|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 THREE ESSAYS ON LEVERAGE AND DEBT CONTRACTS By HUGH MARBLE III A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2007
2 2007 Hugh Marble III
3 ACKNOWLEDGMENTS I thank my parents and my friends for supporting me in this endeavor. I am grateful to David T. Brown, Joel Houston, Mike Ryngaert and Bipin Ajinkya for guiding and coaching me along the way.
4 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................3 LIST OF TABLES................................................................................................................. ..........6 ABSTRACT....................................................................................................................... ..............8 CHAPTER 1 INTRODUCTION..................................................................................................................10 2 INVESTMENT INCENTIVES AND THE RECOURSE STRUCTURE OF DEBT: THEORY AND EVIDENCE.................................................................................................12 Introduction................................................................................................................... ..........12 Debt Recourse and Investment Incentives..............................................................................18 Model Set-Up..................................................................................................................19 Debt Recourse and Acquisition Incentives......................................................................21 Debt Recourse and Investment Incentives......................................................................23 Debt Recourse and T echnology Choice Incentives.........................................................26 Equilibrium Recourse Strategies.....................................................................................27 Recourse Strategies: Evidence from Real Estate Investment Trusts......................................31 Real Estate Investment Trusts and Their Debt Contracts................................................31 Sample Description a nd Summary Statistics...................................................................35 Evolution of the Recourse Structure of REIT Debt.........................................................37 Analysis of Recourse Structure Transitions....................................................................39 Secured Debt Financing and the Underinvestment Problem..................................................41 Conclusion..................................................................................................................... .........42 3 SECURED DEBT FINANC ING AND LEVERAGE: TH EORY AND EVIDENCE...........49 Introduction................................................................................................................... ..........49 Secured Debt and Debt Capacity............................................................................................55 Model Set Up...................................................................................................................55 Underinvestment..............................................................................................................56 Asset Substitution Problem.............................................................................................57 Model Discussion and the Equilibrium Le verage and Level of Secured Debt...............59 Comparison to Other Models..........................................................................................60 Data and Preliminary Analysis...............................................................................................61 Regression Analysis............................................................................................................ ....64 Analysis of Leverage and the Fractio n of the Debt That is Secured...............................64 Analysis of the Seniority Structure of Debt....................................................................66 Instrumental Variables Analysis......................................................................................68 Conclusion..................................................................................................................... .........69
5 4 ANATOMY OF A RATINGS CHANGE..............................................................................81 Introduction................................................................................................................... ..........81 Data and Methodology...........................................................................................................82 Results........................................................................................................................ .............86 Conclusion..................................................................................................................... .........89 LIST OF REFERENCES............................................................................................................. ..98 BIOGRAPHICAL SKETCH.......................................................................................................101
6 LIST OF TABLES Table page 2-1 Summary leverage and secure d debt to total debt statis tics for REITs and industrial firms.......................................................................................................................... .........44 2-2 Evolution of the recourse structure of REIT debt..............................................................45 2-3 Recourse structure transitions............................................................................................46 2-4 Probit model of the factors influencing the transition to unsecured debt financing..........47 2-5 Probit model of increases in secured debt financing.........................................................48 3-1 Summary statistics......................................................................................................... ....71 3-2 Mean leverage and number of firm-year observations conditional on fraction secured and market-to-book............................................................................................................72 3-3 Mean leverage and number of firm-year observations conditional on fraction secured and fixed assets ratio......................................................................................................... .73 3-4 Determinants of market leverage using cluster-robust OLS regressions...........................74 3-5 Determinants of leverage by market a ccess subsamples using cluster-robust OLS regressions.................................................................................................................... ......75 3-6 Determinants of leverage by market-t o-book ratio subsamples using cluster-robust OLS regressions................................................................................................................ .76 3-7 Cluster-robust OLS model explaini ng the fraction of debt secured..................................77 3-8 Model of unsecured debt to firm valu e using cluster-robust OLS regressions..................78 3-9 Model of secured debt to firm va lue using cluster-robust OLS regressions......................79 3-10 Two-stage least squares m odel of leverage using net co mponents of property, plant and equipment as instrume nts for fraction secured............................................................80 4-1 Credit rating change data by year......................................................................................91 4-2 Potential causes of credit rating changes...........................................................................92 4-3 Summary of overall scoring criteria..................................................................................93 4-4 Causes of downgrades by invest ment versus speculative grade........................................94 4-5 Causes of upgrades by investment versus speculative grade.............................................95
7 4-6 Frequency of substantial management infl uence by credit quality and direction of ratings change................................................................................................................. ...96 r4-7 Incidence of selected specific causes of downgrades and upgrades..................................97
8 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THREE ESSAYS ON LEVERAGE AND DEBT CONTRACTS By Hugh Marble III August 2007 Chair: David T. Brown Major: Business Administration My studies considered three th ings: (1) the choice between n on-recourse secured debt and recourse debt (unsecured debt or secured debt with recourse) by firms that are sequentially acquiring assets and then making investment choi ces once those assets have been acquired, (2) how secured debt financing impacts both the a sset substitution and underinvestment problems, and (3) the frequency with which credit rating changes result from ch anges in the firms operating environment versus changes in cap ital structure controlled by management. First, non-recourse secured debt is shown to be optimal for firms engaged in the acquisition of assets which have little need for non-contractib le ongoing investment. Unsecured debt provides superior post-acquisition incenti ves for owners of asse ts that require ongoing investment or that can be easily modified. Empi rical tests using Real Es tate Investment Trusts provide evidence supporting the model a nd consistent with previous work. Second, the underinvestment problem does not depend on the proportion of the original debt that is secured and the asset substitution pr oblem decreases in the proportion of the original debt that is secured. Debt capac ity increases with the proportion of debt that is secured as the asset substitution problem is lower for a given leve l of debt. An analysis of a large sample of firms with COMPUSTAT data supp orts the model predictions.
9 Third, I found that management action plays a significant role in credit rating changes. Twenty-four percent of downgrad es and 41% of upgrades have a substantial management influence. The frequency of management impact on credit ratings shows the limitations of credit risk modeling using structural models th at assume constant capital structure.
10 CHAPTER 1 INTRODUCTION This analysis of leverage and debt contra cts comprises the following three studies: (1) Investment Incentives and the Recourse Structur e of Debt: Theory and Evidence, (2) Secured Debt Financing and Leverage: Theory and Evid ence and (3) Anatomy of a Ratings Change. Investment Incentives and th e Recourse Structure of Debt: Theory and Evidence provides a model of the choice between non-re course secured debt and recour se debt (unsecured debt or secured debt with recourse) by firms that are sequentially acquiring a ssets and then making investment choices once those asse ts have been acquired. Non-reco urse secured debt is shown to be optimal for firms engaged in the acquisiti on of assets which have little need for noncontractible ongoing investment. Unsecured debt provides superior post -acquisition incentives for owners of assets that require ongoing inve stment or that can be easily modified. The recourse structure, mix of rec ourse and non-recourse debt, of R eal Estate Investment Trusts, publicly traded entities that use significant amounts of non-re course debt, are shown to be consistent with this model and with predic tions of the Stulz and Johnson (1985) model. Secured Debt Financing and Leverage: Theory and Evidence models how secured debt financing impacts both the asset substitution and underinvestment problems and documents empirical evidence in support of the model. The model considers a firm with a mix of secured and unsecured risky debt claims against an asse t in place and the opportunity to acquire another asset. I show that (1) the underinvestment problem does not depend on the proportion of the original debt that is secured a nd (2) the asset substitution proble m decreases in the proportion of the original debt that is secured. Debt capacity increases with the proportion of debt that is secured as the asset substitution problem is lower for a given level of debt. An analysis of a large sample of firms with COMPUSTA T data supports the model pred ictions. First, leverage is
11 positively and significantly related to the fraction of the debt that is secured, controlling for other variables known to affect leve rage. Second, attaching collateral to the debt, unlike shortening maturity (Johnson (2003)) or incl uding protective cove nants (Billett, King and Mauer (2007)), does not increase debt capacity by mitigating the underinvestment problem. Anatomy of a Ratings Change documents the frequency with which credit rating changes result from changes in the firms operating e nvironment versus changes in capital structure controlled by management. The analysis uses a sa mple of 604 credit rating changes about which sufficient information is available to attribute the cause of the rating change to management, operational or economic causes. I find that management action plays a significant role in credit rating changes. Twenty-four percent of downgrades and 41% of upgrades have a substantial management influence. The frequency of management impact on credit ratin gs shows the limitations of credit risk modeling using structural models that a ssume constant capital structure.
12 CHAPTER 2 INVESTMENT INCENTIVES AND THE RECOURSE STRUCTURE OF DEBT: THEORY AND EVIDENCE Introduction This analysis models a firms choice of the recourse structure of its debt: the mix of recourse and non-recourse debt. The firm in this model has investment opportunities that arrive sequentially. The firm chooses wh ich assets to acquire and whet her to finance the acquisition with non-recourse secured debt (pro ject finance) or debt that has recourse to the other projects: unsecured debt or secured debt with recourse. Th e firm then chooses how to deploy the acquired assets. The firm makes a non-contractible choice of how much to invest in the asset that determines the expected net present value of the project, and a non-c ontractible choice of technology that determines the volati lity of the project returns. The recourse structure of debt impacts the payof fs to the firm (1) when assets are acquired ( acquisition incentives ) and (2) from investments that change the expected net present value and volatility of the acquired project ( investment incentives ). The optimal recourse structure minimizes the risky debt related distortions to the acquisition and investment incentives. Some firms use all recourse debt, others only non-reco urse debt and some a mix of recourse and nonrecourse debt. The models predictions about the recourse stru cture of debt financi ng are supported by an analysis of the recourse structures of Real Es tate Investment Trusts (REITs) debt. While nonrecourse, project financing, is common for privat e firms, its use by public firms is limited, see Esty and Sesia (2005).1 The empirical analysis focuses on RE ITs as (1) there are a large number of publicly traded REITs with considerable vari ation across firms in the mix of non-recourse and 1Some publicly traded firms have subsidiary debt that is not guaranteed by the parent which is effectively project financi ng. See Kolasinski (2006).
