Clientele Effects Explain the Decline in Corporate Bond Maturities

Transcription

1 Clientele Effects Explain the Decline in Corporate Bond Maturities Alexander W. Butler, Xiang Gao, and Cihan Uzmanoglu * January 14, 2019 Abstract The average maturity of newly issued corporate bonds has declined substantially over the past 40 years, and traditional determinants of debt maturity fail to explain the trend. We show that the changing composition of the investors in the corporate bond market resolves this puzzle. The results of a Granger causality test, an instrumental variable approach, a natural experiment, and a regulatory study suggest that a decline in insurance company ownership in bonds leads to shorter bond maturities. These findings illustrate how developments in financial institutions can have real effects on corporations. Keywords: Debt Maturity; Supply of Credit; Demand for Bonds; Insurance Company Ownership; Clientele Effects. JEL Code: G20; G22; G23; G30; G32. * Alexander W. Butler, Rice University, alex.butler@rice.edu; Xiang Gao, Binghamton University, xgao22@binghamton.edu; Cihan Uzmanoglu, Binghamton University, cuzmanog@binghamton.edu. We thank Lee Ann Butler, Murali Jagannathan, Rajesh Narayanan, Shweta Srinivasan, Shikha Jaiswal (FMA discussant), and the seminar participants at 2018 FMA Annual Meeting, Binghamton University, California State University-East Bay, and Rochester Institute of Technology for helpful comments.

2 Clientele Effects Explain the Decline in Corporate Bond Maturities Abstract The average maturity of newly issued corporate bonds has declined substantially over the past 40 years, and traditional determinants of debt maturity fail to explain the trend. We show that the changing composition of the investors in the corporate bond market resolves this puzzle. The results of a Granger causality test, an instrumental variable approach, a natural experiment, and a regulatory study suggest that a decline in insurance company ownership in bonds leads to shorter bond maturities. These findings illustrate how developments in financial institutions can have real effects on corporations.

3 1. Introduction From 1975 to 2015, the average maturity of new corporate bond issues in the U.S. declined from 20 years to 10 years. Surprisingly, the macroeconomic, issuer, or issue level determinants of bond maturity fail to explain this decline (Custódio, Ferreira, and Laureano, 2013). In this paper, we test the hypothesis that changes in the aggregate investor base in the corporate bond market explain the maturity puzzle. Insurance companies are the major investors in the U.S. corporate bond market. This bond capital from insurance companies is an important source of long-term financing for corporations because insurance companies, particularly life insurance companies which supply 85% of this bond capital, invest heavily in long-term bonds to offset their exposure to interest rate risk on their long-term liabilities. Long-term bonds also help insurance companies secure a predictable income and achieve their long-term investment goals. Although insurance companies have the largest ownership share in the U.S. corporate bond market, their ownership share has declined from 40% to 25% over the last four decades. 1 While insurance companies collective market share decreased, mutual funds experienced the most growth because tax and retirement policies led to a rise in their money inflows (Rydqvist, Spizman, and Strebulaev, 2014). Mutual funds ownership share in the corporate bond market was less than 2% in 1970s, but today, they are the second largest bond investors following insurance companies with an ownership share of 15%. Compared to insurance companies, however, mutual funds invest in shorter-maturity bonds to mitigate their redemption risk. Therefore, this decline in the relative participation of insurance companies as investors in the corporate bond market may reduce the demand for long-term bonds. 1 According to the U.S. flow of funds data published by Federal Reserve. 2

4 We construct a sample of over 19,000 new bonds issued by public U.S. industrial firms between 1975 and 2015 to test our hypothesis that the decline in the aggregate ownership share of insurance companies in the corporate bond market explains the aggregate decline in bond maturity. We begin by verifying the microfoundations of our argument (a) that insurance companies ownership relates to bond maturity and (b) that demand from insurance companies can cause firms to choose bond maturity accordingly. Indeed, we find that the percentage of insurance company ownership in a bond is positively related to the bond s maturity, controlling for issuer-level (e.g., firm fixed effects, size, stock returns, leverage, profitability, tangibility), issue-level (e.g., characteristics fixed effects, issue amount, credit ratings), and macro-economic (e.g., year fixed effects, share of long-term government debt, term and default spreads) determinants of bond maturity. This relation between insurance company ownership and bond maturity is economically meaningful: on average, a 10-percentage point decrease in insurance company ownership is associated with a 7%, or 9-month, decrease in bond maturity. The association we observe between insurance company ownership and bond maturity is jointly determined based on the preferences of both insurance companies and bond issuers. But do changes in insurance company demand for a bond have a causal effect on its maturity at issue? To test this causal channel, we exploit a regulatory restriction faced by insurance companies, utilize a natural experiment, and use an instrumental variable approach. The results provide consistent evidence that the direction of causality can run from the insurance company demand for bonds to the maturity of new bond issues. To establish the causal link between bond ownership and maturity, we first explore a regulatory restriction that increases insurance company ownership in certain types of bonds to investigate the direction of causality. The risk-based capital requirements for insurance 3

