Essays on Stock Issuance
Abstract
EconStor is a publication server for scholarly economic literature, provided as a non-commercial public service by the ZBW.
Full text
Kohl, Niklas Doctoral Thesis Essays on Stock Issuance PhD Series, No. 38.2017 Provided in Cooperation with: Copenhagen Business School (CBS) Suggested Citation: Kohl, Niklas (2017) : Essays on Stock Issuance, PhD Series, No. 38.2017, ISBN 9788793579491, Copenhagen Business School (CBS), Frederiksberg, https://hdl.handle.net/10398/9536 This Version is available at: https://hdl.handle.net/10419/209046 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/
Niklas Kohl The PhD School in Economics and Management PhD Series 38.2017 PhD Series 38-2017ESSAYS ON STOCK ISSUANCE COPENHAGEN BUSINESS SCHOOL SOLBJERG PLADS 3 DK-2000 FREDERIKSBERG DANMARK WWW.CBS.DK ISSN 0906-6934 Print ISBN: 978-87-93579-48-4 Online ISBN: 978-87-93579-49-1 ESSAYS ON STOCK ISSUANCE
Essays on Stock Issuance Niklas Kohl Supervisor: Søren Hvidkjær PhD School in Economics and Management Copenhagen Business School
Niklas Kohl Essays on Stock Issuance 1st edition 2017 PhD Series 38.2017 © Niklas Kohl ISSN 0906-6934 Print ISBN: 978-87-93579-48-4 Online ISBN: 978-87-93579-49-1 “The PhD School in Economics and Management is an active national and international research environment at CBS for research degree students who deal with economics and management at business, industry and country level in a theoretical and empirical manner”. All rights reserved. No parts of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage or retrieval system, without permission in writing from the publisher.
Preface This dissertation is the result of my Ph.D. studies at the Department of Finance at the Copenhagen Business School. It consists of summaries in English and Danish, an introduction and three self-contained essays on the long-run performance of rms issuing new equity. The dissertation, and my professional development at large, has bene- ted from the support and advice of many people. First and foremost, I am indebted to my supervisor Søren Hvidkjær for his support and guidance throughout the process. My secondary supervisor Ken Bechmann has read a number of very preliminary draft and helped me sharpen ideas. Lasse Heje Pedersen invited me to teach the course Hedge Fund Strategies together with him, and helped me secure an internship at AQR Capital Management. Moreover, I thank colleges, fellow Ph.D. students, and the numerous master students, I have had the pleasure to teach and supervise, for making my years at Copenhagen Business School so enjoyable. There are things they don't teach you at a Business School - for example how markets really work and how you make money on them. Fortunately, I have spent time, actually a lot of time, hanging out with people who could make op for this. Thorleif Jackson has taught me a lot about how you run a small investment company and has introduced me to his network of investors and fund managers. Numerous discussions with my business partner Jon Forst has sharpened my understanding of, in particular, market making and price dynamics in connection with corporate actions. I hope our joint struggle to keep markets ecient will remain joyful and protable. i
Finally, I thank my family, parents, children and in particular Lene for support throughout the process. Niklas Kohl Copenhagen, September 2017 ii
Summary Summary in English Stock Issuance and the Speed of Price Discovery Firms which issue new equity subsequently have lower returns than other rms, but does the strength of the issuance eect vary in the cross section of rms? The essay shows, that US rms with characteristics that makes them hard to value have returns which are strongly related to their past issuance activity, while the return of easy to value rms are less related to their past issuance activity. In most cases the dierence between hard to value and easy to value rms are signicant. As proxies for hard to value, I use three dierent types of rm characteristics. First, I consider rms for which relatively little information is available as hard to value. Examples are rms covered by few analysts and small rms. Second, I consider rms with high levels of analyst disagreement on stock price target, next quarter earnings per share and share recommendation as hard to value. Third, rms with expected cashows in the more distant future are hard to value. These include rms with low earnings, high asset growth, and low dividend yield. As one possible explanation, consistent with the empirical results, I propose a model with informed investors receiving a noisy value signal, and other investors who infer value from past market prices. I analyze the price dynamics after informed investors have received a new value signal (for instance an issue announcement), and show that prices will converge to fundamental iii
value, but convergence will be slowest when the value signal is most noisy, i.e. for rms which are hard to value. The Issuance Eect in International Markets The issuance eect rst documented in the US market also exists in international markets, but does the strength of the issuance eect vary in the cross section of markets? The essay shows that the issuance eect is stronger in non-developed markets, i.e. markets not classied as developed by MSCI, than in developed markets. If rms listed in non-developed markets are more dicult to value than rms listed in developed markets, then the result is consistent with the hard to value hypothesis advocated in the essay Stock Issuance and the Speed of Price Discovery. The empirical results are inconsistent with those reported by McLean et al. (2009) who nd a stronger issuance eect in more developed markets than in less developed markets. 1 My essay shows, how their results are not robust to minor methodological changes. I propose an alternative approach, which arguably is better suited to explore dierences in the issuance eect in the cross-section of markets. I show that this approach conrms my empirical results in several robustness tests. Issue costs, nancial and otherwise, are likely to be higher in less developed markets than in more developed markets. The essay proposes a model of the relationship between issue costs, issuance behavior and average longrun performance of issuers. Higher levels of issue costs predict lower issuance activity and lower long-run returns for issuers, consistent with the empirical ndings. 1 The list of references is found at the end of the section Introduction. iv
Does Information Asymmetry Explain Issuer Underperformance? A prominent behavioral explanation for the low long-run returns of rms raising new equity through seasoned equity oerings (SEOs) holds, that opportunistic rms exploit information asymmetry at issue time to sell overvalued equity Loughran and Ritter (1995). If this explanation holds, one would expect that the most overvalued issuers, and those which are least constrained in the sense, that they do not need to issue to continue operations or service current debt, have the best opportunities to exploit temporary windows of mispricing. Therefore, issuers with these characteristics should experience the lowest risk-adjusted returns subsequent to SEOs. I derive proxies for overvaluation and issuer constrainedness and show, empirically, that the most overvalued and least constrained US SEO rms have similar or higher risk-adjusted long-run returns relative to issuers without these characteristics. Consequently, I nd no evidence of information asymmetry at issue time as explanation for long-run performance of SEO rms. As an alternative explanation, I propose that information asymmetry is particularly low at event time because of the information requirements on issuing rms and the incentives of issuers, investors, and intermediaries. In this case, a possible explanation for the low returns subsequent to SEOs is, that the marginal investor does not fully utilize all available information. I measure the informational content of the SEO announcement using the event return. Negative event returns are interpreted as bad news while the rarer positive event returns are interpreted as good news. I show that, empirically, event news, and in particular negative event news, predict longrun return. This is consistent with the hypothesis that investors underreact v
3 Data and Variables 79 4 Empirical Results 83 5 Conclusions 93 Does Information Asymmetry Explain Issuer Underperformance? 113 1 Introduction 114 2 What Drives Issuer Underperformance? 118 3 Data and Variables 129 4 Returns Before and After Issue 135 5 Event Returns 136 6 Long-run Returns 137 7 Conclusions 151 xii
Introduction This dissertation consists of three papers on stock issuance by listed rms. The study of stock issuance is important because one of the primary functions of the stock market is to enable rms to raise new equity to nance investments or operations. This takes place through initial public oerings (IPOs), but even more importantly through new equity issues by rms which are already listed. According to Thomson Reuters (2017), global IPO activity in 2016 totaled $131 billion while seasoned equity oerings (SEOs) raised $448 billion. McKeon (2015) shows that US-listed rms raise a similar amount in other issues. In total, global equity issuance activity raised around $1 trillion, and more than 80% of this was raised by listed rms. SEOs refer to cases where the rm oers new shares for cash, usually to a group of selected investors, or pro rata to all current shareholders. Typically, the issue consists of at least 3% new shares, although larger issues are commonplace (McKeon, 2015). SEOs are events in the sense that the issue is announced and one can study return pre-event, when the event occurs, and post-event. Other issues, including the exercise of employee stock options, other warrants and convertible bonds, are much more frequent than SEOs but individually much smaller. These issues are not generally announced when they occur, but can only be inferred from quarterly reports or other lings . New issues also occur in connection with stock-nanced mergers where the acquiring rm purchases all or some stocks in the target rm and pays with its own stocks. It is well known that rms which issue new equity, on average, subsequently have high returns before the issue and low returns. In the third
paper, I show that the average US SEO rms overperform, relative to the stock market, by more than 60% the year before issue and underperform by more than 20% over the three years subsequent to issue. The appreciation before issue has a number of plausible explanations. It could reect improved earnings prospects for the rm. To utilize these, increased investments might be necessary, hence the issue of new equity. Alternatively, the appreciation could be due to a reduction in required return, either market wide or for the particular rm, and either rationally or otherwise. In any case, lower required returns mean that more investment opportunities will move into positive net present value territory, hence the rm will invest more and issue more to nance investments. Finally, if the appreciation reects mispricing, and rm management realize this, opportunistic rms may try to exploit the situation and sell overpriced equity to new investors to the benet of old investors, possibly including themselves. In the case of issues due to the exercise of employee stock options (or other derivatives), average high returns before issue follow from the fact that these are only exercised when they are in-the-money. This is most likely to take place after the stock has appreciated. From an investor's perspective, the appreciation before issue is not interesting, because we do not know which rms will be next year's issuers. The depreciation after issue is much more interesting. A key discussion in nancial economics is to what extent nancial markets are ecient in the sense that prices reect available information. The majority of research on returns subsequent to issue takes a stance on this, either arguing that the low returns subsequent to issue are a puzzle which cannot be explained by a fully rational model or that returns are explained by known risk factors 2
or at least factors known to predict return in the cross-section of stocks, i.e. there is no issuance puzzle. From an investor's perspective, the depreciation after issue is of utmost importance: to the extent it reects a deviation from market eciency, it provides trading opportunities. Even if it reects exposure to rationally priced risk-factors, investors need to decide whether and to what extent they wish to be exposed to this risk. My three papers seek to explore and test existing explanations and propose new explanations for the low returns subsequent to issue. The majority of previous research aims to show that issuers underperform or do not underperform on a risk-adjusted basis subsequent to issue. However, my papers dier, in that I investigate whether there are issuer characteristics which determine which issuers are likely to underperform. This is a useful approach, because the ability to characterize the types of issuers which underperform may help us understand the reasons for the underperformance regardless of whether these are behavioral or explained by risk. From an investment perspective, it is also useful because it highlights the issuers which should be avoided or possibly shorted and the issuers which can safely be purchased. The rst paper Stock Issuance and the Speed of Price Discovery , focuses on the issuance eect, i.e. the extent to which past issuance activity (in SEOs or otherwise) predicts future return in the cross section of listed US rms. This has previously been performed by Ponti and Woodgate (2008) using the Fama and MacBeth (1973) methodology to measure the issuance eect. They report that past issuance activity is a strong and signicant predictor of future return in the cross section of rms. The mentioned papers only control for rm size and rm book-to-market ratio in the Fama-MacBeth regressions. By now, it is well established that other factors predict future 3
return. I add asset growth and protability. This is partly motivated by the incorporation of these factors in the Fama French ve-factor model Fama and French (2015), but also by the fact that issuers and non-issuers are likely to dier substantially in terms of these characteristics. Firms issue for a reason and that reason is often because they need more equity due to poor profitability or because they want to grow their asset base through investments. Controlling for asset growth and protability reduces the issuance eect substantially, i.e. a substantial part of the low return of issuers is explained by the fact that they have high asset growth and low protability. This is partly in line with Bessembinder and Zhang (2013), who nd that issuers and nonissuers dier in return-predicting characteristics beyond market value and book-to-market ratio. However, the important contribution of the paper is to study how the issuance eect varies in the cross-section of rms. The question is whether the issuance eect is stronger for some types of rm than for others. Empirically, I show that the issuance eect is strong and signicant among rms which are hard to value but small and often insignicant among rms which are easy to value. I use three dierent types of proxies for hard to value the amount of information available, the extent to which equity analysts agree on rm valuation, and whether expected cash-ows are in the near or more distant future. As one possible explanation, consistent with the empirical results, I propose a model with informed investors receiving a noisy value signal and other investors who infer value from past market prices. I study the price dynamics after informed investors have received a new value signal (for instance an issue announcement) and show that prices will converge to 4