13 recourse debt with nearly all firms using so me non-recourse debt a nd (2) the popularity of nonrecourse debt among REITs is consistent with th e model. Recourse debt financing distorts acquisition incentives. If an asset pu rchase is financed with recourse debt, the value of the debt is influenced by subsequent asset purchases and ho w they are financed. There are wealth transfers to or from the original lender. However, th ere are no wealth transfers when the first asset purchase is financed with non-recourse secured debt: the value of the non-recourse debt claim depends only on the project financed. Non-recourse secured debt financing eliminates the debt overhang problem of Myers (1977). Potentially offsetting the distortions to acquis ition incentives, the debt-related distortions to investment incentives are smaller when the firms assets are bundled as collateral for recourse debt. A capital structure with r ecourse debt provides better in centives to maintain and manage the acquired assets than a capital structure wh ere the projects are financed with non-recourse debt and thus non-recourse debt is not a domin ant contract. Since the benefits of bundling collateral vary with both the nature of the assets and the size of the firm, some borrowers choose non-recourse debt while others us e debt with recourse or a mi x of non-recourse and recourse debt. The first improvement in investment incentives when a project is financed with recourse debt is that incentives to increase project return variability are smaller. The wealth transfer from the lender(s) to the firm from increased project va riability is greater with non-recourse debt than with recourse debt where the collateral is bund led. When the collateral is bundled with other non-perfectly correlated projects, an increase in project specific volatility is at least partially diversified away by the returns on other assets. Thus, when the collateral is bundled, spending resources on negative NPV modifications to the asset that increase return volatility is less
14 attractive as it results in a smalle r wealth transfer away from th e lender. Consistent with the model prediction that firms have st ronger incentives to modify assets financed with non-recourse debt, I provide industry commentar y indicating that non-recourse RE IT debt covenant protection constrains modifications to the asset much more than REIT unsecured debt contracts and that the resulting limited operating flexibility is a drawback to non-recourse financing.2 The second advantage of recourse debt financin g is that it mitigates the incentives to under-invest in the assets in place. This o ccurs because bundling two less than perfectly correlated assets lowers the promised yield on th e debt per dollar borrowe d relative to when the projects are financed on a standa lone basis with non-recourse debt. This idea is similar to Lewellen (1971) where the diversification effect of conglomerate mergers lowers the probability of default for a given level of debt and increases debt capacity. In this model, the diversification effect occurs when reco urse debt is used. REIT industry commentary is also consistent with idea in this model that recourse financing improves investment incentives. Specif ically, industry commenta ry states that higher quality assets are financed with non-recourse debt while lower quality assets are more likely to collectively provide the collateral for recourse debt. Lower quality assets are more likely to require non-contractible investment s in the future to reposition the asset. The incentive to underinvest in repositioning is greater with non-recour se debt financing. The returns to lower quality assets depend on idiosyncratic factors and thus benefit from the diversification effects of bundling the collateral. 2Along the same lines, John and John (1993) point out that project fina ncing contracts precommit the cash flows to customers, suppliers an d creditors and curtail managerial discretion.
15 The intuition of the model equilibrium starts out with the case where the firm sequentially purchases two assets with either non-r ecourse or recourse debt financing.3 Recourse debt financing provides better incentives to manage the asset and thus the combined value of the assets is greater with recourse debt. The extent of the gain from recour se financing depends on the nature of the asset and is referred to as the assets investment sensitivity. If the assets are sufficiently investment sensitive, then the expect ed gain from bundling the assets as collateral for recourse debt exceeds the expected losses associat ed with positive NPV acquisitions that are not taken because of the debt overhang problem. In th is case, recourse debt financing is optimal. When the assets are not sufficiently investment sensitive the optimal strategy is non-recourse debt financing. Extending past the two asset case, the firm that starts with non-recourse debt may convert some of its non-recourse debt to recour se debt (pledging the most investment sensitive assets as collateral for the recourse debt) af ter acquiring a large number of assets because the gains from recourse financing are increasing in the number assets bundled as collateral for recourse debt. Finally, if the transactions costs of repurch asing the non-recourse debt are high, the firm only uses recourse debt when it purchases a la rge number of assets at the same time. Theoretical works related to this model in clude Kahn and Winton ( 2004) and Flannery et. al. (1993) who model the trade-offs between a se parately capitalized subs idiary structure (nonrecourse debt) and joint incorpor ation (recourse debt). In Fla nnery et. al. (1993) and Kahn and Winton (2004) the borrower transfers wealth from the recourse lender by changing the mix of assets. Separately financed subsidiaries (effectiv ely non-recourse financing) reduce the ability to 3It is easy to show that the fi rm does not finance one asset with non-recourse debt and the other recourse debt.
16 shift assets.4 In our model, non-recourse debt creates a greater asset substitution problem with regard to changing the volatility of the assets in place. I provide evidence that suggests the recourse structure of REIT debt and REIT debt contract design are infl uenced by this problem. The idea in this model that combining assets as collateral for recourse debt lowers the nominal cost of debt and hence improves investment incentiv es is also present in Fl annery et. al. (1993). In John and John (1991) non-recour se project finance creates th e flexibility to optimally assign different debt levels to two separate projects so as to minimize agency costs and taxes.5 Finally, in this model and Berkovitch and Kim (1990) non-recourse debt eliminates the debt overhang problem with regard to purchasing ad ditional assets. In Be rkovitch and Kim (1990) non-recourse debt is not a dominant contract when asymmetric information is introduced. In this model, non-recourse debt is not a dominant c ontract with symmetric information because nonrecourse debt creates greater distortions to the incentives to manage the assets. Our analysis of the recourse structure of RE ITs uses a sample of 1,277 firm years of data from 1990-2003 and provides empirical support fo r a number of the model predictions and complements the anecdotal evidence in support of th e model. There is considerable variation in the mix of non-recourse and r ecourse debt financing by REITs, however, nearly all REITs use some non-recourse debt. The analysis shows that REITs typically start out with all of their debt as non-recourse secured debt. Consistent with the model, the firms that experience large enough growth to capture the diversification bene fit of bundling collatera l eventually use some recourse debt 4In Triantis (1992) securing debt mitigates the asset substitution problem as the borrower cannot sell assets and reinvest the proceeds in more risk y ventures without paying off the debt secured by the asset that is sold. 5Corporate level taxes also influence the mode of incorporation in Fla nnery et al. (1993).
17 (typically unsecured debt). When firms with all non-recourse de bt access unsecure d debt capital, they are left with significant amounts of unsecured debt which is consistent with a larger number of projects being bundled. The estimates of a probit model indicate that th e probability that a firm transitions from all non-recourse debt to some recourse debt financing is increasing in the size of the firm, increasing in the growth during the year and decreasing in leverage. The significantly positive relation between firm size and the probability the firm tran sitions to unsecured debt is consistent with the idea that firms must reach a sufficient size in or der to realize the divers ification benefits of bundling several projects as collateral for unsecur ed debt. The finding that firms are more likely to transition to unsecured debt financing in a year where they grow a lot suggests that they tend not to buy back secured debt. In fact, in the ye ar the firm transitions to unsecured debt the average firm grows by nearly seve nty percent and nearly all of the growth is financed with unsecured debt. In some cases, firms that have transitioned to unsecured debt financing, use non-recourse secured debt to finance subsequent acquisitions. Estimates of a probit model show that firms with unsecured debt are more likely to finance future acquisitions with non-recourse secured debt if they are highly leverage d. This evidence is consistent w ith the predictions of the Stulz and Johnson (1985) model where firms finance add itional investments with secured debt in order to mitigate the wealth transfer to existing lenders as the potential wealth transfers are larger when the firm is highly leveraged. In summary, the model derives two central pred ictions about how the re course structure of debt chosen by a firm varies with agency costs of debt inherent in the assets held by the firm and the number of assets. These predictions are suppo rted by the analysis of REITs. The first
18 prediction is that firms use non-re course debt financing when de bt related distortions to the incentives to invest in assets in place are small. While the potential debt related distortions to investment incentives are difficu lt to measure directly, industr y commentary that (1) lower quality assets are bundled to support recourse debt and (2) non-recourse debt contracts provide tighter protective covenants support this prediction. The second pred iction is that for assets like commercial real estate where non-re course debt financing is attrac tive, recourse debt financing may be used by firms that have acquired a large nu mber of assets. I document patterns in the mix of recourse and non-recourse de bt by REITs that are consis tent with this prediction. I also show that REITs have nearly twice the fi nancial leverage of industrial firms. This is somewhat surprising given (1) REITs do not pay corporate taxes on earnings distributed to shareholders (dividends) and hence debt financ ing does not have the ta x advantage it does for industrial firms (see Howe and Shilling (1988)) and (2) REITs grew a great deal by acquiring assets during the sample period and the presence of growth options should mitigate the use of leverage. Our model provides an explanation fo r this finding. Specifically, our model suggests that REITs rely heavily on non-reco urse debt because it eliminat es the underinvestment problem with regard to asset purchases a nd REIT assets are not investment intensive. It follows that if non-recourse debt financing mitigates the major agency cost of debt, then firms that in equilibrium borrow on a non-recourse basis, REITs, should be more leveraged. Debt Recourse and Investment Incentives This section models the relationship between the recourse structure of debt (i.e., the mix of recourse and non-recourse debt outstanding) and the incentives to (1) acquire assets and (2) invest in the assets that have been acquired. After describing the set up of the model, the optimal debt recourse structure is shown in three differe nt cases: (1) when the firm is acquiring assets that do not require investment to maintain the as sets, (2) when the assets have been acquired and
19 the borrower decides the level of investment to improve and/or maintain the assets, and (3) when the assets have been acquired a nd the borrower decides whether to change project risk. These results lead to a characterization of equilibrium outcomes where the nature of the firms assets, growth options and the number of assets that have been acquire d determine the optimal recourse structure. Model Set-Up Consider a firm with a fixed amount of equity capital and access to debt markets to finance the purchase of projects. The amount of equity is fixed in order to focus on the recourse structure of the debt. Incorporating taxes or asymmetric information to obtain an equilibrium amount of outside equity would needlessly co mplicate the model. Th e project opportunities arrive sequentially and have random payoffs th at are uniformly distributed between V-B and V+B. There are no synergies betw een the assets: liquidating one a sset does not impact the value of the other asset and thus the recourse structure of the debt does not impact the liquidation decision.6 The expected payoff of the project is V, a nd B determines the volatility of the project payoff. The variance of the project payoff is B2/3. The debt financing can be either non-recourse secured debt or debt with recourse to the other assets. The results of this model are largely the same if the recourse debt is unsecured or secured (i.e., has a first lien on an asset). Thus for simplicity I assume that the recourse debt is unsecured and it has a pari passu claim on all unencumbered assets proportional to the face amount of the debt. 6Specifically, if there were synergies associated with the two projects, th e firm would consider these synergies when deciding whet her to liquidate the assets. At another level, if there are synergies between the assets, the notion that they are se parate assets is reduced.
