Job Market Paper. The Real Consequences of Market Segmentation

Transcription

1 Job Market Paper The Real Consequences of Market Segmentation Sergey Chernenko Harvard University Adi Sunderam Harvard University December 2, 2009 Abstract We study the real effects of market segmentation due to credit ratings. We focus on a matched sample of firms just above (BBB-) and just below (BB+) the investment-grade cutoff. These firms are similar on observable characteristics, including average rates of investment. However, flows into high-yield mutual funds only affect the cost of capital for the speculativegrade firms. We show that these cost of capital shocks have an economically significant effect on issuance and investment, especially for firms that are smaller, do not pay dividends, cannot substitute to bank loans, or are unlikely to access the asset-backed securities market. The effect of fund flows is associated with the discrete change in label from investment-grade to speculative-grade, not with changes in continuous measures of credit quality. We also conduct falsification tests showing that the investment-grade cutoff is the only one where the investment of firms just below the cutoff is more sensitive to fund flows than the investment of firms just above the cutoff. Finally, we offer some suggestive evidence that during the recent credit boom CDO flows increased the investment of BB+ firms relative to their BBB- matches. JEL Classifications: G24, G31 Key Words: credit ratings, firm investment, mutual fund flows, market segmentation, comovement We thank Malcolm Baker, Bo Becker, Effi Benmelech, Dan Bergstresser, Lauren Cohen, Fritz Foley, Robin Greenwood, Victoria Ivashina, David Scharfstein, Erik Stafford, Jeremy Stein and Harvard University finance lunch seminar participants for helpful comments and suggestions, Doug Richardson at the Investment Company Institute for providing mutual fund flow data, and Jerome Fons, Martin Fridson, Oleg Melentyev, and Michael Weilheimer for discussing the high yield market with us. Appendix tables can be accessed at ~schernen/papers/credit_ratings_appendix.pdf

2 1 Introduction Capital markets play a critical role in the efficient allocation of capital across firms. Their ability to play this role, however, may be impeded by market segmentation. In this paper, we study one of the most prominent divides in capital markets the distinction drawn between investmentand speculative-grade firms. A large number of regulations, investment charters, and contracts reference this distinction, and recent research suggests that it can affect firms cost of capital (Ellul, Jotikasthira, and Lundblad, 2009; Kisgen and Strahan, 2009). Market segmentation between investment- and speculative-grade firms may also have real effects. When retail investors withdraw capital from high-yield mutual funds, which are large buyers of speculative-grade bonds, arbitrage capital may not immediately offset this negative shock (Mitchell, Pedersen, and Pulvino, 2007; Duffie and Strulovici, 2009). As a result, the supply of capital available to speculative-grade firms will be reduced. Unable to access other sources of financing, some speculative-grade firms will be forced to cut their investment. This paper documents evidence of this mechanism at work. Simple comparisons between investment- and speculative-grade firms would be confounded by differences in fundamentals between them. Instead, drawing on the econometric literature on treatment effects and regression discontinuity, we construct a matched sample of firms just above (BBB-) and just below (BB+) the investment-grade cutoff. These firms are similar on observable characteristics, but high-yield fund flows only affect the cost of capital of speculative-grade firms. As a result, although these firms have the same average rates of investment, fund flows cause the investment of firms just below the cutoff to diverge from the investment of matched firms just above the cutoff. This effect is economically meaningful one standard deviation increase in fund flows increases the investment of BB+ firms relative to their BBB- matches by about 7% of the average investment rate. The effect is stronger for smaller, financially constrained firms that have limited ability to substitute to either bank loans or the asset-backed securities market. Our results indicate that market segmentation causes temporary differences in the investment of similar firms, but that these differences tend to average out over time. A natural objection to our interpretation of these results is that fund flows may be anticipating 1

3 investment opportunities. Our empirical methodology matches BB+ and BBB- firms based on industry and firm characteristics such as size, leverage, Altman s z-score, and Q. Our identifying assumption is that firms close to the cutoff with similar observable characteristics are subject to similar shocks to profitability and investment opportunities. If this assumption holds, then by studying the differential sensitivity of investment to fund flows we difference out any common shocks that may be correlated with fund flows. We can then interpret the differential effect of fund flows on the investment of BB+ firms as evidence of a supply of capital effect. Although our matching methodology differences out any common shocks, investment opportunities may still not be exactly the same, and fund flows may be picking up the differential investment opportunities of less creditworthy firms. We address this concern in a number of ways. First, drawing on the regression discontinuity literature, we control for the interaction of fund flows with continuous measures of credit quality. Our results show that the greater sensitivity of BB+ firm investment to fund flows is due to their speculative-grade label, not their size, leverage, or Altman s z-score. Second, we conduct falsification tests, examining the sensitivity of investment to fund flows for firms around other rating cutoffs. We find that the investment-grade cutoff is the only one where the investment of firms below the cutoff is more sensitive to fund flows than the investment of firms above the cutoff. Thus, greater sensitivity to fund flows is not a general characteristic of lower-rated firms. Moreover, BB+ firms constitute only 10% of all speculative-grade firms, so fund flows are likely to be driven by the performance and investment prospects of lower-rated firms, which are very different on observable characteristics from BB+ firms. This suggests that the differential effect of fund flows on BB+ firms relative to BBB- firms may be due to categorization. If investors emphasize differences between investment- and speculative-grade firms and overlook differences within these broad categories, they may treat two firms in the same category similarly, even if the firms have very different performance and investment prospects (Barberis and Shleifer, 2003). Our evidence is consistent with this idea. We find that even after controlling for the investment of matched BBBfirms, the investment of BB+ firms comoves with the investment of lower-rated speculative-grade firms in unrelated industries. Another objection to our identification strategy is that firms could be selecting into different 2