fundamental value, but convergence will be slowest when the value signal is most noisy, i.e. for rms which are hard to value. The second paper The Issuance Eect in International Markets , considers the issuance eect in international markets. If the issuance eect, at least partly, reects some sort of market ineciency or friction, this might be detectable in the cross section of international markets. It is natural to hypothesize that the issuance eect should be stronger in less developed, and presumably less eciently priced, markets than in more developed markets. However, this hypothesis is at odds with the ndings of McLean et al. (2009), who nd that the issuance eect is strongest in the most developed markets, suggesting that this is because rms in developed markets can easily issue and repurchase equity. Therefore, in developed markets, it is easy to be opportunistic and exploit temporary mispricings. In less developed markets, issues and repurchases are more expensive and issues will only occur for primary reasons, i.e. not to exploit mispricings. I nd this result troubling for two reasons. First, the reasoning assumes that rms get away with opportunistic behavior on a large scale in the most developed markets. Second, it is not at all clear that rms will refrain from opportunistic issues just because it is expensive to issue. The paper addresses both these concerns. Theoretically, I show that issue costs do reduce the frequency at which issues occur but do not prevent rms from attempting opportunistic issues. In fact, theoretically, the relation is opposite. In markets with high issue costs long-run issuer underperformance should be stronger than in markets with low issue costs. Empirically, I show that the methodology employed by McLean et al. (2009) is highly sensitive to seemingly arbitrary methodological choices. I suggest an alternative methodology, 5
one which is arguably more suited to analyzing the issuance eect in the cross section of markets. The empirical result is that the issuance eect is significantly stronger in non-developed markets than in developed markets. This may be because of higher issue costs in non-developed markets, but the result is also consistent with the hard to value hypothesis developed in my rst paper. While the rst two papers study the issuance eect, i.e. how issuance activity, whatever the form, predicts future return, the third paper focuses on SEOs. The purpose is to explore whether information asymmetry between rm management and investors at issue time can potentially explain long-run performance. This idea is most explicitly advocated in Loughran and Ritter (1995). If issuer underperformance is explained by opportunistic issues by overvalued issuers this could potentially be detected with suitable proxies for issuer overvaluation and proxies for whether issuers were in a position where they could choose to issue or not to issue. The hypothesis is that rms which are less nancially constrained have more room to be opportunistic in their issuance behavior than rms for which an issue is necessary to nance current operations or service current debt. Empirically, I nd no support for information asymmetry as an explanation for issuer underperformance, because the most overvalued issuers and the least nancially constrained issuers do not have lower risk-adjusted long-run returns than less overvalued and more constrained issuers. The paper also considers the possibility that information asymmetry is low at issue time. This is plausible due to information requirements in connection with issues, rms' incentives to attract interest in the issue, and investors' and intermediaries' interest in conducting their own independent 6
research in connection with issues. Nonetheless, long-run underperformance is possible if the marginal investor does not fully take the available information into consideration. I show that this explanation is consistent with empirical ndings because event returns, and, in particular, negative event return (bad news at event time), predict long-run returns. As always in nancial economics, empirical ndings lend support for dierent interpretations. My empirical ndings are that certain types of issuers, those with little information available, those which analysts disagree about , those with most of their expected cash-ows in the distant future, those which are listed in less developed markets, and those which experience the most negative event returns when they announce a SEO, are more likely to subsequently underperform on a risk-adjusted basis. One possible explanation, developed in the rst paper, is that some investors do not have or do not utilize all available information, and the activities of more sophisticated investors, due to limits of arbitrage, cannot immediately compensate fully for this, in particular when the most sophisticated investors have the most negative valuation. References Bessembinder, H. and Zhang, F. (2013). Firm characteristics and long-run stock returns after corporate events. Journal of Financial Economics , 109 (1), 83 102. Fama, E. F. and French, K. R. (2015). A ve-factor asset pricing model. Journal of Financial Economics , 116 (1), 1 22. Fama, E. F. and MacBeth, J. D. (1973). Risk, return, and equilibrium: 7
Empirical tests. Journal of Political Economy , 81 (3), 607. Loughran, T. and Ritter, J. R. (1995). The new issues puzzle. Journal of Finance , 50 (1), 23 51. McKeon, S. B. (2015). Employee option exercise and equity issuance motives. SSRN. McLean, D. R., Ponti, J., and Watanabe, A. (2009). Share issuance and cross-sectional returns: International evidence. Journal of Financial Economics , 94 (1), 1 17. Ponti, J. and Woodgate, A. (2008). Share issuance and cross-sectional returns. Journal of Finance , 63 (2), 921 945. Thomson Reuters (2017). Global equity capital markets review. Webpage: http://share.thomsonreuters.com/general/PR/ECM_4Q_2016_E.pdf. 8
Stock Issuance and the Speed of Price Discovery Niklas Kohl * Abstract Firms which issue new equity subsequently have lower returns than other rms. In this paper, I show that underperformance by issuers is conned to rms which are hard to value, while issuance activity does not signicantly predict future returns for easy to value rms. Hard to value rms include small cap, rms with high dispersion in analyst estimates and recommendations, and rms with more distant cash-ows, such as rms with low protability, low dividend yield, or high asset growth. Moreover, I show that only the negative component of seasoned equity oering (SEO) event returns signicantly predicts one-year post-SEO returns. These results are consistent with a model in which informed investors receive noisy signals of fundamental value and shorting is constrained or costly. ∗ Department of Finance, Copenhagen Business School, Solbjerg Plads 3, 2000 Frederiksberg, Denmark. E-mail: nk.@cbs.dk. I am grateful for comments and suggestions received from Søren Hvidkjær, Nigel Barradale, Ken Bechmann, Lasse Heje Pedersen, Janis Berzins as well as seminar participants at Copenhagen Business School and the Nordic Finance Network PhD Workshop 2016 in Bergen. Any errors remain mine. 9
(1 −τ)Pt(µu,t −Pt) a σ2 u +τPt(µi,t −Pt) a σ2 i =Pt with the solution Pt=µu,t +stτ−Σaσ2 u Σ(1 −τ) + τ (2) where Σ = σ2 i σ2 u denotes the ratio between variance of valuation of informed investors and uninformed investors. Σ measures the precision of the signal received by informed investors relative to variance perceived by uninformed investors. st=µi,t −µu,t is the time t spread between informed and uninformed investors' expected value of the risky asset. If the signal received by informed investors remains constant, i.e. µi,t =µi for t≥T a necessary and sucient condition for equilibrium is Pt=Pt−1 . Insertion of this condition and the uninformed investors' valuation formula from equation 1 in equation 2 yields Pt=Pt+aσ2 u+stτ−Σaσ2 u Σ(1 −τ) + τ⇒st=aσ2 u(Σ −1) (3) By denition, µi=µu,t +st . Inserting µu,t from equation 1 and st from equation 3 and using the denition of Σ and the equilibrium condition Pt= Pt−1 yields the equilibrium price P∗=µi−aσ2 i . It depends only on informed investors' expected value and variance. In equilibrium investors do not agree on expected value unless Σ=1 , but any disagreement will be oset by disagreement on variance. Consider a situation in which informed investors receive a new and constant value signal µi,t =µi for t≥T . This creates a new equilibrium price, 16
but the question of interest is under what conditions and how fast this equilibrium will be reached. Proposition 1 shows that Pt will always converge linearly to the equilibrium price P∗ . Proposition 1. If µi,t =µi for all t≥T then Pt→P∗ for t→ ∞ The rate of convergence is Σ(1−τ) Σ(1−τ)+τ . Proof. See Appendix A. By proposition 1, the rate of convergence depends only on Σ and τ . Since the partial derivatives ∂γ ∂Σ=τ−τ2 Λ2>0 ∂γ ∂τ =−Σ Λ2<0 convergence is faster for higher fractions of informed investors τ and for lower levels of noise of the value signal Σ received by informed investors. We may augment the model with constraints on shorting. Some investors may be unable or unwilling to short and those who can and will, may face costs associated with shorting and limitations due to margin requirements and lending fees. If P∗> Pt−1 informed investors will be buyers and uninformed investors will be sellers and potential shorters. If unconstrained uninformed investors 17
would have taken short positions, introduction of shorting constraints would increase their demand and thus price. This, in turn, will increase µu,t above what it would otherwise have been, and increase the demand from uninformed investors until the shorting constraints are no longer binding. An equilibrium where only informed investors hold the risky asset is not possible. In such an equilibrium, uninformed investors must have negative demand. This requires µu,t ≤Pt . But by equation (1) µu,t =Pt−1+aσ2 u , so an equilibrium is impossible when a > 0 and σ2 u>0 . Consequently, shorting constraints on uninformed investors will decrease their impact on prices, and thus increase the speed of price discovery. If P∗< Pt−1 the potential shorters are informed investors. If shorting constraints are binding, prices will be higher than they would otherwise have been, and the speed of price discovery will decrease . Even if shorting is impossible, an equilibrium where only uninformed investors hold the risky asset, and price discovery does not occur, is impossible. If uninformed investors hold all risky assets, market clearing implies that Pt=µu,t −aσ2 u 1−τ= Pt−1−τaσ2 u 1−τ . Consequently, the price will decline provided a > 0 , σ2 u>0 , and τ∈]0,1[ . Summing up, the model predicts that price discovery will always occur but be slowest for shares traded by few informed investors and for shares which are hard to value by informed investors. Shorting constraints will increase the speed of price discovery for good news, i.e. when P∗> Pt , but decrease the speed of price discovery for bad news, i.e. when P∗< Pt . 18
2.2 Application to Issuance Large share issues, as well as share repurchases, are known to be informationconveying events. This has been documented in numerous event studies showing that SEO announcements, on average, are greeted with negative abnormal event returns, whereas repurchase announcements are greeted with positive abnormal event returns (see Eckbo et al. (2007) for a survey of studies of SEOs and Peyer and Vermaelen (2009) for repurchases). For the case of share issuance, McKeon (2015) shows that 90% of quarters in which rms issue new shares, the issuance was not initiated by the rm but rather by investors, in particular through the exercise of employee stock options. These issues are generally small and unlikely to convey much information. In contrast, larger issues, often associated with SEOs or stock nanced acquisitions, are rm-initiated and likely to convey information. The model outlined in Section 2.1, predicts that larger share issues will be positively associated with future negative abnormal returns, because they on average convey negative information. Smaller issues are less likely to be associated with abnormal returns, as the information conveyed by smaller issues, in particular investor-initiated issues, is limited. Empirically, this is consistent with Fama and French (2008a) who nd that large issues are associated with signicant negative future abnormal returns, whereas small issues are associated with insignicant positive future abnormal returns. Repurchase announcements may convey substantial positive information, but the model predicts that it will be absorbed by the market faster than negative information. Hence, it is less likely that share repurchases will be associated with signicant future abnormal returns. A novel prediction of the model is that the speed of price discovery will be 19
slowest for hard to value rms trading above their fundamental value, such as hard to value rms with large equity issues. As hard to value is not directly observable, I consider three types of proxies for this property. First, I consider rms for which less information is publicly available. I measure the amount of public information by the rm's market value, because small rms disclose less information, and by the number of equity analysts following a rm. Second, I consider rms with high disagreement in analyst opinion. Here, I calculate dispersion in analyst price target, recommendation, and next quarter EPS estimate. Third, partly inspired by Baker and Wurgler (2007), I consider rms with more distant cash-ows. Firms with more distant cash- ows are harder to value, because there is more uncertainty associated with the more distant future. Firms with distant cash-ows are rms with low protability, measured as return on equity, rms with low dividend yield, rms with high asset growth, and rms with low earnings to price ratio. All these measures may arguably be proxies for diculty to value, but may also be correlated with other characteristics known to predict return. In particular, market value, protability, asset growth and the earnings to price ratio are all known to predict return. As an example, the model predicts that low protability issuers will underperform relative to issuers with higher protability because they are harder to value. But the underperformance may also be caused directly by the lower protability. I address these concerns in two ways. First, I also use proxies which are not obviously correlated with return-predicting characteristics. Second, and more importantly, in the Fama-MacBeth regressions in Section 4, I control for all the returnpredicting characteristics of the Fama and French (2015) ve factor model as well as momentum and in the double sorted portfolio regressions reported in 20
Section 5, I regress returns on the Fama French ve factor returns. 3 Empirical Strategy 3.1 Measures of Issuance My gross sample consists of all shares on the monthly CRPS le during the period from 1985 to 2014 for which price prc or alternate price altprc and monthly return with and without dividends ( ret and retx ) are available. 3 Following some previous research (including Eckbo et al. (2007), Fama and French (2008a), and Bessembinder and Zhang (2013)), I leave out nancial rms. 4 To measure issuance activity, I monthly calculate the adjusted number of shares using the number of shares outstanding ( shrout ) and the cumulative factor to adjust shares ( cfacshr ) reported by CRSP. Observations for which the number of shares and cumulative factor to adjust shares are not available are dropped from the sample. Following Daniel and Titman (2006) net issue over the past year is dened as NetIssuet,t−12 =ln(AdjustedSharest)−ln(AdjustedSharest−12) where AdjustedShares t is the time t adjusted number of shares. To distinguish between positive issuance and negative issuance (repurchases), I dene Issuet,t−12 =max(NetIssuet,t−12,0) 3 Here and in the following variable names in CRSP and Compustat and other databases are given in courier . 4 Some papers, including Loughran and Ritter (1995) and Daniel and Titman (2006) leave out utilities. 21