20 The sequence of events and information structur e are as follows. At t=0, the first of two projects arrives. Each project is described by values for V a nd B. At t=1, the second investment opportunity (project) arrives. At t=2, the firm makes non-contrac tible investment choices on all acquired assets are made. The project outcomes ar e realized and the cash fl ows distributed to the lender(s) and the firm at t=3. All debt contract s mature at t=3 and do not make coupon payments. All agents are risk neutral and the risk free rate of interest is zero. The firm makes two investment choices. First, it can expend resources to increase the net present value of the project. Sp ecifically, it can incr ease the project V. Second, the firm can expend resources to alter th e volatility of the project returns: ch ange B. Since we are interested in how the recourse structure of the firms borro wings impacts investment incentives (i.e., the returns to equity as V and B change) I do not needlessly comp licate the model by specifying a cost function for changing V or B. A recourse strategy that results in a higher return to increasing V or lower return to increasing B pr ovides better investment incentives than an alternative recourse structure where the returns to increasing V are lower or the returns to increasing B are higher. Solving the three pieces of the model demonstr ates the relative advantages of non-recourse and recourse debt. The first problem, the acquisition strategy problem, assumes that the V and B are fixed. Projects arrive at t=0 a nd t=1 drawn from a distribution of V and B. This allows us to isolate the impact of the recourse struct ure of debt on acquisition incentives. The second part of the analysis, the investment incentives problems, derives the returns to the investment choices at t=2. First, the margin al return to the firm from increasing V with recourse debt and non-recourse de bt are derived. Second, the firms incentives to engage in asset
21 substitution, the marginal valu e to the firm from increasing B, with recourse debt and nonrecourse debt are derived. Debt Recourse and Acquisition Incentives Using non-recourse debt eliminates wealth tr ansfers across lenders as assets are acquired. The value of the non-recourse debt depends onl y on the amount borrowe d and the collateral. Subsequent acquisitions and how they are financ ed have no impact on the value of non-recourse debt outstanding. The agency costs of debt resu lting from these wealth transfers are well known. An important issue for the equilibrium in this model is the extent to which the firm passes up positive NPV acquisitions with recourse debt. I start with the following simple example wher e the firm invests $3 in equity and borrows $12 to finance a zero-NPV acquisition: a project that has an expected total payoff, or V, of $15. With B=10, a non-recourse lender would require a face amount of approximately $14.05 to have an expected return of zero. If a second project with the same parameter values arrives the next period, the firm could raise $3 in fairly priced outside equity and borrow $12 in fairly priced non-recourse debt purchase the project. Alternatively, the firm could attempt to fina nce the first project purchase with recourse debt financing. With a correlation between the project payoffs of zero and assuming the lender knows the second project will be taken and financed with unsecured debt, the fair face value of recourse debt for the first project would be $12.54 When the second proj ect arrives, the second lender would similarly provide $12 worth of rec ourse debt with a face amount of $12.54. Both lenders would have an expected return of zero an d the firm would also have an expected return of zero given the project parameters. The next section shows that lo wer face value of the recourse debt ($12.54 versus $14.05 w ith non-recourse debt) induces better incentives to invest in the acquired assets.
22 However, it is not rational for the firm to inve st in the second projec t. Passing up the zeroNPV second project results in a wealth transfer away from th e original lender. Knowing the second project will not be taken, the fair face value of the origin al recourse loan is $14.05, the non-recourse face amount. Extending the example above, assume the first proj ect is financed with recourse debt and a non-negative NPV second project w ill arrive with certainty. Howeve r, the NPV of the project is uncertain: the second project V wi ll be drawn from a uniform distribution between 15 and 20. All of this information is known to the first lender. Lending is perfectly competitive and the second lender will price the loan to have an e xpected NPV of zero consid ering the original loan and project. The firm only takes the second project if it increases the value of the equity stake. The equilibrium outcome is that the firm rejects all second period projects with V less than V* and V* equals $15.81. Although all projects with V>$15 are positive NPV on a standalone basis, the projects with values of V between $15 and $15.81 reduce the value of the equity stake: the wealth transfers to the original lender exceed the project NPV. In this example, the firm knows the distributi on of the second project s V and all projects have the same B. If the projects arrive with different mean return s (V) and return volatilities (B), then the moral hazard problem associated with re course debt financing is considerably larger. The firm also has an incentive to pass up high V projects in order take lower V projects with high levels of B.7 7Chapter 3 shows that the priority of the fi rst loan influences the moral hazard problem. Specifically, the extent of the underinvestment problem (passing up positive NPV) projects is identical if the original debt is secured or unsecured. However, the asset substitution problem (picking lower NPV projects that have large values of B) is mitigated, but not eliminated if the original debt is secu red without recourse.
23 Considering only the acquisition incentives, non -recourse debt is the dominant contract. The case for recourse debt financing is shown in the next section when the investment decisions are endogenous. With recourse debt, firms chose levels of V that are closer to first best and have weaker incentives to spend resources to increase B. Debt Recourse and Investment Incentives The impact of the recourse stru cture of the firms debt on inve stment incentives is shown by deriving the relation between the ma rginal return to the firm fro m investments that change the projects payoff distribution and th e recourse structure of debt. Th e investment is made after two projects have been acquired. The recourse struct ure of the firms debt liabilities only has an impact on investment decisions if the firm has more than one project. For simplicity, the two projects are assumed to be identical. The case where each project is financed by non-recourse secured debt is compared to the case where each project is financed by unsecured debt: the two assets serve as collateral for the unsecured debt. Specifically, I compare the returns to the firm associated with an increase in V (the expected value of the project) and B (the volatility of the project re turns) for two firms that both have acquired two projects. Rather than rationall y price the debt, I assume each firm has the same face value of debt outstanding. This allows us focus on how the recourse structure of the debt influences investment incentiv es holding leverage constant. From the setup described above, if each projec t is financed by non-recourse secured debt with a face value of F, the probabil ity of solvency for each project is B B V F 2 ) ( 1 (1-1) The lender receives F when the project is solvent and, on average, receives the midpoint of [V-B, F] when the project is insolvent. Ther efore, the expected payment to the lender is
24 2 ) ( 2 ) ( 2 ) ( 1 B V F B B V F B B V F F (1-2) The firm receives V less the expected payment to the lender. 2 ) ( 2 ) ( 2 ) ( 1 B V F B B V F B B V F F V (1-3) In the case where two projects are collate ral for unsecured debt, I assume the project payoffs are uncorrelated assets for tractability. However, the qualitative results hold for any correlation less than one. If the two projects ar e perfectly correlated, there is no distinction between non-recourse debt and bundling the collateral (i.e., using unsecured debt). The total payoffs to the first and second projects are drawn from uniform di stributions between V1-B1 and V1+B1 and between V2-B2 and V2+B2. Together, the two project s have a total payoff drawn from a symmetric triangular di stribution with a minimum at V1-B1+V2-B2 and a maximum at V1+B1+V2+B2. For simplicity, I assume that each proj ect is expected to be solvent: F
25 The manager receives the total expected payo ff less the expected payoff to the lender from 1-7 above. 2 2 1 3 2 1 2 1 2 1 2 1 2 2 1 3 2 1 2 2 1 2 2 1 2 1 2 1 2 1 2 1) ( 6 ) ( ) ( ) ( 3 ) ( 2 ) ( 2 )] ( ) ( ) [( 1 ) ( ) ( B B B B V V B B V V F F F F B B B B V V F F F F V V (1-8) Differentiating 1-3 with respect to V yields the marginal value to the firm of increasing the projects expected payo ff with non-recourse secu red debt financing. B V F B 2 (1-9) With unsecured debt, differentia ting (8) with respect to V1 gives: 2 2 1 2 2 1 2 1 2 2 1 2 2 1 2 1 2 1 2 1 2 16 3 3 1 B B B B V V F F B B B B V V F F F F (1-10) The partial derivatives of the payoff to the fi rm from increasing the mean project payoff of one project are larger in the re course (unsecured) case than in the non-recourse secured case for all plausible parameter values. It follows that unsecured debt financing (bundling the collateral) improves the firms incentives to invest in its assets in place. The intuition behind why investment incentives are better when the co llateral is bundled is straightforward. When two less th an perfectly correlated projects are bundled, the probability of default on the unsecured debt is lower than the probability of default with two separate nonrecourse debt financed projects holding the total f ace value of debt constant. Thus, the returns to the firm of increasing the expected return to the project are greater as the wealth transfer to the lender is smaller. Further, the improved investment incentives with recour se debt grow with the number of uncorrelated a ssets the firm owns. The idea that combining two less than perfectly correlated assets lowers the probability of default for a given amount of debt dates back to Lewellen (1971) who argues that conglomerate mergers could generate value by increasing debt capacity. The point that debt capacity is enhanced because investment distortions are lower for a given level of debt when two assets are
26 combined is also made by Flannery et. al. (1993) in a model where a firm decides whether to jointly or separately incorporate two assets. The point here is that thes e potential improvements in investment incentives are lost when the firm uses non-recourse debt to finance its assets. Debt Recourse and Technology Choice Incentives Risky debt creates the potential for the asset substitution problem because the firm has an incentive to increase the risk of the asset returns in order to transfer wealth from existing lenders. Changing B changes the variance of the payo ff without changing the expected payoff. Differentiating 1-3, the expected payoff to the fi rm with non-recourse secured debt financing, with respect to B yields 24 ) )( ( B B V F V F B (1-11) Differentiating 1-8, the payoff to the firm with recourse (unsecured) debt financing, with respect to B1 gives: 3 2 1 2 2 1 2 1 2 1 2 1 2 1 2 16 2 2 2 2 B B B B V V F F B B V V F F (1-12) The partial derivative of the payoff to the firm from increasing the variability of the project payoff of one project is smaller in the unsecured ca se than in the secured case for all plausible parameter values. It follows that using unsecure d debt reduces the firms incentive to pursue activities that increase the vari ance of the project payoff. The intuition behind why recourse debt fina ncing (bundling the two projects) lowers the incentive to increase asset volatility is more subtle. When two less than perfectly correlated assets are bundled, the expected wealth transfer away from bondholders associated with an increase in the volatility of retu rns to either asset is smaller. The borrower walks away from the poorly performing asset when it is financed by non-recourse debt. When the asset performs poorly and is bundled with another asset, in ma ny states of the world the other asset performs