5 companies place bonds with AAA through A- ratings in the same risk category and assign them the same risk factor. Insurance companies that reach for yield will prefer to invest in higher yield and hence, higher risk bonds within the AAA through A- rating spectrum (Becker and Ivashina, 2015). We find that insurance company ownership is indeed abnormally high in A rated (A+ through A-) bonds. We next examine whether these A rated bonds also have abnormally long maturities and find that they do. This result is consistent with the direction of causality running from ownership to maturity when a regulatory restriction increases insurance company demand for bonds. Next, we use natural disasters as an exogenous shock to the demand for bonds from insurance companies. Massa and Zhang (2011) show that insurance companies exposed to losses due to Hurricane Katrina liquidated their positions in corporate bonds, and that the effects of this liquidation lasted for several months. Costly natural disasters create shocks to the demand for bonds from insurance companies, but not from other investors. We examine the impact of disaster-induced reductions in demand from insurance companies particularly, property and casualty (P&C) insurers on the maturity of newly issued bonds. We find that insurance company ownership in new bond issues decreases significantly following natural disasters, validating the demand shock argument. We then test whether these bonds with lower insurance company ownership due to disaster losses also have shorter maturities. We find that they do, consistent with the direction of causality running from ownership to maturity when disasterinduced losses decrease insurance company demand for bonds. An instrumental variable approach provides additional evidence on the direction of causality. One plausibly valid instrument for insurance company ownership is based on the physical distance between bond issuers and insurance companies. Numerous studies show that 4

6 investors overweight local securities (e.g., Coval and Moskowitz, 1999; Huberman, 2001; Ivković and Weisbenner, 2005). We expect our distance-based instrument to be correlated negatively with the insurance company holdings in the issuers bonds: more distance between the investor and the issuer, less ownership. Below, we elaborate on the validity of the instrument and how we construct it. Using this instrumental variable approach, we find that a 10-percentage point decrease in (instrumented) insurance company ownership results in a 35%, or 3.5-year, decrease in bond maturity. This economic magnitude, which is similar to the economic magnitudes of the results from our regulation test and natural experiment, implies that the substantial decline in aggregate ownership share of insurance companies in the bond market can explain the 50% decrease in bond maturities between 1975 and This result is consistent with the direction of causality running from ownership to maturity via a home bias channel affecting insurance companies demand for bonds. Overall, our findings thus far provide evidence that causality runs from ownership to maturity. We now turn to our main tests and findings: examining the aggregate trend in bond maturity. We first reproduce the aggregate maturity trend tests in Custódio, Ferreira, and Laureano (2013). Like them, we find that after controlling for its determinants, bond maturity declines by a statistically significant and economically large amount: an average of 1.5 months each year, totaling 5 years of maturity shortening during our sample period. We then include the aggregate ownership share of insurance companies in the corporate bond market (or, the insurer market share ) as an additional regressor. We find not only that insurer market share is significantly positively related to bond maturity, but also and more importantly, the declining trend disappears after controlling for it in bond maturity regressions. This finding, which is robust to using firm fixed effects, controlling for other trending variables, and studying 5

7 alternative periods, suggests that the maturity puzzle can be fully explained by the declining ownership share of insurance companies in the corporate bond market. Next, we use a Granger causality test to examine the direction of (Granger) causality between insurer market share and bond maturity in the aggregate. Although above we make a case that causality runs from ownership to maturity, causality could also run from maturity to ownership. A Granger causality test provides some evidence as to whether causality appears to run from aggregate insurance company ownership share to average maturities, from average maturities to aggregate ownership share, or both. We find that the lagged insurer market share is a positive and significant determinant of the average bond maturity in the subsequent quarter, but the lagged average bond maturity is insignificant in predicting insurer market share. These findings suggest that insurer market share Granger-causes bond maturity. We then examine cross-sectional variation in maturity trends across subsamples of categories of bonds. If the decline in insurer market share is driving the decline in bond maturities, then this effect should be more pronounced for bonds that insurance companies prefer due to institutional or regulatory restrictions. We study the determinants of insurance company ownership in bonds and provide evidence that insurance company demand is higher for investment grade bonds without conversion, put, or floating coupon features. We then find that the declining trend in maturities is significant only among these types of bonds that face high demand from insurance companies. Moreover, we find that the declining trend in bond maturities is more closely associated with the ownership share of life insurance companies, which tend to have longer-term liabilities (and accordingly invest more heavily in longer-term bonds) than do P&C insurance companies. These findings provide additional evidence that the decline in corporate bond maturities is associated with the decline in insurer market share. 6

