The job rating game: Revolving doors and analyst incentives

ARTICLE IN PRESS

JID: FINEC

[m3Gdc;May 24, 2019;7:11]

Journal of Financial Economics xxx (xxxx) xxx

Contents lists available at ScienceDirect

Journal of Financial Economics journal homepage: www.elsevier.com/locate/jfec

The job rating game: Revolving doors and analyst incentivesR Elisabeth Kempf Booth School of Business, University of Chicago, 5807 S. Woodlawn Ave., Chicago, IL 60637, USA

a r t i c l e

i n f o

Article history: Received 22 July 2017 Revised 16 October 2018 Accepted 7 November 2018 Available online xxx JEL classification: G14 G24 G28 G30

a b s t r a c t Investment banks frequently hire analysts from rating agencies. While many argue that this “revolving door” creates captured analysts, it can also create incentives to improve accuracy. To study this issue, I construct an original data set, linking analysts to their career paths and the securitized finance ratings they issue. First, I show that accurate analysts are more frequently hired by underwriting investment banks. Second, I exploit two distinct sources of variation in the likelihood of being hired by a bank. Both indicate that, as this likelihood rises, analyst accuracy improves. The findings suggest policymakers should consider incentive effects alongside capture concerns. © 2019 Published by Elsevier B.V.

Keywords: Revolving door Career concerns Analysts Credit ratings Securitized finance

1. Introduction Revolving doors—the frequent hiring of monitors by the industries they monitor—are widespread in financial

R I am grateful to my PhD advisers Alberto Manconi, Luc Renneboog, and Oliver Spalt for their valuable guidance and support. I would like to thank Christian Opp (the referee) for the many valuable comments, and Lieven Baele, Fabio Braggion, Fabio Castiglionesi, Jess Cornaggia (discussant), Kim Cornaggia (discussant), Joost Driessen, Cesare Fracassi (discussant), Rik Frehen, William Goetzmann, Abhiroop Mukherjee (discussant), Miriam Schwartz-Ziv (discussant), Amit Seru, Joel Shapiro (discussant), Amir Sufi, Thomas Wollmann, Han Xia (discussant), as well as participants at AFA 2017, Berkeley, Bocconi, Boston College, Carnegie Mellon, CBS Top Finance Graduate Award Conference 2016, CQA, CUHK, Columbia, CU Boulder, Duke, EFA Doctoral Tutorial 2015, EFA 2016, HBS, HEC Paris, Oxford, Ohio State, Princeton, Stanford, Tilburg University, UCLA, University of Amsterdam, University of Chicago, University of Michigan, University of Rochester, University of Toronto, University of Washington, USC, WFA 2016, Wharton, and Yale for helpful discussions. I also thank William Harrington for having shared his knowledge of the credit rating industry. Part of this research was completed while I was visiting Yale University. E-mail address: [email protected]

markets. Bank regulators join financial institutions, risk controllers join trading floors, and analysts join rated entities. Despite their prevalence, revolving doors are often seen as a source of economic distortion. For example, a prominent narrative of the financial crisis is that conflicts of interest due to the revolving door contributed to weak regulatory oversight and distorted credit ratings.1 The main concern is that the revolving door encourages laxity among monitors who hope to exchange leniency in return for well-paying future jobs. If monitors get hired as an explicit quid pro quo for favors to their future (or potential) employers, then they may indeed be encouraged to be lenient (Stigler, 1971; Peltzman, 1976; Eckert, 1981). However, if monitors are also hired for their expertise, they will have a greater incentive to invest in their industry qualifications or to signal

1 See, for example, the final report by the Financial Crisis Inquiry Commission, January 2011; “Lure of Wall Street cash said to skew credit ratings,” Bloomberg News, February 25, 2015; “Credit raters join the rated,” Wall Street Journal, December 2, 2011.

https://doi.org/10.1016/j.jfineco.2019.05.012 0304-405X/© 2019 Published by Elsevier B.V.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 2

ARTICLE IN PRESS

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

their expertise during their employment as monitors (Che, 1995; Salant, 1995; Bar-Isaac and Shapiro, 2011). In this paper, I will refer to the former effect as the “capture effect” and to the second effect as the “incentive effect.” While they are not mutually exclusive, gauging the economic importance of both effects is crucial for determining optimal regulation of revolving doors and, more broadly, for understanding how concerns about future career prospects affect performance incentives. In contrast to previous work on revolving doors in the context of rating agencies, which has focused on capture concerns, this paper aims at shedding light on the magnitude of potential incentive effects. I identify incentive effects in the context of job transitions between credit rating agencies and underwriting investment banks. In particular, I analyze whether analysts’ ratings accuracy increases the likelihood of being hired by an underwriting investment bank in the securitized finance market and how much analyst accuracy improves due to the possibility to go work for an underwriter. The market for securitized finance instruments is of first-order economic importance with more than $10 trillion of outstanding debt in the US by the end of 2016, which is 1.2 times the size of the US corporate bond market.2 Analysts at rating agencies assess the relative credit risk of these products and are shown to have substantial influence on the final rating outcome (Fracassi et al., 2016). Their decisions therefore have the potential to affect investor demand, e.g., through regulatory reliance on credit ratings (White, 2010; Acharya and Richardson, 2009; Acharya et al., 2013), prices (He et al., 2012; Hand et al., 1992), and investment decisions (Becker and Ivashina, 2015). In addition, the rating inflation of securitized finance products, documented by Benmelech and Dlugosz (2009) and Ashcraft et al. (2010), has been identified as being at the root of the last financial crisis,3 and it has at least partially been attributed to the revolving door between rating agencies and investment banks. In the attempt to address this concern, the Dodd– Frank Act of 2010 introduced new provisions that require credit rating agencies to disclose analyst transfers to rated entities as well as to implement mandatory look-back reviews.4 Using these new disclosures, recent research supports the revolving door concerns by showing that the corporate bond ratings of firms who hire credit rating analysts are inflated prior to the employment transfer (see Cornaggia et al., 2016). My data set links the career paths of 245 credit rating analysts at Moody’s to 24,406 ratings of securitized finance

2 Securities Industry and Financial Markets Association (SIFMA). Reports available at http://www.sifma.org. 3 The Financial Crisis Inquiry Commission concluded that “the failures of credit rating agencies were essential cogs in the wheel of financial destruction. The three credit rating agencies were key enablers of the financial meltdown. The mortgage-related securities at the heart of the crisis could not have been marketed and sold without their seal of approval.” Report available at https://www.govinfo.gov/content/pkg/ GPO- FCIC/pdf/GPO- FCIC.pdf. 4 See section 932 of the Dodd–Frank Wall Street Reform and Consumer Protection Act of 2010 (“Dodd–Frank”), which adds a new paragraph to section 15E(h)(5) of the Securities Exchange Act of 1934. Available on the SEC’s website at https://www.sec.gov/divisions/marketreg/ ratingagency/wallstreetreform- cpa- ix- c.pdf.

securities issued between 20 0 0 and 2009.5 Twenty-seven percent of the analysts in my sample subsequently join a prestigious investment bank that underwrites deals rated by Moody’s, and 13% join a prestigious investment bank whose deals they personally rated. The data set allows me to circumvent one of the most important difficulties for empirical studies of revolving doors: the availability of micro data on employee decisions and career paths. Credit ratings represent a publicly observable and relatively frequent measure of output by individual analysts. Subsequent corrections of the initial ratings issued by these analysts provide a useful proxy for analyst (in)accuracy.6 Another attractive institutional feature of Moody’s organization is that subsequent rating adjustments are performed by a separate internal surveillance team and are therefore not under the direct influence of the analyst who assigned the initial rating. Having access to reliable measures of output quality is crucial for studying the incentive effects of revolving doors and represents an important advantage of studying credit rating analysts over many public sector functions, where individual output quality is difficult to assess.7 Even with micro data on individual performance, a second empirical challenge is that observed performance may be confounded by differences in the task environment faced by different individuals. By comparing performance across analysts at the same rating agency for the same type of product and at the same point in time, my research design ensures that the compared individuals face the same organizational environment, task difficulty, objectives, and internal career concerns. I begin my analysis by documenting a novel finding: accurate credit rating analysts are more likely to be hired by underwriting investment banks. Specifically, analysts who are one standard deviation more accurate are 78% (5.5 percentage points) more likely to be hired by a prestigious underwriting investment bank, compared to the average analyst who rates similar products at the same point in time. The positive relation also holds for departures to underwriters personally rated by the analyst and is robust to various alternative measures of ratings accuracy, including a measure based on realized tranche losses, which cannot be influenced by Moody’s surveillance team. This finding is consistent with a model in which observed rating performance is a relevant factor in investment banks’ hiring policies. Qualitatively, most standard models of optimal incentive provision would therefore predict greater analyst effort in the presence of the possibility to move to an underwriting investment bank. Quantitatively, the magnitude of the effect of investment bank hiring on ex-ante

5 The main sample ends in 2009 to restrict the analysis to the unregulated period that is not confounded by the Dodd–Frank provisions. 6 A potential concern regarding the sample period studied is that expost rating adjustments are greatly affected by recessions. In fact, the majority of rating adjustments in my sample are downward adjustments, and pessimistic analysts that were inaccurate ex ante may have turned out to be very accurate ex post. I perform a number of robustness tests aimed at disentangling accuracy and pessimism (see Section 4.1.1). 7 For example, while it is possible to link lawyers and judges to individual cases, or regulators and examiners to individual decisions, the optimal level of harshness or leniency is unknown. This feature makes it challenging to distinguish whether revolvers are hired on the basis of skill or bias.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

analyst effort and observed rating performance is unclear. The magnitude will depend, among other things, on the relative importance of ability and effort, which are difficult to disentangle. To identify the magnitude of the incentive effects generated by the possibility to go work for an underwriting investment bank, I exploit two distinct sources of variation in the likelihood of being hired while controlling for analyst heterogeneity via analyst fixed effects. First, I use discontinuous changes in the demand for investment banking positions as a shock to the likelihood of being hired by an investment bank. Consistent with the prediction of greater analyst effort as opportunities in investment banking arise, performance improves significantly within the same analyst around these events. Moreover, in the cross-section of analysts, the improvement in performance is concentrated among analysts who are ex ante more likely to switch careers. These pronounced cross-sectional differences mitigate the possibility that the change in performance could be induced by changes in asset fundamentals, which would affect the performance of all analysts. Second, I exploit departures of connected analysts as an instrument for the analyst’s own departure to an investment bank. Connected analysts are defined as analysts who started in the same calendar year as the analyst in question but rated different types of securities. Both approaches suggest the possibility to be hired by an underwriting investment bank has a sizable effect on ex-ante analyst effort: they imply a 38–44% improvement in ratings accuracy in response to a one standard deviation increase in the likelihood of being hired by an underwriting bank. To the best of my knowledge, these are the first estimates of the magnitude of the incentive effect in the revolving door literature. They suggest that the possibility to be hired by an underwriting bank represents a strong incentive for credit analysts to perform well, and that restricting these employment transfers without changing other aspects of analyst compensation may lead to lower ratings quality.8 However, I also find evidence consistent with capture, in line with the findings by Cornaggia et al. (2016) on corporate bond ratings. While being optimistic does not increase an analyst’s overall prospect of being hired, being optimistic on a deal underwritten by a specific bank does increase the likelihood of being hired by that same bank. Hence, the revolving door may also encourage analysts to curry favors toward specific employers. Combined, my results suggest that optimal regulation should trade off the benefits of reducing capture against the cost of reducing labor mobility and weakening incentives for analysts to develop and showcase their skill. For example, a broad ban would likely lead to a greater reduction in labor mobility and, hence, to a greater reduction in effort provision by analysts, relative to a more targeted revolving door policy that only prohibits transitions to entities that analysts have

8 Consistent with the findings by Kisgen et al. (2018) for corporate bond analysts, I find that internal promotions at Moody’s are positively related to ratings accuracy (see Table 10), but the relation between accuracy and internal promotions in securitized finance is economically and statistically weaker than the relation between accuracy and departures to investment banks.

[m3Gdc;May 24, 2019;7:11] 3

recently rated. These consequences should be taken into account by policymakers when designing revolving door policies.

2. Related literature Despite the public interest and regulatory concern for the topic, there is little systematic evidence on revolving doors. The few existing studies on the career concerns of financial analysts have focused on documenting capture effects. The study most closely related to mine is Cornaggia et al. (2016), who show that corporate bond ratings of firms who hire former credit rating analysts are inflated prior to the employment transfer. My results are consistent with their findings since I find that an optimistic rating on issues of a particular bank increases the likelihood of being hired by that bank. However, my study differs from Cornaggia et al. (2016) by (i) studying securitized bonds rather than corporate bonds, (ii) documenting that analysts who are hired by underwriting investment banks are more accurate overall, and (iii) showing that the possibility to be hired by a prestigious underwriter leads to sizable improvements in analyst performance. For sell-side equity analysts, Cohen et al. (2012) report that analysts who get appointed as independent directors are overly sympathetic to management and poor relative performers, and Lourie (2019) finds that analysts who get hired by a firm they cover become more optimistic about their future employer while becoming more pessimistic about other firms. Horton et al. (2017) show that banking analysts exhibit a stronger pattern in issuing optimistic forecasts at the beginning of the year and pessimistic forecasts at the end of the year when they are forecasting earnings of potential future employers. This paper is also related to a growing literature on regulatory capture (Stigler, 1971; Peltzman, 1976; Shleifer and Vishny, 1993). Most empirical work on revolving doors in the public sector finds no or limited evidence of capture. For example, Agarwal et al. (2014) and Lucca et al. (2014) find that regulatory lenience is associated with reduced prospects of finding employment in the financial sector. In addition, Forster and Shive (2017) show that financial firms take significantly less risk after hiring former regulators, consistent with the view that they are hired for their expertise in risk management. deHaan et al. (2015) find that harsher Securities and Exchange Commission (SEC) lawyers are more likely to be hired by private law firms. On the other hand, recent research by Tabakovic and Wollmann (2018) shows evidence of capture by showing that US patent examiners grant more patents to firms that hire them and extends this leniency to other firms in close geographical proximity. Finally, related work on lobbyists suggests that networks are valued more than expertise (Blanes i Vidal et al., 2012; Bertrand et al., 2014).