and Repurchaset,t−12 =max(−NetIssuet,t−12,0) To simplify notation Issue , Repurchase , and NetIssue refer to Issue t,t−12 , Repurchase t,t−12 , and NetIssue t,t−12 , respectively. In some empirical tests, rm-month observations are sorted into issuance portfolios on NetIssue value. These portfolios are denoted issue1 , issue2 , issue3 , issue4 , and issue5 , respectively. The breakpoints used are xed to facilitate the interpretation of the portfolios. issue1 consists of net repurchasers with NetIssue < −0.1% . issue2 is zero-issuers with −0.1% ≤NetIssue < 0.1% . issue3 , issue4 , and issue5 are net issuers with NetIssue of at least 0.1% , 3% and 15% , respectively. The 3% breakpoint is motivated by McKeon (2015) who nds that issues of at least 3% are typically rm-initiated. The 15% breakpoint is chosen to separate rm-initiated issues in two groups of approximately same size. The number of rms per NetIssue portfolio is shown in Figure 2. Figure 2 shows that zero-issuers have become less common and that the number of repurchasers varies strongly over time. In particular, it seems that the number repurchasers spikes in the period after major stock downturns, for example year 1988, after the dot-com bubble in year 2000, after the 2008 Financial crisis, and after the August 2011 stock market fall. Since repurchase is measured over the past year, a possible interpretation is that some rms utilize the low valuations to repurchase own equity. [Insert Figure 2 about here] Table 1 provides statistics for each of the ve NetIssue portfolios. In terms 22
of rm-month observations, issue3 , the portfolio with small positive issuance activity, accounts for 36% of all observations. There are 20% repurchasers ( issue1 ), 15% zero-issuers ( issue2 ) and 16% and 12% in issue4 and issue5 , the two groups with high issuance activity. Zero-issuers are, on average, the smallest rms, issuers are larger and repurchasers the largest rms. BM is highest for zero-issuers and lowest for rms with high issuance activity. ROE and EP are, as one would expect, monotonically decreasing in NetIssue while AG in increasing in NetIssue . [Insert Table 1 about here] One of my empirical tests focuses on SEO rms. I obtain information on SEOs from the Thomson One Banker New Issues Database (SDC Platinum). I selected Follow-On equity issues with total proceeds of at least 3% of the total pre-issue market value. Most of the issues eliminated are oerings of shares by major shareholders. These issues may be large but are not rm-initiated and do not change rm equity. The Figure 3% is motivated by McKeon (2015), as discussed above. SEO observations are merged with CRSP observations on cusip number and rm name. 3.2 Proxies for hard to value As discussed in Section 2.2, I use nine dierent proxies for hard to value. These proxies are calculated monthly. Market value, denoted MV , is calculated from CRPS data. For rms ( permco s) with more than one share class (more than one permno ) issued, only the share class with the highest market value is kept, but the rm's market value is aggregated over all share 23
classes. Dividend yield, denoted Yield , over the past 12 months is calculated as CRSP holding period return ( ret ) over the past 12 months less holding period return without dividend ( retx ) over the past 12 months. For the calculation of return on equity ( ROE ), asset growth ( AG ), and earnings to price ratio ( EP ), accounting data from Compustat are used. I use only data from annual reports. The most recent Compustat observation, at least six months old and no more than two years older than the CRSP observation, is used. AG is calculated as the relative change in assets ( at ) over the past 12 months. ROE is calculated as net income ( ni ) divided by book equity ( ceq ) and EP is calculated as net income divided by market value. CRSP observations, for which Compustat accounting information (assets, net income and book equity) is not available, are omitted. Data on equity analysts and their recommendations are from the IBES database. The most recent IBES observation, no more than one year old, is used. The number of analysts with a next quarter earnings per share (EPS) estimate is denoted #Analysts . Three measures of analyst disagreement are calculated for rms with at least two analyst observations. Dispersion in analyst price target ( PTG ) is given by Dptg =σptg µptg where σptg and µptg is the standard deviation and mean of analyst price targets reported by IBES. Dispersion in analyst recommendation ( REC ) Drec is the standard deviation in recommendation, measured on a ve-point scale, reported by IBES. Dispersion in analyst expected next quarter earnings per 24
share EPS is scaled with price, i.e. Deps =σeps P where σeps is the standard deviation of analysts' next quarter EPS estimate and P is the price per share. While CRSP observations without corresponding accounting data are dropped, observations without analyst information are kept in the sample. Figure 1 shows the number of rms for which at least one estimate of next quarter EPS, at least one price target, and at least one recommendation, are available. EPS estimates start around the year 1985 and coverage gradually increases until around year 2000. Analyst recommendations start becoming available from the year 1995 and price targets from year 2000. By the end of the sample, more than 80% of the rms have EPS estimates, recommendations and price targets. [Insert Figure 1 about here] Since analyst recommendations and price targets are not available from 1985, the empirical test using analyst recommendations covers the period 1995-2014 while test using analyst price targets cover the period 2000-2014. 3.3 Empirical Tests In order to explore to what extent the predictions of the model presented in Section 2 can be conrmed empirically, I have performed three types of tests. First, in Section 4, I do one dimensional sorts on each of the nine variables proxying for hard to value and create quintile samples. Portfolios are constructed monthly. As customary breakpoints are calculated using NYSE 25
where NetIssue is one sort variable and the other sort variable is a proxy for hard to value. For each of the double sorted portfolios, value-weighted monthly return is calculated and regressed on the FF3 and the FF5+ UMD market models. The regression intercept αk i,j is the abnormal return of the intersection between NetIssue portfolio i and portfolio j of sort variable k . For example, αMV 1,1 is the abnormal return of a portfolio of small cap share repurchasers ( issue1 ) and αROE 5,1 is the abnormal return on a portfolio of small cap rms with high issuance activity ( issue5 ). The variable of interest is the dierence in regression intercept between a portfolio of high issuers and a portfolio of repurchasers, within the same quintile of the hard to value variable. This dierence is denoted the issuance spread ∆k j=αk 1,j −αk 5,j For example, ∆MV 1 is the dierence in abnormal return between small cap repurchasers and small cap issuers, while ∆MV 5 is the same dierence for large cap rms. I test whether the issuance spread is signicantly dierent from zero for hard to value portfolios as well as for easy to value portfolios. Figure 5 depicts the monthly α 's of value-weighted portfolios sorted on NetIssue and MV regressed on FF5+ UMD . Within the group of small cap rms, repurchasers have an α of 36 bp, while rms with the largest issuance activity ( issue5 ) have an α of -22 bp. The dierence between these is the issuance spread ∆MV 1 = 58 bps, which is signicant with a t -value of 3.13. It can be interpreted as the abnormal return on an investment which is long small cap repurchasers and short small cap large issuers. If α 's are measured relative to FF3, ∆MV 1 = 110 bps with a t -value of 5.51. For large cap ∆MV 5 32
is -14 bps for the FF5+ UMD model and 14 bps for the FF3 model, both of these are insignicant. Table 6 shows the issuance spread for the most easy to value and the most hard to value rms for each of the nine variables proxying for diculty to value and the two market models FF3 and FF3+ UMD . For the easy to value rms, the issuance spread is only signicant in one case, while it is signicant in 13 out of 18 cases for the hard to value rms. Issuance spreads are uniformly larger when returns are regressed on the FF3 model than when regressed on the FF5+ UMD model, again conrming that some of the underperformance of issuers is explained by exposure to the RMW , CMA and UMD factors. [Insert Table 6 about here] 6 Returns Subsequent to SEOs This section focuses on rms which, according to SDC Platinum, have carried out an SEO. One advantage of focusing on SEOs is that we can calculate event returns. Abnormal event returns can be taken as a proxy for the information conveyed in connection with the issue. Most previous research nds that abnormal event returns on average are negative ((Eckbo et al. , 2007)), but occasionally they will be positive. These events convey positive information about the issuing rm. This enables me to test the model prediction, namely that the speed of price discovery is faster for good news than for bad news, cf. Section 2. In the SDC Platinum database I select all SEOs ( Follow-On oerings) 33
by non-nancial rms between 1985 and 2014 where the proceeds from the oering exceed 3% of the market value before the oering. SDC observations are matched with CRSP and Compustat data using the cusip code and the rm name. Return information must be available in CRSP. [Insert Figure 6 about here] Figure 6 shows the value-weighted cumulated return of SEO rms less the market return from 10 trading days before the issue date ( T ) until 10 trading days after the issue date 6 . Before issues, issuers experience positive abnormal returns (relative to the market) of around 1%. This is not necessarily surprising, as rms may chose to issue when they perceive their own shares to be performing strongly. From the day before the issue to two days after the issue, SEO rms experience negative abnormal event returns of about -1.6% followed by a partial rebound. Motivated by Figure 6, I measure abnormal event returns over the three-day period from close on day T−2 to close on day T+ 1 , i.e. ER =RSEO T−2,T −1−RMkt T−2,T −1 where RSEO and RMkt denote the return of the SEO rm and the CRSP value-weighted market return, respectively. Of the 11,481 SEO events, ER is positive in 4,237 cases (37%). As a simple test of whether the news conveyed at issue time is associated with future abnormal returns, I decompose ER into its positive component ER+=max(ER, 0) and its negative component ER−=max(−ER, 0) and regress one-year buy and hold abnormal return 6 If the issue date is a Saturday or a Sunday, T is the Friday before the issue. 34
BHAR on ER+ and ER− . As the relation between ER and BHAR may not be piecewise linear, I also sort SEOs into quintiles based on ER and regress BHAR on dummy variables associated with quintiles 1, 2, 3 and 5. I chose quintile 4 as the base category, as this quintile contains SEOs with zero abnormal event return. BHAR is the return of the SEO rm over some period less the return of a benchmark investment over the same period. The literature on long run abnormal returns has documented that results are very sensitive to the actual calculation of BHAR , i.e. the choice of benchmark (Mitchell and Staord (2000), Eckbo et al. (2007), and Bessembinder and Zhang (2013)). One stream of the literature uses the matched rm approach, in which the benchmark of a SEO rm is another rm, which is similar to the issuer usually in terms of market value and book-to-market ratio, but Bessembinder and Zhang (2013) show that issuers and non-issuers dier in several other characteristics known to predict return. According to the authors, these dierent characteristics explain the observed dierences in post-issue return and controlling for these dierences there is no abnormal BHAR . In the absence of a commonly agreed benchmark for calculation on BHAR , I chose the simplest possible approach to calculating one-year BHAR as BHAR =RSEO T+10,T +1y−RMkt T+10,T +1y As this is likely to be a biased estimate of true one-year abnormal return, it is not suitable for inference on the absolute level of BHAR . However, it may be more suited for making an inference about the relation between event abnormal returns and long-run abnormal returns. Table 7 shows the result 35
of value-weighted regressions of BHAR on ER+ and ER− as well as on ER quintile dummies. A 1% increase in ER− , i.e. a 1% decrease in abnormal return event, when abnormal event return is already negative, is associated with a 76 bps decrease in one-year BHAR ( t -value -5.78), while ER+ is insignicant. In the regression with ER+ and ER− as regressors, the intercept is 1.55%, indicating that zero or positive event return is associated with small positive BHAR . In the regression with ER quintile dummies, issuers with lowest abnormal event returns have a 7.3% lower one-year BHAR ( t -value -5.26). Second and third quintiles also have signicantly lower BHARs of 5.33% and 1.87% than the base category. Firms with the highest abnormal event returns have positive BHARs of 2.8%. This is signicantly dierent from zero but not from the 1.55% intercept reported in the model with ER+ and ER− as regressors. Both regressions support that negative event return is signicantly associated with BHAR , while positive event returns are not. [Insert Table 7 about here] One may be concerned that the results presented above reect that abnormal event returns are correlated with other rm characteristics known to predict returns. It may, for example, be that the least protable SEO rms experience the lowest event returns. To address this concern, as well as the methodological issues concerned with BHAR calculations and their distribution, I construct two calendar-time portfolios as suggested by Mitchell and Staord (2000). One portfolio consists of rms which have carried out an SEO with positive event returns ER during the past year. The other portfolio consists of 36
negative event return SEO rms. In addition, I construct a long-short zeroinvestment portfolio which is long SEO rms with positive ER and short SEO rms with negative ER . The SEO calendar-time portfolios are updated monthly and SEO rms are included from the rst complete month after T+ 10 (10 trading days after the issue) and for a total of 12, 24 or 36 consecutive months. I also construct portfolios of rms which issued 12 to 23 months ago and 24 to 35 months ago, respectively. Value-weighted monthly portfolio returns are regressed on FF3 as well as on FF5+UDM. Table 8 shows the results of these regressions. [Insert Table 8 about here] In panels A and B returns are regressed on FF3. The negative ER portfolios have signicant α s between -49 bps and -56 bps for holdings periods of one, two and three periods. The positive ER portfolios also have negative α 's but the long-short portfolios have signicant positive α 's for holding periods of one and two years. However, α is only signicant for the rst year and insignicant for the second and third year. In panels C and D returns are regressed on FF5+ UMD . Controlling for RMW , CMA and MOM increases α for all SEO portfolios, reecting that all portfolios have signicant negative exposure to RMW and CMA . Again, this conrms the nding that the underperformance of issuers is partly explained by their low protability and high asset growth. When regressed on FF5+ UMD issuers with positive ER have insignicant α s for all holding periods considered. However, negative ER issuers experience signicant negative abnormal returns for two-year holding periods as well as during the second year. The long-short portfolios also have positive, but insignicant, abnormal 37