8 Overall, this paper shows that the declining presence of insurance companies in the corporate bond market leads to shorter bond maturities. It contributes to the literature by explaining the maturity puzzle documented by Custódio, Ferreira, and Laureano (2013). In addition, it illustrates that the institutional environment can influence the maturity of corporate debt, and hence complements the earlier findings that the developments in financial and legal institutions can affect the capital structure of companies (e.g., Rajan and Zingales, 1998; Demirguc-Kunt and Maksimovic, 1999; Giannetti, 2003; Fan, Titman, and Twite, 2012). Moreover, the literature documents that the supply and uncertainty of credit can have real consequences for corporations (e.g., Ellul, Jotikasthira, and Lundblad, 2011; Massa, Yasuda, and Zhang, 2013). Our paper adds to this stream of literature by providing evidence that the type of institutions supplying credit is an important determinant of corporate debt maturity. The rest of the paper is organized as follows. Section 2 describes the sample and variable definitions. Section 3 explains the empirical setting and presents the results. Finally, Section 4 concludes with a summary of findings. 2. Data, Sample Selection, and Variables We construct our sample by combining all U.S. dollar denominated corporate bonds issued by U.S. industrial firms from Mergent FISD and SDC New Issues databases. We study industrial firms because the determinants of debt maturity are likely to be different across industrial and financial firms due to, for instance, the safety net of explicit or implicit government guarantees for financial firms. Custódio, Ferreira, and Laureano (2013), who first document the maturity puzzle, also study industrial firms. We obtain the list of unique bond issues from these databases using their CUSIP numbers. When a CUSIP number is not available, we identify unique bonds using their offering date, maturity date, offering amount, and 7

9 coupon rate information. Our final sample includes 19,101 bonds issued between 1975 and 2015 by 2,988 firms that have information in CRSP and COMPUSTAT databases. Our sample period starts in 1975 because the small number of bond issues (fewer than 10 bonds in each year) that satisfy the screening criteria during the earlier years prevents a meaningful comparison of average bond maturities through time. We compute Bond Maturity as the number of years between a bond s maturity and its issue date. Table 1 reports the summary statistics on the maturity of our bond sample and shows that the mean and median maturities during the entire sample period are 11 years and 10 years, respectively. To examine the trend in maturities through time, Panel A of Figure 1 plots the average bond maturity by years. In each year, there are on average 466 new bonds issued by 237 firms. The average bond maturity is 21 years in 1975, but it declines to 10 years in Panel B of Figure 1 compares the average maturity of new bonds issued by small and large firms. In each year, we define a firm as large (small) if its market value of equity is above (below) the median market value of equity observed in that year. The declining trend in bond maturities appears to be similar across large and small issuers. Over the last four decades, insurance companies relative aggregate market share in the corporate bond market has decreased substantially because the size of the bond market has grown much faster than the size of the insurance sector. To estimate the aggregate ownership share of insurance companies in the U.S. corporate bond market, we use the quarterly data from the U.S. flow of funds accounts published by the Federal Reserve. Accordingly, we define Insurer Market Share as the amount of insurance company ownership in U.S. corporate bonds and foreign entity bonds issued through U.S. dealers and purchased by U.S. residents divided by their total outstanding amount. The ownership data include both U.S. corporate and foreign 8

10 entity bonds as they are reported jointly. Nevertheless, foreign entity bonds make up a small portion of the institutional investor holdings (Becker and Ivashina, 2015), and hence, we interpret the ownership ratios as the shares of institutions in the U.S. corporate bond market. Appendix A provides more details on these variables. Table 1 reports summary statistics for the market share of insurance companies, and Panel A of Figure 1 plots the average market share of insurance companies by years. The average insurance company market share varies substantially through time with its highest value of 41% observed in 1979 and the lowest value of 19% observed in By plotting both the average bond maturity and insurer market share by years, Panel A of Figure 1 allows for a visual comparison of trends in the two series, and shows that, in the absence of any control variables, bond maturity is positively associated with insurer market share. We construct an extensive list of firm, bond, and macro level control variables that are used in the literature to explain bond maturities (e.g., Barclay and Smith, 1995; Guedes and Opler, 1996; Greenwood, Hanson, and Stein, 2010; Custódio, Ferreira, and Laureano, 2013). Firm level variables are Market Value of Equity, Total Debt/Total Assets, Net Income/Total Assets, Tangibility, Market-to-Book Value of Assets, Stock Return, Industry Dummies, and Firm IPO Decade Dummies, and bond level variables are Offering Amount, Callable Dummy, Floating Dummy, Convertible Dummy, Puttable Dummy, Sinking Fund Dummy, Global Dummy, and Credit Rating Dummies. We consider four macro-level control variables in our analyses: Term Spread, Default Spread, Real Short-Term Rate, and Share of Long-Term Government Debt. Appendix A provides a detailed description of these variables. 2 Following are the other major investors with at least 1% ownership in the U.S. corporate bond market as of 2015, and their ownership shares in 1975 and 2015, respectively, presented in parentheses: exchange-traded funds (0%, 2.1%), foreign banking offices in the U.S. (0.1%, 1.5%), government (0%, 1.5%), household (18.8%, 9.7%), mutual funds (1.9%, 15%), other unidentified foreign investors (4.5%, 26.4%), pension funds (30.5%, 10.8%), and U.S.- charted depository institutions (7.5%, 4.4%). 9