3. Theory and data To fix ideas, this section describes my theoretical framework and derives key empirical predictions.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 4

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

3.1. Theoretical framework A key challenge in the revolving door literature is that performance in the absence of the revolving door is unobservable. To evaluate potential policy interventions, one would ideally like to compare the average ratings quality under revolving door scenario A to the average ratings quality under revolving door scenario B. Unfortunately, natural experiments that exogenously vary revolving door intensity are scarce. Moreover, a specific policy experiment might be informative about a particular policy choice (e.g., scenario A versus scenario B), but it might not be very informative about a broader set of potential interventions. Rather than evaluating specific policy experiments, existing empirical studies have therefore exploited the fact that the two main potential effects of the revolving door make different assumptions about the hiring policy of revolving door employers, which can be tested in the data.9 The hiring policy, in turn, will be an important determinant of the optimal behavior of an analyst who is maximizing the chances of advancing her career to the rated industry. Specifically, an analyst who has to rate N investments, and can affect their final ratings (m1,...,N ), will choose to influence these ratings in the direction that maximizes the probability of being hired by a prestigious underwriting investment bank (pIB ):10

max E ( pIB |m1,...,N ).

m1,...,N

[m3Gdc;May 24, 2019;7:11]

(1)

Whether the analyst ultimately chooses to assign accurate or optimistic ratings depends on the sensitivity of the probability to be hired by a prestigious bank to analyst accuracy and optimism, respectively. If underwriting investment banks hire in return for leniency, then aspiring analysts will optimally choose to inflate ratings (capture effect). If, on the other hand, investment banks hire based on skill, then analysts will have an incentive to signal their skill by being more accurate (incentive effect). The two effects are not mutually exclusive. Previous work on corporate bond ratings has focused on identifying potential capture effects. Specifically, Cornaggia et al. (2016) show firms that hire analysts from rating agencies receive more optimistic ratings prior to the employment transfer. However, it is unclear (i) whether revolving door employers also care about the overall accuracy of the analysts they hire and (ii) if they do, whether their hiring policies could generate positive incentive effects. My empirical analysis is therefore divided in two parts. First, I estimate the sensitivity of the likelihood of being hired by an underwriting investment bank with respect to analyst accuracy (see Section 4.1). In the simple framework above, investment banks hiring based on accuracy is a necessary condition for the existence of a positive incentive effect. Quantitatively, how much analyst 9 See, for example, Cohen et al. (2012), deHaan et al. (2015), Cornaggia et al. (2016) and Lourie (2019). 10 Analysts could, for example, exert additional effort to receive a more precise signal about the quality of the rated investment. Alternatively, they could choose to report a high rating even if their private signal suggests the underlying investment is of low quality.

effort and observed rating performance change in response to changes in investment bank hiring is unclear and will depend, among other things, on the relative importance of ability and effort. In the second part of my analysis, I identify the incentive effect by estimating the sensitivity of analyst performance to changes in the probability of leaving to an underwriting bank while controlling for baseline ability via analyst fixed effects (see Section 4.2). I exploit two distinct sources of variation in the likelihood of leaving to an underwriting investment bank. First, I use discontinuous changes in the demand for investment banking positions as a shock to the likelihood of being hired by an investment bank. Second, I exploit departures of connected analysts as an instrument for the analyst’s own departure to an investment bank. The two experiments are informative about the potential effect of revolving door policies that reduce the overall probability to be hired by a prestigious underwriting bank. A mandatory cooling-off period or a ban that prohibits analysts from working for any entity rated by their rating agency would have the most dramatic effect on labor mobility, by effectively reducing it to close to zero. However, even narrower policy interventions may, in equilibrium, result in a lower hiring rate by underwriting investment banks. For example, prohibiting an analyst from working for an entity she has rated in the previous 12 months may already significantly restrict her potential labor market options: in my sample, the median analyst rates four different “Bloomberg Top 20” underwriters in a given year.11 In addition, mandatory look-back reviews and public disclosure of employment transitions to rated entities may increase investment banks’ monetary and reputational cost of recruiting talent from rating agencies. As a result, an investment bank that would have otherwise been indifferent between hiring a former rating analyst and a candidate with a different background may prefer candidates with a non-CRA background post regulation. In sum, by quantifying the sensitivity of average analyst performance to changes in the overall hiring rate, my analysis can speak to a range of potential revolving door regulations. A more comprehensive discussion of the policy implications of my results can be found in Section 5. 3.2. Empirical approach My analysis will consist of two parts. First, I study whether analyst accuracy increases the likelihood of being hired by a prestigious underwriter. This condition is necessary, but not sufficient, for the existence of a positive incentive effect. Second, I identify the magnitude of the incentive effect by exploiting variation in the likelihood of leaving to an underwriting investment bank. Implementing these two sets of tests poses at least two main challenges. First, it requires reliable measures of on-the-job performance at the individual analyst level. Second, differences in performance across analysts may be confounded 11 Note that these four underwriters only refer to underwriters of deals where the analyst serves as the lead analyst or committee chair. The number of underwriters with which the average analyst interacts in other rating committee roles will be higher.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

due to potential nonrandom assignment of analysts to securities. This section describes how my empirical strategy addresses these two key issues. 3.2.1. Measuring rating performance An important advantage of the rating agency context is that individual output, i.e., the ratings issued by a given analyst and their subsequent performance, is observable. However, traditional metrics of ratings accuracy, such as average default rates by rating category or accuracy ratios (see Cornaggia et al., 2017), rely on a large number of sample events to be meaningful. Considering that a given analyst only rates a limited number of securities in each period and defaults are infrequent events, these measures may not be very reliable in gauging analyst-level performance. To circumvent this difficulty, I propose to exploit updated assessments of the expected future default probability by Moody’s surveillance team as an alternative to realized defaults.12 Specifically, I focus on instances where the surveillance team concludes that the initially assigned rating no longer reflects the expected default probability going forward and adjusts the rating. The absolute difference between the initial rating and the subsequent rating by the surveillance team is my main measure of ratings (in)accuracy. Measuring ratings accuracy based on deviations between the initial rating and subsequent ratings has important advantages. First and foremost, it does not require a large number of ratings for each individual analyst to be meaningful. Second, it allows me to capture smaller changes in the expected default probability that may not always lead to a default. Third, to the extent that rating updates are driven by the arrival of idiosyncratic information that is orthogonal to the analyst’s information set at issuance, this would introduce noise in the measurement of analyst inaccuracy and bias me against finding crosssectional differences in my subsequent analysis. One potential concern regarding the sample period studied is that ex-post rating adjustments are greatly affected by the recession. In fact, the majority of rating adjustments in my sample are downward adjustments, and pessimistic analysts that were inaccurate ex ante may have turned out to be very accurate ex post. My measure of accuracy might therefore reflect pessimism rather than accuracy, implying that investment banks were trying to hire harsh, rather than accurate, analysts.13 Note that investment banks hiring based on harshness still contradicts the main assumptions underlying the capture hypothesis and arguments by Dodd–Frank. In fact, such a hiring policy would generate the opposite of capture: analysts would have an incentive to be harsh rather than lenient, which would directly contradict the claim that the rating inflation in securitized finance was (at least partially) driven by analysts who were currying favors to future employers.14 12 The surveillance team is in charge of the ongoing monitoring of ratings at Moody’s. 13 A potential motivation for investment banks to hire harsh analysts could be to reduce the pool of potentially harsh analysts in the future. 14 At the same time, rating agencies would be systematically losing their harshest analysts. The net effect of such a hiring policy on overall rating optimism is therefore ambiguous.

[m3Gdc;May 24, 2019;7:11] 5

The hiring policy still matters, though, for understanding whether the revolving door will incentivize greater bias or greater accuracy at the rating agencies. I perform two sets of tests to disentangle these two possibilities. First, I separate rating adjustments into downward and upward rating adjustments, and I show that ratings issued by analysts who get hired by investment banks do not have to be adjusted more upwards. If anything, they need to be adjusted less upward (see Table 4, Panel A). Second, I estimate the relation between accuracy and departure to an investment bank separately for different subperiods (see Internet Appendix). I find that the sign and magnitude of the coefficient is remarkably stable across different subperiods, even though these periods vary dramatically in terms of the sign and magnitude of the average ex-post rating adjustment. An attractive feature of Moody’s organization in structured finance is that ratings surveillance is performed by a separate team.15 Subsequent ratings are therefore unlikely to be directly influenced by the analyst who rated the security at issuance.16 It is also worth noting that systematic mistakes by the surveillance team would affect the performance of all ratings and cannot bias my cross-sectional comparisons. Systematic mistakes by the surveillance team would only bias my results if they were differentially affecting analysts who join investment banks. For example, Morkötter et al. (2016) show that between 2007 and 2012, securities by issuers with close bilateral business relationship to rating agencies are downgraded less severely and with a greater delay. If analysts who rate issuers with strong business relationships are also more likely to be hired by a prestigious investment bank, my analysis could be biased toward finding higher accuracy for these analysts. I perform a variety of tests to rule out this possibility. First, I show that securities rated by analysts who depart to an investment bank do not differ from securities rated by other analysts along a variety of observable dimensions, including the issuer’s market share with Moody’s (see Internet Appendix). Second, I show that my results are robust to measuring ex-post rating performance based on realized tranche losses and downgrades to default (see Table 4). By definition, realized tranche losses cannot be influenced by the surveillance analysts, and downgrades to default are typically tied to hard events, such as covenant violations, and therefore less subjective than other rating adjustments (see Griffin et al., 2014). The that fact my results are robust to using these alternative measures of accuracy implies that the documented effect is not simply due 15 Michael Kanef, former head of the Asset Backed Finance Rating Group at Moody’s Investors Service, testified before the US Senate in 2007 that “monitoring is performed by a separate team of surveillance analysts who are not involved in the original rating of the securities, and who report to the chief credit officer of the Asset Finance Ratings Group.” His testimony is available on the website of the US Senate at http://www.banking.senate.gov/public/index.cfm?FuseAction=Files. View&FileStore_id=e9c1a464- a73b- 417a- a384- 41c15315f8c2. 16 To rule out the possibility that analysts may move between the ratings issuance and ratings surveillance functions, I also compute a measure of ratings accuracy using only subsequent rating actions that are performed by different analysts than the one responsible for the initial rating. Since there are very few exceptions to Moody’s rule, I obtain a correlation coefficient of more than 98% between the two ratings accuracy measures and a very similar baseline coefficient.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 6

ARTICLE IN PRESS

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

to subjectivity in the ex-post adjustment of ratings at Moody’s. Finally, a potential disadvantage of defining ratings accuracy based on subsequent adjustments is that it represents an ex-post measure of performance that cannot be directly observed in real time. First, I show in the Internet Appendix that my main results are robust to measuring subsequent rating adjustments over various horizons, including a shorter horizon of one year, and to using proxies for ratings quality that can be observed in real time, such as characteristics of the initial press release. Second, investment banks may observe signals about analyst performance that are unobservable to the econometrician but highly correlated with ex-post measures of performance. For example, underwriting investment banks may, from direct interactions with the analyst during the ratings process, learn about her skill by observing the level of preparation, the type of information requested, and the quality of the questions raised. Even if investment banks do not directly interact with an analyst, they may still receive signals about her quality through their professional network, e.g., from conversations with other bankers who have directly worked with the analyst, her colleagues at Moody’s, etc. Finally, since rating reports are publicly available on Moody’s website, investment banks may infer rating quality from the provided rating rationale and characteristics of the deal.

3.2.2. Comparing rating performance Comparing ratings accuracy across analysts is nontrivial because different analysts may be rating different types of products. For example, analysts often specialize in deals of one or few collateral types, which may exhibit different trends in fundamentals that may be correlated with hiring intensity by investment banks.17 To be able to compare analysts’ rating performance on a subset of securities that provide similar levels difficulty, I aggregate ratings (in)accuracy at the analyst × collateral type level instead of at the analyst level. This approach has the additional advantage that Moody’s internal organization follows a similar division (see Appendix B), which ensures that analysts who rate securities of the same collateral type face similar incentives, rating methodologies, and management leadership. In sum, I define analyst inaccuracy as the absolute difference (in notches) between the initial rating and the subsequent surveillance rating, averaged across the set of securities Sizt rated by analyst i in collateral type z and semester t:18

17 As shown in the Internet Appendix, there is substantial heterogeneity in the average ratings accuracy across different collateral types even during the same time period. 18 While aggregating across all securities rated by the same analyst in a given collateral type and semester reflects the idea that individual analysts are the main research subjects in this study and has the advantage of reducing the influence of outliers, it is also possible to run my subsequent analysis at the individual deal level. The results, reported in the Internet Appendix, are both quantitatively and qualitatively very similar. In addition, my results are robust to computing a value-weighted performance measure, where the weights are proportional to the security’s principal amount (see Internet Appendix).