4 ratings based on unobservable characteristics. The distribution of credit ratings reported in Figure 1 suggests that some selection might be taking place, as there are fewer BB+ observations than either BBB- or BB ones. The direction of any bias introduced by selection on unobservable characteristics is ambiguous. While it could be the case that firms with lower sensitivity of investment to high-yield fund flows choose to be investment grade, we think that the more natural selection story goes the opposite way and suggests that our results should be treated as a lower bound on the potential distortions in real behavior due to the investment-grade cutoff. Firms whose investment would be most affected by the volatility of fund flows if they were rated BB+ have the strongest incentives to alter their behavior to achieve a BBB- rating. Thus, firms that do carry a BB+ rating in our data are likely to have a relatively low sensitivity of investment to fund flows. In summary, we find that shocks to the supply of capital of high-yield mutual funds cause the investment of firms just below the investment-grade cutoff to diverge from the investment of similar firms just above the cutoff. Our approach also sheds light on the role of market segmentation in the recent credit boom. By issuing AAA and AA rated securities backed by speculative-grade bonds and loans, collateralized debt obligations (CDOs) allowed speculative-grade borrowers to tap investment-grade sources of financing. 1 We offer some suggestive evidence that CDO flows increased the demand for speculative-grade securities and loans, allowing BB+ firms to increase their investment relative to their BBB- matches. Our work contributes to the literature studying the role of credit ratings in capital markets. 2 Most papers in this literature study the information content of credit ratings and their effects on assets prices and capital structure, while we focus on real investment. Lemmon and Roberts (2007) 1 Benmelech and Dlugosz (2009) study the structures and rating practices of CDOs. In their data, AAA and AA rated tranches constitute 71% and 5% respectively of all CDO issuance. Average credit quality of the underlying collateral, on the other hand, is B+, with only 0.2% rated investment grade. 2 Harold (1938) and Hickman (1958) are among the first to study the use of credit ratings and in particular their ability to forecast defaults. The papers collected in Levich, Majnoni, and Reinhart (eds.) study the various aspects of the role of credit ratings in the global financial markets. More recently, Faulkender and Petersen (2006) find that firms with a credit rating that allows them to access public debt markets have 35% more debt than other similar firms. Kisgen (2006) shows that firm financing decisions are affected by the discrete costs and benefits of different credit ratings. Sufi (2009) shows how the introduction of syndicated bank loan ratings by Moody s and Standard & Poor s in 1995 expanded the set of investors able to invest in syndicated loans and led to increased debt issuance and investment by lower-rated borrowers. Kisgen and Strahan (2009) study the recent certification of ratings from Dominion Bond Rating Service for regulatory purposes. They find that after certification bond yields fell for firms for whom Dominion had a higher rating than the other certified rating agencies. Ellul, Jotikasthira, and Lundblad (2009) find that regulatory constraints can force insurance companies to sell recently downgraded bonds, generating downward price pressure and subsequent reversals. 3

5 are perhaps the most closely related to our work. They investigate the effect on firm investment of the collapse of Drexel Burnham Lambert and of the concurrent regulatory changes. They use a difference-in-differences approach to compare the investment of all speculative-grade firms with unrated firms. We compare the financing and investment behavior of firms on both sides of the investment-grade cutoff, and suggest that the effect of fund flows on the investment of BB+ firms just below the cutoff is a spillover effect due to categorization. Furthermore, while the existing literature focuses on isolated changes in the institutional environment, we study how recurring shocks to the capital of an important investor class interact with market segmentation to affect real investment. In studying the capital supply effects of high-yield fund flows, our work also contributes to the literature on market-driven corporate finance, reviewed recently by Baker (2009). Most of this literature studies capital supply effects in equity markets. 3 For most firms, however, external equity is a relatively minor source of financing compared to retained earnings and debt (Mayer, 1988). It is therefore important to understand the effects of market-driven corporate finance in credit markets. The remainder of the paper is organized as follows. In the next section we review the institutional background that motivates our empirical methodology, which we discuss in more detail in Section 3. Section 4 describes our data and summarizes differences in firm characteristics across credit ratings. Section 5 reports our main results. In Section 6 we discuss some suggestive evidence of the effect of the recent CDO boom on the investment of speculative-grade firms. Section 7 concludes. 2 Institutional Background We begin by briefly describing two institutional features of credit markets and credit ratings that motivate our empirical methodology. First, many regulatory and voluntary conventions restrict the ability of different investor groups to hold speculative-grade securities. Second, rating methodologies introduce noise and inertia in credit ratings. Once we review this institutional background, we go on to describe our empirical methodology. 3 For example, see Baker, Stein, and Wurgler (2003) and Polk and Sapienza (2009). 4