11 Table 1 presents the summary statistics of the control variables in our study. The average issuer in our sample has a market cap of $17 billion, market-to-book ratio of 1.9, net income to assets ratio of 0.3%, leverage ratio of 35%, asset tangibility ratio of 0.45, and an average monthly stock return of 6% that is measured during the 3-month period before the bond offering date. About half of the issuers in our sample went public before 1970, fewer than 10% of them went public during the post-2000 period, and they are evenly distributed across industries. In our sample, we observe bonds that are callable (62%), convertible (17%), global issues (11%), puttable (7%), and floating rate (6%). Bonds with a sinking fund provision account for 5% of the sample and the average bond offering amount is $0.29 billion. Figure 2 reports the distribution of bond characteristics through years and shows that bond features do not appear to follow the same trend as the maturity. About 47% of our sample bonds are investment-grade and 28% are speculative-grade. We are unable to find at-issue credit ratings from Moody s, S&P, or Fitch for the remaining 25% of bonds in our sample. During the analysis period, the mean values of the share of long-term government debt, real short-term rate, term spread, and default spread are 74%, 2.9%, 1.5%, and 1%, respectively. We also construct bond-level ownerships of insurance companies at the time of bond issuance. Bloomberg compiles the bond holding information of institutional investors from the 13F, Schedule D, 10-K, Form 990, and Form 5500 filings since 1998 at a quarterly frequency. 3 Using this database, we define Insurer Ownership as the amount held by insurance companies measured at the end of the issuance quarter divided by the bond s issue amount. We construct 3 The Securities and Exchange Commission (SEC) requires all institutional investment managers that exercise investment discretion over $100 million to report its holdings on Form 13f. The National Association of Insurance Commissioners (NAIC) requires all U.S. insurance companies to file Schedule D to reveal their holdings. Form 990 is a document filled with IRS by certain federally tax-exempt organizations. Form 5500 is filed with the Department of Labor by the sponsor of any employee benefit plans subject to Employee Retirement Income Security Act (EIRSA). Appendix B provides an example of the ownership information for a sample bond. 10

12 these ownership variables since 1999 to eliminate a potential coverage bias in the database inception year, and also perform several quality control checks (e.g., total reported institutional holdings should be less than or equal to the offering amount) to ensure data integrity. We are able to find the bond-level insurance company ownership data for 7,337 bonds in our sample. The average insurance company ownership in bonds is 15% in our sample. To evaluate the quality of Bloomberg s institutional ownership data, we compare our ownership statistics with the statistics reported elsewhere in the literature. For our sample, Bloomberg reports that the aggregate amount of institutional investor holdings (primarily insurance companies, mutual funds, and pension funds) in the quarter of bond issuance is $1.08 trillion, which accounts for about 30% of the total bond offering amount ($3.70 trillion). This level of ownership coverage in bonds is comparable to that reported in the literature. For example, Becker and Ivashina (2015) obtain the institutional bond holdings from Lipper emaxx database and report that the total ownership of insurance companies, mutual funds, and pension funds accounts for 33.7% of the total face value of bonds in Massa, Yasuda, and Zhang (2013) and Dass and Massa (2014) also report similar ownership statistics using Lipper emaxx. Lipper emaxx compiles bond ownership information based on regulatory disclosures to NAIC for insurance companies and to the Securities and Exchange Commission (SEC) for mutual funds, which are similar to the information sources of Bloomberg. Overall, for our bond sample, we conclude that the ownership coverage of Bloomberg s data is at par with that of Lipper emaxx which is used in previous studies. 3. Analyses of Bond Maturity In this section, we first investigate the validity of our premise that a decline in demand for bonds from insurance companies can lead to shorter bond maturities. We then test our 11

13 hypothesis that the decline in the aggregate ownership share of insurance companies in the bond market explains the declining trend in bond maturities Insurance Company Ownership and Bond Maturity To study the relation between insurance company ownership and bond maturity, we run the following regression using our sample of 7,337 new bond issues with available bond-level insurance company ownership data: Log Bond Maturity = + W + Z + X + Insurer Ownership + (1) ' ' ' ( ij ) jt i t i ij, where Bond Maturityij is the maturity of bond i issued by firm j, α is the intercept, Wjt, Zi, and Xt represent firm, bond, and macro level control variables, respectively, Insurer Ownershipi is the percentage of bond i s offering amount held by insurance companies as measured at the end of the offering quarter, and εij is the error term (the earlier section and Appendix A provide the list and detailed definitions of these control variables). In this regression, the coefficient on Insurer Ownership (λ) estimates the relation between the insurance company ownership in a bond and the bond s maturity. We also run regressions with firm fixed effects to examine this relation within firms. This regression and its variants document the correlation between the jointlydetermined maturity and ownership. Section 3.2 is dedicated to establishing, via several disparate approaches, whether there is a causal link running from ownership to maturity. Column (1) in Table 2 reports the coefficient estimate on Insurer Ownership from the above bond maturity regression. 4 We find that the coefficient estimate on Insurer Ownership is 0.67 and significant, indicating that bond maturity is positively associated with insurance company ownership. This coefficient estimate suggests that a 10-percentage point decrease in 4 The sign and significance of untabulated coefficient estimates on the control variables are intuitive. Larger firms and firms with better stock performance, higher profitability, higher tangibility, lower leverage, and lower book-tomarket ratios issue longer maturity bonds. Bond characteristics are also important determinants of maturity: callable, fixed coupon rate, convertible, and puttable bonds have longer maturity. At the macro level, bond maturity increases most notably when the default spread declines. 12