returns for all holding periods. Even though t -statistics are less impressive than for the BHAR regressions, these results conrm that underperformance subsequent to SEOs is stronger when the SEO conveyed bad news than when it conveyed good news. 7 Conclusions Firms which issue new equity subsequently have low returns. Some of this performance can be explained by exposure to other risk factors beyond the classical three Fama French factors, but some abnormal return remains to be explained. I propose a model in which some investors are uninformed , assuming that the latest observed price reects fundamental value, whereas informed investors receive a noisy value signal. Prices are set in competition between uninformed and informed investors and I show that prices converge to an equilibrium price dependent only on the value signal observed by informed investors. The speed of price discovery depends on the noise embedded in the signals received by informed investors, i.e. how easy the rm is to value, and will, in the presence of shorting constraints or limitations, be slowest for bad news. I have applied this model to the case of issuance and derive two predictions. First, the model predicts that underperformance subsequent to stock issues will be strongest for the rms which are hardest to value. As proxies for hard to value, I use measures of information available, analyst disagreement, and more distant cash-ows. Empirically, I show that rms with these characteristics do indeed underperform subsequent to issues, whereas rms 38
without these characteristics do not underperform. Second, the model predicts that underperformance will be strongest when stock issues convey negative news. I use SEO event returns as proxy for the news conveyed at issue, and empirically show that the negative component of abnormal event returns is signicantly associated with negative buy and hold abnormal returns, whereas the positive component of abnormal event returns does not predict buy and hold abnormal returns. Moreover, I show that a calendar-time portfolio of SEO rms with negative abnormal event returns have signicant negative abnormal return over some holding periods, whereas a calendar-time portfolio of SEO rms with positive abnormal event returns have insignicant or numerically lower abnormal return subsequent to the SEO. 39
References Baker, M. and Wurgler, J. (2007). Investor sentiment in the stock market. Journal of Economic Perspectives , 21 (2), 129 151. Barberis, N., Shleifer, A., and Vishny, R. (1998). A model of investor sentiment. Journal of Financial Economics , 49 (3), 307 343. Bessembinder, H. and Zhang, F. (2013). Firm characteristics and long-run stock returns after corporate events. Journal of Financial Economics , 109 (1), 83 102. Daniel, K. and Titman, S. (2006). Market reactions to tangible and intangible information. Journal of Finance , 61 (4), 1605 1643. Due, D. (2010). Presidential address: Asset price dynamics with slowmoving capital. Journal of Finance , 65 (4), 1237 1267. Eckbo, E. B., Masulis, R. W., and Norli, O. (2007). Handbook of Corporate Finance , Elsevier, vol. 1, chap. Security Oerings. Fama, E. F. and French, K. R. (2008a). Average returns, b/m, and share issues. Journal of Finance , 63 (6), 2971 2995. Fama, E. F. and French, K. R. (2008b). Dissecting anomalies. Journal of Finance , 63 (4), 1653 1678. Fama, E. F. and French, K. R. (2015). A ve-factor asset pricing model. Journal of Financial Economics , 116 (1), 1 22. Fama, E. F. and MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical tests. Journal of Political Economy , 81 (3), 607. 40
Fu, F. and Huang, S. (2015). The persistence of long-run abnormal returns following stock repurchases and oerings. Management Science . Hong, H., Lim, T., and Stein, J. C. (2000). Bad news travels slowly: Size, analyst coverage, and the protability of momentum strategies. Journal of Finance , 55 (1), 265 295. Hong, H. and Stein, J. C. (1999). A unied theory of underreaction, momentum trading, and overreaction in asset markets. Journal of Finance , 54 (6), 2143 2184. Loughran, T. and Ritter, J. R. (1995). The new issues puzzle. Journal of Finance , 50 (1), 23 51. Lyandres, E., Le, S., and Lu, Z. (2008). The new issues puzzle: Testing the investment-based explanation. Review of Financial Studies , 21 (6), 2825 2855. McKeon, S. B. (2015). Employee option exercise and equity issuance motives. SSRN. McLean, D. R., Ponti, J., and Watanabe, A. (2009). Share issuance and cross-sectional returns: International evidence. Journal of Financial Economics , 94 (1), 1 17. Mitchell, M. L. and Staord, E. (2000). Managerial decisions and long-term stock price performance. Journal of Business , 73 (3), 287. Myers, S. C. and Majluf, N. S. (1984). Corporate nancing and investment decisions when rms have information that investors do not have. Journal of Financial Economics , 13 (2), 187 221. 41
‐1.00% ‐0.50% 0.00% 0.50% 1.00% 1.50% T‐10 T‐9T‐8T‐7T‐6T‐5T‐4T‐3T‐2T‐1 T T+1T+2T+3T+4T+5T+6T+7T+8T+9T+10 SEOabnormalreturn Figure 6: Value-weighted average cumulative abnormal return of SEO rms before and after the issue date ( T ). Abnormal returns are calculated as SEO rm return less market return. On average issues occur on the backdrop of almost 1% abnormal return between T−10 and T−2 . During the event window from close on T−2 to close on T+1 , issuers have negative abnormal returns of -1.6%. Subsequently, there is a partial recovery. 48
Table 1: Firm characteristics for value-weighted portfolios sorted on NetIssue . Repurchasers have NetIssue below -0.1%, zero-issuers have NetIssue between -0.1% and 0.1%. Small issues, mid issues and large issues, are net issuers with NetIssue of at least 0.1%, 3%, and 15%, respectively. bm is the logarithm of book-to-market ratio, Yield is the past year dividend yield, AG past year asset growth, ROE past year return on equity, EP earning-price ratio, and MV market value in million USD. Portfolio averages are value-weighted, except for MV . NetIssue group N NetIssue bm Yield AG ROE EP MV Repurchasers 271992 -0.033 -1.37 2.2% 9.9% 22.6% 5.6% 4.82 Zero issuers 196354 0.000 -1.17 2.9% 11.7% 15.9% 4.4% 0.98 Small issuers 488819 0.010 -1.34 1.9% 17.7% 13.2% 3.1% 1.95 Mid issuers 220901 0.068 -1.57 1.4% 30.6% 8.0% 2.3% 1.47 Large issuers 163997 0.387 -1.74 1.5% 48.0% 4.2% 0.9% 1.07 49
Table 2: Full-sample Fama-MacBeth regression 1985-2014. Each month rm excess return is regressed on rm characteristic expected to explain return. bm is the logarithm of the book-to-market ratio, mv is the logarithm for rm equity market value, MOM is past year return excluding the last month, ROE is return on equity, and AG is past year asset growth. Reported coecient estimates are time series averages with t -statistics reported in parenthesis. *, ** and *** indicates signicance in a two-sided test at a 10%, 5% and 1% signicance level, respectively. bm -0.02 0.07 (-0.17) (0.9) mv -0.04 -0.04 (-1.02) (-1.02) MOM 0.49** (2.05) ROE 0.41*** (2.94) AG -0.34*** (-2.79) Issue -0.89*** -0.77*** (-2.91) (-2.83) Repurchase 2.46* 1.32 (1.86) (1.17) Adj. R2 4.85% 8.01% 50
Table 3: Level of signicance of Issue , i.e. the positive component, of NetIssue and Repurchase , i.e. the negative component of NetIssue , in Fama-MacBeth regressions with 1-month return as dependent variable. In addition to Issue and Repurchase the independent variables are log book-to-market ratio bm , log market value mv , past year return MOM , return on equity ROE , and past year asset growth AG where all is specied. Easy and Hard refers to the most easy to value quintile and the most hard to value quintile, respectively, for the given sort variable. *, ** and *** indicates signicance in a two-sided test at a 10%, 5% and 1% signicance level, respectively. Details of regressions are reported in Table ?? Issue Repurchase Quintile Easy Hard Easy Hard Independent variables all bm + mv all bm + mv all bm + mv all bm + mv MV *** *** Yield *** *** ** AG *** *** * ** ROE ** *** * ** EP *** *** ** Deps ** #Analysts ** *** *** *** Drec ** Dptg ** *** ** ** 51
Table 4: Fama-MacBeth regressions with 1-month return as dependent variable. For each of the nine variables proxying for hard to value, rm-month observations are sorted in quintile samples, and Fama-MacBeth regressions are carried out for each quintile separately. Each sort variable is reported in separate panels (Panel A to I) In addition to Issue and Repurchase independent variables are log book-to-market ratio bm , log market value mv , past year return MOM , return on equity ROE , and past year asset growth AG . t -statistics are reported in parenthesis. *, ** and *** indicates signicance in a two-sided test at a 10%, 5% and 1% signicance level, respectively. The signicance of Issue and Repurchase reported in this table are summarized in Table 3. Panel A: Portfolios sorted by MV Quintile 1 2 3 4 5 1 2 3 4 5 bm 0.15 0.04 0.07 -0.03 -0.06 0.27*** 0.13 0.18* 0.12 0.06 (1.51) (0.31) (0.58) (-0.3) (-0.52) (3.04) (1.38) (1.85) (1.22) (0.64) mv -0.03 -0.15 -0.24 -0.08 -0.08 -0.03 -0.14 -0.25 -0.05 -0.05 (-0.35) (-1.13) (-1.48) (-0.72) (-1.42) (-0.41) (-1.17) (-1.53) (-0.51) (-1.06) MOM 0.63*** 0.4** 0.52** 0.52** 0.52 (4.58) (2.34) (2.56) (2) (1.62) ROE 0.12 0.21 0.23 0.7*** 0.63** (0.99) (1.36) (1.27) (2.74) (2.17) AG -0.68*** -0.54*** -0.44*** -0.19 -0.26 (-6.88) (-4.51) (-3.12) (-1.31) (-1.4) Issue -2.39*** -1.32*** -1.47*** -1.12** -0.11 -1.83*** -1.1** -1.19*** -0.97** -0.32 (-5.72) (-2.62) (-3.08) (-2.41) (-0.21) (-5.68) (-2.42) (-2.73) (-2.2) (-0.71) Repurchase 1.17 3.38*** 1.43 4.58*** 2.5 0.02 2.65** 0.75 3.94*** 0.78 (1.03) (2.65) (1.22) (3.18) (1.37) (0.02) (2.22) (0.66) (2.91) (0.46) Adj. R2 1.5% 1.8% 2.3% 2.5% 5.4% 2.7% 3.5% 4.8% 6.0% 10.5% 52
Panel B: Portfolios sorted by Yield Quintile 1 2 3 4 5 1 2 3 4 5 bm -0.15 -8.93 0.14 -0.03 0.12 0.04 0.46** 0.25** 0.12 0.29*** (-1.22) (-1.03) (1.01) (-0.29) (1.07) (0.34) (2.29) (1.98) (0.88) (2.59) mv 0.04 -0.01 -0.05 -0.08* -0.02 0.04 0.16 -0.04 -0.04 -0.01 (0.65) (-0.11) (-0.85) (-1.71) (-0.48) (0.67) (1.61) (-0.77) (-0.94) (-0.22) MOM 0.76*** 3.36** 0.51 0.22 0.39 (4.06) (2.11) (1.41) (0.64) (1.13) ROE 0.29** -0.54 1.03** 0.92** 1.18*** (2.42) (-0.76) (2.23) (2.26) (2.62) AG -0.41*** -0.15 -0.21 -0.24 -0.08 (-3.86) (-0.36) (-0.95) (-0.94) (-0.33) Issue -1.85*** -2.62 -0.9 -0.25 -0.15 -1.71*** -3.26* -0.72 0.34 -0.21 (-4.86) (-1.29) (-1.17) (-0.25) (-0.24) (-4.68) (-1.71) (-0.94) (0.37) (-0.38) Repurchase 4.6** 13.11 5.81** 0 3.1 2.15 18.42 3.18 0.42 1.65 (2.47) (0.84) (2.19) (0) (1.38) (1.28) (1.26) (1.27) (0.18) (0.8) Adj. R2 5.0% 10.8% 7.9% 9.7% 9.0% 7.5% 17.8% 13.5% 14.9% 16.4% 53
Panel C: Portfolios sorted by AG Quintile 1 2 3 4 5 1 2 3 4 5 bm 0.11 -0.02 -0.01 0.01 -0.21 0.16* 0.07 0.12 0.21* 0.06 (1.02) (-0.16) (-0.06) (0.06) (-1.64) (1.78) (0.65) (1.06) (1.88) (0.56) mv -0.02 -0.09* -0.08* -0.04 0.04 -0.02 -0.09** -0.05 -0.06 0.06 (-0.39) (-1.76) (-1.69) (-0.89) (0.78) (-0.35) (-1.98) (-1.16) (-1.18) (1.17) MOM 0.31 0.21 0.27 0.42 0.9*** (1.48) (0.74) (0.94) (1.47) (3.27) ROE 0.04 0.85** 0.8* 1.29*** 0.54** (0.27) (2.09) (1.84) (3.53) (2.53) AG 0.56 -1.65 1.1 -0.05 -0.42*** (1) (-0.57) (0.38) (-0.04) (-4.08) Issue -0.02 -1.05 -1.33* 0.04 -1.3*** -0.33 -0.92 -0.83 0 -0.97*** (-0.03) (-1.37) (-1.78) (0.07) (-3.36) (-0.6) (-1.26) (-1.28) (0) (-2.62) Repurchase 3.79** 3.86* -0.68 2.35 3.23 3.3* 2.44 -1.01 1.37 1.84 (2.01) (1.74) (-0.32) (1.03) (1.08) (1.89) (1.16) (-0.52) (0.64) (0.68) Adj. R2 7.1% 8.2% 7.8% 7.3% 7.3% 10.6% 13.0% 13.2% 11.7% 11.3% 54
Panel D: Portfolios sorted by ROE Quintile 1 2 3 4 5 1 2 3 4 5 bm 0.02 -0.1 0.06 0.15 0.06 0.21* 0.01 0.18 0.24 0.2* (0.2) (-0.6) (0.36) (0.82) (0.51) (1.93) (0.08) (1.15) (1.39) (1.71) mv -0.04 -0.11** -0.07 -0.04 -0.03 -0.02 -0.11** -0.04 -0.04 0 (-0.79) (-1.99) (-1.34) (-0.82) (-0.53) (-0.32) (-2.06) (-0.81) (-0.9) (-0.08) MOM 0.79*** 0.43 0.29 0.55* 0.57* (4.16) (1.55) (0.96) (1.95) (1.88) ROE 0.07 2.98 -2.5 1.86 0.66 (0.51) (1.17) (-0.58) (0.61) (1.22) AG -0.54*** -0.29* -0.37* -0.24 0.25 (-4.35) (-1.88) (-1.76) (-0.93) (0.89) Issue -1.19*** -0.98* -0.65 -0.83 0.05 -0.93** -0.6 -0.57 -0.76 -0.24 (-2.73) (-1.75) (-1.07) (-1.05) (0.07) (-2.17) (-1.21) (-1) (-1.1) (-0.38) Repurchase 4.83** 3.35 3.35 1.25 1.2 3.77* 3.42* 2.86 0.48 0.86 (2.25) (1.54) (1.59) (0.62) (0.58) (1.84) (1.7) (1.36) (0.25) (0.47) Adj. R2 5.4% 7.5% 7.5% 9.0% 7.1% 8.6% 11.6% 12.6% 14.0% 13.4% 55
Panel E: Portfolios sorted by EP Quintile 1 2 3 4 5 1 2 3 4 5 bm -0.02 -0.14 -0.08 -0.15 0.05 0.2* -0.05 0.47* 0.43 -0.08 (-0.2) (-1.34) (-0.67) (-1.21) (0.37) (1.93) (-0.29) (1.89) (1.47) (-0.49) mv -0.06 -0.02 -0.07 -0.07 0 -0.04 -0.01 -0.07 -0.06 -0.03 (-0.99) (-0.45) (-1.47) (-1.36) (-0.02) (-0.63) (-0.3) (-1.42) (-1.28) (-0.5) MOM 0.84*** 0.61** 0.2 0.03 0.53 (4.42) (2.37) (0.56) (0.08) (1.28) ROE 0.03 0.15 3.28** 3.11** -0.39 (0.26) (0.16) (2.49) (2.06) (-0.55) AG -0.35*** -0.29 -0.44* -0.07 -0.43* (-2.62) (-1.51) (-1.79) (-0.26) (-1.76) Issue -1.23*** -0.85 0.78 -2.88** -0.48 -1.13*** -0.63 0.7 -2.11** -0.19 (-3.18) (-1.54) (0.95) (-2.23) (-0.55) (-2.95) (-1.24) (0.92) (-2.14) (-0.22) Repurchase 6.03** -1.35 3.25 2.87 -0.02 3.86 -1.31 1.84 2.63 -0.48 (2.37) (-0.48) (1.49) (1.18) (-0.01) (1.62) (-0.52) (0.86) (1.26) (-0.29) Adj. R2 6.0% 6.2% 7.4% 8.1% 8.0% 9.5% 10.8% 12.7% 13.9% 14.2% 56
Panel F: Portfolios sorted by Deps Quintile 1 2 3 4 5 1 2 3 4 5 bm -0.11 0.17 0.43*** 0.35*** 0.04 0.09 0.37*** 0.47*** 0.26** 0.01 (-0.69) (1.4) (3.52) (2.79) (0.23) (0.55) (2.94) (3.82) (2.25) (0.07) mv -0.12** -0.13** -0.06 -0.02 -0.02 -0.08 -0.12** -0.09 -0.03 -0.03 (-2.06) (-2.31) (-1.12) (-0.29) (-0.19) (-1.47) (-2.28) (-1.62) (-0.58) (-0.36) MOM 0.79*** 0.47 0.16 -0.15 -0.01 (2.92) (1.48) (0.5) (-0.48) (-0.02) ROE 0.58 1** 0.76* -0.05 0.2 (1.11) (2.21) (1.83) (-0.15) (0.92) AG 0.14 -0.33 -0.79*** -1.07*** -0.84*** (0.59) (-1.5) (-3.77) (-4.87) (-3.59) Issue -0.19 -0.07 -0.9 -0.75 -1.53** -0.69 0.21 -0.12 -0.2 -1.05 (-0.22) (-0.09) (-1.13) (-1.03) (-2.03) (-0.95) (0.28) (-0.17) (-0.27) (-1.41) Repurchase 0.1 4.25** 2.48 3.81 5.31 0.33 2.33 1.47 3.93* 3.43 (0.04) (2.15) (1.06) (1.62) (1.6) (0.12) (1.29) (0.67) (1.67) (1.11) Adj. R2 8.8% 8.0% 8.0% 7.4% 8.1% 15.1% 13.7% 13.0% 11.8% 12.5% 57
Table 8: Monthly SEO portfolio returns regressed on FF3 (panel A and B ) and FF5+ UMD (panel C and D) factors. ER > 0 is the portfolio of SEO rms with positive event returns, whereas ER < 0 is the portfolio of SEO rms with negative event return. Pos-Neg is a zero-investment portfolio long SEO rms with positive event return and short SEO rms with negative event return. Period refers to the period during which a SEO rm is included in the portfolio, for example [1, 12] means that SEO rms are included in the portfolio from the month after issue until and including the twelfth month after issue. t -statistics are reported in parenthesis. *, ** and *** indicates signicance in a two-sided test at a 10%, 5% and 1% signicance level, respectively. Panel A: One-, twoand three-year SEO portfolio returns regressed on FF3. Period [1, 12] [1, 24] [1, 36] ER > 0 ER < 0 Pos-Neg ER > 0 ER < 0 Pos-Neg ER > 0 ER < 0 Pos-Neg α -0.06 -0.49** 0.43** -0.32* -0.56*** 0.25* -0.31** -0.50*** 0.20 (-0.28) (-2.31) (2.09) (-1.89) (-3.33) (1.73) (-2.02) (-3.15) (1.60) Mkt 1.12*** 1.19*** -0.07 1.13*** 1.17*** -0.04 1.14*** 1.18*** -0.04 (22.43) (24.64) (-1.53) (29.13) (30.06) (-1.12) (32.73) (32.28) (-1.43) SMB 0.47*** 0.54*** -0.07 0.40*** 0.48*** -0.08* 0.34*** 0.45*** -0.11*** (6.75) (7.97) (-1.03) (7.44) (8.84) (-1.68) (7.06) (8.85) (-2.77) HML -0.46*** -0.21*** -0.25*** -0.40*** -0.20*** -0.19*** -0.34*** -0.20*** -0.14*** (-6.34) (-3.01) (-3.65) (-7.04) (-3.60) (-4.12) (-6.80) (-3.78) (-3.53) Adj. R2 74.7% 75.8% 5.6% 82.2% 82.0% 6.3% 84.7% 83.8% 5.8% 64