Inaccuracyizt =

 1   R j,t+h − R j,t , N

(2)

j∈Sizt

where Rj,t refers to the initial rating of security j issued by analyst i in semester t, and R j,t+h refers to the rating of the same security at some future point in time, t + h. In my baseline definition, h will be equal to three calendar years.19 Credit ratings are transformed into a cardinal scale, starting with 1 for Aaa and ending with 21 for C, as in Jorion et al. (2005). I can then implement the idea of comparing analysts rating securities of the same underlying collateral type at the same point in time by including collateral type × semester fixed effects in the regressions. Even within a given collateral type and date, analysts may be assigned to securities with different characteristics, e.g., deals with complex subordination structures. In addition to collateral type × semester fixed effects, I therefore control for a rich set of observable tranche and deal characteristics. Specifically, I control for the logarithm of the combined principal value of all tranches in the deal (“deal size”); the geographical concentration of the collateral, measured as the sum of the squared shares of the top five US states in the deal’s collateral as in He et al. (2016); the level of overcollateralization, computed as the difference between the total collateral value and the combined principal value of the tranches as in Efing and Hau (2015); the weighted average loan-to-value (LTV) ratio and the weighted average credit score of the underlying collateral at issuance; the weighted average life; the fraction of tranches with an insurance wrap; and the logarithm of the number of tranches in the deal. These characteristics are averaged across all securities rated by the analyst in a given collateral type and time period.20 Controlling for this rich set of tranche and deal characteristics takes into account that some securities might be harder to rate and systematically face larger rating adjustments than others. 3.2.3. Can individual analysts influence ratings? A necessary condition for analyst incentives to affect ratings quality is that the ratings process for securitized finance products needs to provide sufficient room for individual analysts to affect the final rating of the security. This assumption is not obviously true given that the final rating decision is taken by a committee. Upon receiving a rating application from a potential customer, Moody’s assigns a lead analyst to the ratings process. The lead analyst meets with the customer to discuss relevant information, which she subsequently analyzes with the help of Moody’s analytical team. She then proposes a rating and provides a rationale to the rating committee, which consists of a

19 In the robustness tests reported in the Internet Appendix, I consider rating adjustments over alternative horizons (one and five years) and find similar effects. 20 Since information on some tranche and deal characteristics (specifically, the weighted average life, insurance wrap, geographical concentration, LTV ratio, and credit score) is available only for a subset of my data, I follow He et al. (2016) by replacing missing observations and including additional indicators equal to one if information on a given variable is not available. My robustness test in the Internet Appendix shows that the approach of replacing and controlling for missing observations does not materially affect my results.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

number of credit risk professionals determined by the analyst in conjunction with the committee chair. Once the rating committee has reached its decision, Moody’s communicates the outcome to the customer and publishes a press release.21 The ratings process at Moody’s therefore provides ample opportunities for individual analysts to influence the final rating, even if the final decision is taken by a committee. Lead analysts guide meetings with the customer, request and interpret information, and play a key role in the rating committee by proposing and defending a rating recommendation based on their own analysis. In addition, the rating committee chair serves a special role by influencing the composition of the rating committee and acting as the moderator. How much individual analysts are able to influence ratings is ultimately an empirical question. Fracassi et al. (2016) attribute a substantial part of the variation in corporate bond ratings to individual analysts: they explain 30% of the within-firm variation in ratings. For securitized finance ratings, Griffin and Tang (2012) provide evidence that collateralized debt obligation (CDO) ratings by a major credit rating agency frequently deviated from the agency’s main model, reflecting room for subjectivity in the ratings process. Note that if individual analysts played no role, my analysis should be biased against finding significant differences across analysts. 3.3. Data Testing the predictions outlined in Section 3.1 requires a data set with two main features. First, it needs to contain performance measures at the individual analyst level. Such a data set is not readily available for credit analysts.22 To overcome this problem, I hand-collect an original data set that links individual analysts to the performance of the ratings they issue. Second, I identify revolving door analysts as analysts who leave to underwriting investment banks after their employment at the rating agency. I collect this information from analysts’ self-reported profiles on a professional networking website and from web searches. I also identify instances where analysts rate securities underwritten by their future employers by manually matching the name of the analyst’s subsequent employer to the lead underwriting banks of the security reported in SDC Platinum. The full data set is described in more detail below. My sample consists of all nonagency securitized finance securities issued in the US and reported in SDC Platinum. Additional deal and tranche information is manually collected from Bloomberg. I restrict my sample to all issues between 20 0 0 and 20 09 that were initially rated by Moody’s because (i) data are sparse prior to 20 0 0, (ii) my goal is to study the analyst labor market prior to potentially confounding effects from the Dodd-Frank regulation, and (iii) Moody’s publicly discloses analyst names in press

21 See Fracassi et al. (2016) and https://www.moodys.com/sites/ products/ProductAttachments/mis_ratings_process.pdf for a description of the ratings process at Moody’s. 22 Standard databases on corporate and securitized finance credit ratings (e.g., Mergent FISD, Bloomberg, or SDC Platinum) do not provide the identity of the individual analysts responsible for a given rating.

[m3Gdc;May 24, 2019;7:11] 7

releases of a new rating action on its website.23 Most of the times, Moody’s press releases list two names, one more junior employee (typically the lead analyst), and one more senior employee (typically the rating committee chair).24 In addition to the analyst names, I also collect data on subsequent rating changes for each security from Moody’s website. The securitized finance data are complemented with hand-collected biographical information from web searches, with the vast majority of cases from analysts’ public profiles on a professional networking website. In particular, I gather information on the date when the analyst left Moody’s and the identity of her first employer following her employment at Moody’s, as well as information on previous employment, graduate, and undergraduate education. I am able to track the career paths of 245 out of 268 analysts. As shown in Table 1, Panel B, 66 out of these 245 analysts subsequently go work for an underwriter that was ranked in the prestigious “Bloomberg 20” investment bank ranking in the year prior to their exit.25 The Bloomberg 20 investment banks capture a large fraction of the underwriting market in securitized finance: they underwrite 88% of the securities in my sample (Table 1, Panel C). Ninety-four analysts leave to other employers, and 85 analysts have not left Moody’s as of December 2015. Half of the analysts who join a top investment bank rate their future employer at some point during their employment at Moody’s, but only 21% (= 14/66) do so during their last year at the rating agency (Table 1, Panel B). Conditional on analysts who leave to an investment bank, the average duration of their employment at Moody’s is 3.5 years (Fig. 1, Panel B). As shown in Table 2, analysts with fewer years of prior work experience, no graduate degree, an undergraduate degree from an institution located in New York City, and a nonlaw undergraduate degree are more likely to leave to an investment bank. Interestingly, graduates from Ivy League institutions are less likely to subsequently work for an investment bank, although this relation is not statistically significant. Overall, this pattern is consistent with the idea that analysts with higher expected increases in lifetime earnings (e.g., younger analysts without graduate degrees) and lower switching costs (e.g., analysts with ties to New York City and nonlaw degrees) are more likely to leave to investment banking. As reported in Table 1, Panel A, my final data set consists of 24,406 tranches from 4979 securitized finance deals. All securities combined account for an aggregate issuance volume of circa $2.7 trillion, which represents at least 40% and therefore a sizable fraction of the aggregate US nonagency securitized finance deal vol23 I am able to find corresponding analyst information from Moody’s website in 71% of the cases. 24 As I show in the Internet Appendix, my results are economically and statistically stronger for the group of lead analysts. 25 Since the ranking is only available from 2004 onward and the composition of the ranked investment banks is fairly stable prior to 2008, I use the 2004 ranking to classify analyst exits prior to 2004. Fig. 1, Panel A provides an overview of the top hiring banks in my sample. In the Internet Appendix, I show that my main findings are robust to alternative definitions of prestigious underwriters.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC 8

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

Table 1 Summary statistics. The table presents summary statistics for my sample, which comprises all US nonagency securitized finance deals rated by Moody’s between 20 0 0 and 20 09 with information identifying the analyst(s) at issuance and information on the analyst’s post-Moody’s employment status. Panel A shows the breakdown of securities by collateral type. Panel B provides an overview of the subsequent career paths of the analysts in my sample and the number of analysts who, at some point during their employment at Moody’s, rate securities underwritten by their future employers. Panel C reports descriptive statistics of key variables. A complete list of variable definitions is provided in Appendix A. Panel A: Sample Number of tranches

Number of deals

Issuance volume ($bn)

Market segment: ABS ABS auto ABS card ABS home ABS other ABS student

1929 487 4540 4826 146

539 246 891 1089 40

433.06 184.55 406.93 566.80 23.15

Market segment: MBS CMBS RMBS

537 11,036

65 1839

71.67 977.29

Market segment: CDO CDO Total

905

270

66.68

24,406

4979

2730.14

Panel B: Number of analysts by subsequent career path Other exit

Number of analysts Full sample Rate future employer Rate future employer during last yr

All

No exit

IB exit

Other bank

Asset mgr.

Insurer

Other

245 33 14

85 0 0

66 33 14

30 0 0

21 0 0

11 0 0

32 0 0

Panel C: Summary statistics

Career path variables IB exit IB exitt+1 Rated IB exit Rated IB exitt+1 Promotiont+1

N

Mean

Std. Dev.

0.25

Median

0.75

1860 1860 1860 1860 1704

0.24 0.07 0.17 0.04 0.06

0.43 0.25 0.37 0.20 0.24

0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

0.00 0.00 0.00 0.00 0.00

2.69 6.34 9.24 0.07 2.61 0.08 -0.04

4.96 6.83 9.33 0.17 4.99 0.27 0.20

0.00 1.02 1.71 0.00 0.00 0.00 -0.07

0.00 3.99 5.03 0.00 0.00 0.00 0.00

2.18 9.56 15.83 0.00 2.09 0.00 0.00

5.92 4.86 0.88 0.61 4.97 0.34 673.58 67.41 0.02 658.24 7.86 0.01

6.07 9.05 0.27 0.96 2.43 0.06 49.82 11.63 0.07 557.08 5.45 0.08

2.00 1.00 0.93 0.00 3.22 0.30 628.32 64.33 0.00 341.77 4.76 0.00

4.00 2.00 1.00 0.25 4.45 0.33 678.79 69.25 0.00 534.45 7.00 0.00

8.00 5.00 1.00 0.86 6.16 0.36 719.97 72.88 0.00 848.85 9.71 0.00

Measures of analyst (in)accuracy and relative optimism Inaccuracy (baseline) 1860 3-yr excess losses (in %) 513 5-yr excess losses (in %) 660 3-yr defaults 1860 3-yr downgrades 1860 3-yr upgrades 1860 Relative optimism 1842 Control variables Tenure (in semesters) Number of deals IB underwriter Issuer market share (in %) Weighted avg. life (in years) Geographical HHI Weighted avg. credit score Weighted avg. LTV (in %) Insurance wrap Deal size (in $m) Number of tranches in the deal Overcollateralization

1860 1860 1807 1860 1750 990 739 925 1748 1860 1,860 1,860

ume over this period reported by the Securities Industry and Financial Markets Association (SIFMA).26 Using similar

26 Since SIFMA does not report agency asset-backed securities separately, I compute the aggregate deal volume as the sum of $4.5 trillion of nonagency mortgage-backed securities and $2.3 trillion of asset-

categories as in Griffin et al. (2014), I classify securities depending on the type of the underlying collateral into eight collateral groups and three broad market segments backed securities (agency and nonagency). Hence, the 40% represent a lower bound estimate of the covered market share.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

[m3Gdc;May 24, 2019;7:11] 9

Fig. 1. Summary plots. Panel A plots the total number of Moody’s analysts hired by each underwriting investment bank over the sample period. An analyst departure is classified as an exit to an investment bank if the subsequent employer was ranked in the Bloomberg 20 ranking in the year prior to the analyst’s departure. Panel B plots the cumulative percentage of analysts departing to an underwriting investment bank for a given number of years of employment at Moody’s.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC 10

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

Table 2 Predicting analyst departures to underwriting investment banks. The table reports the characteristics of analysts who depart to investment banks. IB exit is an indicator equal to one if the analyst departs to an investment bank that was ranked in the Bloomberg 20 ranking in the year prior to her departure, and it is regressed on various analyst characteristics using a probit model. Ivy League indicates whether the analyst has obtained her most recent degree prior to joining Moody’s at an Ivy League institution. Law degree and Tech degree are indicators if the analyst’s undergraduate degree is in law or in a technical field (mathematics/engineering/physics/computer science), respectively. NYC undergrad indicates whether the analyst has obtained her undergraduate degree from an institution located in New York City, and Graduate degree is an indicator equal to one if the analyst has obtained a graduate degree prior to joining Moody’s. Prior work experience refers to the logarithm of one plus the number of years of prior work experience. In column (2), dummies indicating the calendar year of the beginning of the analyst’s employment with Moody’s are included. Robust t-statistics are reported in parentheses. IB exit Ivy league Law degree Tech degree NYC undergrad Graduate degree Prior work experience Female Cohort fixed effects N Pseudo-R2

[m3Gdc;May 24, 2019;7:11]

(1)

(2)

−0.595 (−1.42) −0.812 (−1.40) 0.038 (0.10) 0.845 (2.00) −0.801 (−2.20) −2.923 (−3.59) −0.308 (−0.89)

−0.509 (−1.01) −1.067 (−1.91) 0.342 (0.71) 1.186 (2.01) −1.094 (−2.57) −2.820 (−2.73) −0.514 (−1.36)

No

Yes

99 0.220

79 0.284

(asset-backed securities (ABS), mortgage-backed securities (MBS), and CDO). Classifying all securities by collateral type is important for my empirical approach of comparing analyst performance within collateral type and date. Table 1, Panel C reports descriptive statistics of my sample. On average, ratings issued during my sample period are adjusted by 2.69 notches over a three-year horizon. There is a high degree of heterogeneity in rating adjustments, with a median of zero and a standard deviation of 4.96 notches. As shown in the Internet Appendix, I find little evidence of statistically significant differences between the securities rated by revolving and nonrevolving analysts along a rich set of tranche and deal characteristics. This fact reduces potential concerns that the securities rated by revolving analysts may differ on some unobserved dimension. While my main tests below are designed to address identification issues, Fig. 2 shows that two important insights emerge even from the raw data. In each collateral type and semester, analysts are sorted into quartiles based on their inaccuracy—computed as in Eq. (2). The graph plots the average relative inaccuracy as well as the average exit rate to underwriting investment banks by inaccuracy

quartile. First, there is substantial variation in analyst performance even within a given collateral type and semester. Ratings issued by analysts in the lowest inaccuracy quartile have to be adjusted by more than three notches less relative to the mean compared to ratings issued by the average analyst in the highest inaccuracy quartile. Second, the probability of being hired by a top underwriting investment bank is more than twice as high in the bottom inaccuracy quartile than in the top inaccuracy quartile.27 Hence, the raw data suggest that accuracy is associated with a significantly higher likelihood of being hired by a prestigious underwriter. 4. The incentive effect This section presents my main results. First, I show that analyst accuracy is a strong predictor for being hired by a prestigious underwriter. This relation holds both for departures to prestigious underwriters rated by Moody’s, as well as for departures to prestigious underwriters directly rated by the analyst, and is robust to various measures of accuracy. Second, to identify the incentive effect, I show that increases in the likelihood of being hired by a prestigious underwriter lead to sizable improvements in ratings accuracy. 4.1. Analyst accuracy and hiring by prestigious underwriters To test whether analyst accuracy matters for the likelihood of being hired by a prestigious underwriter, I first compute analyst inaccuracy in a given collateral type and semester as the average number of adjustments that are made to the ratings issued by analyst i in collateral type z and semester t—see Eq. (2). Then I estimate different versions of the following regression:

yizt = λzt + δ Inaccuracyizt + β  Xizt + izt ,

(3)

where yizt refers to indicators for analysts who are hired by a prestigious underwriter (see below). Inaccuracyizt stands for analyst inaccuracy as computed in Eq. (2) and λzt for collateral type × semester fixed effects. Vector Xizt includes the average deal characteristics listed in Section 3.2.2. In addition to average deal characteristics, I also control for the logarithm of the total number of deals rated by analyst i in collateral type z and semester t, the logarithm of one plus the analyst’s tenure at Moody’s (in semesters), and the fraction of tranches underwritten by investment banks rated in the Bloomberg 20 ranking,28 as well as the average issuer market share with Moody’s.29 All variables are defined in Appendix A. Standard errors are clustered at the analyst level. I consider four different specifications of the dependent variable to reflect alternative definitions of revolving door transitions to underwriters: IB Exit, which is an