14 insurance company ownership is associated with a 7% decrease in bond maturity, which corresponds to 9 months for the average bond in our sample. Column (2) in Table 2 reports that the coefficient estimate on Insurer Ownership maintains its sign and significance controlling for firm fixed effects, showing that the positive relation between insurance company ownership and bond maturity also persists within firms. Our results in this section show that insurance company ownership in a bond is positively associated with the bond s maturity. This finding provides suggestive evidence that the decline in the aggregate ownership share of insurance companies in the corporate bond market may lead to the decline in bond maturities observed since 1970s. In the next section, we further explore how bond maturity varies with insurance company ownership using alternative identification strategies Estimating the Causal Effect of Insurance Company Ownership on Bond Maturity The earlier section documents that higher insurance company ownership in bonds is associated with longer bond maturities. In this section, we implement a regulation test, a natural experiment, and an instrumental variable regression approach to examine whether changes in the demand for bonds from insurance companies can have a causal effect on the maturity of new bond issues Capital Regulation Test Insurance companies in the U.S. have capital requirements established by NAIC s riskbased capital guidelines. These guidelines assign bonds with AAA through A- ratings the same risk factor, but assign an individual risk factor to all other rating categories. 5 Becker and Ivashina (2015) document that this capital regulation incentivizes insurance companies to invest 5 The NAIC s capital requirements for credit ratings are presented in parentheses as follows: AAA-A (0.4%), BBB (1.3%), BB (4.6%), B (10%), CCC (23%), CC or below (30%) (Obersteadt, 2017). 13

15 in riskier bonds within the AAA and A- rating spectrum to reach for yields while minimizing their regulatory capital requirements. Accordingly, we predict this capital regulation to cause insurance company ownership to be higher in A rated (A+ through A-) bonds than in AAA-AA rated (AAA through AA-) bonds. Consistent with Becker and Ivashina (2015), we observe in Figure 3 that insurance company ownership is generally lower in riskier bonds, but within the investment grade range, it is higher in riskier bonds. 6 If higher insurance company ownership leads to longer bond maturities, then we would also expect A rated bonds to have longer maturities than AAA-AA rated bonds. An empirical challenge in implementing this test is that, because credit ratings proxy for risk, the differences in maturities of AAA-AA and A rated bonds would reflect the influence of both regulation-related ownership differences and credit risk differences. To separate the two, we construct a numerical Credit Rating variable (e.g., AAA=1, AA+=2, AA=3, and so forth) to control for credit risk associated with credit rating assignments. We then estimate insurance company ownership and bond maturity controlling for Log(Credit Rating), its second and third powers, and credit rating dummies (e.g., A Rated, BBB Rated, BB Rated). 7 We omit the dummy variable for AAA-AA rating from the list of controls. By conditioning on rating and its higher ordered terms, we are partialing out the credit risk that the categorical rating dummies would otherwise reflect, and so they capture regulatory, rather than credit risk, reasons for the maturity to be different or for insurers to own the bonds. If the NAIC s capital regulations incentivizes insurance companies to invest in A rated bonds more than AAA-AA rated bonds 6 Figure 3 also shows that, although capital requirements are higher for BBB rated bonds than for A rated bonds, insurance company ownerships in BBB and A rated bonds are qualitatively similar, suggesting the capital requirements for BBB rated bonds are not stringent enough to reduce insurance company holdings in them. 7 We reach similar conclusions using the non-logged credit rating measure and omitting the higher order terms, but we prefer this specification because including higher order terms controls for non-linear aspects of credit risk on ownership and maturity. 14

16 and higher insurance company demand leads to longer bond maturities, we expect to find a positive and significant coefficient estimate on A Rated Dummy in both ownership and maturity regressions. Columns (1) and (2) in Table 3 report the estimation results from OLS and fixed effects regressions of Insurer Ownership, respectively, using our sample of rated bonds. Consistent with our regulatory arbitrage prediction, we find that the coefficient estimate on A Rated Dummy is indeed positive and significant in both specifications. We also find that the coefficient estimate on BBB Rated Dummy is positive and significant, suggesting that, despite BBB rated bonds having a higher risk factor than AAA-AA rated bonds, insurance companies reach for yields within the entire investment grade rating spectrum. The coefficient estimates on the other credit rating dummies are insignificant in both models, implying that higher risk factors assigned to riskier bonds deter insurance companies from reaching for yields in the junk rating territory. We next run the above regression of insurance company ownership using Log(Bond Maturity) as the dependent variable and report the results from OLS and fixed effects models in Columns (3) and (4) of Table 3, respectively. The coefficient estimates on A Rated Dummy and BBB Rated Dummy variables are positive and significant in both regression models, suggesting that abnormally high insurance company ownership in A and BBB rated bonds is associated with abnormally long bond maturities. The magnitudes of the coefficient estimates suggest that, depending on the specification, a 10-percentage point decline in insurance company ownership would lead to a 25% to 40% decline in bond maturity. Therefore, it is plausible that the decline in the ownership share of insurance companies in the corporate bond market from 40% in 1970s to 25% in 2000s causes the 50% decline in bond maturities observed during the same period Natural Experiment 15