Panel B: Secondand third-year SEO portfolio returns regressed on FF3 [13, 24] [25, 36] ER > 0 ER < 0 Pos-Neg ER > 0 ER < 0 Pos-Neg α -0.59*** -0.73*** 0.14 -0.24 -0.27 0.03 (-2.97) (-3.83) (0.74) (-1.24) (-1.29) (0.16) Mkt 1.14*** 1.15*** -0.01 1.12*** 1.17*** -0.05 (25.05) (26.30) (-0.13) (25.10) (23.92) (-1.04) SMB 0.33*** 0.35*** -0.01 0.19*** 0.45*** -0.26*** (5.26) (5.70) (-0.21) (3.11) (6.68) (-3.81) HML -0.25*** -0.19*** -0.06 -0.13** -0.21*** 0.08 (-3.77) (-2.98) (-0.97) (-1.99) (-2.92) (1.09) Adj. R2 75.4% 76.7% 0.4% 73.4% 74.1% 7.4% 65
Panel C: One-, twoand three-year SEO portfolio returns regressed on FF5+ UMD [1, 12] [1, 24] [1, 36] ER > 0 ER < 0 Pos-Neg ER > 0 ER < 0 Pos-Neg ER > 0 ER < 0 Pos-Neg α 0.06 -0.20 0.25 -0.08 -0.28* 0.20 -0.05 -0.17 0.12 (0.27) (-0.99) (1.19) (-0.45) (-1.71) (1.36) (-0.33) (-1.13) (0.95) Mkt 1.10*** 1.11*** -0.02 1.06*** 1.08*** -0.02 1.06*** 1.08*** -0.02 (20.92) (23.22) (-0.32) (26.00) (27.30) (-0.60) (29.49) (29.26) (-0.49) SMB 0.34*** 0.35*** -0.02 0.29*** 0.35*** -0.06 0.23*** 0.32*** -0.09** (4.69) (5.39) (-0.24) (5.21) (6.52) (-1.26) (4.60) (6.33) (-2.17) HML -0.15 0.26*** -0.40*** -0.10 0.16** -0.26*** -0.08 0.13* -0.21*** (-1.48) (2.83) (-4.11) (-1.34) (2.12) (-3.83) (-1.14) (1.91) (-3.60) RMW -0.50*** -0.72*** 0.22** -0.45*** -0.51*** 0.07 -0.45*** -0.52*** 0.08 (-4.93) (-7.78) (2.20) (-5.72) (-6.77) (0.96) (-6.48) (-7.42) (1.28) CMA -0.30** -0.56*** 0.26* -0.40*** -0.51*** 0.11 -0.36*** -0.49*** 0.13 (-2.14) (-4.38) (1.88) (-3.70) (-4.85) (1.14) (-3.73) (-4.94) (1.54) UMD 0.21*** 0.21*** 0.00 0.08** 0.10*** -0.02 0.05* 0.04 0.00 (4.84) (5.21) (0.08) (2.28) (2.94) (-0.66) (1.65) (1.45) (0.19) Adj. R2 77.6% 81.0% 6.7% 84.3% 85.0% 6.3% 86.8% 86.7% 6.8% 66
Panel D: Secondand third-year SEO portfolio returns regressed on FF5+ UMD [13, 24] [25, 36] ER > 0 ER < 0 Pos-Neg ER > 0 ER < 0 Pos-Neg α -0.20 -0.51*** 0.30 -0.04 0.08 -0.12 (-1.03) (-2.6) (1.55) (-0.22) (0.35) (-0.54) Mkt 1.01*** 1.06*** -0.05 1.07*** 1.06*** 0.00 (21.12) (22.43) (-1.02) (21.87) (20.46) (0.04) SMB 0.26*** 0.34*** -0.08 0.11 0.35*** -0.25*** (3.93) (5.29) (-1.30) (1.62) (4.97) (-3.29) HML 0.00 -0.02 0.02 -0.07 -0.08 0.02 (-0.02) (-0.20) (0.17) (-0.72) (-0.86) (0.17) RMW -0.35*** -0.10 -0.25*** -0.29*** -0.37*** 0.09 (-3.80) (-1.08) (-2.76) (-3.05) (-3.72) (0.81) CMA -0.51*** -0.43*** -0.08 -0.05 -0.22 0.17 (-3.96) (-3.40) (-0.62) (-0.39) (-1.62) (1.19) UMD -0.10** -0.07* -0.03 -0.06 -0.14*** 0.08* (-2.56) (-1.79) (-0.80) (-1.45) (-3.18) (1.73) Adj. R2 78.1% 78.0% 3.3% 74.4% 76.4% 9.0% 67
Appendix A Proposition 1. If µi,t =µi for all t≥T then Pt→P∗ for t→ ∞ The rate of convergence is Σ(1−τ) Σ(1−τ)+τ . Proof Let P∗=µi−Σaσ2 u dene the equilibrium price. By (2) the market clearing price is Pt=µt+stτ−Σaσ2 u Σ(1−τ)+τ,∀t≥T . Since st+1 =µi,t+1 −µu,t+1 =µi−Pt−aσ2 u and µu,t+1 =Pt+aσ2 u insertion in (2) yields Pt+1 =µt+1 +st+1τ−Σaσ2 u Σ(1 −τ) + τ=Pt+aσ2 u+(µi−Pt−aσ2 u)τ−Σaσ2 u Σ(1 −τ) + τ with the denition Λ = Σ(1 −τ) + τ Pt−P∗=µt+stτ−Σaσ2 u Λ−µi−Σaσ2 u = (µt−µi) + stτ Λ+ Σaσ2 u1−1 Λ =stτ Λ−1+ Σaσ2 u1−1 Λ and 68
Pt+1 −P∗=Pt+aσ2 u+(µi−Pt−aσ2 u)τ−Σaσ2 u Λ−µi−Σaσ2 u =Pt1−τ Λ+aσ2 u1+Σ−Σ + τ Λ+µiτ Λ−1 =µt+stτ−Σaσ2 u Λ1−τ Λ+aσ2 uΣ−Στ Λ−µi1−τ Λ =−st+stτ−Σaσ2 u Λ1−τ Λ+ Σaσ2 u1−τ Λ =1−τ Λstτ Λ−1+ Σaσ2 u1−1 Λ The rate of convergence is dened as γt=|Pt+1 −P∗| |Pt−P∗|= 1 −τ Λ=Σ(1 −τ) Σ(1 −τ) + τ Since γt is constant for t≥T , Pt converges linearly to P∗ provided that |γt|<1 which is the case for all τ∈]0,1[ . 69
70
The Issuance Eect in International Markets Niklas Kohl * Abstract Equity issuance predicts future low returns, but the reasons for this underperformance are disputed. I use an international sample and show that the underperformance by issuers is smaller in developed markets than in other markets. This empirical result is consistent with the hard to value theory which holds that underperformance is strongest for rms which are hard to value because informed investors are more constrained in their ability to express negative information (Kohl, 2016). However, the result contradicts the ndings of McLean et al. (2009), who argue that the underperformance of issuers is strongest in developed markets because lower issue costs induce issuers to exploit mispricings more frequently. To analyze this, I have developed a model with issue costs and information asymmetry between issuer and investors, where opportunistic issuers, to some extent, manage to sell overpriced equity. The model predicts that issuer underperformance is increasing in issue cost in line with the ndings in this paper. ∗ Department of Finance, Copenhagen Business School, Solbjerg Plads 3, 2000 Frederiksberg, Denmark. E-mail: nk.@cbs.dk. I am grateful for comments and suggestions received from Søren Hvidkjær, Nigel Barradale, and Ken Bechmann as well as seminar participants at Copenhagen Business School. Any errors remain mine. 71
1 Introduction Firms which issue new equity underperform subsequently, relative to other rms. This phenomenon was rst studied in the context of seasoned equity oerings (Loughran and Ritter (1995)). Later research has generalized this result to issuance in general and shown that rms which issue equity, on average, subsequently underperform (Daniel and Titman (2006), Ponti and Woodgate (2008), Fama and French (2008b), Fama and French (2008a)). McLean et al. (2009) show that this result also applies in international markets. While the issuance eect , i.e. the underperformance by issuers, is well documented, the reasons for underperformance are disputed. The classical behavioral explanation suggested by Loughran and Ritter (1995) is that rms announce issues when their equity is grossly overvalued and the market does not revalue the stock appropriately, and the stock is still substantially overvalued when the issue occurs. Ponti and Woodgate (2008), more cautiously, conclude that ... it appears doubtful that these results can be explained solely by a risk-based asset pricing model but do not suggest any particular behavioral explanation and do not rule out that the underperformance could be explained by a transaction cost model. 1 A risk-based explanation is given by Bessembinder and Zhang (2013) who nd that SEO underperformance is explained by risk factors including idiosyncratic volatility, liquidity, momentum and investment. When controlling for these factors, the issuance eect becomes insignicant. A related 1 Unfortunately, Ponti and Woodgate (2008) do not specify what transaction cost model they have in mind. In this paper, I show that issue transaction costs increase subsequent underperformance, but only in the presence of a deviation from rational expectations on the side of investors. 72
result is Lyandres et al. (2008) who report that about 75% of SEO underperformance is explained by an investment factor. Fu and Huang (2015) report negative, but insignicant, abnormal returns following SEOs during the period of 2003-2013. The authors suggest that this is because the pricing of stocks has become more ecient and rms less opportunistic in their issuance behavior. McLean, Ponti, and Watanabe (2009), in the following MPW, study the issuance eect in the cross-section of countries. According to MPW, the issuance eect is strongest in countries with greater issuance activity, greater stock market development, and stronger investor protection. The authors propose that this is because stock issuance and repurchases are cheaper in these more developed markets. This enables rms to be more opportunistic in their issuance activity, whereas In less developed markets where share issuance is more costly, the benets of market timing are exceeded by issuance costs, and share issuance occurs only for primary reasons. Regardless of whether one views the issuance factor as a priced risk or a mispricing, it may be a surprise that it is stronger in more developed, and presumably more ecient, markets than in less developed markets. This surprise is the starting point of the present paper. The empirical contribution of this paper is to reconrm the existence of the issuance factor in an international sample and, more importantly, to show that the issuance factor is stronger in non-developed markets than in developed markets. In order to do this, I proceed as follows. First, I reproduce some of MPW's ndings, which appear to document that the issuance eect is strongest in developed markets. Second, I discuss and demonstrate how these results are not robust to changes in methodological choices. With 73
and accounting data were obtained from Compustat. For international rms, these data were obtained from Thomson Datastream. The international sample consists of all rms followed by Worldscope. 7 To determine the country of rms in the Datastream sample the variable ISIN_ISSUER_CTRY is used. Firms for which this variable is not available and rms for which it is US are discarded from the Datastream part of the sample. All calculations are in USD. Some rms have issued more than one class of ordinary equity. For these rms only the share class with highest aggregate market value is used, but the market value of the rm is calculated as the sum of market values for all issued classes of ordinary shares. Following Daniel and Titman (2006), Ponti and Woodgate (2008), and McLean et al. (2009), I calculate net issue over the past year as NetIssuet,t−12 =ln(AdjustedSharest)−ln(AdjustedSharest−12) where AdjustedShares t is the time t adjusted number of shares. For US data, the adjusted number of shares is calculated using the the CRSP variables shrout and cfacshr whereas the adjusted number of shares in the international sample is calculated from the Datastream variables nosh and af . In both cases, the number of shares is adjusted for stock splits, stock dividends as well as other corporate actions such as rights issues. Net issue over the past two and three years, denoted NetIssue t,t−24 and NetIssue t,t−36 , respectively, is similarly calculated. McKeon (2015) shows that the majority of issues are small investor7 Worldscope is an international database of accounting and other fundamental data provided by Thomson Reuters. 80
initiated issues, for example in connection with the exercise of employee stock options. These issues are less likely to convey any information about the future stock return than rm-initiated issues. He suggests distinguishing between investorand rm-initiated issues using a 3% threshold. Previous research shows that this is indeed important. Fama and French (2008a) document that rms conducting small issues have slightly higher returns than zero-issuers while rms with large issuance activity underperform signicantly. Consequently, I dene the indicator variables issueri as 1 if net issue exceeds 0.03 over the past i years, and 0 otherwise. In addition to NetIssue , a number of variables are known to predict future return. MOM is the return over the past year excluding the past month. cap is the logarithm of rm market value. 8 bm is the logarithm of the ratio between book equity and market value. In some cases, I include observations for which bm cannot be calculated. To facilitate this, I follow MPW and dene bm = 0 and bmdummy = 1 when bm cannot be calculated while bmdummy = 0 for rms with known bm value. Asset growth AG is calculated as the relative change in book value of assets over the past year and return on equity ROE is calculated as the net income relative to book equity for US data. For international data, the Worldscope ROE variable wc08301 is used. All accounting variables are lagged by at least six months and only annual accounting values are used. 9 Throughout this paper, it is required that cap , MOM , NetIssue and next month return can be calculated. (rm, month) observations for which this is not possible are discarded from the sample. In some cases, observations 8 In general, lower case variables are the logarithm of the original variable. 9 Since wc08301 reports current scal year ROE it is lagged by 18 months. 81
without bm , AG and ROE are also discarded. Figure 3 shows the number of rms in the full sample per month while Figure 4 shows the distribution between the largest countries. In the 80s US rms accounted for around three quarters of the all rms and six developed markets (US, Japan, UK, Germany, France, and Canada) accounted for 95% of all rms. By 2014 the US accounted for less than 15% of all rms while the six mentioned countries accounted for a total of about 40%. This development primarily reects the growing coverage of the Datastream and Worldscope databases and shows that the early part of the sample is heavily biased toward a small number of markets. [Insert Figure 3 about here] [Insert Figure 4 about here] MPW study the relationship between the issuance factor and a number of country-specic variables proxying for market development and investor protection, including the frequency of stock issues, stock market liquidity, GDP per capita, and a number of proxies for investor protection and earnings management. I use some of these variables as reported by MPW. Moreover, I distinguish between developed markets , which are the markets identied as MSCI as developed and non-developed markets , which are all other markets. 10 10 MSCI dene the following markets as developed; Canada, United States, Austria, Belgium, Denmark, Finland, France, Germany, Ireland, Israel, Italy, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, Australia, Hong Kong, Japan, New Zealand, and Singapore. The majority of the other, i.e. non-developed, markets considered in this study are classied as emerging by MSCI. 82
4 Empirical Results The empirical analysis consists of three parts. First, I reproduce some of MPW's results on the issuance eect in the cross-section of markets. The purpose is to demonstrate that, through the use of my dataset, I can largely reproduce their results. Second, I show how the signicance and even sign of the results presented by MPW are sensitive to particular methodological choices. Finally, I propose a dierent methodology, which arguably is more suited to analyze the issuance eect in the cross-section of countries. Using this methodology, I show that the issuance eect is stronger in non-developed countries than in developed countries. 4.1 Reproduction of Selected Results from MPW According to MPW, the issuance eect is stronger in developed markets than in other markets. To show this, they use proxies for market development and corporate governance. They report Fama-MacBeth regressions with next month and next year return as dependent variables and factors known to predict return, cap , bm , bmdummy and MOM , as well as NetIssue , a proxy for market development or corporate governance and an interaction term ( NetIssue times the proxy for market development or corporate governance). The regressor of primary interest is the interaction term. MPW show that it estimates the marginal impact of the proxy in the issuance eect. If, for example, a proxy for market liquidity times NetIssue is negative and signicant, the inference is that an increase in the proxy for market liquidity is associated with a signicantly stronger (more negative) issuance eect. In MPW, the interaction term is generally signicant; for example, market 83