6 2.1 Regulations restrict holdings of speculative-grade securities Many rules and regulations restrict the ability of different investor groups to hold speculative-grade securities. Commercial banks have been prohibited from holding bonds rated BB+ and below since The Financial Institutions Reform, Recovery, and Enforcement Act of 1989 extended the ban on commercial bank holdings of speculative-grade bonds to thrifts. 5 Most state insurance regulations follow the guidelines established by the National Association of Insurance Commissioners, which sets both higher risk charges and a hard cap on holdings of speculative-grade bonds. 6 In addition, the net capital rule for broker-dealers requires larger haircuts for speculative-grade securities (U.S. Securities and Exchange Commission, 2003). 7 Although not subject to regulatory restrictions, most bond mutual funds specialize in either investment- or speculative-grade bonds. Investment-grade funds typically limit their holdings of speculative-grade bonds to 5-10% of fund assets. For instance, PIMCO Total Return Fund, the largest corporate bond fund with $169 billion in assets, limits holdings of high-yield bonds to 10% of fund assets (PIMCO Total Return Fund Prospectus). As of June 2009, PIMCO Total Return Fund invested 5% of its assets in speculative-grade and unrated securities. Vanguard Short-, Intermediate-, and Long-Term Investment-Grade Funds ($46 billion in total assets) are more restrictive and invest exclusively in securities rated BBB- and above (Vanguard Bond Funds Prospectus). Some investment-grade funds, such as Fidelity U.S. Bond Index Fund ($10 billion in assets), do not have explicit limits on the credit quality of their portfolio holdings. Instead, they seek to provide investment results that correspond to the total return of the bonds in the Lehman Brothers U.S. Aggregate Index (Fidelity U.S. Bond Index Fund Prospectus). Since the index consists exclusively of investment-grade securities, any speculative-grade holdings would expose such funds to signifi- 4 West (1973) argues that these regulations increased the spreads for speculative-grade borrowers relative to investment-grade ones. 5 The act prohibited purchases of speculative-grade bonds and mandated existing holdings to be liquidated by As a result, thrifts share of the corporate bond market fell from around 7% in 1988 to 1% by 2008 (Table L212, Corporate and Foreign Bonds, of the Flow of Funds Accounts of the United States). 6 Risk charges for A, BBB, BB, and B rated bonds are 0.4%, 1.3%, 4.6%, and 10% respectively. The portfolio share of all non-investment grade bonds is capped at 20%. As a result of these restrictions, insurance companies share of all speculative-grade bonds is only 8.5%, one-fourth of their 34% share of all investment-grade bonds (Ellul, Jotikasthira, and Lundblad, 2009). 7 Haircuts for investment-grade nonconvertible debt securities paying a fixed interest rate vary between 2% and 9% depending on maturity. Haircuts for speculative-grade bonds are generally 15%. 5

7 cant tracking error and would be thus quite costly. Not surprisingly, as of June 2009, Fidelity U.S. Bond Index Fund invested only 0.5% of its assets in speculative-grade bonds. Conversely, high-yield funds specify a minimum share of their assets to be invested in speculativegrade securities. Vanguard High-Yield Corporate Fund ($11 billion in assets) must invest at least 80% of its assets in corporate bonds that are rated below Baa by Moody s Investor Service, Inc. (Moody s); have an equivalent rating by any other independent bond-rating agency (Vanguard High-Yield Corporate Fund Prospectus). As of June 2009, Vanguard High-Yield Corporate Fund held 91% of its bond portfolio in speculative-grade and unrated bonds. The 80% minimum on holdings of speculative-grade bonds is typical as of June 2009, high-yield mutual funds on average had 87% of their bond portfolios invested in speculative-grade bonds. 2.2 The muddled origins of investment grade Given the large number of restrictions on investing in speculative-grade securities, one may worry that the specific location of the investment-grade cutoff between BBB- and BB+ is endogenous. In particular, differences in firm characteristics and investment opportunities may be especially stark across this ratings transition. We examine differences in observable firm characteristics below, but the origins of the distinction between investment and speculative grades may also mitigate some of these concerns. 8 When Moody s published the first credit ratings in 1909, it did not use the terms investment grade and speculative grade. Instead the agency used the term grade to refer to three groups of credit ratings: AAA, AA, and A bonds constituted the first-grade, BBB and BB bonds the second-grade, B and lower rated bonds low grade. Thus in contrast to the modern distinction between BBB and BB bonds, Moody s originally thought of them as being of similar quality. 9 It was not until 1931 that the modern distinction between speculative- and investment-grade bonds began to emerge. On September 11, 1931, the Comptroller of the Currency ruled that com- 8 See Fons (2004) and Harold (1938) for more on the history of credit ratings and the distinction between investment and speculative grades. 9 As Poor s Publishing, Standard Statistics, and Fitch Publishing companies entered the credit ratings market in 1916, 1922, and 1924, they generally followed similar characterizations. Appendix Table A1 reports the rating symbols used by the principal rating agencies in the early 1930s. All agencies described BB as either Good or Fair, and none referred to it as speculative (Harold, 1938). 6