17 We study the variation in the maturity of new bonds issued following natural disasters to investigate the direction of causality between insurance company ownership and bond maturity. Massa and Zhang (2011) show that insurance companies affected by Hurricane Katrina liquidated their corporate bond holdings and that the fire-sale effects of this demand shock lasted for several months. We extend their findings and utilize the ten natural disasters that led to the largest insured losses during our sample period as an exogenous shock to insurance company ownership and investigate whether the maturities of new bonds issued during the disaster periods have shorter maturities. 8 As most insurance companies offer both life insurance and P&C insurance products, we expect natural disasters to reduce the average insurance company ownership in bonds. We define a Disaster Dummy variable that equals one for 627 bonds issued during the calendar quarter in which a large natural disaster occurred, and zero otherwise. Columns (1) and (2) in Table 4 report the results from OLS and fixed effects regressions of Insurer Ownership, respectively. The coefficient estimate on Disaster Dummy is and significant in both specifications, indicating a 2-percentage point decline in insurance company ownership in bonds following natural disasters (measured at the end of the natural disaster quarter). Hence, natural disasters appear to serve as a negative shock to the demand for bonds from insurance companies. Next, we test whether these bonds that have low insurance company ownership due to natural disasters also have shorter maturities. Column (3) in Table 4 reports the estimates from the OLS regression of Log(Bond Maturity) and shows that the coefficient on Disaster Dummy is 8 We identify the ten largest natural disasters based on the insured losses provided by Swiss Re and Insurance Information Institute. The disasters with event quarter and their insured losses (in billions) in parentheses, are Charley (2004Q3; $10), Frances (2004Q3; $6), Ivan (2004Q3; $16), Katrina (2005Q3; $81), Rita (2005Q3; $13), Wilma (2005Q4; $15), Ike (2008Q3; $23), Super Outbreak (2011Q2; $8), Irene (2011Q3; $6), and Sandy (2012Q4; $30). 16

18 -0.07 and significant, indicating an average of 7% decline in maturities of bonds issued during the natural disaster periods. 9 Column (4) shows that the results are similar controlling for firm fixed effects. These findings suggest that a 10-percentage point decline in insurance company ownership would result in a 35% decline in bond maturity, which is consistent with the economic magnitude of the insurance company effect estimated from the earlier regulation test Instrumental Variable Approach Prior literature documents that investors prefer securities issued by geographically proximate companies (e.g., Coval and Moskowitz, 1999; Huberman, 2001; Ivković and Weisbenner, 2005). Thus, the geographical distance between an insurance company and an issuer is expected to be negatively related to the amount it invests in the issuer s bonds. This distance, however, may violate the exclusion restriction condition for a valid instrument because industry characteristics of issuers located in close distance to insurance companies that are mostly located in large metropolitan areas may influence the maturity of their bond issues. Hence, an instrument based on firm location would reflect both the familiarity of insurance companies with an issuer and the fixed effects of firm location on bond maturity. To address this concern, we construct our geographical distance variable at the bondlevel as the average distance between a bond s issuer and its insurance company investors. This bond-level distance varies within a firm and allows us to control for firm fixed effects in instrumental variable regressions, thereby eliminating the concern that the instrument violates the exclusion restriction condition because an issuer s location may influence its bond maturity. In 9 We find similar results when we use natural disasters as an instrument for insurance company ownership and estimate the effect of insurance company ownership on bond maturity following in a two-stage least squares framework. We omit these results because although this natural disaster instrument is a significant (at 1%) and negative determinant of insurance company ownership in bonds, its F-statistic of 5.52 is lower than 10 (the conventional threshold for strong instruments). This could be because there are few bonds issued during the disaster quarters, reducing the power of the instrument in predicting insurance company ownership in bonds. 17

19 addition, because we construct this distance measure using insurance companies that hold the bond, the measure accounts for firm and bond characteristics that make a certain bond undesirable for certain insurance companies. Alternatively, using all insurance companies to construct the distance measure would require us to control for their investment opportunity sets that are unobservable. Another potential concern with using this distance variable as an instrument could be that insurance companies located in closer distance to issuers can better monitor them and accordingly invest more in their longer maturity bonds. This concern is unlikely to influence our estimations because the monitoring role is more closely associated with banks who are concentrated creditors with insider information than with bondholders who are dispersed creditors facing holdout problems and have access to only public information (e.g., Fama, 1985). Thus, we argue that our distance-based instrument is likely to satisfy the exclusion restriction condition. To measure the geographical distance between issuers and insurance companies, we obtain county level ZIP codes for the headquarters of issuers and insurance companies in our sample from COMPUSTAT and A.M. Best Rating Service, respectively. Eliminating issuers and insurers that are located outside the continental U.S. results in a sample of 6,235 bonds with non-missing information. We then identify the latitude and longitude for each county from the U.S. Census Bureau s Gazetteer Place and ZIP Code database, and use the standard formula to compute the geographic distance between any two counties We compute the distance between county a and b as: d(a, b) = arccos[cos(a1) cos(a2) cos(b1) cos(b2) + cos(a1) sin(a2) cos(b1) sin(b2) + sin(a1) sin(b1)]r, where a1 and b1 (a2 and b2) are the latitudes (longitudes) of the two counties (expressed in radians), respectively, and r is the radius of the Earth (3,963 miles). 18