liquidity times NetIssue is signicantly negative. The inference is that the issuance eect is more negative (stronger) in countries with higher market liquidity. In the reproduction, I follow MPW and winzorize regressors within country at the 1st and 99th percentiles. These percentiles are calculated for each time period. 11 For international data, but not US data, observations with return below the 1st or above the 99th percentile are trimmed, i.e. removed from the sample. McLean et al. (2009) motivate this with ... many of these extreme observations appear to be the result of coding errors. The time period is June 1981 to July 2006. As MPW, for each (country, date) combination, I require at least 50 observations. Otherwise, the observations associated with the (country, date) combination are discarded. This concentrates the sample even further than suggested by Figure 4. In fact, the early sample, in my reproduction, consists of only eight developed countries and, by 1990, these eight countries still correspond to more than 96% of the sample. MPW report a broader initial sample including ve more countries of which two (Philippines and South Africa) are not developed markets. One reason for this dierence may be that I have limited the sample to rms covered by Worldscope. MPW propose a number of variables, broadly categorized under the headings issuance activity, market development and governance, which may be related to the magnitude of the issuance eect. I reproduce their results using most of these with values as reported in MPW. 12 Percentage with non-zero issuance is the fraction of (rm, month) observations with non-zero NetIs11 McLean et al. (2009) do not specify whether they calculate the 1st and 99th percentiles for each time period or for all observations for each country. 12 McLean et al. (2009), Table 7. 84
sue and is considered a proxy for the cost of share issuance. A number of marked development variables were originally introduced by La Porta et al. (2006). Liquidity is the value of stocks traded scaled by GDP for the period of 1996-2000 while Turnover is the value of stocks traded scaled by the total market value. GDP is the logarithm of GDP per capita. Governance variables used have their origin in La Porta et al. (1998). Common law is an indicator variable with the value 1 for common law countries and 0 otherwise, measuring the degree of investor protection. accounting is an index of accounting standards, where a higher value implies higher standards, liability and criminal measure how easily accountants, directors and distributors can be pursued in civil and criminal courts, respectively. Higher values indicate that this is easier. Investor protection is compiled from several measures with higher values indicating higher levels of investor protection. Higher values of earnings management indicate less reliable accounts. MPW show that these measures, except for earnings management , are almost uniformly positively correlated, while earnings management is negatively correlated to the other measures. Countries with more developed markets (more issuance activity, higher liquidity, higher turnover, and higher per capital GDP) have higher legal standards, higher levels of investor protection and less earnings management. Table 1 reports the results of Fama-MacBeth regressions with next month return as dependent variable and the same regressors as in MPW. Consistent with MPW I nd cap , bm and MOM to be signicant in all regressions. The regressor of primary interest is the interaction between NetIssue and variables proxying for market development and corporate governance. Figure 5 compares the t -value of the interaction variable reported by 85
MPW with my ndings. As MPW, I nd the interaction between NetIssue and market development variables - fraction of non-zero issuers, market liquity, market turnover and gdp - to be negative. Though my t -statistics are lower, the interaction is also signicant in my sample except for non-zero issuance. This suggests that higher level of market development is associated with a stronger (more negative) issuance eect. [Insert Table 1 about here] [Insert Figure 5 about here] In contrast to MPW, I do not nd common law and accounting standards to be signicant. For the legal variables Criminal and Liability as well as for investor protection and earnings management my results have same sign as those of MPW but only Criminal and investor protection are signicant. 13 Taken in their entirety, the results are qualitatively aligned with the results of MPW. 4.2 Sensitivity of the MPW results The second part of my empirical results examine the impact of a number of the methodological choices in MPW. The purpose is to demonstrate that results, and consequently inference, is sensitive to these choices. Alternative choices, which may be as justiable as those made by MPW, destroy the signicance, and in some cases even reverses the sign of results. 13 The interaction between NetIssue and Criminal is positive and signicant in MPW as well as my ndings. This implies, that the issuance eect is weaker (less negative) in countries where directors etc. can more easily can be pursued in criminal court. This result is inconsistent with the predictions of MPW. 86
I consider the following empirical choices; First, MPW only report equalweighted Fama-MacBeth regressions. Since the vast majority of rms are small, especially in the MPW sample which includes rms outside the Worldscope universe, equal-weighted results may reect phenomena which only or primarily exist for small cap rms. Second, the early sample is dominated by US rms and the representation of rms from non-developed economies is particularly low in the early sample. For this reason, it is natural consider whether the results hold in a later period with broader international coverage. Third, the winzorization methodology employed by MPW may be disputed. Though winzorization is justied to eliminate the impact of outliers and coding errors, it is not given that winzorization should be performed independently for each country. A particular concern is that winzorization at the 1st and 99th percentiles will have no impact on samples (countries) with at most 100 observations. This also holds for the trimming of the sample at the 1st and 99th percentile next month return applied to non-US data. Trimming is justied as Ince and Porter (2006) show that Datastream contains errors, resulting in implausibly large returns, but trimming independently by country does not aect small sample countries. Further trimming of the international data at the 99th percentile is likely to introduce a downwards bias on international returns. On average, international observations are trimmed above a monthly return of about 61%. Though this is a high return, it is far from implausibly high, thus the vast majority of observations removed are likely to be genuine observations. Fourth, MPW only control for cap , bm and MOM . By now, it is well established that controlling for investment and protability is appropriate 87
when measuring the issuance eect ((Lyandres et al. , 2008), (Bessembinder and Zhang, 2013)) since a part of the issuance eect is explained by these factors 14 . In tests for robustness with respect to the four issues mentioned above, I redo the regressions reported in Section 4.1 changing one issue at a time as well as changing all four. For each regression, three t -statistics are presented: The t -value report by MPV, the t -value reported in subsection 4.1 and t -value with modied methodological choices. Figure 6 shows results with value-weighted Fama MacBeth regressions as the only change while Figure 7 shows the period January 1990 to December 2014 as the only change (i.e. equal-weighted regressions). Both these changes almost uniformly reduce the signicance of results, though the sign generally remains unchanged. [Insert Figure 6 about here] [Insert Figure 7 about here] Several modications of the winzorization and trimming scheme are possible. In Figure 8, regressors are trimmed at their global 1st and 99th percentiles, instead of national calculation of winzorization values. As above, international return observations are trimmed at their global 1st and 99th percentiles. This approach, which also ensures winzorization and trimming for countries with less than 100 observations, reduces the signicance of results. 14 A fth debatable choice by MPW is the choice to exclude observations from countries with less than 50 observations. Since country is not used in the Fama MacBeth regressions, there is no particular reason to exclude observations from countries with few observations. However, including these observations only has a minor impact on results. 88
Changes in trimming methodology may also increase signicance. Figure 9 shows results if all countries, including the US, are treated equally in the sense that regressors are winzorized at their national 1st and 99th percentiles while return observations are trimmed at their national 1st and 99st next month return percentiles. This approach, which ensures that extreme returns are also trimmed for US data, increases the signicance of the interaction variables, in most cases to the levels reported by MPV. 15 [Insert Figure 8 about here] [Insert Figure 9 about here] In Figure 10, the sample is limited to observations for which bm , ROE and AG are known, and these are used as controls in the Fama MacBeth regressions (in addition to the regressors used in the previous regressions). In most cases, this reduces the signicance of results. Finally, in Figure 11, results with four simultaneous changes are reported. Regressions are value-weighted using all controls including bm , ROE and AG . The period spans January 1990 to December 2014. Regressors are winzorized at their global 1st and 99th percentiles and international observations are trimmed at their global 1st and 99th next month return percentiles. With these choices, the sign of all interaction variables, except one, is reversed relative to the results reported by MPW and the only signicant result is that higher levels of earnings management are associated with a more negative (stronger) issuance eect. 15 Since percentiles are calculated per country no winzorization and trimming takes place for countries with less than 100 observations, as discussed above. 89
Lyandres, E., Le, S., and Lu, Z. (2008). The new issues puzzle: Testing the investment-based explanation. Review of Financial Studies , 21 (6), 2825 2855. McKeon, S. B. (2015). Employee option exercise and equity issuance motives. SSRN. McLean, D. R., Ponti, J., and Watanabe, A. (2009). Share issuance and cross-sectional returns: International evidence. Journal of Financial Economics , 94 (1), 1 17. Myers, S. C. and Majluf, N. S. (1984). Corporate nancing and investment decisions when rms have information that investors do not have. Journal of Financial Economics , 13 (2), 187 221. Ponti, J. and Woodgate, A. (2008). Share issuance and cross-sectional returns. Journal of Finance , 63 (2), 921 945. 96
Figure 1: The issuance decision when rm management knows the per share value of assets in place a and the investment opportunity b and investors only know the distribution of these (Myers and Majluf, 1984). The solid line depicts the case without issue costs. The issue price is p and the equity to be raised to invest is E . The rm issue if b > E pa−E , the upper-left region M' and do not issue otherwise (region M ). If b = 0, the rm will issue if a<p . The dashed line depicts the case with issue cost c . Everything else being equal, the value of the investment opportunity must be higher than in the case without issue costs before the rm choses to issue. 97
Panel A Panel B Figure 2: Equilibrium issue price p∗ , probability of issue, event return, average long-run return, and average a and b if the rm issues, all as function of issue costs c (on the x-axis). a is uniformly distributed on [20,40] , b on [0,10] , and equity to be raise in the issue E = 20. Qualitatively similar results are obtained with dierent a and b intervals and dierent E values. 98
0 5000 10000 15000 20000 25000 30000 35000 Figure 3: Number of monthly rm observations in the full dataset. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% SouthKorea India China Australia France Germany UK Canada Japan US Figure 4: The largest countries as fraction of the full sample. 99
‐8 ‐6 ‐4 ‐2 0 2 4 6 Non‐zero issuance Marketliquidity Turnover GDP Commonlaw Accounting Criminal Liability Investor protection Earnings management MPW MPWreproduction Figure 5: Comparison of t -statistics of the interaction between NetIssue and variables proxying for issuance activity, market development and corporate governance reported by MPW and the results reported in this paper in Table 1 (where details of the regressions are reported). Taken in their entirety, the ndings of MPW are conrmed. 100
‐8 ‐6 ‐4 ‐2 0 2 4 6 Non‐zero issuance Market liquidity Turnover GDP Commonlaw Accounting Criminal Liabililty Investor protect Earnings management MPW MPWreproduction ValueWeighted Figure 6: Comparison of t -statistics for the interaction term, as reported by MPW, from the reproduction of MPW reported in Table 1 and from a reproduction with value-weighted observations otherwise identical to the reproduction of Table 1. Value weighting generally reduces the signicance, but not the sign, of results. 101
‐8 ‐6 ‐4 ‐2 0 2 4 6 Non‐zero issuance Market liquidity Turnover GDP Commonlaw Accounting Criminal Liabililty Investor protect Earnings management MPW MPWreproduction 1990‐2014 Figure 7: Comparison of t -statistics for the interaction term, as reported by MPW, from the reproduction of MPW reported in Table 1 and from a reproduction using only data from the period of 1990-2014 (where international coverage is broader) and otherwise identical to the reproduction of Table 1. Using a broader sample and shorter time period generally reduces the signicance, but not the sign, of results. 102
‐8 ‐6 ‐4 ‐2 0 2 4 6 Non‐zero issuance Market liquidity Turnover GDP Commonlaw Accounting Criminal Liabililty Investor protect Earnings management MPW MPWreproduction Int'lWinzorization Figure 8: Comparison of t -statistics for the interaction term, as reported by MPW, from the reproduction of MPW reported in Table 1 and from a reproduction where regressors for international observations are winzorized at their global (instead of national) 1st and 99th percentiles and otherwise identical to the reproduction of Table 1. This generally reduces the signicance, but not the sign, of results. 103
‐8 ‐6 ‐4 ‐2 0 2 4 6 Non‐zero issuance Market liquidity Turnover GDP Commonlaw Accounting Criminal Liabililty Investor protect Earnings management MPW MPWreproduction UStrimmed Figure 9: Comparison of t -statistics for the interaction term, as reported by MPW, from the reproduction of MPW reported in Table 1 and from a reproduction where all regressors, also for US observations, are winzorized at their national 1st and 99th percentiles and returns are trimmed at their national 1st and 99th percentiles, and otherwise identical to the reproduction of Table 1. In most cases this restores signicance to the level reported by MPW. 104
‐8 ‐6 ‐4 ‐2 0 2 4 6 Non‐zero issuance Market liquidity Turnover GDP Commonlaw Accounting Criminal Liabililty Investor protect Earnings management MPW MPWreproduction Allcontrols Figure 10: Comparison of t -statistics for the interaction term, as reported by MPW, from the reproduction of MPW reported in Table 1 and from a reproduction limited to observations with known bm , ROE , and AG where the latter two are included in the set of regressors, and otherwise identical to the reproduction of Table 1. This generally reduces the signicance, but not the sign, of results. 105
112
