economics 1, 181–205.
Balogh, A., U. Creedy, and D. Wright (2020). Time to acquire: Regulatory burden and M&A activity. Working Paper.
Baugness, S. W., J. John W. Cooney, and K. W. Hanley (2018). Investor demand for information in newly issued securities. Working Paper.
Ben-David, I., F. Franzoni, and R. Moussawi (2018). Do ETFs increase volatility? Journal of Fi- nance 73 (6), 2471–2535.
Bernstein, S. (2015). Does going public affect innovation? Journal of Finance 70 (4), 1365–1403.
Bond, P., A. Edmans, and I. Goldstein (2012). The real effects of financial markets. Annual Review of Financial Economics. 4 (1), 339–360.
117 BIBLIOGRAPHY 118
Booth, J. R. and R. L. Smith, II (1986). Capital raising, underwriting, and the certification hypothesis. Journal of Financial Economics 15, 261–281.
Bortolotti, B., W. Megginson, and S. B. Smart (2008). The rise of accelerated seasoned equity under- writings. Journal of Applied Corporate Finance 20 (3), 35–57.
Calomiris, C. W., M. Larrain, S. L. Schmukler, and T. Williams (2019). Search for yield in large international corporate bonds: Investor behavior and firm responses. NBER Working Paper No. 25979 .
Cattaneo, M. D., M. Jansson, and X. Ma (2020). Simple local polynomial density estimators. Journal of the American Statistical Association 115 (531), 1449–1455.
Chaplinsky, S., K. W. Hanley, and S. K. Moon (2017). The JOBS Act and the cost of going public. Journal of Accounting Research 55 (4), 795–836.
Chava, S. and M. R. Roberts (2008). How does financing impact investment? The role of debt covenants. Journal of Finance 63 (5), 2085–2121.
Chemmanur, T., J. J. He, X. S. Ren, and T. Shu (2020). The disappearing IPO puzzle: New insights from proprietary U.S. Census data on private firms. Working Paper.
Chen, H., L. Cohen, U. Gurun, D. Lou, and C. Malloy (2020). IQ from IP: Simplifying search in portfolio choice. Journal of Financial Economics 138 (1), 118–137.
Chen, Z., A. A. Lookman, N. Schurhoff, and D. J. Seppi (2014). Rating-based investment practices and bond market segmentation. Review of Asset Pricing Studies 4 (2), 162–205.
Corwin, S. A. (2003). The determinants of underpricing for seasoned equity offers. The Journal of Finance 58 (5), 2249–2279.
Crane, A., K. Crotty, and T. Umar (2019). Hedge funds and public information acquisition. Working Paper.
Da, Z. and S. Shive (2017). Exchange traded funds and asset return correlations. European Financial Management 24 (1), 1–33.
Dambra, M., L. C. Field, and M. T. Gustafson (2015). The JOBS Act and IPO volume: Evidence that disclosure costs affect the IPO decision. Journal of Financial Economics 116, 121–143.
Dannhauser, C. D. (2017). The imapct of innovation: Evidence from corporate bond exchange-traded funds (ETFs). Journal of Financial Economics 125, 537–560.
Department of the Treasury, U. (2017, Oct). A financial system that creates economic opportunities - Capital Markets.
Dick-Nielsen, J. and M. Rossi (2019). The cost of immediacy for corporate bonds. The Review of Financial Studies 32 (1), 1–41.
Doidge, C., K. M. Kahle, G. A. Karolyi, and R. M. Stulz (2018). Eclipse of the public corporation or eclipse of the public markets? Journal of Applied Corporate Finance 30 (1), 8–16. BIBLIOGRAPHY 119
Doidge, C., G. A. Karolyi, and R. M. Stulz (2013). The U.S. left behind? Financial globalization and the rise of IPOs outside the U.S. Journal of Financial Economics 110 (3), 546–573.
Doidge, C., G. A. Karolyi, and R. M. Stulz (2017). The U.S. listing gap. Journal of Financial Eco- nomics 123 (3), 464–487.
Drake, M. S., D. T. Roulstone, and J. R. Thornock (2016). The usefulness of historical accounting reports. Journal of Accounting and Economics 61 (2-3), 448–464.
Eckbo, B. E., R. W. Masulis, and O. Norli (2007). Security offerings. In B. E. Eckbo (Ed.), Handbook of Corporate Finance: Empirical Corporate Finance, Volume 1, Chapter 6. North-Holland/Elsevier.