27

The difference is significant at the 1% level. Griffin et al. (2014) show that securities issued by high-reputation investment banks have higher default rates. 29 Efing and Hau (2015) show issuers that maintain strong business relationships with a rating agency receive more favorable ratings. 28

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

[m3Gdc;May 24, 2019;7:11] 11

Fig. 2. Analyst accuracy and hiring by underwriting investment banks. The graph plots by analyst inaccuracy quartile the average relative inaccuracy and the fraction of analysts who leave to underwriting investment banks. Analyst inaccuracy is computed as the average number of notches by which ratings issued by a given analyst in a given collateral type and semester need to be adjusted within three years after issuance—see Eq. (2). Using this measure, analysts are sorted into quartiles within a given collateral type and semester. The dotted line plots, for each quartile, the average analyst inaccuracy relative to the mean in the same collateral group and semester. The gray bars show the fraction of all analysts within a given quartile who depart to an investment bank ranked in the Bloomberg 20 ranking by the end of the following semester.

indicator equal to one for analysts who eventually join an underwriting investment bank whose deals are rated by Moody’s; IB Exitt+1 , an indicator equal to one for analysts who, by the end of the following semester, join an underwriting investment bank whose deals are rated by Moody’s; Rated IB Exit, an indicator equal to one for analysts who eventually join an underwriting investment bank whose deals they personally rated; and Rated IB Exitt+1 , an indicator equal to one for analysts who, by the end of the following semester, join an underwriting investment bank whose deals they personally rated. Note that since the performance measure is analyst inaccuracy, the assumption that investment banks value analyst accuracy implies δ < 0 in the above regression. Table 3, Panel A reports the results for departures to prestigious underwriters whose deals are rated by Moody’s. Confirming the results from the simple sorts presented in Fig. 2, more accurate analysts are significantly more likely to be hired by an underwriting investment bank compared to less accurate analysts who rate securities of the same collateral type in the same semester. A one standard deviation decrease in inaccuracy increases the probability of being hired by an investment bank by 7.9 percentage points, or 33% (= 0.016 × 4.96/0.24). The economic effect is even larger when I focus on the likelihood of departing to an investment bank in the near future— columns (3) and (4)—which is a one standard deviation decrease in inaccuracy increases the probability of being hired by an investment bank by the end of the following semester by 78% (= 0.011 × 4.96/0.07), equivalent to 5.5 percentage points. In Panel B, I study departures to underwriters whose deals the analyst has personally rated. Average analyst accuracy continues to be an important predictor for being hired: a one standard deviation increase in accuracy in-

creases the probability of being hired by an underwriter whose deals the analyst has rated by 87% (= 0.007 × 4.96/0.04).30 Hence, analyst skill appears to be an important predictor for being hired by an underwriter, including for underwriters whose securities the analyst has personally rated. In sum, my findings are consistent with the view that issuing accurate ratings significantly increases analysts’ prospects of advancing their careers to a prestigious underwriting investment bank. In the following, I show that this result is robust to alternative measures of ratings accuracy, including a measure based on realized losses. 4.1.1. Robustness Table 4 presents robustness tests that use alternative measures of ratings (in)accuracy. Unless otherwise mentioned, I report results for the specifications in Table 3, columns (2) and (4), and suppress all control variables for brevity. First, I measure inaccuracy based on excess tranche losses, which dramatically reduces the sample size but still yields results of similar economic magnitude. Excess losses are computed as the absolute difference between the realized tranche loss and Moody’s expected loss benchmark for the initial rating category.31 While the statistical significance of the results is lower when excess losses are measured over a three-year horizon, it is high for excess losses computed over a five-year horizon. Next, I focus on instances where securities are downgraded all the way to default—a rating action that is typically tied to hard events 30 The coefficient in Panel B is mechanically lower than in Panel A because the mean of the dependent variable is smaller. 31 See Moody’s Investor Service, 2001. A users guide for Moody’s Analytical Rating Valuation by Expected Loss (MARVEL) – a simple credit training model.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC 12

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx Table 3 Analyst accuracy and hiring by underwriting investment banks. The table reports results from regressing indicators for analyst departures to investment banks on ratings accuracy. In Panel A, columns (1) and (2), IB exit is an indicator equal to one if, at the end of her employment with Moody’s, the analyst departs to an investment bank that was ranked in the Bloomberg 20 ranking in the year prior to her departure. In columns (3) and (4), IB exitt+1 is an indicator equal to one if the analyst departs to a top investment bank by the end of the following semester. In Panel B, the dependent variables are defined analogously but refer only to departures to underwriters that the analyst has rated. Analyst inaccuracy is measured as the average number of notches by which the ratings issued by the analyst need to be adjusted within three years of issuance. All variables are defined in Appendix A. A table with coefficients for the complete set of control variables is shown in the Internet Appendix. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. Panel A: Departures to prestigious underwriters rated by Moody’s IB exit Inaccuracy Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. N R2

IB exitt+1

(1)

(2)

(3)

(4)

−0.014 (−2.77) −0.034 (−0.81) −0.045 (−2.36) −0.065 (−0.88) 0.026 (1.34)

−0.016 (−3.19) −0.031 (−0.78) −0.055 (−2.64) −0.033 (−0.45) 0.016 (0.86)

−0.010 (−2.27) 0.003 (0.25) −0.019 (−2.65) −0.044 (−1.41) 0.015 (1.57)

−0.011 (−2.50) 0.008 (0.63) −0.027 (−3.11) −0.022 (−0.66) 0.009 (0.90)

No Yes

Yes Yes

No Yes

Yes Yes

1807 0.116

1807 0.138

1807 0.109

1807 0.119

Panel B: Departures to prestigious underwriters rated by the analyst Rated IB exit (1) (2) Inaccuracy Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. N R2

Rated IB exitt+1 (3) (4)

−0.010 (−2.44) −0.003 (−0.08) −0.024 (−1.29) 0.003 (0.04) 0.022 (1.37)

−0.012 (−2.96) 0.001 (0.02) −0.037 (−1.84) 0.039 (0.63) 0.009 (0.60)

−0.006 (−1.94) 0.004 (0.34) −0.009 (−1.49) −0.012 (−0.66) 0.010 (1.45)

−0.007 (−2.24) 0.008 (0.73) −0.017 (−2.38) 0.012 (0.65) 0.002 (0.35)

No Yes

Yes Yes

No Yes

Yes Yes

1,807 0.096

1,807 0.121

1807 0.092

1,807 0.109

such as covenant violations (see Griffin et al., 2014) and therefore less subjective than other rating adjustments. The results imply that securities rated by revolving analysts are significantly less likely to be downgraded to default. These findings are important because they are based on measures of ratings accuracy that are either impossible or very unlikely to be influenced by Moody’s surveillance team, suggesting that the documented effect in Table 3 is not simply due to subjectivity in the ex-post adjustment of ratings. I also define accuracy based on upgrades and downgrades only. If investment banks hire pessimistic rather than accurate analysts, potentially because they want to hire away harsh analysts, then one would expect that the ratings issued by revolving analysts need to be adjusted less

downward and more upward. The latter does not seem to be the case. If anything, ratings issued by analysts who get hired by investment banks need to be adjusted less upward. To further alleviate concerns that my measure of accuracy could be reflecting pessimism rather than accuracy, I show in the Internet Appendix that the relation between accuracy and the probability of being hired is remarkably consistent across different subperiods, even though the average number downgrade adjustments varies dramatically across subperiods. I conduct additional robustness checks in the Internet Appendix. For example, I show that my baseline measure of ex-post rating adjustments is strongly correlated with proxies for ratings quality that can be observed by

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

13

Table 4 Robustness. The table presents robustness tests using alternative measures of analyst inaccuracy. The baseline regression refers to columns (2) and (4) from Table 3. For brevity, I only report coefficients of interest and suppress control variables. Economic effects are calculated as the reported coefficient multiplied by the standard deviation of the key independent variable, divided by the mean of the dependent variable. Excess losses are computed as the absolute difference between the tranche’s cumulative losses after three (five) years and Moody’s expected loss benchmark for the initial tranche rating category. Securities are considered to be in default when Moody’s assigns a rating below Ca within three years after issuance. For accuracy based on downgrades and upgrades, only downward or upward rating adjustments are considered. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. Panel A: Departures to prestigious underwriters rated by Moody’s IB exit

Baseline 3-yr 5-yr 3-yr 3-yr 3-yr

excess losses excess losses defaults downgrades upgrades

IB exitt+1

Coeff

t-stat

Econ. effect (%)

N

Coeff.

t-stat

Econ. effect (%)

N

−0.016

(−3.19)

−33.1

1807

−0.011

(−2.50)

−77.9

1807

0.009 0.015 0.152 0.016 0.011

(−1.71) (−3.06) (−1.91) (−3.17) (0.20)

−27.0 −59.6 −10.8 −33.9 1.3

511 655 1807 1807 1807

−0.003 −0.006 −0.108 −0.011 −0.042

(−1.33) (−2.64) (−1.69) (−2.41) (−2.33)

−30.2 −86.1 −26.3 −79.0 −16.5

511 655 1807 1807 1807

t-stat

Econ. effect (%)

N

Panel B: Departures to prestigious underwriters rated by the analyst Rated IB exit

Baseline 3-yr 5-yr 3-yr 3-yr 3-yr

excess losses excess losses defaults downgrades upgrades

Coeff

t-stat

−0.012 0.007 0.009 0.181 0.013 0.055

Rated IB exitt+1

Econ. effect (%)

N

Coeff.

(−2.96)

−35.0

1807

−0.007

(−2.24)

−86.8

1807

(−1.52) (−2.01) (−2.96) (−3.01) (1.04)

−26.8 −48.0 −18.4 −38.4 9.0

511 655 1807 1807 1807

−0.001 −0.003 −0.109 −0.007 −0.012

(−0.88) (−1.80) (−2.79) (−2.21) (−0.90)

−24.0 −74.4 −45.7 −87.4 −7.9

511 655 1807 1807 1807

investment banks at the time of issuance. Overall, I conclude that the positive relation between analyst accuracy and the likelihood of being hired by a prestigious underwriter is robust to a large set of alternative specifications.

tinuous increase in the likelihood of being hired by an investment bank. Second, I exploit departures of socially connected analysts as an instrument for the analyst’s own departure to an investment bank.

4.2. Estimating the magnitude of the incentive effect

4.2.1. Exploiting changes in the demand for investment banking positions My first approach exploits changes in job opportunities at investment banks as a proxy for changes in the probability of being hired by an underwriting bank. Specifically, I use investment bank entries in a particular collateral type as a positive shock to the demand for investment banking positions in this product area. As event dates, I define the filing dates of all prospectuses that list an investment bank as lead underwriter that has not previously been underwriting securities in the collateral group and does not simultaneously enter other product areas. Investment bank entries provide useful events for at least three reasons. First, prospectus filings are publicly observable, and the entry of a new underwriting bank represents a discontinuous event that may signal heightened interest by investment banks in the product area. Second, the event affects the employment prospects of analysts who rate securities in that particular collateral group disproportionally more than those of analysts who rate other products. I can therefore study how the performance of analysts in the affected collateral group changes relative to the performance of a control group. The event is also likely orthogonal to analysts’ learning paths and other career concerns. Third, it allows me to test whether, in the cross-section of analysts

The results reported so far suggest that analyst accuracy matters positively for being hired by an underwriting investment bank. This could happen for at least two reasons. Either investment banks hire based on unobservable characteristics of the analysts that are correlated with accuracy, such as IQ, or they hire based on analysts’ rating accuracy. In the latter case, analysts who aspire to be hired by a prestigious underwriter should have an incentive to improve their accuracy by exerting greater effort. How much analyst performance improves because of the possibility to be hired by a prestigious underwriter is an empirical question. The goal of this section is to provide the first quantitative estimates of this relevant sensitivity in the revolving door literature. To identify the magnitude of the incentive effect, I estimate how much analyst performance changes in response to variation in the likelihood of departing to an investment bank. Since analyst effort is inherently unobservable, I use analyst accuracy as a proxy for effort while controlling for analyst ability via analyst fixed effects. I exploit two distinct sources of variation in the likelihood of leaving to an underwriting investment bank. First, I use changes in the demand for investment banking positions as a discon-