Does Information Asymmetry Explain Issuer Underperformance? Niklas Kohl * Abstract Firms which issue new equity have lower returns than other rms subsequent to issue. A prominent behavioral explanation holds that opportunistic rms exploit information asymmetry at issue time to sell overvalued equity (Loughran and Ritter, 1995). However, this paper shows that the most overvalued issuers, and those which are least constrained in the sense that they do not need to issue to continue operations or service current debt, have as high or higher long-run run returns than other issuers. Instead, I show that event returns, and in particular negative evnet returns (bad news) at event time predicts long-run abnormal return. This result is consistent with investor underreaction to available information, rather than information asymmetry at event time. ∗ Department of Finance, Copenhagen Business School, Solbjerg Plads 3, 2000 Frederiksberg, Denmark. E-mail: nk.@cbs.dk. I am grateful for comments and suggestions received from Søren Hvidkjær and Ken Bechmann. Any errors remain mine. 113
1 Introduction Listed rms which issue new equity subsequently have low returns. This has been documented in numerous studies, initially by Loughran and Ritter (1995), in the context of seasoned equity oerings (SEOs), and later in the broader context of rms which issue or retire equity, regardless of reason ((Daniel and Titman, 2006), (Ponti and Woodgate, 2008), (Fama and French, 2008b), (Fama and French, 2008a), (McLean et al. , 2009)). The reasons for the low returns are, however, disputed. Loughran and Ritter (1995), suggest that rms announce issues when their equity is grossly overvalued and the market does not revalue the stock appropriately, and the stock is still substantially overvalued when the issue occurs. According to the authors, their ... evidence is consistent with a market where rms take advantage of transitory windows of opportunity by issuing equity when, on average, they are substantially overvalued. Several more recent papers, including Ponti and Woodgate (2008), also fail to nd risk-based explanations for low returns post-issue. A competing stream of literature argues that the apparent underperformance subsequent to issue is due to exposure to known risk factors or at least known priced factors, which may or may not proxy for risk. Examples include Eckbo et al. (2000), Lyandres et al. (2008), and Bessembinder and Zhang (2013). Billett et al. (2011) nd that repeating issuers, regardless of the type of security issued, underperform signicantly, while rare issuers do not. A recent paper by Fu and Huang (2015) argues that signicant issuer underperformance has ceased to exists during the period of 2003-2012 because rms 114
become less opportunistic in stock repurchases and oerings due to more ecient pricing of stocks. This paper contains two main results. First, I show that a large portion of issuer long-run performance is explained by exposure to priced factors beyond the Fama-French three factor model (Fama and French (1992), Fama and French (1993)). However, some signicant underperformance remains unexplained. Second, and more importantly, I investigate whether issuer long-run underperformance, as suggested by Loughran and Ritter (1995), is due to opportunistic rms' exploitation of information asymmetry at event time. If information asymmetry is high at event time opportunistic rms may attempt to exploit this and, unless investors have rational expectations, rms may be successful at it. Thus, information asymmetry in combination with opportunistic issues and deviation from rational expectations at event time may explain long-run underperformance. As an alternative to this explanation, I consider the possibility that information asymmetry is low at event time. If information asymmetry is low at event time, rms will have less opportunity to be opportunistic in their issuance behavior. Nonetheless, long-run underperformance is possible due to investor underreaction at event time. There may be several reasons for investor underreaction. Barberis and Thaler (2003) survey a number of psychological biases which may aect how investors form their beliefs. In particular, conservatism, belief perseverance, and anchoring may all explain why investors do not fully incorporate new information in prices immediately. Alternatively, delayed price reactions (underas well as overreaction) can occur in models with gradual diusion of information (Hong and Stein, 1999) and 115
models with inattentive investors (Due, 2010). Empirically delayed price reaction is found in a number cases, including post earnings announcement drift (Bernard and Thomas, 1989) and post dividend change announcement drift (Michaely et al. , 1995). For these two possibilities, information asymmetry and investor underreaction , I derive testable implications. Empirical results are consistent with the investor underreaction hypothesis but not the information asymmetry hypothesis. I nd no empirical evidence of the exploitation of information asymmetry because the most overvalued issuers and the least constrained issuers, which do not need to issue to nance operations or service current debt, overperform or have similar performance compared to less overvalued issuers and more constrained issuers. In contrast, I nd that the market does not fully absorb information conveyed at event time, in particular bad news at event time. This causes long-run return predictability, in particular when event returns are negative. To the best of my knowledge, this is a new nding. 1 While early research focused on the performance of seasoned equity offering (SEO) rms, most recent work on the relation between issuance and return considers the full cross-section of rms to capture the impact of equity issues and repurchases, regardless of reason. This approach has the advantage of a much larger sample than studies focused on SEOs and, according to Ponti and Woodgate (2008), ... results are essentially unaected by data associated with seasoned equity oerings ..., documenting that the low returns subsequent to SEOs is part of a broader issuance eect. 1 Apart from the preliminary version of this result found in the rst paper of this dissertation. 116
However, McKeon (2015) shows that in around 90% of rm-quarters where new equity is issued, the rm itself did not initiate any stock issues. These are issues due to, for example, utilization of employee options and other decisions beyond the control of the rm. McKeon denotes these investor-initiated issues as opposed to rm-initiated issues, typically in the form of SEOs, where the rm takes initiative to issue new equity. McKeon (2015) shows that relative issue size, i.e. proceeds of the issue relative to rm market value, is an empirically strong indicator of rm-initiated issues since quarters with a relative issue of at least 3% nearly always contain a rm-initiated issue whereas quarters with a relative issue of less than 2% almost never include a rm-initiated component. Since the objective of this paper is to test whether information asymmetry explain subsequent issuer performance, I limit the sample to situations where this may possibly have occurred, i.e. to rm-initiated issues. Limiting the sample to SEOs has the further advantage that an announcement event can be clearly identied. I reconrm that SEO announcements do, in fact, convey information, since event returns are, on average, strongly signicantly negative. To measure the sign of the information conveyed, I interpret negative event return as bad news and the less frequent positive event return as good news. This enables me to determine to what extent information conveyed at event time is fully incorporated in prices at event time. The outline of the remainder of the paper is as follows. In Section 2, I discuss what could drive issuer underperformance. I show how information asymmetry at event time can cause underperformance at event time and subsequently, and I present an alternative - that underperformance is caused by underreaction to news at event time. Section 3 presents data and vari117
ables used. Section 4 presents average issuer returns before and after issue, conrming high abnormal return pre-issue, negative abnormal event return, and negative abnormal return post-issue. In Section 5, event returns are regressed on characteristics hypothesized to explain event return. Section 6 examines the explanations behind issuer long-run abnormal returns. Two dierent methodologies are applied. First, buy and hold abnormal returns, calculated relative to dierent factor models, are regressed on characteristics hypothesized to explain long-run return. Second, issuers are sorted on these characteristics and calendar-time portfolios of issuers, and long-short calendar-time portfolios of issuers matched with non-issuers, are constructed. Finally, Section 7 concludes. 2 What Drives Issuer Underperformance? This section discusses possible reasons for issuer underperformance. Section 2.1 explains how information asymmetry at event time may drive subsequent negative abnormal returns, while Section 2.2 explores how investor underreaction, in the absence of information asymmetry, may drive subsequent underperformance. Based on these two sections, hypotheses are presented in Section 2.3. Two of the hypotheses concern the relation between issuer overvaluation and whether the issuer is constained, respectively, and long-run abnormal return. Proxies for issuer overvaluation and issuer constrainedness are presented in Section 2.4 and 2.5. 118
2.1 Information Asymmetry Information asymmetry as a driver of returns in connection with stock issues was rst proposed by Myers and Majluf (1984), henceforth MM, and is also the driver for negative long-run returns in Loughran and Ritter (1995). 2 To understand the relationship between information asymmetry, issuance event return and long-run return, consider the MM model. Firms have assets in place with per share value a≥0 and an investment opportunity with per share net present value b≥0 , which cannot be postponed and which must be nanced with equity E , per share, raised from new investors. 3 Firm management acts in the interest of old shareholders. At event time rm management knows the realization of a and b , while investors only know the distribution of a and b . Figure 1 depicts the rms' issuance decision. The rms' choice of whether to issue or not depends on a linear combination of the realization of a and b . In particular, rms will issue if a is suciently low or b is suciently high. It is natural to think of this as two dierent reasons to issue: to exploit overvaluation or to pursue attractive investment possibilities. The issue price is P0 and new investors will experience positive post-issue returns if a+b exceeds P0 (the region in M0 above the dotted line) and negative post-issue returns otherwise (the region in M0 below the dotted line). If investors have rational expectations, the equilibrium price P0 must be E(a+b|(a, b)∈I) and 2 While these two papers share the assumption that rm management knows much more than investors at event time, i.e. after an issue has been announced, they dier in terms of whether investors realize this. In MM investors realize their ignorance and purchase new equity at its expected value, taking into account that rms are more likely to issue when they are overvalued than when they are undervalued. In Loughran and Ritter (1995) investors do not fully realize their ignorance and overpay for new equity. 3 In MM, these values are not per share but for the entire rm. Without loss of generality, I assume the rm has one share outstanding before issue and that the rm can issue any fraction of shares. 119
average post-issue abnormal return will be 0. Accordingly, the MM model predicts negative event return because the expected value of a+b conditioned on (a, b)∈I is lower than the unconditional expectation of a+b , but does not predict negative long-run returns because investors have rational expectation. In order to generate long-run underperformance due to information asymmetry, a deviation from rational expectations is required, i.e. if average postissue return is negative, it implies that investors pay too much for new equity and, consequently, that the marginal investor does not fully incorporate the impact of information asymmetry. Several behavioral biases may generate this result including those which may generate underreaction, as discussed in Section 1. Regardless of whether issuers, on average, underperform, the MM model predicts that long-run return will be lowest for issuers with low a+b , i.e. overvalued issuers. Moving beyond the MM model, some issuers may also have the ability to issue during periods where the market value of their equity is particularly high, as suggested by Loughran and Ritter (1995), McLean et al. (2009) and Greenwood and Hanson (2012). Empirically, these issuers have particularly low subsequent returns. [Insert Figure 1 about here] 2.2 Investor Underreaction In the above model information asymmetry, in combination with opportunistic issuers and some deviation from rational expectations on the side of investors, explains negative average post-issue return. However, what if 120
information asymmetry between rm management and issuers is low when issues occur? There are both theoretical and empirical reasons to consider this possibility. Miller and Rock (1985), henceforth MR, consider a situation where investors know the distribution of current earnings but rm management know the actual realization. Future earnings depend on current investments and the production function is concave. They develop a fully revealing signaling equilibrium where rms signal earnings with payouts. 4 Since stock issues are negative payouts, an issue signals low earnings. Investors interpret the signal correctly and announcement returns will be positive or negative, depending on whether the earnings signaled are higher or lower than investors' (unobserved) expectations. In terms of issuance, MR and MM dier in two important ways. First, while MM always predicts negative event returns, event returns may by positive in MR provided that the issue is smaller than expected by investors. In this case, MR investors will interpret the issue as god news. Second, while both models are rational expectations models with on long-run issuer underperformance, the equilibrium in MR is fully revealing in the sense that 4 Period t earnings are given by Xt=f(It−1) + t , where f is the production function, It−1 previous period investments and t a random increment with zero mean. It=Xt+ Bt−Dt where Bt is time t nancing and Dt payouts (dividends, stock repurchases or stock issues), respectively. At time t rm management knows Xt but investors know only f(It−1) . Firm management acts partly in the interest of investors who will sell their equity before Xt is revealed, i.e. those who want to maximize current share price, and partly in the interest of investors who will stay invested, i.e. those who want the rm to invest optimally. In the fully revealing equilibrium rms use Dt to signal Xt . Payouts will be higher and investments lower than under the Fisher rule, but because f00 <0 smaller deviations from the Fisher rule will be cheaper, in term of lost future earnings, than larger deviations. Hence, it will not be optimal for rms with low earnings to pay as large payouts as rms with high earnings. 121