27 sufficiently well that the borrower has no incentive to default. Two related implications of this analysis are (1) assets that can be transformed relatively eas ily are not good collateral for nonrecourse (project) debt financing and (2) debt covenants limiting what can be done with the collateral will be much more restrictiv e with non-recourse project financing. Equilibrium Recourse Strategies The analysis shows that non-recourse debt fina ncing provides better acquisition incentives than recourse debt financing. However, recourse debt financing provides (1) better incentives to invest in the assets in place and (2) limits incen tives to engage in negative NPV increases in the volatility of the project returns. Using non-recour se debt versus recour se debt trades better acquisition incentives off agains t the better ongoing investment in centives of using recourse debt. The intuition behind the equilibrium outcomes in this model are easily returning to the numerical example in the Debt Recourse and Acquisition Incentives section. Assume that when the first asset purchase is finan ced with either recourse debt or non-recourse debt the second asset is financed with same type of debt. Later I show that if the first asset is financed with recourse debt then there is no ga in from financing the second asset with non-recourse debt and if the first asset is financ ed with non-recourse debt then there is no gain from financing the second asset with recourse debt. In the numerical exam ple, all positive NPV projects (V> $15) are taken with non-recourse debt and onl y projects with V > $15.81 are taken with recourse debt. Focusing only on the level of V (ignoring B for simplicity), suppose that the levels of V in this example reflect the level of investment ma de with non-recourse debt. The value of both assets is higher if they are both financed by r ecourse debt and the gain in value from using recourse debt is larger when the assets are investment sensitive. Now suppose when both assets are financed with recourse debt the value each project is $0. 41 higher than with non-recourse
28 debt. In this case, all projects th at are taken with non-recourse debt (V > $15) are taken with recourse debt financing as well. With recourse debt financing the project with V = $15 is also undertaken by the recourse lender because the marginal value of the second project is worth $15.82 to the recourse debt borrower (the V of th e second project is effectively $15.82) and the recourse debt borrower takes all projects with V > $15.81. When it is rational to purchase all positive NPV second assets, there is no cost to using recourse debt as (1) both projec ts are undertaken, (2) the debt is priced assuming both projects are undertaken and (3) the firm captures the better investment incentives associated with recourse financing. In fact, it is the better investment incentive s that make the second asset purchase rational. If recourse debt financing increases each pr oject value by at least $0.41, then recourse debt obviously dominates non-recourse de bt. Likewise, if recourse debt has a small impact on project values, the expected losses from passing up positive NPV projects exceeds the expected enhancement to project values. Thus, recourse debt financing occurs when the assets are sufficiently investment sensitive. Since non-recourse debt is only used to finan ce assets with low inve stment sensitivity, one would expect firms that use non-recourse debt financing to be highly leveraged. Non-recourse debt financing is used by firms that have small i nvestment sensitivity so th e agency costs of debt financing with regard to investments in the asset are small. Non-recourse debt financing eliminates the agency costs of debt with regard to acquisition incentives. Thus, the agency costs of debt are lower for all levels of debt for firm s that in equilibrium choose non-recourse debt and leverage should also be higher for firms that choose non-recourse debt.
29 If the first asset is purchased with non-recourse debt financing, in this two asset example, the second asset is de facto non-recourse debt fina ncing because there is only one asset available to pledge as collateral. More importantly, in this equilibrium, if the firm financed the purchase of the first asset with recourse de bt, it would not purchase the sec ond asset with non-recourse debt because then there would be no improvement in investment incentives associated with bundling the collateral. However, if over time, perhaps as the result of poor operating performance, the firms leverage became higher than expected at the time the initial debt financing was arranged, the firm might use non-recourse debt financing to limit the wealth transfer to the original lender. This point is made by Stul z and Johnson (1985). The firm that finances its asset purchases w ith non-recourse debt might convert its nonrecourse debt to recourse debt as the probability the firm expects to purchase more assets in the future falls since the role of non-recourse debt is to provide better incen tives. Likewise, as the firm acquires assets using non-recourse debt fi nancing, it may find it optimal to bundle some of the assets as collateral (particularly investment sensitive assets). Since the improved incentives to invest in the asset and limited incentives to increa se asset volatility associat ed with recourse debt occur because of the diversifica tion effect of bundling uncorrela ted assets, these benefits are increasing in the number of assets the firm owns. The optimal recourse structure of the debt would be a mix of rec ourse and non-recourse debt. As the firm uses recourse debt to buy b ack non-recourse debt, it enjoys benefits from improving the investment incentives on the assets pl edged as collateral for the recourse debt but increases the underinvestment probl em. Provided that the firm e xpects to continue purchasing assets, an internal optimal mix of recourse and non-recourse debt wi ll exist for firms that initially
30 start out with non-recourse debt. Obviously, the fi rm will pledge the most investment sensitive assets as collateral for the recourse debt. It is important to note that th e optimal recourse structure for firms that start out with nonrecourse financing is either (1 ) all non-recourse debt or (2) a mix of non-recourse debt and a sufficiently large amount of recour se debt (assets ple dged as collateral for recourse debt) to realize significant benefits of bundling the collat eral. Put another way, as the firm uses more recourse debt to buy back more non-recourse debt, the benefits of bundling the initially grow at an increasing rate with the number of assets pledged as collateral for recourse debt. This is an important testable prediction of th e model. I would expect that firm s in an industry that uses nonrecourse debt will either have (1) all non-recourse debt or (2) a mix of recourse and non-recourse debt with a significant proportion of the debt non-recourse. Further, small firms with few projects will be more likely to have all non-recourse debt and firms with se veral projects will be more likely to have a mix of r ecourse and non-recourse debt. The size of the transaction co sts associated with buying b ack secured debt play an important role in how the firm that starts wi th all non-recourse debt transitions to using recourse debt. If the costs of buying back the debt (including the lenders ab ility to extract rents by holding out) are large, the firm is less likely to transition to recourse debt by buying back the debt. However, if the firm purchases several as sets in one period, the firm can effectively use recourse debt without buying back non-recourse debt. The transa ctions costs associated with repurchasing the debt are likely to be increasing in the risk of the debt because the bargaining space is limited by the firms ability to buy the bonds at par. The non-recourse lender is able to extract some rents from the borrower when the debt is repurchased. T hus, the potential amount the non-recourse lender can extract increases with the risk of the debt. Thus, I would expect (1)
31 highly leveraged firms to be less likely to transi tion to recourse debt financing because of the potential wealth transfers, and (2) firms that grow a lot in particular period to be more likely to transition to recourse debt financing. Recourse Strategies: Evidence from Real Estate Investment Trusts This section provides an empirical examination of the recourse structure of REIT debt. I focus on Real Estate Investment Trusts because it is an industry with a large number of publicly traded firms and an industry that relies signifi cantly but not exclusiv ely on non-recourse debt financing. Specifically, while REIT s may have both secured and unsecured debt, virtually all of the secured debt is mortgages (non-recourse secured debt) against specific income earning properties (see Moodys Investor Services (2002)). While the recourse of the debt is not explicitly stated in SEC filings our spot check of REIT 10-K filings reveals that most of the secured debt is listed as either mortgage debt or trust deed debt both of forms of non-recourse debt. Less commonly, the filings explicitly indi cate that the secured lender does not have recourse to other assets of the firm. Project financing is mostly a source of debt capital for private firms (see Esty and Sesia (2005)). RE ITs are the only public firms, and hence have systematic data available on the recourse structure of the entitys debt, that I am aware of that borrow extensively on a non-recourse basis. Real Estate Investment Trus ts and Their Debt Contracts Since they were initially created under the ta x code, REITs allow broad ownership of real estate with deferred corporate taxes if the ma jority of income is paid as dividends to shareholders. REITs engage in two broad lines of business: owning equity in terests in real estate and owning mortgages and mortgage-backed securities Our analysis eliminates REITs that hold exclusively mortgages in their asset portfolio: mortgage REITs. Instead I focus on REITs that
32 own income producing commercial real estate: equity REITs.8 The Tax Reform Act of 1986 made key changes in REIT legislation, facilitati ng a fundamental shift in the business activities of REITs. Before 1986, REITs were essentially passi ve holding vehicles for real estate and real estate mortgages. After the tax law change, RE ITs were allowed to actively manage and operate the real estate they owned. By the early 1990s nearly all REITs activ ely managed their real estate holdings and hence our analysis focuses on this time period. The nature of the business of REITs and the assets they hold fit the model conditions for firms that would borrow on a non-recourse basis. First, REITs have grown rapidly through the acquisition of real esta te assets held by private investor s (see Riddiough and Wu (2006)). With REITs natural advantage of raising capital in liquid public markets, such growth was not surprising and was probably anticipated by manage rs. Thus, REIT borrowers would naturally value the improved acquisition incen tives associated with non-reco urse debt financing. Second, the REIT assets are distinct projects with very little synergies across the assets. Thus, the REIT has limited concerns about any lost synergies that might occur when a secured lender forces the liquidation of an asset. Third, commercial real estate is not an invest ment intensive asset in the context of this model and thus the diminished incentives to maximize the value of the asse t through investments to maintain the asset are less problematic in co mmercial real estate. Co mmercial real estate requires significant investment to maintain the asset. However, such investment is easily contractible. For example, Ling and Archer ( 2005) report that non-rec ourse lenders require borrowers to set aside funds each year (typica lly $250 per year for apartment units and $0.15 per 8Health-care and lodging REITs are excluded, as is common, because these tend to operate in fundamentally different ways than RE ITs owning all other types of property.