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 14

ARTICLE IN PRESS

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

within the same collateral group, the performance of analysts with certain characteristics changes more than that of others. Specifically, the performance of analysts who are ex ante less likely to leave to investment banks (e.g., because of a higher career switching cost or lower expected lifetime benefits from working in investment banking) should respond less to fluctuations in the demand for investment banking jobs. Exploiting these cross-sectional differences is important to mitigate the possibility that my findings are a result of unobservable factors that are driving both investment bank entry and the overall rating performance in a collateral group, or by other changes that are directly induced by the entry of a new investment bank, such as underwriter competition or average analyst workload. The following thought experiment illustrates my empirical approach. Consider two collateral groups, student loan ABS and auto loan ABS. Suppose now that a new investment bank starts to underwrite securities in student loan ABS but remains absent in auto loan ABS. My conjecture is that this event will signal improved future employment prospects in investment banking for analysts rating student loan ABS.32 In contrast, and by construction, the prospect of future employment in auto loan ABS is not affected. I can therefore estimate the effect of variation in the likelihood of being hired by an underwriting investment bank on analyst incentives by comparing changes in the performance of analysts in student loan ABS and in auto loan ABS around the announcement of the investment bank entry while controlling for analyst fixed effects. To identify collateral group and semester observations where a new investment bank enters the underwriting market, I use the following approach. I consider as an event all collateral-group-semester observations where a prospectus is filed that lists as lead underwriter an investment bank that has not previously been underwriting securities in that collateral group and does not simultaneously start to underwrite other collateral types.33 These criteria yield 16 investment bank entry events in seven collateral groups. To verify that these events are indeed associated with increased prospects to be hired by an underwriting investment bank, Fig. 3 plots the difference in the frequency of analyst departures to investment banks between the event group and the control group. The frequency of analyst departures increases by 7.6 and 7.8 percentage points in the two semesters following the event (equivalent to an increase of 0.86 and 0.88 standard deviations), respectively. Hence, the entry of a new underwriter seems to be a reasonably good proxy for increases in the demand for investment banking positions and in the likelihood of being hired by a top investment bank. Next, I investigate whether average analyst performance reacts positively to the news about improved future employment prospects in investment banking. The following

32 Improved employment prospects may be driven by hiring by the entering investment banks themselves as well as from other investment banks. 33 I restrict the set of potential investment bank entries to all investment banks that appear in the Bloomberg 20 investment bank ranking in the majority of the sample period.

regression is estimated:

Inaccuracyizt = λst + λi +

τ =+3

δτ I (New IB ztτ ) + β  Xizt + izt ,

τ =−3

(4) where Inaccuracyizt stands for analyst inaccuracy as computed in Eq. (2), and λst and λi are market segment × semester and analyst fixed effects, respectively.34 I (New IB τzt ) is a set of seven event-time dummy variables labeled τ = −3, τ = −2, ..., τ = +2, τ = +3, where my convention is that dummy τ = 0 takes on the value one in the collateral group and semester where a prospectus lists a new underwriting investment bank. Vector Xizt includes the same set of control variables as in Table 3. If analyst incentives respond positively to improvements in investment banking opportunities, then one would expect δ τ < 0 for τ = 0 and possibly for periods shortly following the announcement. Table 5, column (1) and the dotted line in Fig. 3 report the results. Two things are worth noticing. First, analysts in the event group and in the control group perform similarly in the pre-event window, alleviating potential concerns about unobserved differences between these two groups. Second, and more important, the performance of analysts in the event group reacts positively to the announcement of the investment bank entry, creating a gap of 0.93 and 1.05 notches vis-à-vis the control group in semesters τ = 0 and τ = +1, respectively. The results imply that a one standard deviation increase in the departure frequency leads to a performance improvement of 40%–44% (= 0.931/0.86/2.69 and 1.053/0.88/2.69, respectively), and it is therefore economically sizable. As analysts leave to investment banks at a higher rate in semesters τ = +1 and τ = +2, the performance difference between the event and the control group returns to zero. Overall, the pattern is consistent with analysts exerting additional effort when opportunities in investment banking arise. These estimates are suggestive of the following rough timeline. Suppose an investment bank starts to underwrite student loan ABS in January 2003, leading to more intense hiring by investment banks over the following two semesters (until 1H2004). Anticipating the improved prospects of moving to an investment bank, analysts rating student loan ABS start to work harder, rating securities more precisely for about two semesters, during 1H2003 and 2H2003. The timing raises two potential questions. First, given that rating adjustments are only observed ex post, is it reasonable to assume that investment banks could already understand in 2H2003 and 1H2004 that the ratings issued by these analysts would be more accurate? Second, even if they were able to observe signals about an analyst’s accuracy, would they update their beliefs about her skill after only one or two semesters? Regarding the first question, I show in the Internet Appendix that ex-post rating adjustments are highly 34 Since the event-time dummies do not vary within the same collateral type and semester, I only include market segment × semester fixed effects in this part of the analysis. See Table 1, Panel A for the list of market segments.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

[m3Gdc;May 24, 2019;7:11] 15

Fig. 3. Event study of changes in the demand for investment banking jobs. The graph plots the frequency of analyst departures to investment banks and average analyst inaccuracy around changes in the demand for investment banking positions. An event is defined as instances where an investment bank enters a new collateral group as a lead underwriter. The gray bars show the difference in the frequency of analyst departures to investment banks between the event group (i.e., the collateral group where a new investment bank enters) and the control group in the window (−3, +3) around the event. For each collateral type and semester, the IB departure frequency, measured as the fraction of analysts departing by the end of the following semester, is regressed on a set of seven event-time dummy variables labeled τ = −3, τ = −2, ..., τ = +2, τ = +3, where my convention is that dummy τ = 0 takes on the value one in the collateral group and semester in which the investment bank entry occurs, market segment × semester and collateral type fixed effects, as well as the same control variables as in Table 5. Each bar shows the coefficient on one of the seven dummy variables, and asterisks ∗ ∗ ∗ , ∗ ∗ , ∗ indicate statistical significance at the 1%, 5%, and 10% level. Standard errors are calculated following the method of Newey–West (1987) with two lags. The dotted line plots the coefficient estimates reported in Table 5, column (1), i.e., the difference in the average analyst inaccuracy between the event and the control group while controlling for analyst fixed effects, over the same event window.

correlated with proxies for ratings quality that can be observed in real time, such as characteristics of the initial press release. Moreover, investment banks may observe signals about analyst performance that are unobservable to the econometrician but highly correlated with ex-post measures of performance. For example, they may receive signals about an analyst’s quality through their professional network, e.g., from conversations with other bankers who have directly worked with the analyst, her colleagues at Moody’s, interactions at practitioner conferences, etc. Regarding the second question, the Internet Appendix provides evidence that investment banks appear to place a high weight on recent (i.e., current semester’s and previous semester’s) performance when making hiring decisions. In light of these results, the timeline implied by the estimates in Table 5 seems plausible. A potential concern could be that investment banks may not randomly decide to enter student loan ABS, and that a new entry is more likely when banks anticipate the asset class to perform well. However, improved asset fundamentals do not necessarily translate into increased ratings accuracy for new ratings. First, analysts who observe an investment bank entry can update their beliefs about asset fundamentals and incorporate these in the new ratings they issue. Second, if analysts fail to update, the entry event would lead to lower, rather than higher, ratings accuracy. Third, under the alternative explanation of improved asset fundamentals, the performance of all analysts should improve. I therefore test whether the performance of some analysts reacts more strongly to the event than the performance of other analysts. More concretely, I use the predicted values from the probit regression of IB exit on

ex-ante analyst characteristics, presented in Table 2, column (2), as a measure of the analyst’s ex-ante likelihood of switching careers. Intuitively, analysts who are ex ante more likely to switch to investment banking should be more sensitive to changes in investment banking opportunities. Table 5, columns (2) and (3) report results. While the performance of analysts who are ex ante less likely to leave does not change significantly—column (2)—the performance of analysts who are ex ante more likely to leave to investment banking improves—column (3). The fact that the improvement in performance is concentrated in this subset of analysts raises the bar for alternative explanations. Most importantly, it mitigates the possibility that investment bank entries coincide with periods during which ratings accuracy is generally high. Another potential concern could be that my results are confounded by correlation between the location of analysts who are ex ante likely to leave to an investment bank and the location of the entering investment bank. Suppose, for example, that analysts located in New York City are ex ante more likely to be hired by an investment bank and also more likely to rate securities by investment banks in located in New York. If investment banks that are located in New York enter a collateral group because of positive expectations about or superior access to assets of that collateral type, then this could potentially explain the improvement in rating performance for this set of analysts.35

35 That said, as discussed above, good asset fundamentals do not necessarily translate into higher ratings accuracy.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC 16

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx Table 5 Event study of changes in the demand for investment banking positions. The table presents results from analyzing analyst inaccuracy around changes in the demand for investment banking positions. An event is defined as instances where an investment bank enters a new collateral group as a lead underwriter. The inaccuracy of analysts in the event collateral group (i.e., the collateral group where an investment bank enters) is compared to the inaccuracy of analysts in other collateral groups in the same market segment (control group) around the event. Analyst inaccuracy is regressed on a set of seven event-time dummy variables labeled τ = −3, τ = −2, ..., τ = +2, τ = +3, where my convention is that dummy τ = 0 takes on the value one in the collateral group and semester in which an investment bank is listed as lead underwriter in a prospectus for the first time. Columns (2) and (3) show the same regression for IB exit ) refers to the analyst’s ex-ante predicted probtwo subgroups of analysts. Pr ( ability of leaving to an investment bank, estimated as the predicted values from the probit model in Table 2, column (2), and is split at the median across all analysts in my sample. All regressions include market segment × semester fixed effects, analyst fixed effects, collateral type dummies, and the same controls as in Table 3. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. Inaccuracy

New IB τ = −3 New IB τ = −2 New IB τ = −1 New IB τ = 0 New IB τ = +1 New IB τ = +2 New IB τ = +3 Controls included Segment × semester f.e. Analyst f.e. N R2

All analysts (1)

Low Pr ( IB exit ) (2)

High Pr ( IB exit ) (3)

−0.119 (−0.30) −0.268 (−0.75) −0.296 (−0.96) −0.931 (−2.62) −1.053 (−3.61) −0.067 (−0.20) −0.190 (−0.61)

0.505 (0.85) −0.156 (−0.33) −0.172 (−0.37) 0.245 (0.49) −0.362 (−0.75) 0.578 (0.93) 0.688 (1.28)

−0.338 (−0.53) −0.185 (−0.34) −0.396 (−0.98) −1.612 (−3.63) −1.267 (−3.60) −0.344 (−0.78) −0.660 (−1.49)

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

1752 0.800

850 0.856

817 0.717

To address this potential concern, I reestimate the regressions presented in Table 5, after replacing market segment × semester fixed effects by market segment × semester × office location fixed effects. This modification produces qualitatively similar results (unreported for brevity) and confirms that differences in analysts’ locations cannot explain the pattern in Table 5.36 Finally, I also estimate the analysis in Table 5 at the individual deal level (results are reported in the Internet Appendix). This approach allows me to distinguish between deals underwritten by the entering investment bank and other deals.37 I continue to find a strong and statistically significant improvement in analyst performance on deals that are not underwritten by the entering bank. For deals underwritten by the entering bank, ratings also tend to be

36 I am able to determine the office location for 212 out of 245 analysts, using location information from the press releases on Moody’s website. Out of these 212 analysts, 195 are located in New York. Hence, in my sample, the analysts’ location has little heterogeneity. 37 Note that I am only able to observe analyst accuracy on deals underwritten by the entering investment bank after the entry has occurred.

more accurate, but they are not statistically different from the control group. Hence, I do not find evidence of less precise ratings on the new securities underwritten by the entering investment bank. Overall, the results support the interpretation that the observed improvement in analyst performance is due to greater analyst effort in anticipation of improved employment opportunities in investment banking. 4.2.2. Exploiting peer effects in occupational choices The second source of variation in the likelihood of being hired by an investment bank that I exploit comes from peer effects. Several studies have documented the importance of peer effects in educational and occupational choices.38 Building on this existing literature, I hypothesize that an analyst is more likely to pursue a job at a top investment bank if one of her peers has made the same

38 For example, Nanda and Sørensen (2010), Lerner and Malmendier (2013), and Field et al. (2016) have documented the importance of peer effects in individuals’ decision to become an entrepreneur.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

17

Table 6 Instrumental variable regression using peer departures to investment banks. This table shows results from instrumental variable regressions. Columns (1) and (2) report the first-stage results and columns (3) and (4) the second stage. The instrument is an indicator equal to one if a connected analyst has departed to a top investment bank by the end of the previous quarter, and zero otherwise. A connected analyst is an analyst who started in the same year as the analyst in question but never rated securities of the same collateral type. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. First stage IB exitt+1 Peer exitt

(1)

(2)

0.276 (4.45)

0.278 (4.46)

0.041 (1.18) 0.0 0 0 (−0.07) −0.002 (−0.09) 0.002 (0.32)

0.044 (1.36) −0.006 (−1.08) 0.001 (0.05) −0.002 (−0.24)

−4.338 (−2.36) 0.542 (1.53) 0.194 (2.12) −0.068 (−0.23) 0.203 (2.57)

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

1759 19.82 0.0 0 0

1759 19.91 0.0 0 0

1759

1759

IB exitt+1 Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. Analyst f.e. N F-statistic p-value

Second stage Inaccuracy (3) (4)

career transition. If this is the case, then peer departures to top investment banks can be used as an instrument for the likelihood of the analyst’s own departure. The instrument is constructed as follows. For each analyst, I determine her peers as the set of analysts who (i) started working for Moody’s in the same calendar year and (ii) never rated securities of the same collateral type as the analyst during their employment at Moody’s. The second criteria is aimed at reducing potential confounding effects if analysts who rate the same type of products compete for the same jobs. For analyst i in semester t, I then define the instrument as an indicator equal to one if at least one of the analyst’s peers has departed to a top investment bank by the end of the previous semester.39 To be a valid instrument, peer departures must satisfy the relevance and exclusion conditions. First, regarding the relevance criterion, my hypothesis that the choice of a peer to join an investment bank should be an important factor in an analyst’s decision to switch career builds on the existing peer effects literature cited above. The first-stage results, reported in Table 6, columns (1) and (2), confirm that peer departures are indeed a strong determinant of an analyst’s own decision to switch, resulting in a reasonably strong F-statistic. Second, regarding the exclusion restriction, it is important that peer departures are uncorrelated with other factors determining analyst performance. One such factor could be analyst baseline ability, which I can control for through analyst fixed effects. In addition, one

39 My results are robust to using the number of peer departures to investment banks rather than the indicator variable.

−4.119 (−2.36) 0.586 (1.78) 0.028 (0.30) 0.065 (0.22) 0.083 (1.08)

would like to rule out that peers are assigned similar securities, and that the performance of these securities as well as investment bank hiring are correlated with those rated by the analyst. Considering only departures of peers who rated securities in different collateral groups substantially reduces this potential issue. In addition, since the number of peer departures varies across analysts within the same collateral type and date, I am able to control for any collateral-wide changes in fundamentals by including collateral type × date fixed effects. Table 6, columns (3) and (4) report the second-stage results. They suggest a strong causal effect of an increase in the likelihood of being hired by an investment bank on analyst performance. The effect is also economically sizable: a one standard deviation increase in the probability of leaving to an investment bank leads to a 1.03 notch (= −4.119 × 0.25) reduction in inaccuracy. Relative to the average inaccuracy of 2.69 notches, this change represents a sizable improvement in performance of 38.3%. Hence, the results from the instrumental variable regressions confirm that the possibility to move to an investment bank represents an important incentive to be accurate and yield an economic magnitude that is remarkably consistent with the analysis of increases in the demand for investment banking positions. 4.2.3. Economic magnitude of the incentive effect The estimates in the previous two sections suggest an accuracy improvement of circa 41% for a one standard deviation increase in the probability to be hired by a prestigious underwriter (taking the midpoint of the 38–44%

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 18

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

estimated range). This improvement is equivalent to a reduction in the average ex-post rating adjustment of 1.1 (= 0.41 × 2.69) notches. In other words, if the average probability to be hired by a prestigious underwriter had been one standard deviation higher, the average tranche in my sample would have been initially rated AA– instead of AA. In the Internet Appendix, I show that a one-notch-lower rating between the AA and A rating categories is associated with an 11 (=(46.6–12.4)/3) basis points (bps) higher yield spread. Using the average effective duration of the Barclays US MBS Index between 20 0 0 and 20 09 as a proxy for the duration of my sample securities, I estimate that an 11 bps increase in the average initial yield spread is consistent with a decline in the average initial bond price of approximately 36 bps. Extrapolating this 36 bps price drop to the entire sample implies an approximate loss of $9.7 billion in the aggregate issuance value. Of course, these numbers are coarse and have to be taken with a grain of salt. They are nevertheless useful for gauging the economic magnitude of the incentive effect.