may reect limited coverage in SDC before approximately 1990. Despite this the correlation between market excess return and number of issuers is 0.40 ( t -value 9.1). Second, since overvalued rms are more likely to issue, issuers will, on average, have past positive abnormal returns. In Section 4, I show that issuers have, on average, very high abnormal returns before issue. Both these patterns have been known since Loughran and Ritter (1995). [Insert Figure 2 about here] 2.5 Proxies for Constrainedness Testing Hypothesis 2 requires a proxy for issuer constrainedness. Constrainedness implies that the issuers had to issue to continue operations, i.e. to nance operations or to service current debt. At the extreme, these rms must raise new equity to avoid bankruptcy and protect some value for current shareholders. I denote these defensive issuers because they must issue to survive, whereas other issuers choose to issue to pursue attractive investment possibilities or exploit overpricing. In Section 3, I present a number of measures for the extent to which an issuer is defensive. Most of these consider to what extent the rm can pay o current debt with existing cash and cashow from operations and the level of cashow from operations. 11 Hypothesis 2 implies that defensive issuers will have higher long-run abnormal returns than other issuers. One may be concerned that other returnpredicting characteristics of defensive issuers systematically dier from nondefensive issuers. In particular, defensive issuers may have lower protability. Since high protability, for example, measured as exposure to the Fama and 11 Current debt is debt due within a year. 128
French (2015) RMW factor, is a strong return predictor, it is essential to control for this in abnormal return calculations. Another concern with the testability of Hypothesis 2 is that Baker et al. (2003) show that ... the eects of stock market valuations (ecient or otherwise) on investment are greater for more nancially constrained (equity dependent) rms. Since rms which issue when market valuations are high subsequently have particular low returns (Loughran and Ritter, 1995), this could drive underperformance of defensive issuers. Therefore, it is important to control for past market return when the relation between defensiveness of the issuer and long-run return is analyzed. 3 Data and Variables 3.1 SEO Sample Daily and monthly stock returns are sourced from CRSP. Only ordinary equity is selected. 12 The CRSP Compustat merge was used to obtain accounting data. Only annual accounting data are used and all accounting data are lagged by at least six months. Daily and monthly market excess returns as well as factor returns (SMB, HML, RMW, CMA, and WML), were collected from Kenneth French homepage. SEO events were gathered from the SDC Platinum database available through Thomson One Banker. The sample contains all completed SEOs (in SDC denoted Follow-On oerings) with issue date in 2015 or earlier. The relative issue size, i.e. proceeds raised divided by the market value before the 12 First digit of the CRSP shrcd code is 1. 129
oer, both as reported in SDC, is denoted Issue . 13 Observations where Issue cannot be calculated are discarded. Moreover, I require Issue to be at least 3%. This is to eliminate very small issues with limited information content. It reduces the sample by around 8%, with limited impact on empirical results. Issuers with pre-oer valuation of less than $100 million in CPI adjusted December 2015 prices are removed from the sample. This is to eliminate phenomena which only exist in micro caps. It reduces the sample by around 11% with limited impact on most empirical results. Finally, nancials and insurance companies are eliminated from the sample. 14 The SEO announcement date is the Original date reported by SDC. I have, for a small sample of records, veried that this date is indeed the day the issue was announced. Occasionally, SDC records more than one SEO event for a given rm with a given announcement date. These dierent records typically represent issues in dierent markets, to dierent investor groups or using dierent issuance methods. In any case, these records are merged into one SEO observation. CRPS and SDC observations are matched based on their cusip number. The matched sample consists of 11,106 SEO observations. Figure 2 shows how the number of observations, satisfying the criteria mentioned, has evolved since 1980. 3.2 Abnormal return Issuer abnormal returns are used in a number of sorts and regressions as a dependent as well as an explanatory variable. In all cases, abnormal return 13 Issue is calculated using the SDC variables Proceeds__Amt___sum__of_all_Mkts and Market_Value_Before_Offer____mil . The latter is also used together with the CRSP CPIIND variable to calculate rm market value at December 2015 prices. 14 Issuers with Standard Industrial Classication Code between 6000 and 6499. 130
over period p is calculated as rp abn,MM =rp excess −X i∈MM ˆ βp irp i (1) rp excess is the return in excess of the risk-free rate. Market model MM is a set of return-predicting factors, rp i is the return of factor i over period p and ˆ βp i is the estimated exposure to factor i . I primarily consider the Fama-French ve factor model (Fama and French, 2015), denoted FF5, but occasionally also consider other models including the Fama-French three factor model, denoted FF3, and CAPM. For abnormal returns before and at issuance announcement event time, factor loadings ˆ βp i are estimated using daily excess returns from one year before announcement to one month before announcement. For abnormal returns after issue, factor loadings are estimated using data from one month after announcement to one year after announcement. 15 In both cases, excess returns are regressed on factor returns with three daily lags and the estimated loading ˆ βp i is the sum of estimated loadings on the contemporaneous factor return and the three lagged factor returns (the Dimson (1979) method). Factor loadings are only estimated with at least 200 degrees of freedom in the regressions, i.e. when rm returns are available almost daily. 16 Table 2 reports average FF5 estimated factor loadings before and after issue. Consistent with Loughran and Ritter (1995), issuer market betas are slightly above 1. ˆ βMkt is 1.10 before issue, increasing to 1.13 after. The increase of 0.03 is small but statistically signicant, suggesting that issuers, 15 In most cases the issue date is either the announcement data or the day after. 16 Estimating FF5 loadings requires 221 observations: one for the model intercept, ve for contemporaneous factor returns, 15 for lagged factor returns plus 200 degrees of freedom. 131
on average, do not issue to strengthen their balance sheet but rather to invest. Consistent with results reported by Greenwood and Hanson (2012), the average issuer is small cap. Issuers load heavily on SMB with ˆ βSMB of 0.81 before and 0.76 after. The decrease is as expected, since issues increase the market value of the rm. ˆ βHML decreases from -0.07 to -0.13, again suggesting that issuers invest rather than strengthen their balance sheet. This also applies to the decrease in the asset growth factor ˆ βCMA from - 0.07 to -0.15, i.e. issuers become more aggressive post-issue. Greenwood and Hanson (2012) also report low issuer protability. The table reects this, with ˆ βRMW loadings of -0.31 before and -0.40 after issue. [Insert Table 2 about here] 3.3 Dependent and Explanatory Variables in Regressions In the event regressions, Section 5, the dependent variable is abnormal event returns, revent abn,F F 5 calculated using equation 1 and event excess return revent excess . Regressors are past market excess return ( r−n year Mkt , n ∈ {1,2,3} ), measured from one, two and three years before announcement to one month before announcement and issuer abnormal return prior to announcement ( r−n year abn,F F 5, n ∈ {1,2,3} ), measured from one, two and three years before to one month before announcement. Further regressors are issue =log(1 + Issue) and the logarithm of equity market value (denoted mv ). In long-run abnormal return regressions reported in Section 6.1 one-, twoand three-year post-announcement abnormal returns, denoted r1year abn,F F 5 , 132
r2year abn,F F 5 and r3year abn,F F 5 , respectively, are regressed on factors hypothesized to explain (and predict) long-run return. The abnormal returns are calculated from one month after announcement to one, two and three years after announcement using factor exposures estimated post-announcement and market models FF5. In addition to the regressors mentioned above, abnormal event return is used as regressor in long-run abnormal return regressions. Further abnormal event return is decomposed into its positive and negative component revent+ abn,F F 5=max(revent abn,F F 5,0) and revent− abn,F F 5=max(−revent abn,F F 5,0) . The purpose of this is to determine whether positive event returns, i.e. good news conveyed at event time, aect long-run returns dierently than negative event returns, i.e. bad news conveyed at event time. In the entire sample, 26% of the events have positive event return and the annual fraction is almost always between 20% and 40% with a downward sloping trend over the past 15 years. In Section 6.1.2 long-run abnormal return is regressed on proxies for being a defensive issuer, in addition to the explanatory variables mentioned above. Table 3 summarizes characteristics related to whether the issue is likely to be defensive. In all cases, low values are associated with more defensive issues. The cash ratio CR measures the ratio between cash and current debt, i.e. debt due within one year. 17 To eliminate the impact of extreme observations, the calculated cash ratio is projected on the interval [0,5] . CR1 is a binary variable measuring whether cash exceeds current debt. [Insert Table 3 about here] 17 CR is calculated using the Compustat variables ch and dlc lagged by at least six months. 133
The cash and cashow ratio CCF is the ratio between cash plus operating cashow and debt. 18 CCF is projected on the interval [−5,5] . The binary variable CCF1 is one if cash plus operating cashow exceeds current debt, and zero otherwise. The cashow yield CFY is the ratio between operating cashow and issuer market value. The binary variable PosCF is 0 if operating cashow is negative, and 1 otherwise. Firms without any long-term debt are also likely to be defensive issuers because the lack of any debt may be caused by inability to borrow. The binary variable LTDebt is 0 for issuers without any long-term debt and 1 otherwise. 19 Finally, dividend paying issuers, i.e. issuers which have paid a dividend over the year before issue, are likely to be less defensive than non-dividend paying issuers. PosDiv is 0 for issuers which have not paid a dividend, and 1 otherwise. A substantial number of issuers delist within three years after issue. In long-run abnormal regressions, proceeds including delisting returns, as reported by CRSP, is assumed to be reinvested in the market portfolio. In unreported results, delisting rms were omitted from the sample. This does not change the results substantially. In the calendar-time portfolios, delisting issuers are removed from portfolios at the rst monthly rebalancing after the delisting. Delisting returns, as reported by CRSP, are included in the last monthly return. 18 CCR is calculated using the Compustat variables ch , dlc , and oancf lagged by at least six month. 19 The Compustat variable dltt is used for long-term debt. 134
4 Returns Before and After Issue This section briey reviews average issuer abnormal return before and after issue, conrming previous ndings of high abnormal returns before issue and negative abnormal returns post-issue. Figure 3 shows the average cumulated abnormal return, using CAPM as market model, of issuers from 20 trading days before announcement (day -20) to 20 trading days after announcement (day 20). From day -20 to the day before announcement, cumulated abnormal returns exceeds 5%. On the two subsequent days, i.e. the announcement date and the day after, average abnormal returns are below -2% followed by a rebound of about 0.75% over the next two weeks. All these results are strongly signicant and the pattern does not change much if abnormal return is calculated with respect to FF3 or FF5. The gure motivates calculating event returns over the two-day time window consisting of the announcement date and the subsequent day. 20 [Insert Figure 3 about here] Figure 4 shows the abnormal return index of issuers, i.e. issuers hedged with their exposures to the CAPM, FF3 and FF5 factors, respectively, from 12 months before announcement to 36 months after announcement normalized at 100 on announcement date. Abnormal returns are almost 60% (from around index 63 to index 100) the year before announcement regardless of market model. After issue performance depends on market model. Controlling for market exposure only, issuers record abnormal returns of -22% on average over three years in line with results reported by Loughran and Ritter 20 In some cases, the announcement was made after close on the announcement date. Hence, the negative return on day t+ 1 cannot be interpreted as a delayed reaction. 135
(1995), but adding further controls, in particular the CMA and RMW factors of FF5 changes the picture somewhat. Over one year issuer abnormal returns are insignicant 1% relative to FF5 and only -10% over three years. As hinted by Table 2, the dierence is chiey explained by issuers negative loading on the protability factor RMW. [Insert Figure 4 about here] 5 Event Returns As motivated in Section 3, abnormal announcement event returns revent abn,F F 5 are calculated as the abnormal return on the announcement date and the subsequent trading day. Table 4 shows the results of regressions of abnormal event returns (specication 1 to 6) as well as excess event returns (speci- cation 7) on 1, 2, and 3 past years' market excess return, 1, 2 and 3 past years' issuer abnormal return as well as log relative issue size issue and log market value mv . Since event return periods may overlap, standard errors are calculated using the Newey-West correction of standard errors for heteroscedasticity and autocorrelation using three lags. All regressors are normalized ( z -scores), hence regression intercepts can be interpreted as event abnormal return. Across specications (1) to (6), FF5 abnormal event return is about -2.3% and highly signicant. Past years' market excess return is only signicant for market excess return over three years (specication (3)) with a coecient estimate of 0.2, i.e. a one standard deviation change in past three-year market excess return is associated with about 20 bps higher event returns. If high 136
market return pre-issue is a proxy for attractive investment possibilities, one would expect the coecient to be positive, as it is in all specications. Past years' abnormal return is signicant in most specications with coecient estimates around 0.15, showing that one standard deviation higher past year(s) abnormal return is associated with about 15 bps higher event returns. Relative issue issue is insignicant in all specications, i.e. proceeds raised relative to market value is not signicantly related to event return. Issuer market value is signicant, with estimates between 0.32 and 0.35, i.e. larger issuers have higher (less negative) event returns. This is consistent with larger issuers being more analyzed, hence, on average, less information is conveyed at announcement time. All these conclusions are largely unchanged in regressions of raw event returns (specication 7) instead of abnormal event returns. [Insert Table 4 about here] 6 Long-run Returns This section analyzes to what extent issuer long-run returns can be explained and predicted. There are two dierent approaches frequently applied in the literature. The Buy and Hold Abnormal Return (BHAR) method involves measuring the buy and hold abnormal return issue by issue and trying to explain these by issuer characteristics and other variables in regressions or sorts. Issuer BHAR is either measured as the return dierence between the issuer and a comparable rm (the matched rm approach) or between 137