8 mercial banks could carry bonds rated BBB or higher at cost, but that they had to mark to market lower rated and defaulted bonds. 10 On February 15, 1936, the Comptroller and the Federal Reserve went further and completely prohibited commercial banks from purchasing investment securities in which the investment characteristics are distinctly or predominantly speculative (Harold, 1938). The ruling caused significant confusion regarding the precise definition of speculative securities. American Banker concluded that the regulation limits investments practically to those with an A rating (Harold, 1938). Faced with an outcry from the banking community, the Comptroller backtracked, and by 1938 Moody s persuaded the regulators that bonds rated BBB are not distinctly or predominantly speculative. Subsequent regulations continued to use the same BBB vs. BB cutoff, but its origins suggests that the cutoff could have just as easily been drawn at A vs. BBB or BB vs. B. This history suggests that the investment-grade cutoff was not originally drawn to distinguish between firms with sharply different fundamentals. Over time, however, market institutions may have evolved around the cutoff to make its location more correlated with firm characteristics and investment opportunities. Below we provide evidence that this is not the case, showing that differences in observable firm characteristics at the investment-grade cutoff are similar to differences across other rating transitions. 2.3 Noise and inertia in credit ratings Credit ratings do carry information about firms credit quality and potentially about their investment opportunities. 11 However, if ratings are subject to noise and inertia, we will be able to find pairs of firms that are similar on observable characteristics but are on different sides of the cutoff. There is a number of reasons to believe that credit ratings are noisy, lagging measures of credit quality. First, ratings methodologies explicitly aim to provide stability. The agencies trade off 10 Although the ruling applied only to national banks, many state banking departments followed the Comptroller s lead and introduced similar restrictions for state chartered banks. 11 There is still considerable debate, however, as to whether credit ratings contain any information not already available to investors. Market-based measures of credit quality tend to be better predictors of default than credit ratings, at least at short- and medium-term horizons (Moody s, 2006). Kliger and Sarig (2000) and Tang (2009) argue that Moody s refinement of its rating system revealed new information about rated firms. Jorion, Liu, and Shi (2005) find greater informational effects of credit rating changes after Regulation Fair Disclosure prohibited companies from selectively disclosing nonpublic information, but excluded rating analysts from the new regulations. 7

9 rating accuracy versus stability and are reluctant to upgrade or downgrade firms if such changes might have to be reversed in the future (Moody s, 2006). This is particularly true at the investmentgrade cutoff, as the agencies are aware that their decisions affect the ability of market participants to hold certain bonds. Moreover, even when the agencies do adjust credit ratings, the adjustment is likely to be only partial, followed by additional changes (Altman and Kao, 1992). As a result, market-based measures such as yield spreads are more accurate than credit ratings in forecasting defaults at short- and medium-term horizons (Moody s, 2006). 12 In addition to the inertia in ratings generated by the explicit goal of stability, credit rating agencies organizational structures may create incentives for analysts to be conservative in upgrading or downgrading firms. One such organizational practice, exemplified by Moody s Leverage Finance Group, is having separate groups analyzing investment- and speculative-grade credits. This organizational structure could create conflicts of interest as the group covering a particular firm would lose fee revenue if it downgraded or upgraded the client across the investment-grade cutoff. Rating outlooks introduce a further wedge between the information and regulatory content of credit ratings. These outlooks assess the potential for an issuer rating change but are not necessarily a precursor for a rating change (Standard & Poor s, 2008). On average, issuers with a positive outlook default at the same rate as issuers rated one notch higher (Moody s, 2005). 13 Thus, although the information content of a BB+ rating with a positive outlook is the same as, if not better than, that of an unconditional BBB- rating, their regulatory implications are quite different until they are actually upgraded, BB+ firms with positive outlooks cannot access the investment-grade market. 12 Appendix Table A3 provides some suggestive evidence of credit ratings inertia in our data. Panel A compares the characteristics of BB+ firms in the two years before they are upgraded with the characteristics of all BBB- firms. Other than being smaller, BB+ firms that are subsequently upgraded perform much better than an average BBBfirm. They have higher valuations, lower market leverage, higher profitability and cash holdings, and invest more than BBB- firms. Conversely, Panel B shows that BBB- firms that are downgraded in the next two years perform significantly worse than the average BB+ firm. They have significantly higher leverage, lower values of Altman s z-score and interest coverage, much worse profitability and they invest less than BB+ firms. 13 In fact, over the period, BB+ firms with a positive outlook had a 5-year default rate of only 0.95%, significantly lower than the 3.88% default rate of BBB- firms with stable outlook (Moody s, 2005). 8

10 3 Empirical Methodology The previous section suggests two stylized facts that drive our empirical approach. First, regulatory and voluntary conventions restrict the ability of many investor groups to hold speculative-grade securities. Thus, if there are limits to arbitrage, arbitrageurs might be unable to fully offset liquidity or noise shocks suffered by traditional speculative-grade investor groups. In particular, since highyield mutual funds hold about 20% of speculative-grade bonds, flows into high-yield mutual funds can have a significant effect on the investment of speculative-grade firms. Second, noise and inertia in credit ratings imply that there are BB+ firms that are similar to BBB- in terms of observable firm characteristics and investment opportunities. That is, there are firms rated BB+ that should be BBB- and vice versa. Our empirical strategy is to match BB+ and BBB- firms based on industry and firm characteristics such as size, leverage, Altman s z-score, and Q, and to compare the investment sensitivities of the matched firms to high-yield fund flows. Our benchmark matching procedure uses industry and size. 14 Within each industry-quarter we match each firm rated BB+ with a firm rated BBB- that is of similar size. We consider firms to be of similar size if the ratio of their book assets is within [0.5, 2]. Starting with the Fama-French 48 industries, we match firms using 38, 30, and 17 industry definitions until we either have a successful match or are unable to match a given BB+ firm. In addition to our benchmark matched sample, we examine two subsets of firms closer to the investment-grade cutoff. The first subset consists of less levered BB+ firms. Excluding the top 25% of BB+ firms by market leverage results in a matched sample of BB+ and BBB- firms with virtually no differences in observable characteristics. The second subset consists of BB+ firms with positive outlooks. These firms have observable characteristics and default rates similar to BBBfirms, but their cost of capital is still subject to shocks associated with high-yield fund flows. Our approach is similar in spirit to the pseudo-experimental approaches like regression discontinuity and differences-in-differences. Consider a regression of the investment of firms that have a true, unobservable rating R on flows into high-yield mutual funds and standard investment regression 14 We obtain similar results when matching on leverage, z-score, or Q in addition to size. 9