33 square foot of industrial, office or retail space) in a reserve account. The accumulated balances in these reserve accounts are used to fund non-recu rring capital costs like replacements of (1) carpets, (2) roofs and (3) air conditioning. Fu rther, the models of Brown, Ciochetti and Riddiough (2006) and Williams (2001) explain the stylized fact that REITs are more likely to own assets that are not investment sensitive while private owner-managers own the investment intensive assets. An important aspect of our model is the pred icted relation between the recourse structure of debt and the borrowers incentives to modify the technology. Modifying the technology in the context of commercial real esta te is referred to as repositioning the asset. The limitations on reconfiguring commercial re al estate assets observed in debt contracts are consistent with the model. Specifically, there are much stronger restri ctions on reconfiguring assets in non-recourse debt financing and this is an important drawback to non-recourse debt financing. A discussion of the merits of secured (non-recourse) versus unsecured debt financing by Moodys Investors Service (2002) specifically mentions (1) the prev alence of restrictive covenants in secured debt contracts and (2) the lost flexibility associated with these restrictions: mortgage agreements typically inhibit or restrict the ability of an owner to reposition properties, if it result s in cash flow disruption. In addition, secured le nders may balk at an owner reformatting a property, such as reconfiguring and expanding space, even if it desperately needs it, if doing so entails ejecting key tenants. These factors can make the mortgaged asset less attractive to a buyer, t hus impairing asset liquid ity and constraining a firms ability to strategically reposition or even manage its portfolio. REITs that fund with unsecured debt do not face these challenges. REITs consistently tell us that this strategic flexibility issue is one of th e biggest drawbacks of mortgage debt. These points are consistent with our models prediction that th e incentives to modify assets in a manner that transfers wealth away from lenders are greater with non-recourse debt financing. Thus, it follows that we would expect, as observed, ti ghter covenant restrictions on asset repositioning in non-reco urse debt contracts.
34 The same report goes on to state that the nature of the assets influences whether they are financed with non-recourse or recourse debt: The best assets tend to get mortgaged, with the weaker ones left un encumbered to support the unsecured bonds. This happens because it is more efficient to pledge high quality assets. Asset quality refers to the cash flows genera ted from the asset and the likelihood that these cash flows will be maintained (see Ling and Archer (2005)). High quality assets are located in strong markets and have long term leases with te nants that are unlikely to default on the lease. Further, the age of the building is important. High quality assets are new buildings with modern designs that are unlikely to require a reconfigurat ion to attract tenants in the future. Further, Avalon Bays 10-K filing for 2002 reports We e xpect to continue to fund development costsfrom retained operating cash and borrowings under the unsecu red credit facility (p.48). Avalon Bay uses unsecured (recourse debt) to fund investment in real estate assets that are being developed (investment intensive assets) and funds stabilized assets with non-recourse mortgage debt. The observed relationship between asset quality and debt recour se is consistent with the model in two regards. First, the model suggests that it is inefficient to use non-recourse financing when non-contractible investments are likely to be required. Thus, the model suggests it is inefficient to fund lower quality assets with non -recourse debt as they are likely to require investments to reconfigure the asset in the future and it is difficult to wr ite a debt contract exante that specifies that the borrower will engage in the optimal reconfiguring of the asset. Second, the returns to low quality assets are more variable and idiosyncratic as the asset cash flows depend on local market dynamics. Thus, the diversification be nefit of bundling the
35 collateral (using these assets to support unsecured, non-recourse, debt) is higher for low quality assets.9 Sample Description and Summary Statistics This section describes the sample of REITs that are used in the anal ysis of the recourse structure of REIT debt. This se ction also compares the usage of secured debt and leverage of REITs to industrial firms. The mix of secured and unsecured borrowing by firms is initially examined using annual COMPUSTAT data from 1990 through 2003. I compare equity REITs, excluding healthcare and lodging RE ITs, to industrial firms: SIC codes from 2000 through 5999. Data are available from COMPUSTAT to calcula te leverage ratios for industrials for 63,662 firm-years and for REITs for 1,277 firm-years. Calc ulating secured debt to total debt ratios reduced the available firm-years to 55,493 for industrials and 1,202 for REITs because the secured debt ratio is undefined for firms with no debt. Those observations, which have a leverage ratio of zero, are included in the data fo r calculating descriptive statistics about leverage ratios, but not in the data for calculating descri ptive statistics about s ecured debt ratios. Book leverage is defined as the sum of Total Long-Term Debt (Item 9) and Debt in Current Liabilities (Item 34) divi ded by Total Assets (Item 6). The numerator is adjusted by Debt Due in One Year (Item 44) when a footnote fl ag indicates that Item 44 is included in Item 9 Because Item 44 is a part of Item 34, failing to make this adjustment double counts debt due in one year. The secured debt ratio is defined as Secured Debt and Mortgages (Item 241) divided by total debt. Total debt is the numerat or from the debt ratio calculation. Before presenting the comparison of REITs and industrial firms it is important to note that our subsequent analysis of REIT debt seniority strategies incor porates data available from the 9 I thank David Ling for raising this point.
36 SNL DataSource to calculate the secured debt ra tio and to add some firms that do not have COMPUSTAT data. The initial comparison of REITs and industrials relies on COMPUSTAT data for comparison. Later, I combine the SNL DataSource and COMPUSTAT because the COMPUSTAT data tends to understate the ratio of secured debt to total debt.10 However, this bias is relatively small and does not qualitatively impact the su mmary statistics presented in Table 2-1. As shown in Table 2-1, the mean (median) leverage of 47.4% (48.4%) of REITs is much greater than the mean (median) leverage of 27.5% (22.4%) for i ndustrial firms. The usage of secured debt shows a dramatic difference in the recourse structure of debt: REITs have median secured debt to total debt ratio of 78.1% versus 16.3% for industrial firms and REIT secured debt is generally non-recourse while th e secured debt of industrial firm s has recourse to other assets. Information about whether secured debt has recour se to the other assets of the firm is not indicated by COMPUSTAT. However, as mentione d above it is widely known that the secured 10 Both SNL DataSource and COMPUSTAT report total debt and secured debt but the definitions are not identical. The SNL DataS ource definition does not include mandatorily redeemable preferred stock in its debt measures while COMPUSTAT does. For most firm-years, the debt ratios calculated using SNL DataSour ce and COMPUSTAT are the same within onetenth of one percent. Spot checking 10-K filings for firm-y ears where there were larger disagreements traced differences to (1) treatment of mandatorily redeemable preferred stock, (2) issues with short-term debt or (3) very rare ly, legitimate ambiguity of classification in the original filing. SNL DataSource defines secured debt as being debt backed by real property while COMPUSTAT includes debt backed by letters of credit and similar instruments. The SNL DataSource data includes all secure d debt backed by real propert y, regardless of maturity, while COMPUSTAT only includes long-term secured debt. There is no direct measure of short-term secured debt in COMPUSTAT. Comparing secu red debt ratios across SNL DataSource and COMPUSTAT resulted in more frequent disagree ments of material magnitude. The differences overwhelmingly came from the values for secure d debt. The most common difference is that SNL DataSource would reflect a firm-year ha ving a secured debt ratio of 100% and COMPUSTAT would reflect a lower number. Co mmonly, this difference would be would be exactly explained if COMPUSTATs Debt in Cu rrent Liabilities was added to COMPUSTATs Secured Debt and Mortgages. Spot-checki ng 10-K filings overwhelming verified this explanation.