Erel, I., B. Julio, W. Kim, and M. S. Weisbach (2012). Macroeconomic conditions and capital raising. Review of Financial Studies 25 (2), 341–376.
Ewens, M. and J. Farre-Mensa (2020). The deregulation of the private equity markets and the decline in IPOs. Review of Financial Studies 33, 5463–5509.
Fama, E. F. and K. R. French (2005). Financing decisions: Who issues stock? Journal of Financial Economics 76, 549–582.
Fedyk, A. (2021). Disagreement after news: Gradual information diffusion or differences of opinion? Reivew of Asset Pricing Studies Forthcoming.
Field, L. C. and J. Karpoff (2002). Takeover defenses of IPO firms. Journal of Finance 57 (5), 1857–1889.
Gao, X. and J. R. Ritter (2010). The marketing of seasoned equity offerings. Journal of Financial Economics 97, 33–52.
Gao, X., J. R. Ritter, and Z. Zhu (2013). Where have all the IPOs gone? The Journal of Financial and Quantitative Analysis 48 (6), 1663–1692.
Gargano, A., A. G. Rossi, and R. Wermers (2017). The Freedom of Information Act and the race toward information acquisition. Review of Financial Studies 30 (6), 2179–2228.
Gibbons, B. (2019). Passive debt ownership and corporate financial policy. Working Paper.
Graham, J. R. and M. T. Leary (2011). A review of empirical capital structure research and directions for the future. Annual Review of Financial Economics 3, 309–345.
Grinblatt, M. and C. Y. Hwang (1989). Signalling and the pricing of new issues. Journal of Finance 44 (2), 393–420.
Gunay, E. and N. Ursel (2015). Underwriter competition in accelerated seasoned equity offerings: Evi- dence from Canada. Journal of International Markets, Institutions and Money 34, 94–110.
Gustafson, M. (2018). Price pressure and overnight seasoned equity offerings. Journal of Financial and Quantitative Analysis 53, 837–856.
Gustafson, M. T. and P. Iliev (2017). The effects of removing barriers to equity issuance. Journal of Financial Economics 124, 580–598. BIBLIOGRAPHY 120
Hadlock, C. J. and J. R. Pierce (2010). New evidence on measuring financial constraints: Moving beyong the KZ index. Review of Financial Studies 23 (5), 1909–1940.
Hamm, S. J. (2014). The effect of ETFs on stock liquidity. Working Paper.
Henry, T. R. and J. L. Koski (2010). Short selling around seasoned equity offerings. Review of Financial Studies 23 (12), 4389–4418.
Holden, C. W. and J. Nam (2019). Market accessibility, corporate bond ETFs, and liquidity.
Ibbotson, R. G. (1975). Price performance of common stock new issues. Journal of Financial Eco- nomics 2 (3), 235–272.
Israeli, D., C. M. Lee, and S. A. Sridharan (2017). Is there a dark side to exchange traded funds? An information perspective. Review of Accounting Studies 22, 1048–1083.
Jenkinson, T., H. Jones, and F. Suntheim (2018). Quid pro quo? What factors influence IPO allocations to investors? Journal of Finance 73 (5), 230–2341.
Jenkinson, T., A. D. Morrison, and W. J. Wilhelm Jr. (2006). Why are European IPOs so rarely priced outside the indicative price range? Journal of Financial Economics 80 (1), 185–209.
Karpoff, J. M., A. Koester, D. S. Lee, and G. S. Martin (2017). Proxies and databases in financial misconduct research. The Accounting Review 92 (6), 129–163.
Kashyap, A. K., N. Kovrijnykh, J. Li, and A. Pavlova (2018). The benchmark inclusion subsidy. NBER Working Paper No. 25337 .
Kothari, S. and J. B. Warner (2007). Econometrics of event studies. In B. E. Eckbo (Ed.), Handbook of Corporate Finance, Volume 1. Elsevier.
Kutsana, K. and R. Smith (2004). Why does book building drive out auction methods of IPO issuance? Evidence from Japan. Review of Financial Studies 17 (4), 1129–1166.
La Porta, R., F. Lopez-de Silanes, A. Shleifer, and R. W. Vishny (1997). Legal determinants of external finance. Journal of Finance 52 (3), 1131–1150.
Leary, M. T. and M. R. Roberts (2005). Do firms rebalance their capital structures? Journal of Finance 60 (6), 2575–2619.