11 controls: Inv R i,t = α R + β Q Q R i,t 1 + β CF CF R i,t + β R Flows Flows t 1 + InvOpportunities R t + ε R i,t where common shocks to investment opportunities InvOpportunities R t are unobservable and where we assume the coefficients on Q and cash flow to be the same across ratings. 15 Individual firms are too small for aggregate fund flows to be correlated with the firm-specific shocks ε R i,t. Fund flows are potentially correlated with the common shocks InvOpportunities R t, which would bias the coefficient βflows R upward. However, if our matching procedure is effective in identifying BB+ and BBB- firms that are subject to the same common investment opportunities shocks (BB+ firms that should be BBB- and vice versa), we can difference the two equations to obtain Inv BB+ i,t Inv BBB j,t +β Q (Q BB+ i,t 1 Q BBB j,t 1 )+β CF (CFi,t BB+ =(α BB+ α BBB )+(β BB+ Flows βbbb Flows ) Flows t 1 CFj,t BBB )+(ε BB+ i,t ε BBB j,t ) or more compactly Inv i,t = α + β Flows Flows t 1 + β Q Q i,t 1 + β CF CF i,t + η i,t where X = X BB+ X BBB is the difference in firm characteristic X between matched BB+ and BBB- firms. By differencing the investment of matched firms, we thus remove any correlation between fund flows and investment opportunities. Finding a positive and statistically significant coefficient β Flows is then evidence of a capital supply effect of fund flows on the investment of BB+ firms In untabulated results we show that the assumption that β BB+ Q = β BBB Q and β BB+ CF = β BBB CF cannot be rejected in our data. 16 Note that the reasons we expect high-yield fund flows to affect the investment of speculative-grade firms do not extend to investment-grade fund flows. Investment-grade funds account for a much smaller share of the investmentgrade corporate bond market than the share of high-yield mutual funds in the speculative-grade bond market. Furthermore, investment-grade firms have access to more alternative sources of financing. Therefore we would not expect investment-grade flows to have any effect on the difference in the investment rates of BB+ and BBB- firms, and including investment-grade flows would only make the statistical inference more difficult. Untabulated results indicate that when investment- and speculative-grade fund flows are in fact included together, the coefficient on investment-grade fund flows is small and not statistically significant. The coefficient on speculative-grade fund flows retains its magnitude; its statistical signficance is somewhat lower due to the correlation between investment- and 10

12 Our matching procedure selects firms that have similar observable characteristics but are on different sides of the investment-grade cutoff, and documents the differential sensitivity of the investment of BB+ firms to fund flows. An alternative econometric approach to implement our empirical strategy is propensity score weighting, which is a panel regression that uses the full sample of all BB+ and BBB- firms but gives more weight to firms close to the cutoff (Imbens and Wooldridge, 2009; Hirano, Imbens, and Ridder, 2003). This approach delivers similar results, which are reported in Appendix Table A8. Both matching and propensity score weighting assume that BB+ and BBB- firms are subject to the same investment opportunities shocks. One might still be concerned that investment opportunities are not quite the same and that flows are picking up the differential investment opportunities of less creditworthy firms. We address this concern in a number of ways. First, drawing on the regression discontinuity design literature, we include the interactions of firm characteristics with fund flows to show that it is the discontinuous change in credit rating and not changes in continuous measures of credit quality that determines the sensitivity of firm investment to fund flows. Controlling for up to five powers of firms characteristics interacted with flows, there is still a discontinuous change in the sensitivity of investment to fund flows as firms are rated BBB- versus BB+. Second, the basic version of the differential investment opportunities story predicts that investment of lower-rated firms around any rating cutoff is more sensitive to fund flows than investment of higher-rated firms. For example, it predicts that the investment of B+ firms is more sensitive to fund flows than the investment of BB- firms. This is not what we find in falsification tests comparing the sensitivity of investment to fund flows around other rating cutoffs the investment-grade cutoff is the only one where the investment of firms below the cutoff is more sensitive to fund flows than the investment of firms above the cutoff. Therefore, one requires a more complicated story according to which the cutoff endogenously arose at the place of dramatic change in the nature of investment opportunities. Comparing the characteristics of matched firms suggests that this is not the case, as do the historical origins of the investment-grade cutoff. speculative-grade fund flows. 11