20 Next, we construct the distance variable for each bond by averaging the distances between the bond s issuer and the bond s insurance company investors. We define a Closer Dummy variable that equals one for bonds with the bond-level distance less than the median bond-level distance (806 miles) in our sample, and zero otherwise, and use this dummy variable as the instrument for insurance company ownership in bond maturity regressions. Our cutoff for defining Closer Dummy is the median bond-level distance because using a short distance cutoff (say, 100 miles) identifies few bonds (0.4% of bonds in our sample) as being below the cutoff. 11 Columns (1) and (2) in Table 5 report the first-stage estimation results from OLS and fixed effects regressions of Insurer Ownership, respectively, using Closer Dummy as an instrument. The coefficient estimate on Closer Dummy is positive and significant in both specifications, verifying that, conditional on investing, insurance companies located in closer distance to issuers invest more in their bonds. Closer Dummy is also unlikely to be a weak instrument as its F-statistic is greater than 10 in both regression specifications. Columns (3) and (4) in Table 5 report the estimates from the second-stage regression of Log(Bond Maturity) using OLS and fixed effects models, respectively. The coefficient estimate on Instrumented Insurer Ownership variable is around 4 and significant in both regression models, indicating that a 10-percentage point decline in insurance company ownership would lead to a 35% decline in bond maturity, which is consistent with the economic magnitude of the insurance company effect estimated from both the earlier regulation and natural disaster tests. Overall, the results from this instrumental variable test, and the earlier regulation and natural experiment tests show that changes in insurance company ownership in bonds can have a 11 We could also use the actual bond-level distance as the instrument but we find in untabulated tests that our median-based Closer Dummy is a stronger predictor of insurance company ownership than the actual bond-level distance. This finding suggests that the relation between the bond-level distance and insurance company ownership is non-linear. 19

21 causal effect on their maturities. These findings validate the underlying mechanism for our hypothesis that a decline in demand for bonds from insurance companies can lead to shorter bond maturities. In the next section, we test our hypothesis that the decline in the aggregate ownership share of insurance companies in the U.S. corporate bond market explains the declining trend in bond maturities observed since 1970s Bond Maturity and the Aggregate Ownership Share of Insurance Companies Having established the strong and potentially causal relation between insurance company ownership and bond maturity, we now examine our hypothesis that accounting for the aggregate ownership share of insurance companies in the corporate bond market explains the declining trend in the maturity of new bond issues. We test this hypothesis by running the following regression equation using our sample of 19,101 bond issues during the period: Log Bond Maturity = + W + Z + X + Trend + Insurer Market Share + (2) ' ' ' ( ij ) jt i t t t ij, which is similar to Equation (1) except that it includes Trendt as an additional regressor and replaces Insurer Ownershipi with Insurer Market Sharet. Trendt is the difference between the year of bond issuance and the year when our sample period starts (1975), and Insurer Market Sharet is the aggregate ownership share of insurance companies in the U.S. corporate bond market (Appendix A provides detailed variable definitions). We estimate this regression with and without controlling for Insurer Market Share to see whether it explains the trend. Column (1) in Table 6 reports the coefficient estimate on Trend from the above OLS regression of bond maturity without controlling for Insurer Market Share. We find that the coefficient estimate on Trend (multiplied by 100) is and significant, indicating a 1.12% unexplained decline in bond maturity in each year. This is equivalent to an annual decline in maturity of 1.5 months for the average bond in our sample, and a total unexplained decline of 5 20

22 years during the sample period, which is about half of that reported in Custódio, Ferreira, and Laureano (2013). Hence, although our comprehensive set of firm, bond, and macro level covariates improve the explanatory power of the existing model, it still fails to explain the 10- year decline in bond maturities observed since 1970s. Column (2) reports that the coefficient estimate on Trend maintains its sign and significance controlling for firm fixed effects, showing that the declining trend in bond maturities also persists within firms. Next, we introduce Insurer Market Share as an additional control variable to our maturity regression and report the results from the OLS and fixed effects specifications in Columns (3) and (4) of Table 6, respectively. The coefficient estimate on Insurer Market Share is 1.8 and significant in both specifications, indicating that a 10-percentage point increase in Insurer Market Share is associated with a 20% increase in bond maturity. Hence, bond maturity is positively related to insurance company presence in the market. More importantly, the coefficient estimate on Trend becomes indistinguishable from zero after controlling for Insurer Market Share. Therefore, accounting for the share of insurance companies in the corporate bond market is sufficient to eliminate the declining trend in bond maturities observed since 1970s. As an additional test, we investigate whether this decline in the insurance company presence in the corporate bond market influences firms cost of borrowing. If insurance companies demand for bonds decreases, firms may need to offer a higher yield to issue longerterm bonds. To test this prediction, we run a regression of bond offering yields controlling for Insurer Market Share, Log(Bond Maturity), and their interaction term in addition to the other controls in our baseline maturity regression. Columns (5) and (6) in Table 6 show that the coefficient estimate on the interaction term between Insurer Market Share and Log(Bond Maturity) is negative and significant in both OLS and fixed effects specifications. The 21