Leland, H. E. and D. H. Pyle (1977). Informational asymmetries, financial structure, and financial intermediation. Journal of Finance 32 (2), 371–387.
Levin, J. (2001). Information and the market for lemons. RAND Journal of Economics 32 (4), 657–666.
Li, F. W., X. Liu, and C. Sun (2018). The real effects of exchange traded funds. Working Paper.
Ljungqvist, A. (2007). IPO underpricing. In Handbook of Empirical Corporate Finance, pp. 375–422. Elsevier.
Loughran, T. and J. R. Ritter (2004). Why has IPO underpricing changed over time? Financial Management 33 (3), 5–37. BIBLIOGRAPHY 121
Marta, T. (2019). Fixed income ETFs, bond liquidity, and stressed markets. Working Paper.
Massa, M., U. Peyer, and Z. Tong (2005). Limits of arbitrage and corporate financial policy.
Masulis, R. W. and A. N. Korwar (1986). Seasoned equity offerings: An empirical investigation. Journal of Financial Economics 15, 91–118.
McCrary, J. (2008). Manipulation of the running variable in the regression discontinuity design: A density test. Journal of Econometrics 142, 698–714.
Mola, S. and T. Loughran (2004). Discounting and clustering in seasoned equity offering prices. Journal of Financial and Quantitative Analysis 39 (1), 1–23.
Murfin, J. (2012). The supply-side determinants of loan contract strictness. The Journal of Fi- nance 67 (5), 1565–1601.
Myers, S. C. and N. S. Majluf (1984). Corporate financing and investment decisions when firms have information that investors do not have. Journal of Financial Economics 13 (2), 187–221.
Pandes, J. A. (2010). Bought deals: The value of underwriter certification in seasoned equity offerings. Journal of Banking and Finance 34, 1576–1589.
Ritter, J. R. (2003). Investment banking and securities issuance. In G. Constantanides, M. Harris, and R. M. Stulz (Eds.), Handbook of the Economics of Finance, Volume 1A. Elsevier.
Rock, K. (1986). Why new issues are underpriced. Journal of Financial Economics 15 (1-2), 187–212.
Rubenstein, A. and A. Wolinsky (1987). Middlemen. The Quarterly Journal of Economics 102 (3), 581–594.
Safieddine, A. and W. J. Wilhelm, Jr. (1996). An empirical investigation of short-selling activity prior to seasoned equity offerings. The Journal of Finance 51 (2), 729–749.
Securities and Exchange Commission (2017a, Jul). Remarks at the Economic Club of New York by SEC Chairman Jay Clayton.
Securities and Exchange Commission (2017b, Jun). SEC’s Division of Corporation Finance expands popular JOBS Act benefit to all companies.
Securities and Exchange Commission (2019a, Sept). Remarks at the Economic Club of New York by SEC Chairman Jay Clayton.
Securities and Exchange Commission (2019b, Feb). Solicitations of interest prior to a registered public offering.
Securities and Exchange Commission (2013). Rule 105 of Regulation M: Short selling in connection with a public offering.
Shearman and Sterling LLP (2010). Finra adopts Rule 5141 (Sale of Securities in a Fixed Price Offering) to replace ‘Papilsky’ rules. Client Publication.
Stambaugh, R. F. (2014). Presidential address: Investment noise and trends. Journal of Finance 69 (4), 1415–1453. BIBLIOGRAPHY 122
Stulz, R. M. (2019). Public versus private equity. European Corporate Governance Institute – Finance Working Paper No. 640/2019 .
Sushko, V. and G. Turner (March 2018). The implications of passive investment for securities markets. BIS Quarterly Review, 113–131.
Tirole, J. (2010). The Theory of Corporate Finance. Princeton University Press.
Vanguard (2017). Vanguard Total Bond Market Index Fund Summary Prospectus.
Welch, I. (1989). Seasoned offerings, imitation costs, and the underpricing of initial public offerings. Journal of Finance 44 (2), 421–449.
Wurgler, J. (2011). On the economic consequences of index-linked investing. In G. Rosenfeld, J. W. Lorsch, and R. Khurana (Eds.), Challenges to Business in the Twenty-First Century, Chapter 3, pp. 20–34. Cambridge, Mass.: American Academy of Arts and Sciences.
Yavas, A. (1992). Marketmakers versus matchmakers. Journal of Financial Intermediation 2, 33–58.
Ye, S. (2019). How do ETFs affect the liquidity of the underlying corporate bonds? Working Paper.