37 debt of REITs generally does not have recourse to other assets. Non-recourse secured debt financing by industrial firms is very rare (see Esty and Sesia (2005)). The fact that REITs are more leveraged than industrial firms is not consistent with traditional models of leverage: (1) REIT debt fi nancing does not have the tax advantage it does for industrial firms because REITs do not pay corporate taxes on earnings distributed to shareholders as dividends and, and (2) REITs gr ew a great deal by acquiring assets during the sample period and the presence of growth op tions has been shown to reduce leverage in numerous studies. Our model suggests that REITs rely heavily on non-recourse debt because (1) it virtually eliminates the underinvestment problem associated with acquisitions and (2) REIT assets are not investment intensive. It follows that if non-recourse debt financing mitigates the major agency cost of debt, then firms that in equilibrium borrow on a non-recourse basis, REITs, should be more leveraged. Evolution of the Recourse Structure of REIT Debt This analysis examines the model predictions regarding how the recourse structure of REIT debt evolves over time. In this analys is, I rely exclusively on data from the SNL DataSource because the treatment of secured debt is more accurate than COMPUSTAT. Using the SNL DataSource, the sample grows to 1,440 firm years of data from 146 different firms. The model predicts that firms that have assets that are likely to be funded with non-recourse debt, will start out using non-recourse debt financing and may use some unsecured debt when they have a sufficiently large number of assets to realize the dive rsification benefits of bundling unencumbered collateral to support non-recourse debt. Our analysis uses the IPO year as the year the firm starts out. Our sample of REITs includes some firms that accumulated assets in the 1960s and 1970s and that have been publicly traded for decades although most of the sample REITs went public in the 1990s. As noted in the
38 discussion of REITs, the industry took its pr esent shape in the early 1990s both through new REIT offerings and through the transition of REIT s from older-style passive holding vehicles to the present form of owners and managers of real assets. Firms with IPO dates before 1990 are treated as having conducted their IPOs in 1990. Table 2-2 shows the recourse structure of REIT debt for a sample of 146 REITs by year relative to the IPO. Consistent with the model of firms acquiring assets with secured debt and transitioning to unsecured debt to improve investment incentives, REITs often start out with only secured debt. Among firms with a defined secured debt ratio repor ted in the year prior to the IPO, 35 (75%) are entirely secured or have de minimis unsecured debt while only 12 have more than 5% of their debt unsecured. At the end of the IPO year, 62% of the firms with debt are entirely secured. Two patterns emerge. First, the fraction of fi rms with all of their debt secured shrinks steadily in the years immediatel y following the IPO. Second, beginning in the year of the IPO, the mean size of the firms using unsecured debt is much larger than th at of firms using only secured debt. Consistent with the models prediction that firms will transition to recourse debt when they become sufficiently large, only 34% of firms are using only secured debt four years after the IPO. In that year, the mean size of secured firms is $594 million while the mean size of firms using unsecured debt is $1,833 million. The firms that do not rely exclusively on non-re course debt have considerable amounts of unsecured debt outstanding. The mean percentage of total debt that is secured, conditional on not relying entirely on secured debt, is around 50%. Consistent with mode l, firms either rely exclusively on secured debt financing or have a sizeable amount of unsecured debt outstanding. The advantage of using unsecured debt in this model comes from the having a diversified set of
39 projects serving as colla teral for the unsecured debt. This advant age is only realized when a large number of projects serve as collateral for the uns ecured debt. Thus, our model predicts that firms will use large amounts of unsecured debt when th ey deviate from using all secured debt. Analysis of Recourse St ructure Transitions The data presented in Table 2-2 indicate that REITs typically initially borrow on a nonrecourse basis. At some point, some these firms borrow on an unsecured basis. The following analysis looks more closely at these recourse structure transitions. Specifically, the firm-year observations are coded according to whether the firms debt was (a ) entirely secured or (b) had some unsecured debt.11 The 1165 firm-years with necessary data are broken down into four categories: all non-recourse secured observations: firm-years where the firms debt was entirely secured (391 firm years), event year observations: 92 firm-years where the firm has some recourse debt in its capital structure and had all non-recourse debt the prior year (i.e., the firm transitioned into borrowing on an unsecured basis), post-event observations: 655 firm-years following an event year where the firm continued to use at leas t some unsecured debt, second event observations: 27 firm-years where a firm transitioned back to borrowing exclusively on a non-recourse ba sis after having used at least some unsecured debt. Table 2-3 presents mean and medians for each of the four categories for leverage (total debt to total assets), s ecured debt ratio (secured debt to total debt), the percent change in the total dollar amount of secured debt outstanding from th e year prior and the percent change in total 11 A firm with a nominally unsecured line that was small re lative to either the size of the firm or the value of a single asset is characterized as having all of its debt secured.
40 assets from the year prior. By construction, the pe rcentage of the debt that is secured is 100% for the all non-recourse and second event, observations and less than 100% in the other categories. Consistent with the model prediction that th e advantages of unsecure d debt result from the diversification effect of bundling assets, the transition to unsecured debt financing leaves firms with a significant amount of unsecured debt. The mean proportion of the debt that is unsecured is 31.3% at the end of the year of the transition to unsecured debt financing. By definition 100% these firms debt was secured at the prior year end. The nature of the transition to using unsecured debt suggests that the tr ansactions costs associated with retiring secured debt are high. During the year in which firms transition to using unsecured debt, the median change in the value of secured debt is 1.7% Most firms do not buy back an y non-recourse secured debt. However the median change in total assets is 44.4%. Thus, the typical firm transitions to recourse debt financing by financ ing the purchase of a large number assets with unsecured debt. The results of a systematic anal ysis of the factors that influence the probability that a firm transitions to using unsecured de bt are reported in Table 2-4. Specifically I estimate a probit model that includes all the observa tions (483 firm-years) where the fi rm started the year with all non-recourse secured debt. These firms are assigne d the value of one if they transition to unsecured debt during the year and zero if their debt remains entirely non -recourse secured. If firms transition to unsecured debt to take advantage of the dive rsification effect of unsecured debt then we should expect the likelihood of a transition to usi ng unsecured debt to be positively related to the firm size as firm size is a proxy fo r the number of projects. Recognizing that some firms may find it costly to retire their secured debt we would expect th at the likelihood of a transition to using unsecured debt to be positively related to grow th during the year (the percent change in firm size) and leverage. If the firm is more leveraged, the secured debt is riskier and
41 the cost of buying back the debt should be highe r. The probit model incl udes year and propertytype dummy variables as control variables. The results of the probit analysis reported in Table 2-4 are consistent with the predictions of the model. The coefficients on the firm size a nd change in firm size variables are positive and statistically significant at the 1% and 5% leve ls respectively. The coefficient on the leverage variable is negative and signi ficant at the 1% level. Secured Debt Financing and the Underinvestment Problem The final analysis uses this data to test th e implications of the Stulz and Johnson (1985) model. Stulz and Johnson (1985) sh ow that financing an investment with secured debt when the firm has unsecured debt outstanding mitigates the underinvestment problem by avoiding wealth transfers to existing unsecured lende rs. It follows that firms with more financial leverage have a greater debt overhang problem and thus a greater incentive to finance acquisitions with secured debt. I test this conjecture by estimating a probit model on the sample of all firms that start a particular year with at least some unsecured de bt. The sample includes the 646 firm-years where the firm started the year with some unsecured de bt. These firms are assigned the value of one if they increase the fraction of their debt that is se cured debt and zero if th ey reduce the fraction of their debt that is secured. The explanatory variables include the fi rms financial leverage and the fraction of the firms debt that is secured and year and property type dummy variables. The results reported in Table 2-5 support the predictions of the St ulz and Johnson (1985) model. The probability that a firm increases its reliance on secured debt financing is an increasing function of its leverage at the start of the year. The leverage variable is positive and statistically significant at the 1% level. Firms w ith large amounts of leverage face a greater debt
42 overhang problem and they mitigate this problem by financing additional investments with secured debt. Conclusion I provide a model of the choi ce between non-recourse secure d debt and recourse debt (unsecured debt or secured debt w ith recourse) by firms that are se quentially acquiring assets and then making investment choices once those asse ts have been acquired. Non-recourse secured debt financing eliminates the underinvestment problem. However, uns ecured debt provides superior post-acquisition incentives for owners of assets that might re quire non-contractible ongoing investment or that can be easily modi fied. Specifically, with unsecured debt the individual projects are bundled as collateral. B undling the collateral lowers the face value of the debt and improves incentives to invest in the as sets in place. When the collateral is bundled the incentives to modify the assets in pl ace (asset substitution) are reduced. The relative merits of recourse and non-reco urse debt lead to the prediction that nonrecourse debt is optimal for firms engaged in the acquisition of assets which have little need for non-contractible ongoing investment. Consistent with this prediction REITs use significant amounts of non-recourse secured debt. Since the advantages of using uns ecured debt increase with the number of projects that are bundled as collateral, the m odel predicts that firms which find non-recourse secured debt financing optimal will transition to unsecured debt when they acquire a sufficient number of assets. The recourse structure, mix of recourse and non-recourse debt, of R eal Estate Investment Trusts, publicly traded entities that use significan t amounts of non-recourse debt, are shown to be consistent with this model along several dimens ions. Consistent with the borrowers stronger incentive to modify the asset, industry comme ntary indicates that co venant protection is considerably tighter in commercial real mortga ge contracts (non-recour se secured financing)
43 relative to REIT unsecured debt contracts and that the resulting limited operating flexibility is a drawback to non-recourse financi ng. Further industry commentary indicates that higher quality assets are financed with secured debt while lo wer quality assets provide the collateral for unsecured debt. Lower quality assets (1) are mu ch more likely to requ ire non-contractible investment to reposition the asset (the model show s that investment incentives for these kinds of investments are distorted with non-recourse debt financing) a nd (2) have returns driven by idiosyncratic factors and thus benefit from the di versification effects of bundling the collateral. The formal statistical analysis provides addi tional evidence consistent with the model. First, most REITs go public with all non-recourse secured debt. Second, when firms transition to unsecured debt they end up with a large amount of unsecured debt so as to capture the diversification benefits of bundli ng the collateral. Third, the probabi lity that a firm transitions to unsecured debt financing is increasing in the size of the firm, the grow th during the year and decreasing in leverage. The fact that firms are more likely to transi tion to unsecured debt financing in year were they grow a lot and when they are not highl y leveraged is consistent with the notion that it is expensive to buy back secu red debt when the firm is highly leveraged. Finally, I show that firms with unsecured debt outstanding ar e more likely to borrow in the future using secured debt if they are highly le veraged. This evidence is consistent with the predictions of the Stulz and J ohnson (1985) model. Firms finan ce additional investments with secured debt in order to mitigate the wealth transfer to existing le nders and the wealth transfer is larger when the firm is highly leveraged.
44 Table 2-1. Summary leverage and secured debt to total debt stat istics for REITs and industrial firms. N Mean Std. dev. Percent at min. 25th pctile. Median 75th pctile. Percent at max. Industrials total debt to total assets 63,662 27.5% 26.2%12.9%4.6%22.4% 40.9% 3.7% REITS total debt to total assets 1,277 47.4% 23.1%5.9%36.1%48.4% 61.9% 2.4% Industrials secured debt to total debt 55,493 33.0% 36.2%28.0%0.0%16.3% 65.0% 3.6% REITS secured debt to total debt 1,202 66.8% 32.5%4.9%40.6%78.1% 96.5% 8.1% Data is for the 1990 to 2003 timeframe and is drawn from COMPUSTAT. Firms with no total debt have undefined secured debt to total debt ratios. The industrial firms are all firms with a current SIC code (DNUM) between 2000 and 5999 with data available to cal culate the ratios. There are 63,662 firm-year observations for whic h leverage ratios can be computed and 55,493 firm-year observations for which secured debt to total debt ratios can be computed for industrials. For REITs, there are 1,277 firm -observations for leverage ratios and 1,202 observations for secured debt to total debt ratios. Ratios greater than 100% are set equal to 100%.