23 magnitudes of the coefficient estimates suggest that, if Insurer Market Share decreases by a 10- percentage point, the cost of increasing the maturity of a new bond issue by a year increases from 5 bps to 8 bps for the representative firm in our sample. 12 These findings provide evidence that the cost of issuing longer maturity bonds increases as the ownership share of insurance companies in the bond market decreases. Overall, the results of this section show that accounting for the share of insurance companies in the corporate bond market explains the declining trend in bond maturities observed since 1970s. In the next section, we examine whether insurer market share Granger-causes bond maturity Granger Causality Test of Bond Maturity and the Market Shares of Insurance Companies Granger causality tests allow us to examine the direction of the relation between the insurer market share and bond maturity in the aggregate. To construct the data for this test, we first take the average of Log(Bond Maturity) in each quarter from 1975 to Next, we match this quarterly maturity series with Insurer Market Share observed at the beginning of each quarter, assuming that issuers and underwriters observe the demand for bonds from insurance companies at the beginning of the quarter and determine the maturity of new bond issues accordingly. Using these two time-series, we then implement a Granger causality test. Column (1) in Table 7 reports the results from a regression where the dependent variable is the quarterly average Log(Bond Maturity), and the independent variables are the one-quarter lagged values of average Log(Bond Maturity) and Insurer Market Share. The coefficient 12 The average Insurer Market Share and Bond Maturity in our sample are 31.54% and 11.2 years, respectively. Accordingly, the coefficient estimates in Column (6) of Table 6 suggest that, when Insurer Market Share is 31.54%, increasing bond maturity from 11.2 years to 12.2 years is associated with a 5 bps increase in offering yield for the average bond in our sample [log(12.2/11.2) ( )]. However, when Insurer Market Share is 21.54%, the same increase in bond maturity is associated with an 8 bps increase in offering yield [log(12.2/11.2) ( )]. 22

24 estimate on both variables are positive and significant, indicating that lagged insurer market share and maturity are significant determinants of maturity in the subsequent quarter. Column (2) in Table 7 reports the results from an identical regression specification with the exception that the dependent variable is Insurer Market Share. In this regression, the coefficient estimate on lagged Insurer Market Share is positive and significant, but that on lagged average Log(Bond Maturity) is insignificant. These findings indicate that insurer market share Granger-causes bond maturity. Table 7 also reports the results from a Wald test that further confirms this conclusion Insurance Company Preferences and the Declining Trend in Bond Maturity In this section, we investigate whether the declining trend in bond maturities varies predictably based on bond and insurance company characteristics that are related to the maturity preferences of insurance companies. If the decline in insurance company ownership drives the declining trend in bond maturities, this effect should be more pronounced among bonds whose primary investors are insurance companies. For instance, because they are subject to risk-based capital regulations, insurance companies invest in safer and non-convertible bonds, and hence, their declining market share should lead to a greater decline in maturity for these types of bonds. To investigate whether this is the case, we first predict bond-level insurance company ownership using the subsample of 7,337 bonds with available ownership information and report the results in Appendix C. The regression results show that insurance company ownership is higher for safe (e.g., investment grade bonds issued by profitable and low leveraged firms), non-convertible, non-puttable, and fixed coupon bonds. Using the coefficients from this regression, we then predict the level of insurance company ownership for the entire sample of 19,101 bonds. 23

25 Because our model also controls for the macro-level variables (e.g., default and term spreads), the predicted ownership levels take into account the state dependent risk preferences of insurance companies. Finally, we classify each bond into facing high or low insurance company demand subsamples based on whether its predicted insurance company ownership is above or below the sample s median predicted insurance company ownership of 11.98%. Columns (1)-(2) and (3)-(4) of Table 8 report the coefficient estimates from the regression of Log(Bond Maturity) for the high and low insurance company demand subsamples, respectively. The coefficient estimate on Trend is negative and significant for the high insurance company demand subsample, and it is negative but insignificant for the low insurance company demand subsample. The difference in these coefficient estimates is statistically significant, indicating that the declining trend in bond maturity is indeed more pronounced for the subsample of bonds that face high demand from insurance companies. As an additional test, we split bonds into facing high and low insurance company demand subsamples by their credit ratings. As Figure 3 shows, insurance company ownership is higher in investment grade bonds than in non-investment grade bonds. The results of the ownership regression in Appendix C also show that insurance companies avoid convertible, puttable, and floating rate bonds. Accordingly, we drop bonds with these features and examine the maturity trends for investment grade and non-investment grade bonds. Columns (5)-(6) and (7)-(8) in Table 8 report the results for these investment and non-investment subsamples, respectively. In the OLS setting, the trends of investment and non-investment grade subsamples are similar. When firm fixed effects are included, however, the trend is only significant within the subsample of investment grade bonds. These results are broadly consistent with our expectation that the 24