5. Capture The previous section establishes that the possibility to go work for an underwriting bank generates an incentive for analysts at credit rating agencies to work harder. However, this finding does not rule out that investment banks may also reward analysts for being optimistic, which could lead to capture. Establishing whether, conditional on accuracy, optimism increases the likelihood of being hired by a prestigious underwriter is the purpose of this section. First, I repeat the analysis presented in Table 3 by simultaneously controlling for relative analyst optimism in Eq. (3). Specifically, I define Relative optimism as the average difference (in notches) between the ratings issued by the analyst and the average ratings by Standard & Poor’s (S&P) and Fitch, multiplied by minus one. If analysts are hired by underwriters in return for leniency, then the coefficient on Relative optimism should be positive. Table 7 presents the results. Issuing more optimistic ratings overall does not significantly increase an analyst’s prospect of being hired by an underwriting investment bank. The importance of relative bias is also smaller in economic terms than the importance of accuracy: a one standard deviation increase in relative optimism increases the likelihood of being hired by 13.4% (= 0.047 × 0.20/0.07). The economic magnitude is even smaller when predicting departures to prestigious underwriters directly rated by the analyst (Panel B). Even if underwriters do not place any emphasis on analysts’ optimism on deals underwritten by other banks, they may still reward analysts for being optimistic on their own deals. To test this conjecture, I study the relation between relative optimism and the probability of being hired by the underwriter of a particular deal. Specifically, I estimate the following regression at the individual deal level:

T his IBExitikzt =

[m3Gdc;May 24, 2019;7:11]

λzt + δ1 Inaccuracyk + δ2 Avg. Inaccuracyizt +γ1 Rel. optimismk +γ2 Avg. Rel. optimismizt + β  Xikzt + ikzt , (5)

where This IB exitikzt is an indicator equal to one if one of the analysts i rating deal k is eventually hired by one of the deal’s lead underwriting banks, and zero if the analysts stay at the rating agency. Inaccuracy on a given deal k is computed as the absolute difference (in notches) between the initial rating and the rating three years after issuance, averaged across all rated tranches of deal k. Rel. optimismk is computed for a particular deal k as the difference between Moody’s rating and the average rating by S&P and Fitch, multiplied by minus one and averaged across all rated tranches of the deal. Avg. Inaccuracyizt and Avg. Rel. optimismizt refer to the average inaccuracy and the average relative optimism of the analysts i rating deal k, measured across all deals they rate in the same collateral type z and semester t, excluding deal k itself. λzt are collateral type × semester fixed effects, respectively, and Xikzt represents the same vector of additional controls as in Eq. (3) but measured at the deal level. If underwriting investment banks reward analysts for being optimistic on their own deals, then γ 1 > 0. Table 8 reports the results. For reasons of easier comparison, all variables are standardized to have zero mean and a standard deviation of one. Consistent with the results documented in Section 4.1, the analyst’s average accuracy across all other deals continues to be a significant predictor for being hired by a prestigious underwriter, including by the underwriter of a deal that the analyst is rating. However, there is also evidence that analysts are rewarded for optimism: relative optimism on a particular deal positively predicts being hired by one of the underwriting investment banks—see column (4). In terms of economic magnitude, the effect of optimism corresponds to circa one-fourth of the effect of overall accuracy. Consistent with underwriters placing a greater weight on optimism, the analyst’s accuracy on deal k is less important for the probability of being hired by one of the deal’s underwriters, compared to the probability of being hired by any other underwriter. The difference in coefficients on the inaccuracy variable across specifications is statistically significant at the 1% level between columns (3) and (4). The positive relation between analyst optimism and the likelihood of joining a lead underwriter of the deal is consistent with the findings by Cornaggia et al. (2016), who study corporate bond ratings and show that the ratings of firms are inflated in the quarters before they hire a credit rating analyst. Combined with the previous results in Section 4 on the importance of the incentive effect, these results are informative for policymakers. Consider, for example, the choice between an outright ban of transitions to any underwriter and a more targeted revolving door policy that only prohibits transitions to entities that analysts have recently rated.40 While both policies would reduce conflicts of interest, a broad regulation would likely lead to a greater reduction in labor mobility and, hence, to a greater reduction in effort provision by analysts. In addition, the findings 40 Examples of broad bans include the 2008 Defense Authorization Act and a recent bipartisan bill, which proposes a lifetime ban on lobbying for members of Congress. See “US senators seek lifetime ban on ex-Congress members lobbying,” Reuters, May 18, 2017.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

19

Table 7 Analyst optimism and hiring by underwriting investment banks. The table reports results from regressing indicators for analyst departures to investment banks on ratings accuracy and relative optimism. In Panel A, columns (1) and (2), IB exit is an indicator equal to one if, at the end of her employment with Moody’s, the analyst departs to an investment bank that was ranked in the Bloomberg 20 ranking in the year prior to her departure. In columns (3) and (4), IB exitt+1 is an indicator equal to one if the analyst departs to a top investment bank by the end of the following semester. The dependent variables in Panel B are defined analogously but refer only to departures to underwriters that the analyst has rated. Analyst inaccuracy is measured as the average number of notches by which the ratings issued by the analyst need to be adjusted within three years of issuance. Relative optimism is equal to (minus) the difference between the initial rating assigned by the analyst and the average rating by S&P and Fitch, averaged across all tranches rated by the analyst. All variables are defined in Appendix A. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. Panel A: Departures to prestigious underwriters rated by Moody’s IB exit Inaccuracy Relative optimism Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. N R2

IB exitt+1

(1)

(2)

(3)

(4)

−0.014 (−2.69) 0.017 (0.24) −0.035 (−0.83) −0.047 (−2.43) −0.061 (−0.84) 0.027 (1.34)

−0.016 (−3.07) 0.059 (0.93) −0.031 (−0.76) −0.056 (−2.66) −0.031 (−0.42) 0.017 (0.90)

−0.010 (−2.22) 0.025 (0.74) 0.004 (0.32) −0.020 (−2.79) −0.044 (−1.39) 0.015 (1.57)

−0.012 (−2.45) 0.047 (1.33) 0.008 (0.67) −0.028 (−3.19) −0.022 (−0.66) 0.009 (0.93)

No Yes

Yes Yes

No Yes

Yes Yes

1792 0.116

1792 0.138

1792 0.111

1792 0.123

Panel B: Departures to prestigious underwriters rated by the analyst Rated IB exit (1) (2) Inaccuracy Relative optimism Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. N R2

Rated IB exitt+1 (3) (4)

−0.011 (−2.44) −0.071 (−1.17) −0.004 (−0.10) −0.027 (−1.39) 0.010 (0.15) 0.022 (1.36)

−0.012 (−2.86) −0.031 (−0.55) 0.001 (0.03) −0.038 (−1.85) 0.043 (0.69) 0.010 (0.67)

−0.006 (−1.92) −0.011 (−0.40) 0.004 (0.40) −0.010 (−1.66) −0.010 (−0.56) 0.010 (1.44)

−0.007 (−2.20) 0.009 (0.30) 0.009 (0.77) −0.017 (−2.43) 0.013 (0.67) 0.003 (0.41)

No Yes

Yes Yes

No Yes

Yes Yes

1792 0.098

1792 0.119

1792 0.094

1792 0.111

in Table 8 suggest that incentives for quid pro quo behavior are most pronounced when analysts join entities they recently personally rated. Hence, a revolving door policy that focuses only on transitions to investment banks with which analysts have recently interacted may already be sufficient for reducing capture. Overall, the results in this paper suggest that a targeted policy intervention would be relatively more advisable, and that policymakers should take into account unintended consequences of revolving door policies on labor mobility and analyst incentives.

6. Extensions This section investigates three remaining questions related to the results in Section 4.1. First, could accuracy be less important for employment transitions of senior employees? Second, could the positive relation between accuracy and the likelihood of being hired by a prestigious underwriter be driven by Moody’s punishing accurate analysts? Third, are analysts influenced by prior work experience at investment banks?

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC 20

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx Table 8 Analyst optimism on specific deals and hiring by underwriting investment banks. The table reports results from deal-level regressions of indicators for analyst departures to investment banks on measures of rating accuracy and relative optimism. Other IB exit is an indicator equal to one if an analyst who rates the deal is hired by an investment bank that is not among the lead underwriters of the deal, and zero if the analyst stays at the rating agency. This IB exit is an indicator equal to one if an analyst who rates the deal is hired by one of the lead underwriters of the deal, and zero if the analyst stays at the rating agency. Other IB exitt+1 and This IB exitt+1 are defined analogously but are equal to one only if the analyst is hired by the end of the following semester. Analyst inaccuracy is measured as the average number of notches by which the ratings of the tranches of a deal need to be adjusted within three years of issuance. Relative optimism is equal to (minus) the difference between the initial rating assigned by the analyst and the average rating by S&P and Fitch, averaged across all rated tranches in the deal. Average analyst inaccuracy and average relative optimism refer to the same respective measures, averaged across all other deals rated by the analysts in the same collateral type and semester, excluding the deal itself. For reasons of easier comparison, all variables are standardized to have zero mean and a standard deviation of one. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the lead analyst level.

Inaccuracy Avg. naccuracy Relative optimism Avg. elative optimism Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × sem. f.e. N R2

Other IB exit (1)

This IB exit (2)

Other IB exitt+1 (3)

This IB exitt+1 (4)

−0.075 (−1.86) −0.277 (−2.83) 0.006 (0.35) 0.038 (1.10) 0.020 (0.26) −0.001 (−0.01) −0.025 (−0.58) 0.063 (1.13)

−0.055 (−1.51) −0.092 (−1.90) 0.017 (0.65) 0.006 (0.25) −0.041 (−1.25) 0.016 (0.30) 0.064 (3.01) −0.022 (−1.20)

−0.169 (−3.56) −0.378 (−3.33) 0.008 (0.44) 0.046 (1.57) 0.138 (2.23) −0.134 (−1.74) −0.052 (−1.67) 0.084 (1.66)

−0.022 (−1.10) −0.103 (−1.86) 0.025 (1.66) −0.017 (−0.71) −0.008 (−0.20) −0.084 (−1.14) 0.031 (2.35) −0.004 (−0.28)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

4098 0.182

2972 0.185

3185 0.306

2890 0.240

6.1. The role of analyst seniority Despite the fact that accuracy is correlated with the likelihood of being hired by a prestigious underwriter for the average credit analyst, there could be substantial heterogeneity across analyst ranks. In particular, it would be problematic if investment banks’ hiring of accurate junior analysts were masking distorted hiring decisions at a more senior level. Since Moody’s press releases disclose both the name of the lead analyst and the name of a more senior member of the rating committee (typically the committee chair), my sample consists of Moody’s employees of various ranks. This feature of the data allows me to look at the performance sensitivity of investment bank hiring as a function of analyst seniority. I define three proxies for senior analysts, identifying analysts with a job title of vice president, senior vice president, and managing director or higher, respectively. Then I estimate the baseline regression from Eq. (3) while adding an interaction term between analyst inaccuracy and the indicator for senior analysts. Table 9 reports the results. Overall, the likelihood to depart to an investment bank is as sensitive to ratings inaccuracy for senior analysts

as it is for junior analysts—columns (1)–(3). However, all three proxies indicate that, for senior analysts, the likelihood to be hired by an investment bank by the end of the following semester is less sensitive to recent performance than for junior analysts. This pattern is consistent with investment banks learning about an analyst’s ability over time and placing a lower weight on the most recent performance signal if they have observed the analyst over a longer period of time. Importantly, the point estimates suggest that the sign of the relation between inaccuracy and the likelihood of being hired does not change for senior analysts, i.e., even for more senior analysts, inaccuracy negatively predicts a transfer to an underwriting investment bank.