45 Table 2-2. Evolution of the rec ourse structure of REIT debt. Leverage (TD/TA) Secured debt ratio (SD/TD) conditional on SD/TD>=95% Secured debt ratio (SD/TD) conditional on SD/TD<95% N Mean leverage Median leverage N Mean secured debt ratio Median secured debt ratio Mean total assets ($mm) N Mean secured debt ratio Median secured debt ratio Mean total assets ($mm) Percent of firms all secured Pre-IPO 49 69.0% 58.9% 3599.5%100.0%933.71263.8%87.0% 464.5 74.5% IPO 146 40.5% 40.8% 8499.9%100.0%409.95156.9%62.2% 748.2 62.2% IPO+1 145 45.1% 46.4% 8299.7%100.0%433.55955.9%58.7% 1,077.0 58.2% IPO+2 141 46.0% 46.2% 6899.5%100.0%452.66756.2%58.8% 1,127.7 50.4% IPO+3 129 45.2% 47.6% 5399.6%100.0%597.07247.8%51.9% 1,383.4 42.4% IPO+4 127 46.5% 47.5% 4199.7%100.0%594.08047.0%46.8% 1,833.0 33.9% IPO+5 124 46.9% 47.0% 4199.5%100.0%472.07747.7%45.8% 2,298.8 34.7% IPO+6 118 49.7% 48.8% 4299.7%100.0%559.47145.1%40.7% 2,585.8 37.2% IPO+7 105 49.5% 49.2% 3699.4%100.0%800.66544.2%39.0% 2,146.0 35.6% IPO+8 89 52.3% 51.7% 2899.4%100.0%789.65941.1%33.8% 2,535.0 32.2% IPO+9 87 51.5% 52.5% 2699.4%100.0%863.05943.4%38.2% 2,739.6 30.6% IPO+10 61 52.0% 53.5% 1899.7%100.0%796.04145.0%37.1% 2,718.4 30.5% IPO+11 42 48.3% 51.4% 1999.8%100.0%990.92243.6%35.9% 2,248.3 46.3% IPO+12 40 49.7% 53.6% 1499.7%100.0%662.72550.3%50.4% 2,589.0 35.9% IPO+13 37 48.6% 52.0% 1699.5%100.0%679.71946.4%47.7% 3,170.7 45.7% All FirmYears 1,440 47.9% 48.7% 60399.6%100.0%595.377948.3%47.0% 1,959.2 43.6% Descriptive statistics for leve rage ratios and secured debt to total debt ratios are drawn from SNL DataSource for equity REITs excludi ng healthcare and lodgi ng. Statistics are reported relative to the IPO year, which is se t to 1990 for firms that went public prior to 1990. Leverage ratios are reported for all the fi rms in the sample. Secured debt ratios are only reported for firms with posit ive leverage. The percentage of the debt that is secured (secured debt ratio) is reported for the full samp le and for the sub-sample of firms that do not rely exclusively on secured debt.
46 Table 2-3. Recourse st ructure transitions. Leverage (TD/TA) Secured debt ratio (SD/TD) Change in secured debt Change in total assets N Mean Median Mean Median Mean Median Mean Median All secured 391 52.5% 52.4%99.3%100.0%12.3%5.7% 24.9%9.8% Event year 92 46.5% 44.7%68.7%75.1%9.0%1.7% 69.1%44.4% Post-event 655 49.6% 48.6%46.3%42.9%5.0%1.3% 21.8%11.0% Second event year 27 53.0% 54.8%96.7%100.0%20.5%12.9% 25.0%12.8% Data is 1,165 firm-years from SNL DataSource. Firm -years are categorized by whether their liability structure includes uns ecured debt. Observations are e liminated when (1) there is insufficient data to calculate changes in secured debt and total assets or (2) the firms total assets are less than $50 million. Observations are classifi ed as all secured if 100% of their debt is secured. Event year observations represent th e year during which firms began using some unsecured debt. Post-event firms are those that have continuously maintained some unsecured debt in their capital structures after starti ng to use unsecured debt. Second event year observations are the firm-years where the REIT revert ed to an entirely secured liability structure after having had some unsecured debt. Change in secured debt and change in total assets are the changes in the book value of those cate gories relative to the previous year.
47 Table 2-4. Probit model of the factors influencin g the transition to unsecured debt financing. Variable Coefficient Intercept -8.89 ** (-5.63) Size 0.70 ** (6.04) Growth 0.31 (2.48) Leverage -1.02 (-2.07) Pseudo R2 0.235 The sample includes 483 observations where the fi rm started the year w ith all secured debt. These firms are assigned the value of one if they transition to unsecured de bt during the year and zero if their debt remains entirely secured. The explanat ory variables are size (log of total assets in thousands of dollars at the beginning of the year), growth (percent change in total assets during the year), and leverage (total debt to tota l assets at the start of the year). The model includes unreported year and pr operty-type dummy variables. Z statistics are reported in parentheses. Coefficients that are different fr om zero at the 5% level are denoted with and coefficients that are different from zer o at the 1% level are denoted with **.
48 Table 2-5. Probit model of increa ses in secured debt financing. Variable Coefficient Intercept -0.44 (-0.83) TD-1/TA-1 0.81* (2.18) SD-1/TD-1 -0.08 (-0.39) Pseudo R2 0.035 This probit model predicts whethe r a firm increases its use of secured debt. The sample includes the 646 firm-years where the firm started the year with unsecured debt: obse rvations are all firmyears with a lagged secured debt to total debt ratio of less than 90% and a lagged total asset value of more than $50 million. These fi rms are assigned the value of one if they increase the fraction of their debt that is secured debt and zero if they reduce the fracti on of their debt that is secured. The explanatory variables are levera ge and fraction of the debt that is secured at the start of the year. Firm year and property type dummies ar e included but not reporte d. Z statistics are reported in parentheses and denotes coefficients that are different from zero at the 5% level.
49 CHAPTER 3 SECURED DEBT FINANCIN G AND LEVERAGE: THEORY AND EVIDENCE Introduction This analysis models how the priority structur e of a firms debt (the mix of secured and unsecured debt) impacts both the asset substitu tion and underinvestment problems and empirical evidence in support of the model. The model c onsiders a firm that has secured and unsecured risky debt claims (original debt) against an asse t in place. The firm has the opportunity to acquire other assets in the future. Ris ky debt outstanding creates an incen tive to over-invest in negative NPV high-risk projects, the asset substitution prob lem, because of the w ealth transfers away from existing lenders. Wealth transf ers to the original le nder lead the firm to under-invest, that is, pass up some low-risk positive NPV projects. The model shows how the extent of the underinvestment and asset substitution problems de pend on the proportion of the original debt that is secured. If the original debt is secured by the asset in place a nd does not have recourse to other assets of the firm, the debt claim is essentiall y project financing, and th e value of the original debt is independent of the nature of any assets acquired in the fu ture and how they are financed. There are no wealth transfers and thus risky debt does not distort the acquisition decision as in the analysis in Berkovitch and Kim (1990) and Ch apter 2 of this study. Chapter 2 shows that non-recourse secured debt financing limits acqui sition incentive distor tions but creates poor incentives to maintain existing assets. Consiste nt with their model, the use of non-recourse secured debt by publicly traded firm s is limited to select industries su ch as real estate investment trusts (see Esty and Sesia ( 2005) and Kolasinski (2006)). Secured debt with recourse to the other asse ts of the firm is the more common form of secured debt financing for industrial firms and is the focus of the model and empirical analysis.
50 When the original debt is secured with recour se, it has a first lien on the asset in place and a claim on the unencumbered portion of any new a sset acquired along with the new lenders that finance the new asset purchase. Underinvestment o ccurs because of the wealth transfer to the original lender in states of nature where the valu e of the original asset is insufficient to make the original debt payment and hence the return on the acquired asset serves as collateral (i.e., the new asset co-insurers the original debt). The ma gnitude of this wealth transfer does not depend on the priority structure of the debt. Secured de bt financing with recourse does not mitigate the underinvestment problem: investments in low ri sk and positive NPV projects co-insure the original debt to the same extent whether the origin al debt is unsecured or secured with recourse. In contrast, the acquisition of a high risk project is likely to re sult in states of nature where the acquired asset return is insufficient to pay th e new lenders. In this case, the new lenders have a claim on the original asset when the original de bt is unsecured and a wealth transfer from the original lender occurs: there is an asset subst itution problem. However, if the original debt is secured by the assets in place, there is no weal th transfer. Thus, borrowing on a secured basis mitigates the asset substitution problem. In summ ary, secured debt with recourse financing does not mitigate the underinvestment problem but does mitigate the asset substitution problem. The analysis extends this insi ght into a model of the relati on between the proportion of the debt that is secured and debt capacity. Speci fically, I show that the borrower can increase leverage without increasing the incentive to invest in high risk assets by increasing the proportion of the original debt th at is secured. That is, if a fi rm converts some of its unsecured debt into secured debt, increasing the proportion of its debt that is secured, financial leverage can be increased without incr easing the wealth transfer away from the original lender associated with the purchase of risky asset (i.e., the asset substitution problem). The borrowers debt capacity is
51 positively related to the proportion of the debt that is secured as the agency costs of debt are lower for any given level of leverage when a greater proportion of the debt is secured. The model in this paper is related to other m odels where secured debt reduces debt related investment distortions. The models of Ber kovitch and Kim (1990) and Chapter 2 focus on nonrecourse debt. In the model closest to this one, Stulz and Johnson (1985), financing investments with secured debt mitigates the underinvestment problem when a firm has unsecured debt in place. Their model does not obviously offer a co mpeting explanation for one of the central predictions of this model: the pr oportion of debt that is secure d should be positively related to leverage. More importantly, in Stulz and J ohnson (1985) follow-on secured debt financing mitigates the underinvestment problem. An important testable implication of our model is that secured debt is not used to mitigate the underinvestment problem. In Triantis (1992), secured debt mitigates an asset substitution problem because with secured debt the borrower cannot sell an asset and buy another without repaying the debt. Our model does not consider this possibility. In Triant is (1992), the fact that the asset is pledged as collateral eliminates the borrowers ability to asset substitute. It would follow that the extent that the asset substitution problem is eliminated, and hence debt capacity increased, depends on the proportion of assets that are encumbered. In this model, the asset substi tution problem mitigation depends on the proportion of the debt that is secured.12 The models two predictions about the relations hip between the prior ity structure of the borrowers debt and leverage are supported by an analysis of firms w ith data available on COMPUSTAT between 1985 and 2004. First, I find that the fraction of the firms debt that is secured is positively related to financial leve rage controlling for the variables found to be 12Other models provide a role of secured debt in a framework where the borrower has private information: Besanko and Thakor (1987a), Be sanko and Thakor (1987b) and Bester (1985).