13 4 Data In this section we describe our sample construction and variable definitions and address three datarelated issues before turning to our results. First, we discuss which of a firm s potentially numerous credit ratings determines whether it can access the investment-grade market. Second, we examine differences in firm characteristics across credit ratings to show that there is no abrupt change in firm characteristics around the investment-grade cutoff. Third, we explain how we measure fund flows, and show that flows are large relative to the capital and investment of speculative-grade firms. 4.1 Sample construction Our sample is the quarterly CRSP/Compustat merged data set covering the 1986Q1-2007Q4 period. The sample period is determined by the availability of S&P domestic long-term issuer credit ratings in Compustat starting in December Some of our specifications use rating outlooks and bank loan ratings from the S&P RatingsDirect and Ratings IQuery databases. We measure investment as CAPX i,t PPE i,t 1, the ratio of capital expenditures in quarter t to net property, plant, and equipment at the end of quarter t Our regressions include standard controls: cash flow normalized by lagged capital, items plus depreciation. 18 CF i,t PPE i,t 1, and Q i,t 1. Cash flow is income before extraordinary Q is the market value of equity from CRSP plus assets minus the book value of stockholder equity, all divided by assets. We also control for size, leverage, Altman s z- score, and interest coverage. Market leverage is book debt divided by the sum of book debt and market value of equity from CRSP. Altman s z-score is 1.2 WC i,t /Assets i,t +1.4 RE i,t /Assets i,t EBIT i,t /Assets i,t + Sales i,t /Assets i,t, where WC i,t is working capital, i.e., current assets minus current liabilities, and RE i,t is retained earnings (Altman, 1968). We use the modified version of Altman s z-score, which excludes leverage, because we control for leverage directly. Interest coverage is the ratio of EBIT to interest expense, calculated using four-quarter moving averages of EBIT 17 Because our proposed mechanism links fund flows and investment through debt issuance, changes in assets could offset the effect of flows on investment if we scaled capital expenditures by assets instead of PPE. Nevertheless, we obtain similar but weaker results when we do so. Appendix Table A4 reports the results of our investment regressions when investment is alternatively scaled by assets, assets net of cash, or PPE. We also try scaling by assets net of cash to mitigate the mechanical effect of debt issuance on the measured rate of investment. The results confirm our intuition the effect of fund flows on investment is strongest when investment is scaled by PPE, weakest when investment is scaled by assets, and in between when investment is scaled by assets net of cash. 18 To make our results more comparable with papers using annual data, we annualize investment and cash flow. 12

14 and interest expense. To reduce the effect of outliers we winsorize all variables at the first and ninety-ninth percentiles. 4.2 Measuring access to the investment-grade market While regulations distinguish between investment and speculative grades at the security level, investment activity occurs at the firm level. We therefore need a firm-level measure of access to the investment-grade market. The senior secured credit rating is typically the highest rating a firm can achieve on an individual security and is therefore the right measure of access to the investmentgrade market. A firm with a BB+ senior secured rating has no way to access the investment-grade market during periods of low or negative flows into high-yield mutual funds. 19 In comparison, a firm with a BB+ senior unsecured rating that has unencumbered collateral may still be able to access investment-grade market by issuing senior secured debt. We use S&P long-term issuer credit rating, which is a current opinion of an issuer s overall creditworthiness, apart from its ability to repay individual obligations and corresponds closely to the senior secured rating (Standard & Poor s, 2008). S&P may notch up rate individual issues above the issuer credit rating when it can confidently project recovery prospects exceeding 70% (Standard & Poor s, 2008). Since few firms are in position to issue senior secured bonds with recovery prospects exceeding 70%, S&P long-term issuer credit rating is a good measure of firm s ability to access the investment-grade market. 20 And to the extent that some firms with BB+ senior secured rating are able to issue higher rated securities, we will be less likely to find any effect of fund flows on the investment of speculative-grade firms. 4.3 No break in firm characteristics at the investment-grade cutoff Our identification strategy and falsification tests require that differences in firm characteristics at the investment-grade cutoff be similar to differences across other rating cutoffs. Table 1 reports the 19 The two primary exceptions are obtaining a guarantee from another entity and issuing asset-backed securities. 20 During the period, average recovery rates for senior secured bonds were 52.3% when measured by post-default trading prices and 63.6% when measured by ultimate recoveries (Moody s, 2009). In untabulated results we estimate that less than 15% of all non-convertible bond issues by non-financial firms are notched up. When weighted by total proceeds, a bit more than 10% are notched up. The vast majority of noteched up issues are assetand mortgage-backed bonds. 13

15 means of firm characteristics by credit rating. 21 As there are few AAA and AA+ firms, we combine these firms into one category. We do the same for firms rated CCC+ through CCC-. Lower-rated firms are generally smaller and more levered. In addition, they have lower values Altman s z-score and interest coverage than higher-rated firms. The ratio of net property, plant, and equipment to assets is relatively constant across credit ratings. Q varies from 2.5 for the most highly rated firms to 1.4 for CCC rated firms. Higher-rated firms are more profitable than lower rated firms, whether one looks at operating margins, ROA, or cash flow. Despite significant differences in Q across credit ratings, firms appear to engage in similar levels of capital expenditures. Importantly, the investment-grade cutoff does not stand out compared to other rating cutoffs. BB+ firms are on average 27.8% smaller than BBB- firms, but there are similar differences in size around other lower rated cutoffs, and our empirical methodology matches on size to produce a sample of comparably sized firms. The market leverage of BB+ firms is 11.6% larger than the market leverage of BBB- firms, but there are only three other cutoffs with smaller percentage differences in market leverage. BB+ firms have higher operating margins but lower ROA and cash flow than BBB- firms. The investment rate of both BB+ and BBB- firms is around 22%. We also examine normalized differences in firm characteristics X = X 1 X 0, where S 2 S 2 0 +S1 2 g is the sample variance of firm characteristic X in group g. Imbens and Wooldridge (2009) advocate using the normalized difference as a scale-free measure of the difference in distributions. They suggest that researchers should check that all normalized differences are below 0.25 because linear regression methods can be sensitive to the specification otherwise. At the investment-grade cutoff, all normalized differences are comfortably below the 0.25 threshold Flows are large relative to the investment of BB rated firms The time series of aggregate flows into high-yield corporate bond mutual funds are from the Investment Company Institute, the national association of U.S. investment companies. At the end of 2008, the Investment Company Institute collected information on assets and flows from 8,889 mutual funds with $9,601 billion in assets under management. 21 Appendix Table A2 reports differences in firm characteristics across adjacent credit ratings. 22 Panel B of Appendix Table A2 reports normalized differences across adjacent credit ratings. 14