6.2. Disincentives at Moody’s A second potential concern could be that my results reflect disincentives within Moody’s organization as opposed to investment banks hiring based on observed performance. For example, if Moody’s was strongly focused on expanding its market share, as suggested by the

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

21

Table 9 The role of analyst seniority. The table presents results for interactions with three different measures of analyst seniority. The regressions presented in Table 3, columns (2) and (4), are estimated with an interaction term between analyst inaccuracy and an indicator for senior analysts. In columns (1)–(3), IB exit is an indicator equal to one if, at the end of her employment with Moody’s, the analyst departs to an investment bank that was ranked in the Bloomberg 20 ranking in the year prior to her departure. In columns (4)–(6), IB exitt+1 is an indicator equal to one if the analyst departs to a top investment bank by the end of the following semester. Analyst inaccuracy is measured as the average number of notches by which the ratings issued by the analyst need to be adjusted within three years of issuance. All variables are defined in Appendix A. In columns (1) and (4) ((2) and (5)) [(3) and (6)], senior analysts are defined as analysts with a title of vice president (senior vice president) [managing director] or higher. Analyst job titles are obtained from press releases on Moody’s website and categorized as in Kisgen et al. (2018). t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. Panel A: Departures to prestigious underwriters rated by Moody’s

≥ VP (1) Inaccuracy Senior Inaccuracy × senior Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. N R2

IB exit IB exitt+1 Definition of analyst seniority ≥ SVP ≥ MD ≥ VP ≥ SVP (2) (3) (4) (5)

≥ MD (6)

−0.020 (−2.53) −0.221 (−3.72) 0.006 (0.79) −0.015 (−0.38) −0.051 (−2.68) −0.011 (−0.14) 0.010 (0.52)

−0.019 (−2.53) −0.220 (−3.75) 0.006 (0.75) −0.015 (−0.38) −0.051 (−2.68) −0.012 (−0.16) 0.010 (0.52)

−0.016 (−2.48) −0.130 (−3.75) 0.001 (0.07) −0.036 (−0.95) −0.048 (−2.52) −0.028 (−0.38) 0.009 (0.51)

−0.017 (−3.36) −0.077 (−2.44) 0.009 (2.33) 0.012 (0.98) −0.026 (−3.13) −0.016 (−0.51) 0.008 (0.84)

−0.017 (−3.37) −0.079 (−2.53) 0.009 (2.37) 0.012 (0.99) −0.026 (−3.13) −0.017 (−0.51) 0.007 (0.84)

−0.014 (−2.93) −0.064 (−3.11) 0.006 (1.63) 0.006 (0.49) −0.025 (−2.97) −0.020 (−0.60) 0.006 (0.70)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

1807 0.176

1807 0.176

1807 0.157

1807 0.131

1807 0.132

1807 0.129

Panel B: Departures to prestigious underwriters rated by the analyst

Inaccuracy Senior Inaccuracy × Senior Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. N R2

≥ VP (1)

Rated IB exit ≥ SVP (2)

≥ MD (3)

≥ VP (4)

Rated IB exitt+1 ≥ SVP ≥ MD (5) (6)

−0.017 (−2.52) −0.200 (−3.44) 0.009 (1.18) 0.015 (0.38) −0.033 (−1.80) 0.058 (0.90) 0.004 (0.24)

−0.017 (−2.48) −0.194 (−3.38) 0.009 (1.13) 0.014 (0.37) −0.033 (−1.81) 0.057 (0.88) 0.004 (0.25)

−0.015 (−2.54) −0.129 (−3.81) 0.005 (0.70) −0.004 (−0.10) −0.030 (−1.69) 0.044 (0.70) 0.003 (0.21)

−0.012 (−3.09) −0.052 (−1.93) 0.008 (2.38) 0.011 (0.98) −0.016 (−2.36) 0.015 (0.77) 0.002 (0.31)

−0.012 (−3.08) −0.051 (−1.90) 0.008 (2.37) 0.011 (0.96) −0.016 (−2.36) 0.015 (0.75) 0.002 (0.32)

−0.011 (−3.04) −0.054 (−2.88) 0.007 (2.15) 0.007 (0.63) −0.015 (−2.24) 0.014 (0.73) 0.001 (0.15)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

1807 0.158

1807 0.156

1807 0.141

1807 0.120

1807 0.120

1807 0.121

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC 22

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx Table 10 Analyst accuracy and internal promotions. The table presents results from regressing an indicator for analyst promotion on measures of analyst accuracy. Promotiont+1 is an indicator equal to one if the analyst is internally promoted at Moody’s (i.e., moves to a higher job title) during the following semester. Analyst job titles are obtained from press releases on Moody’s website and categorized as in Kisgen et al. (2018). Analyst inaccuracy is measured as the average number of notches by which the ratings issued by the analyst need to be adjusted within three years of issuance. In columns (1) and (2), the regression is estimated using all analysts. In columns (3) and (4), only junior analysts with a job title below vice president are included. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the analyst level. Promotiont+1 All analysts Junior analysts ( < VP) (1) (2) (3) (4) Inaccuracy Tenure Number of deals IB underwriter Issuer market share

−0.004 (−1.42) 0.004 (0.62) −0.020 (−3.86) 0.003 (0.19) 0.007 (0.99)

−0.004 (−1.62) 0.005 (0.64) −0.022 (−3.91) 0.002 (0.12) 0.005 (0.61)

−0.008 (−1.73) 0.019 (0.76) −0.002 (−0.17) 0.015 (0.60) 0.008 (0.86)

−0.008 (−1.77) 0.020 (0.80) −0.003 (−0.25) 0.012 (0.43) 0.004 (0.46)

No Yes

Yes Yes

No Yes

Yes Yes

1656 0.132

1656 0.136

648 0.210

648 0.220

Deal controls Collateral type × semester f.e. N R2

Financial Crisis Inquiry Commission (2011),41 it may have punished analysts who issued accurate ratings by not promoting them. This interpretation could explain why accurate analysts may choose to seek employment elsewhere. However, it cannot explain why analysts hired by investment banks outperform and not the average analyst who transitions to other employers. The evidence reported in the Internet Appendix, where I show that departures to other types of employers are at best weakly negatively correlated with inaccuracy, is therefore inconsistent with this story.42 To further investigate this potential concern, I look at the relation between analyst performance and internal promotions at Moody’s. I identify promotions based on changes in the analyst’s job title mentioned in the press releases from Moody’s website, which are classified into the same categories as in Kisgen et al. (2018). The results, shown in Table 10, do not support the conjecture that Moody’s punishes analysts for being accurate. Consistent with the findings by Kisgen et al. (2018), who study promotions of corporate bond analysts at Moody’s, accurate

41 The Financial Crisis Inquiry Commission (2011) reports that “a strong emphasis on market share was evident in employee performance evaluations” at Moody’s. 42 Hiring by investment banks may be more sensitive to rating performance than hiring by other employers because (i) credit rating skill may be particularly valuable for tasks required by investment banks, such as structuring securitized finance deals ahead of public offerings (see BarIsaac and Shapiro, 2011); (ii) investment banks may have superior ability to access or interpret information about analysts’ performance (Cetorelli and Peristiani, 2012); and (iii) investment banks may be more prestigious employers.

analysts in structured finance are also more likely to be promoted. However, the relation between performance and internal promotions is weaker, both in economic and in statistical terms, than the previously documented relation between performance and departures to underwriting investment banks: a one standard deviation decrease in inaccuracy is associated with a 33% increase in the promotion probability (= 0.004 × 4.96/0.06). It is slightly stronger for more junior analysts, i.e., analysts below the rank of vice president—columns (3) and (4). 6.3. Inbound revolvers In my last test, I study whether analysts with former ties to investment banks (“inbound revolvers”) may be conflicted. Relative to the debate about potentially conflicted outbound revolvers, the role of inbound revolvers has attracted less attention in the context of rating agencies. Yet, it is not uncommon for Moody’s to employ analysts with former work experience at investment banks: I identify 53 analysts. Out of these 53 inbound revolvers, 22 rate their former employer at some point during my sample period. As the test on future employers in Section 5, the analysis is run at the individual deal level to be able to distinguish deals that involve a former employer from other deals. Analyst inaccuracy (Panel A) and relative optimism (Panel B) are regressed on an indicator variable equal to one if the analyst has previously worked at a top investment bank (Past IB), and zero otherwise, as well as on an interaction term between Past IB and an indicator equal to one if one of the lead underwriting banks of the deal

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

23

Table 11 Inbound revolvers. The table reports results from deal-level regressions of ratings accuracy and relative optimism on an indicator for past investment banking experience. Past IB is an indicator equal to one if one of the analysts rating the deal has worked for an investment bank that was ranked in the Bloomberg 20 ranking prior to her employment at Moody’s. Past employer is an indicator equal to one if one of the lead underwriting banks of the deal is the former employer of one of the analysts involved in the rating. In Panel A, the dependent variable is analyst inaccuracy, measured as the average number of notches by which the ratings of the deal’s tranches need to be adjusted within three years of issuance. In Panel B, the dependent variable is relative optimism, which is equal to (minus) the difference between the initial rating assigned by the analyst and the average rating by S&P and Fitch, averaged across all rated tranches in the deal. t-statistics, reported in parentheses, are based on standard errors that allow for clustering at the lead analyst level. Panel A: Inaccuracy

Past IB

(1)

(2)

0.113 (0.72)

0.136 (0.87)

Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. Analyst × coll. type × sem. f.e. N R2

(4)

(5)

0.161 (1.02) −0.431 (−1.84) −0.171 (−1.26) 0.035 (0.45) −0.087 (−0.61) 0.098 (2.11)

−0.650 (−2.12) −0.305 (−1.94) 0.078 (0.51) −0.076 (−0.79) 0.073 (1.80)

−0.209 (−1.50) 0.108 (1.54) −0.104 (−0.73) 0.124 (2.78)

−0.172 (−1.27) 0.037 (0.48) −0.098 (−0.68) 0.098 (2.11)

0.135 (0.84) −0.378 (−1.56) −0.208 (−1.50) 0.108 (1.53) −0.095 (−0.68) 0.124 (2.79)

No Yes No

Yes Yes No

No Yes No

Yes Yes No

Yes No Yes

3784 0.800

3784 0.802

3784 0.800

3784 0.802

3784 0.802

(1)

(2)

0.002 (0.16)

−0.001 (−0.07)

Past IB × past employer Tenure

Inaccuracy (3)

Panel B: Relative optimism

Past IB

Relative optimism (3) (4)

0.014 (1.56) −0.029 (−4.16) 0.008 (0.45) 0.006 (1.53)

0.011 (1.27) −0.024 (−3.59) 0.010 (0.62) 0.008 (2.03)

−0.001 (−0.07) 0.039 (1.67) 0.014 (1.56) −0.029 (−4.15) 0.007 (0.40) 0.006 (1.52)

No Yes No

Yes Yes No

No Yes No

Yes Yes No

Yes No Yes

3748 0.079

3748 0.091

3748 0.080

3748 0.091

3748 0.091

Past IB × Past employer Tenure Number of deals IB underwriter Issuer market share Deal controls Collateral type × semester f.e. Analyst × coll. type × sem. f.e. N R2

(5)

is a former employer of one of the analysts involved in rating the deal (Past employer). Table 11 presents results. On average, analysts with former investment banking experience do not perform statistically differently from other analysts—columns (1) and (2). However, they are 16% (= 0.431/2.69) more accurate when rating deals underwritten by their former employers—column (4). The difference gets even larger once I include analyst × collateral type × semester fixed effects—column (5). Hence, the same

−0.003 (−0.29) 0.040 (1.75) 0.011 (1.25) −0.024 (−3.57) 0.009 (0.55) 0.008 (2.02)

0.045 (1.24) −0.005 (−0.37) −0.022 (−2.01) 0.010 (0.59) 0.001 (0.25)

analyst is more accurate on deals underwritten by her past employer, compared to deals of the same collateral type that she rates at the same point in time. Overall, the pattern is consistent with an informational advantage of analysts who rate their former employers and inconsistent with conflicts of interest. While the ratings on deals by former employers tend to be slightly more optimistic relative to S&P and Fitch (see Panel B), this does not appear to hurt overall accuracy. In sum, I do not find

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 24

ARTICLE IN PRESS

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

evidence of sizable distortions in the performance of inbound revolvers. 7. Conclusion My paper contributes to the ongoing debate on how revolving doors affect monitoring incentives. The results indicate that the possibility for analysts to go work for an underwriting investment bank improves ratings accuracy via an incentive effect. I begin my analysis by showing that underwriting investment banks hire analysts who are substantially more accurate relative to peers who rate similar securities at the same point in time. In addition, using two distinct sources of variation in the likelihood of being hired by an underwriter, I show that the possibility to go work for a prestigious underwriter leads to economically sizable improvements in ratings accuracy. However, consistent with prior work on corporate bond ratings, I also find that being optimistic on specific deals increases the likelihood of being hired by one of the underwriters of the deal. Overall, my results suggest that optimal regulation of revolving doors should balance both capture and incentive effects. How well the findings in this paper can be translated to other revolving door settings will depend on a number of factors, including the observability of individual performance signals and the level of competition and skill intensity of the industry in which hiring firms operate. Further research is needed to improve our understanding of the importance of capture and incentive effects in different professions. My paper also contributes to the debate about the causes of poor performance of securitized finance ratings

prior to the financial crisis. Many observers have identified conflicted individual analysts as one of the drivers of poor ratings accuracy, and regulators have responded by imposing enhanced disclosure requirements on rating agencies in cases where employees transfer to rated entities. My results imply that analysts likely issued more accurate ratings because of concerns about their outside labor market options. Restricting the possibility to join prestigious underwriters may therefore have the undesirable effect of discouraging rating analysts from developing and showcasing their expertise while employed at the rating agency. An excessive regulatory focus on conflicted individual analysts may further be detrimental if it shifts the regulator’s attention away from addressing first-order drivers of poor ratings performance in securitized finance.43 This paper focuses on performance incentives and is silent on other channels through which the revolving door may affect ratings quality. For example, credit ratings quality may suffer if rating agencies systematically lose their more experienced or talented staff to investment banks, reducing their incentives to train new analysts (see BarIsaac and Shapiro, 2011). In addition, former analysts may help investment banks game the rating system once they have left the rating agency (Jiang et al., 2018). On the other hand, there may be other positive aspects of the revolving door that I am not capturing in my analysis. For example, the option to move to investment banking may positively affect the quality of the pool of applicants at rating agencies, and many motivated applicants may no longer apply if career mobility is reduced.

Appendix A. Variable descriptions

43 The academic literature has, for example, pointed to distortions created by the “issuer pays” business model of credit rating agencies, such as an excessive focus on issuer relationships (He et al., 2012; Efing and Hau, 2015; Morkötter et al., 2016), rating shopping (Benmelech and Dlugosz, 2009; Mathis et al., 2009; He et al., 2016), and rating catering (Griffin et al., 2013; He et al., 2016). In addition, interactions of the business model with the lack of investor sophistication (Skreta and Veldkamp, 2009; Bolton et al., 2012), regulatory arbitrage (Opp et al., 2013), and the business cycle (Bar-Isaac and Shapiro, 2013) have been identified as potential drivers of poor ratings quality in securitized finance.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

ARTICLE IN PRESS

JID: FINEC

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

[m3Gdc;May 24, 2019;7:11] 25

Table A.1 Variable descriptions. Variable

Description

Career path variables IB exit

Indicator function equal to one if the analyst departs to an investment bank following her employment at Moody’s. Investment banks are employers that were ranked in the Bloomberg 20 ranking in the year prior to the analyst’s departure. For departures prior to 2005, the Bloomberg ranking from 2004 is used. Post-Moody’s employer information is obtained from public profiles on a professional networking website and web searches. Indicator function equal to one if the analyst departs to an investment bank by the end of the following semester. IB exitt+1 Investment banks are employers that were ranked in the Bloomberg 20 ranking in the year prior to the analyst’s departure. For departures prior to 2005, the Bloomberg ranking from 2004 is used. Post-Moody’s employer information is obtained from public profiles on a professional networking website and web searches. Rated IB exit Indicator function equal to one if the analyst departs to an investment bank following her employment at Moody’s and has rated a deal underwritten by the future employer. Investment banks are employers that were ranked in the Bloomberg 20 ranking in the year prior to the analyst’s departure. For departures prior to 2005, the Bloomberg ranking from 2004 is used. Post-Moody’s employer information is obtained from public profiles on a professional networking website and web searches. Indicator function equal to one if the analyst departs to an investment bank by the end of the following semester and has Rated IB exitt+1 rated a deal underwritten by the future employer. Investment banks are employers that were ranked in the Bloomberg 20 ranking in the year prior to the analyst’s departure. For departures prior to 2005, the Bloomberg ranking from 2004 is used. Post-Moody’s employer information is obtained from public profiles on a professional networking website and web searches. Indicator function equal to one if the analyst is internally promoted (i.e., moves to a higher job title) during the following Promotiont+1 semester. Analyst job titles are obtained from press releases on Moody’s website and categorized as in Kisgen et al. (2018). Measures of analyst (in)accuracy and relative optimism The average absolute difference (in notches) between Moody’s initial rating of the security and the rating three years Inaccuracy (baseline) following the issuance, averaged across all ratings issued by a given analyst in a given collateral type and semester by taking the arithmetic mean. Initial ratings are obtained from Bloomberg and rating adjustments from Moody’s website. 3(5)-yr excess losses The average absolute difference between the cumulative tranche losses, i.e., the principal balance write-offs due to default, after three (five) years following the issuance and Moody’s expected loss benchmark for the tranche’s initial rating category, averaged across all ratings issued by a given analyst in a given collateral type and semester by taking the arithmetic mean. Cumulative tranche losses are obtained from Bloomberg and Moody’s expected loss benchmarks are retrieved from Moody’s website (available at https://www.moodys.com/sites/products/productattachments/marvel_user_guide1_V1.pdf). Initial ratings are obtained from Bloomberg. 3-yr defaults The average fraction of securities in default within three years after issuance, averaged across all ratings issued by a given analyst in a given collateral type and semester by taking the arithmetic mean. Securities are considered in default when Moody’s assigns a rating below Ca. Rating adjustments are obtained from Moody’s website. The average absolute difference between Moody’s initial rating of the security and the rating three years following the 3-yr downgrades (upgrades) issuance, provided that the new rating is lower (higher) than the initial rating and averaged across all ratings issued by a given analyst in a given collateral type and semester by taking the arithmetic mean. Initial ratings are obtained from Bloomberg and rating adjustments from Moody’s website. Relative optimism The difference (in notches) between Moody’s initial rating of the security and the average rating assigned by S&P and Fitch, multiplied by minus one and averaged across all ratings issued by a given analyst in a given collateral type and semester by taking the arithmetic mean. Initial ratings are obtained from Bloomberg. Control variables Tenure Logarithm of one plus the number of semesters since the beginning of the analyst’s employment at Moody’s, which is the earlier date of the analyst’s reported start date on the professional networking website and her first appearance in the data set. Number of deals Logarithm of one plus the number of deals rated by the analyst in a given collateral type and semester. IB underwriter The fraction of tranches rated by the analyst in a given collateral type and semester that are underwritten by an investment bank that was rated in the Bloomberg Top 20 ranking in the year prior to ratings issuance. For ratings issued prior to 2005, the Bloomberg ranking from 2004 is used. Lead underwriter information is obtained from SDC Platinum. Issuer market share The average market share of the issuer with Moody’s, based on the dollar volume of deals across all collateral types originated in the previous calendar year, averaged across all tranches rated by the analyst in a given collateral type and semester. Weighted avg. life The number of years that are expected to elapse from the closing date until each dollar of the tranche’s principal is repaid to the investor, averaged across all tranches rated by the analyst in a given collateral type and semester. Geographical HHI The average sum of the squared shares of the collateral within a deal across each of the five US states with the largest aggregate amount of loans, with the aggregation of all the other states as the sixth category, averaged across all tranches rated by the analyst in a given collateral type and semester. Weighted avg. credit The weighted average FICO score of the borrowers in the underlying collateral at issuance, averaged across all tranches rated by the analyst in a given collateral type and semester. score Weighted avg. LTV The weighted average loan-to-value ratio of the loans in the underlying collateral at issuance, averaged across all tranches rated by the analyst in a given collateral type and semester. Insurance wrap The fraction of tranches rated by the analyst in a given collateral type that have a financial guaranty insurance. Deal size The principal amount of all tranches belonging to a given deal at issuance, averaged across all tranches rated by the analyst in a given collateral type and semester. Number of tranches The number of tranches belonging to the same deal, averaged across all tranches rated by the analyst in a given collateral type and semester. Overcollateralization The difference between total collateral value and the combined principal value of the tranches at issuance, averaged across all tranches rated by the analyst in a given collateral type and semester.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC 26

ARTICLE IN PRESS

[m3Gdc;May 24, 2019;7:11]

E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

Appendix B. Moody’s organizational structure

Fig. B.1. Moody’s organizational structure in structured finance. The chart shows the organizational structure of the Structured Finance team at Moody’s as reported on Moody’s website (retrieved on December 2, 2015, from https://www.moodys.com/research/Structured- Finance- Ratings- Quick- Check- Newsletter- - PBS_SF161380). The dashed line highlights the division of interest for this paper, i.e., new ratings in the Americas region.

References Acharya, V., Richardson, M., 2009. Causes of the financial crisis. Crit. Rev. 21 (2–3), 195–210. Acharya, V., Schnabl, P., Suarez, G., 2013. Securitization without risk transfer. J. Financ. Econ. 107 (3), 515–536. Agarwal, S., Lucca, D., Seru, A., Trebbi, F., 2014. Inconsistent regulators: evidence from banking. Q. J. Econ. 129 (2), 889–938. Ashcraft, A., Goldsmith-Pinkham, P., Vickery, J., 2010. MBS ratings and the mortgage credit boom. Unpublished working paper. Bar-Isaac, H., Shapiro, J., 2011. Credit ratings accuracy and analyst incentives. Am. Econ. Rev. Papers Proc. 101 (3), 120–124. Bar-Isaac, H., Shapiro, J., 2013. Ratings quality over the business cycle. J. Financ. Econ. 108, 62–78. Becker, B., Ivashina, V., 2015. Reaching for yield in the bond market. J. Finance 70 (5), 1863–1901. Benmelech, E., Dlugosz, J., 2009. The credit rating crisis. NBER Macro Ann. 24, 161–207. Bertrand, M., Bombardini, M., Trebbi, F., 2014. Is it whom you know or what you know? An empirical assessment of the lobbying. Am. Econ. Rev. 104 (12), 3885–3920. Blanes i Vidal, J., Draca, M., Fons-Rosen, C., 2012. Revolving door lobbyists. Am. Econ. Rev. 102 (7), 3731–3748. Bolton, P., Freixas, X., Shapiro, J., 2012. The credit ratings game. J. Finance 67 (1), 85–111. Cetorelli, N., Peristiani, S., 2012. The role of banks in asset securitization. Federal Reserve Bank NY Econ. Policy Rev. 18 (2), 47–64. Che, Y.-K., 1995. Revolving doors and the optimal tolerance for agency collusion. RAND J. Econ. 26 (3), 378–397.

Cohen, L., Frazzini, A., Malloy, C.J., 2012. Hiring cheerleaders: board appointments of “independent” directors. Manag. Sci. 58 (6), 1039–1058. Cornaggia, J., Cornaggia, K.J., Hund, J., 2017. Credit ratings across asset classes: a long-term perspective. Rev. Finance 21 (2), 465–509. Cornaggia, J., Cornaggia, K.J., Xia, H., 2016. Revolving doors on Wall Street. J. Financ. Econ., 120 (2), 400–419. deHaan, E., Kedia, S., Koh, K., Rajgopal, S., 2015. The revolving door and the SEC’s enforcement outcomes: initial evidence from civil litigation. J. Account. Econ. 60 (2), 65–96. Eckert, R.D., 1981. The life cycle of regulatory commissioners. J. Law Econ. 24 (1), 113–120. Efing, M., Hau, H., 2015. Structured debt ratings: evidence on conflicts of interest. J. Financ. Econ. 116 (1), 46–60. Field, E., Jayachandran, S., Pande, R., Rigol, N., 2016. Friendship at work: can peer effects catalyze female entrepreneurship? Am. Econ. J. Econ. Policy 8 (2), 125–153. Financial Crisis Inquiry Commission, 2011. The Financial Crisis Inquiry Report. Washington, DC: US Government Printing Office. Forster, M., Shive, S., 2017. The revolving door for financial regulators. Rev. Finance 21 (4), 1445–1484. Fracassi, C., Petry, S., Tate, G., 2016. Does rating analyst subjectivity affect corporate debt pricing? J. Financ. Econ. 120 (3), 514–538. Griffin, J.M., Lowery, R., Saretto, A., 2014. Complex securities and underwriter reputation: do reputable underwriters produce better securities? J. Finance 27 (10), 2872–2925. Griffin, J.M., Nickerson, J., Tang, D.Y., 2013. Rating shopping or catering? An examination of the response to competitive pressure for CDO credit ratings. Rev. Financ. Stud. 26 (9), 2270–2310.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012

JID: FINEC

ARTICLE IN PRESS E. Kempf / Journal of Financial Economics xxx (xxxx) xxx

Griffin, J.M., Tang, D.Y., 2012. Did subjectivity play a role in CDO credit ratings? J. Finance 67 (4), 1293–1328. Hand, J.R., Holthausen, R.W., Leftwich, R.W., 1992. The effect of bond rating agency announcements on bond and stock prices. J. Finance 47, 733–752. He, J., Qian, J., Strahan, P.E., 2012. Are all ratings equal? The impact of issuer size on pricing of mortgage-backed securities. J. Finance 67 (6), 2097–2137. He, J., Qian, J., Strahan, P.E., 2016. Does the market understand rating shopping? Predicting MBS losses with initial yields. Rev. Financ. Stud. 29 (2), 457–485. Horton, J., Serafeim, G., Wu, S., 2017. Career concerns of banking analysts. J. Account. Econ. 63 (2–3), 231–252. Jiang, X., Wang, I.Y., Wang, K.P., 2018. Revolving rating analysts and ratings of mortgage-backed and asset-backed securities: evidence from LinkedIn. Management Science 64, 5461–5959. Jorion, P., Liu, Z., Shi, C., 2005. Informational effects of Regulation FD: evidence from rating changes. J. Financ. Econ. 76, 309–330. Kisgen, D.J., Osborn, M.G., Reuter, J., 2018. Analyst promotions within credit rating agencies: accuracy or bias? J. Financ. Quant. Anal. forthcoming. Lerner, J., Malmendier, U., 2013. With a little help from my (random) friends: success and failure in post-business school entrepreneurship. Rev. Financ. Stud. 26 (10), 2411–2452. Lourie, B., 2019. The revolving door of sell-side analysts. Account. Rev. 91 (1), 249–270.

[m3Gdc;May 24, 2019;7:11] 27

Lucca, D., Seru, A., Trebbi, F., 2014. The revolving door and worker flows in banking regulation. J. Monet. Econ. 65, 17–32. Mathis, J., McAndrews, J., Rochet, J.-C., 2009. Rating the raters: are reputation concerns powerful enough to discipline rating agencies? J. Monet. Econ. 52, 657–674. Morkötter, S., Stebler, R., Westerfield, S., 2016. Revenues versus reputation: warring incentives in the rating industry. Unpublished working paper. Nanda, R., Sørensen, J.B., 2010. Workplace peers and entrepreneurship. Manag. Sci. 56 (7), 1116–1126. Opp, C.C., Opp, M.M., Harris, M., 2013. Rating agencies in the face of regulation. J. Financ. Econ. 108, 46–61. Peltzman, S., 1976. Toward a more general theory of regulation. J. Law Econ. 19 (2), 211–240. Salant, D.J., 1995. Behind the revolving door: a new view of public utility regulation. RAND J. Econ. 26 (3), 362–377. Shleifer, A., Vishny, R.W., 1993. Corruption. Q. J. Econ. 108 (3), 599–617. Skreta, V., Veldkamp, L., 2009. Ratings shopping and asset complexity: a theory of ratings inflation. J. Monet. Econ. 56 (5), 678–695. Stigler, G.J., 1971. The theory of economic regulation. Bell J. Econ. Manag. Sci. 2 (1), 3–21. Tabakovic, H., Wollmann, T.G., 2018. Effects of regulatory capture: evidence from patent examiners. Unpublished working paper. White, L.J., 2010. The credit rating agencies. J. Econ. Persp. 24 (2), 211–226.

Please cite this article as: E. Kempf, The job rating game: Revolving doors and analyst incentives, Journal of Financial Economics, https://doi.org/10.1016/j.jfineco.2019.05.012