16 The appropriate measure of flows should capture their magnitude relative to the capital of firms close to the investment-grade cutoff, and also account for the time lag between fund flows and bond issuance on one hand and issuance and investment on the other hand. To accomplish these goals, we calculate cumulative flows over the four quarters [t 4,t 1] and scale flows by the total PPE of firms rated BBB+ through BB-, PPE t 1. Our results are robust to calculating flows over other windows and using alternative scalings, in particular scaling flows by total net assets (TNA) of high-yield mutual funds. Figure 2 shows the time-series of high-yield fund flows relative to PPE and capital expenditures of firms rated BBB+ through BB-. Flows vary significantly over time and are large relative to the investment of these firms. In our regressions, we standardize flows so that the coefficients can be interpreted as the effect on investment of a one standard deviation increase in scaled flows. 5 Results 5.1 Characteristics of matched BB+ and BBB- firms Table 3 reports the characteristics of matched BB+ and BBB- firms. We match 2,818 out of 3,519 firm-quarter observations rated BB+ to 1,992 unique firm-quarter observations rated BBB-. We report both the difference in means and the normalized difference. Our matching procedure successfully picks BB+ and BBB- firms that have similar size. Although in the full sample, BB+ firms are on average 27.8% smaller than BBB- firms, in the matched sample, BB+ firms are actually slightly larger than matched BBB- firms. The difference in size, however, is not statistically significant. BB+ firms are still more levered than matched BBB- firms. The differences in book and market leverage are around BB+ firms also have lower Altman s z-score and interest coverage than matched BBB- firms. However, the normalized differences for these measures of leverage are all below the 0.25 cutoff suggested by Imbens and Wooldridge (2009). With the exception of ROA, none of the other differences are statistically or economically significant. Overall, our matching procedure selects a sample of BB+ and BBB- firms that are similar along most dimensions. In our 15

17 regressions, we are careful to control for the remaining differences in observable characteristics such as leverage. In addition to the full sample of BB+ firms, we consider two subsamples of firms that are closer to the investment-grade cutoff. The first subsample excludes the top 25% of BB+ firms by market leverage. The second subsample consists of BB+ firms with positive outlooks. 23 The differences in leverage are now greatly reduced. In the subsample that excludes the top 25% of BB+ firms by market leverage, the only remaining statistically significant difference is in terms of the z-score. 24 And in the subsample that consists of BB+ firms with positive outlooks, the only statistically significant difference is in terms of the ROA, with speculative-grade firms being in fact more profitable than investment-grade ones. BB+ firms with positive outlooks also have higher values of Q and cash flow than matched BBB- firms. Overall, the two subsamples of BB+ firms select firms that are even closer to the investment-grade cutoff than the ones in the full sample and result in an even better match. 5.2 Flows increase the investment of BB+ firms relative to BBB- firms Table 4 reports the results of our baseline regressions. We regress the difference in the investment rates of matched BB+ and BBB- firms on high-yield mutual fund flows and differences in firm characteristics CF i,t CAP X i,t = α + β Flows Flows t 1 + β Q Q i,t 1 + β CF + ε i,t PPE i,t 1 PPE i,t 1 where X = X BB+ X BBB is the difference in firm characteristic X between matched BB+ and BBB- firms. We use Thompson (2006) to adjust the standard errors for clustering by both firm and quarter. Examining the results in column 1, the coefficient on flows is positive and statistically significant. One standard deviation increase in flows increases the investment of BB+ firms relative to the 23 In the subsample that excludes the top 25% of BB+ firms by market leverage, we match 2,130 BB+ observations to 1,603 unique BBB- firm-quarter observations. In the subsample of BB+ firms with positive outlook, we match 253 BB+ observations to 247 unique BBB- first-quarter observations. 24 Of the four components of z-score, the only statistically significant difference is in retained earnings. There are no statistically significance differences in profitability, sales, or working capital. 16

18 investment of matched BBB- firms by 0.014, or about 7% of the mean investment rate. The constant term is close to zero and not statistically significant, indicating that on average matched firms have similar investment rates. Controlling for size, market leverage, z-score, and interest coverage in column 2 does not affect the magnitude or the statistical significance of the coefficient on fund flows. The number of observations drops by almost a quarter, primarily due to missing values of z-score. Because neither z-score nor interest coverage have significant explanatory power, we omit them from the regressions in column 3. In columns 4-6, we estimate the same regressions as in columns 1-3 but using the subsample of less levered BB+ firms. Although statistical significance is somewhat weaker due to smaller sample sizes, the economic magnitudes remain the same. In columns 7-9, we use the sample of BB+ firms with positive outlooks. The sample size is about one tenth of the full sample, and the coefficient on flows is no longer statistically significant. The point estimates, however, are remarkably similar. Overall, our results in Table 4 indicate a statistically significant and economically meaningful effect of fund flows on the investment of BB+ firms relative to similar firms rated BBB Results are robust to alternative matching procedures Table 5 shows that our results are robust to alternative matching procedures. Our benchmark matching procedure successively matches BB+ and BBB- firms within Fama-French 48, 38, 30, and 17 industries until we either have a successful match or are unable to match a given BB+ firm. Columns 2-4 show that we get even stronger results when we stop the match process at 30, 38, or 48 industries. In particular when, in column 4, we match inside 48 industries only, the coefficient on fund flows is 0.018, more than a quarter larger than the coefficient of in our benchmark specification. In columns 5 and 6, we continue matching inside the 48 industries, but adjust the size constraints. Specifically, in column 5 we relax the size constraints to require the ratio of book assets to be within [1/3, 3]. The coefficient on fund flows is 0.016, somewhat smaller than in column 4, but larger than in the benchmark specification of column 1. In column 6 we tighten the size constraints to require 17

19 the ratio of BB+ to matched BBB- firm assets to be within [0.5, 1.5]. The effect of fund flows on investment is stronger for this sample. In columns 7-9, we match firms on assets and one of three other firm characteristics: leverage, z-score, and Q. 25 When matching on assets and leverage or assets and z-score, the coefficient on fund flows is very similar to the benchmark specification. Matching on assets and Q results in a coefficient on fund flows of that is significant at better than 10%. 5.4 Results are robust to alternative calculations of flows Our results are also robust to alternative constructions of fund flows. In Panel A of Table 6, we calculate fund flows over different windows. In the first four columns, we calculate flows over the four quarters [t 3 lag, t lag]. In the last four columns, we calculate flows over the two quarters [t 1 lag, t lag]. Our results are robust to measuring flows over two quarters. Consistent with there being a lag between flows and bond issuance and a similar lag between issuance and investment, results are actually stronger when using longer lags. Nor do the results depend on how we scale fund flows. Panel B of Table 6 shows that we get similar results when scaling flows by a) total PPE of all speculative-grade firms, b) total assets of firms rated BBB+ through BB-, or c) total net assets of high-yield mutual funds. 5.5 Rating, not size or leverage, determines the sensitivity to flows Our results hold for a subsample of less levered BB+ firms for which there are virtually no differences in observable characteristics between BB+ and BBB- firms. To further dispel any concern that our results might be driven by the investment of smaller and more levered firms, we control for interactions of firm characteristics with fund flows. Table 7 presents the results. Column 1 is identical to column 2 in Table 4, controlling for Q, cash flow, size, leverage, z-score and interest coverage, but not their interactions with fund flows. As we interact firm characteristics with fund flows in column 2, the direct effect of flows is not affected. For compactness, we do not report the 25 We use Fama-French 48, 38, 30, and 17 industries to look for the closest match according to the normalized Euclidean distance. We require the ratio of assets to be within [0.5, 2] and the absolute differences in leverage, z-score, and Q to be less than 0.2, 0.6, and 0.9. These values correspond to roughly one standard deviation of each one of these three variables. 18

20 coefficients on the interactions of firm characteristics with fund flows. Only the interaction of size with flows is consistently negative and statistically significant. Later on we examine more closely whether the investment of small speculative-grade firms is more sensitive to fund flows. In columns 3-6 we control for up to five powers of firm characteristics and their interactions with fund flows. The direct effect of flows is not affected and actually gets stronger. In particular, controlling for five powers of firm characteristics and their interactions with flows increases the coefficient on flows from in column 1 to in column There is no differential sensitivity to flows around other cutoffs Our results so far indicate that the investment of BB+ firms is more sensitive to flows into highyield mutual funds than the investment of matched BBB- firms. Our identifying assumption is that firms close to the investment-grade cutoff are subject to similar investment opportunities shocks. If this assumption holds, the differential sensitivity of investment to fund flows is evidence of the real effects of capital supply shocks in the presence of market segmentation. Our empirical approach of matching on industry and size and controlling for other firm characteristics, and their interactions with fund flows, is designed to ensure that the identifying assumption holds so that our interpretation is valid. There may still be concerns, however, that investment opportunities are not quite the same, and that fund flows are driven by the differential investment opportunities of lower-rated firms. If this were the case, we would expect investment of lower-rated firms around any rating cutoff to be more sensitive to fund flows than investment of higher-rated firms. To test this hypothesis, we conduct falsification tests using matched firm pairs around other rating cutoffs. For each credit rating cutoff from A through B, we match firms just below the cutoff with firms just above the cutoff that are of similar size and in the same industry. 26 For example, we match firms rated A with firms rated A+. As there are few firms rated above A or below B, we do not report the results for cutoffs above A and below B. 27 To focus on firms closer to each cutoff and get a better match on both size and 26 As before, we successively match firms within Fama-French 48, 38, 30, and 17 industries until we either have a successful match or are unable to match a given BB+ firm. We consider firms to be of similar size if the ratio of their book assets is within [0.5, 2]. 27 We find similar results, i.e., that investment of lower rated firms is not more sensitive to fund flows than 19