52 significantly related to leverage in previous work. The positive relation between the fraction of the debt that is secured and leve rage adds to the literature that documents a relationship between the nature of a firms debt and its financial le verage. Faulkender and Pe tersen (2006) find that firms that have access to public bond markets ha ve higher financial leverage. Johnson (2003) finds that the negative impact of growth opti ons on leverage is atte nuated by shortening the maturity of the debt. Billet, King and Mauer (200 7) find evidence consistent with tighter debt covenants mitigating the impact of growth options on leverage. The second prediction of the model is that, in contrast with the documented impact of debt maturity and covenant protect ion on leverage, securing debt does not increase debt capacity by mitigating the impact of growth options on leve rage. This prediction is also supported by the data. Specifically, the model of firm leverage is estimated for sub-samples of the data sorted by market-to-book ratio. The coefficient on the fracti on of debt that is secured is lower for the higher market-to-book ratio sub-samples and statis tically insignificant for the highest market-tobook sub-sample. Our analysis also sheds light on the relations hip between tangible assets and leverage documented in the literature originally by Titm an and Wessels (1988). Titman and Wessels (1988) point out that the fixed na ture of tangible assets makes them better collateral to loan against. Tangible assets may also increase debt capacity because they can be redeployed by the lender (Williamson 1988 and Pulvino (1998)). Since it is easier to perfect liens against tangible assets one would expect, and I find that, the propor tion of debt that is secured is related to the amount of the borrowers fixed assets. This calls into question whether the relationship between secured debt and leverage is an artifact of the impact of fixed assets on debt capacity or vice versa.
53 Our analysis suggests that the relationship betw een the fraction of the debt that is secured and leverage is not an artifact of the impact of fixed assets on debt cap acity. Specifically, I find that (1) fixed assets increase leverage when they are not secured and (2) actually securing fixed assets further increases leverage. The proportion of a firms assets that are fixed and the proportion of the debt that is secured are both po sitively related to leverage in models that include both variables. It is possible that some assets labe led as fixed on accounting statements are more tangible and easily redeployed than ot hers and these assets are more often used as collateral for secured debt. To address this I deco mpose the fraction of debt that is secured into an annual industry mean component and a firm specific deviation from that mean. Both components are positively related to leverage. To the extent that th e nature of the assets within an industry are similar, the firm year deviat ion from the industry mean, which is positively related to leverage, is less likely to be driven by th e nature of the assets an d more likely to be the result of the firms desire to minimize the agency costs of debt associated with higher leverage. Finally, I run separate regression s of secured debt to total asse ts and unsecured debt to total assets: the components of leverage. The point estimates on the market-to-book ratio and tax variables are very similar in the two models. Th e fixed asset ratio variable is significantly positive in both the secured debt to total assets and unsecured debt to total assets model. Firms borrow more against hard assets even when they are not pledged as collateral. The point estimate on the fixed asset ratio is approxi mately twice as large in the secured debt model. When the amount of secured as a fraction of total assets is added to the unsecured debt model, the coefficient is significantly negati ve but greater than negative one : as the model predicts, secured debt crowds out unsecured de bt but not one for one.
54 Estimating separate models of secured debt to total assets and unsecured debt to total assets provides insights into th e relationship between profitability and leverage. The coefficient on profitability is negative and significant, the relationship found in other work analyzing leverage, in the model of unsecured debt to tota l assets. However, profit ability is positive and significant, the more theoretically plausible relations hip, in the secured debt to total assets model. Our results shed some light on the findings of Faulkender and Petersen (2006) that firms with access to public debt markets are more leve red. Specifically, I estimate the leverage model on the sub-sample of firms that have public de bt and the sub-sample that do not. The point estimate on the fraction of debt secured variable is almost three times as large for the sub-sample of firms with access to public debt markets. Further, the public debt dummy is positive and significant in the model of unsecured debt to to tal assets and negative and significant in the model of secured debt to total assets. Firms with public debt access rely less on secured debt financing but secured debt financ ing has a greater impact on levera ge for the firms with access to public debt markets. Secured debt does not cr owd out unsecured debt financing as much for firms with public debt access. This suggests pub lic lenders may be more willing than private unsecured lenders to provide debt ca pital on subordinated basis. The final analysis reports the re sults of an instrumental vari ables regression to control for the potential endogeneity of the fraction of the debt secured. In our model the costs of perfecting the collateral impact the amount of secured debt financing. Thus, I use the components of property, plant and equipment as instruments for th e fraction of debt that is secured given the costs of perfecting different components of ta ngible assets should vary. These data are only available for the early part of the sample period. For the sub-sample where data are available, the estimates of the instrumental variables regr ession confirm the findings of the OLS results.
55 Secured Debt and Debt Capacity This section models the relationship between th e proportion of a firms debt that is secured and the extent of risky debt inve stment distortions. The firm ha s an asset in place with a risky payoff. The purchase of the original asset is financed with some combination of recourse secured debt and/or unsecured debt. Subsequently, the firm has an opportunity to acquire a second asset with a stochastic payoff. I then derive the relationship be tween the proportion of the firms debt that is secured debt and th e extent of the asset substitution problems and underinvestment problems. Model Set Up. At t=0 the firm purchases an asset with some combination of secured debt, unsecured debt and equity. The secured debt has recourse to other assets subsequently purchased. New investments do lead to wealth transfers when non -recourse debt financing, project financing, is used. This analysis focuses on secured debt with recourse as non-recourse debt financing is not a common source of funding for public firms (see Esty and Sesia (2005)). The model from Chapter 2 motivates why non-recourse debt financing is limited. The outstanding debt matures at t=2 and no coupon payments are required prior to maturity. The face value of the original secured debt is denoted DS and the face value of the original unsecured debt is denoted DU. The face value of the firms total debt at t=0, denoted D0, equals D0S + D0U. At t=1, the firm has the opportunity to purchase a second asset. The original debt contracts do not contain cont ractual provisions constraining th e nature of the second asset. Further, once the first asset is ac quired it cannot be sold and replaced with another asset. All debt contracts mature at t=2 so there is risky debt outstanding when the firm purchases the second asset. The risk free rate of interest is zero and all agents are risk neutral.
56 For simplicity of exposition, the debt contracts are not priced. We are concerned with the relation between the agency costs of risky debt and the proportion of the debt that is secured versus unsecured. In particular, I examine how wea lth transfers to or away from the original lenders associated with new i nvestment depend on the seniorit y structure of the debt and leverage. The wealth transfers depend on th e face value of debt: the values of D0U and D0S. Solving for D0S and DU as functions of the amount initially borrowed, the distri bution of payoffs to the original asset, and the distribution of projects available to the firm would needlessly complicate the analysis. The value of the asset purchased at t= 0, the original asset, is denoted A0. I assume that the original asset is a tangible asset and the entire value of the asset can be pledged as collateral. The value of the original asset at t=2 is either A0 + X or A0 X. The firms original debt is risky: A0 + X > D0 > A0 X. The value of the second asset, which can be purchased at t=1, is denoted A1. The value of the second asset at t=2 is either A1 + Z or A1 Z. Let D1 represent the face value of the debt used to finance the purchase of the second asset. The bo rrowing at t=1 can be further decomposed into D1U + D1S = D1. Assume that D1 > A1 Z, that is, the ne w debt would be risky if the second asset were a standalone project. Underinvestment The original debt has a claim against the sec ond asset when the first asset turns out to be worth A0 X. This claim creates the underinvestmen t problem. Specifically, the original lenders have a claim of D0U + D0S (A0 X) on the second asset. The wealth transfer associated with an investment in the second asset depends on the si ze of the claim on the second asset and how the second asset is financed. The size of the claim depends on the amount borrowed, D0U + D0S, but not on the proportion of the debt th at is secured or unsecured. The proportion of the original debt
57 that is secured does not impact the wealth transfer to the original lender. Thus, the extent of the underinvestment problem depends on the amount borro wed against the original asset and not the proportion of the debt that is secured. If the second project is equity financed, the original lenders claim is a first lien on the second asset. If the second asset is financed with equal priority debt, the original lenders claim is pari passu with subsequent bond holders and th e wealth transfer to th e original lenders is smaller. If the new debt is secured by the sec ond asset, then the orig inal bondholders have a subordinated claim against the seco nd asset. In this case, as shown by Stulz and Johnson (1985), the underinvestment problem is mitigated by secu red debt financing of new investments when the existing debt has recourse to subsequently acquired assets. Asset Substitution Problem The asset substitution problem comes about becau se the lenders that fund the second assets share a claim with the original unsecured lenders against the first asset in the good state that is valuable when the second asset performs poorly. Th is claim can result in a wealth transfer away from the original unsecured lender The extent of the wealth tran sfer is limited by the amount of secured debt used to finance the first asset. Spec ifically, note that before the acquisition of an additional project, the original uns ecured lender would be paid in full in when the first asset performs well. If the second asset is acquired and performs poorly, then the second le nder has a claim of D1U (A1 Z) against the first asset to the extent th at it is not encumbered by secured debt. The first unsecured lender and the second unsecured lender will share the total payoff not encumbered by secured debt claims, A0 + X + A1 Z D0S D1S, in proportions D0U / (D0U + D1U) and D1U / (D0U + D1U). If the second asset is acquired, the value of the first unsecured lenders claim becomes:
58 S S U U UD D Z A X A D D D1 0 1 0 1 0 0 (3-1) Equation 3-1 states that the total payoff to the two projects less any secured debt claims is allocated pro rata to the unsecured claimants in proportion to the face am ounts of their unsecured claims. If the first asset performs well, withou t a pari passu claim by the second lender, the first unsecured lender is paid in full. Secured debt against the first asset naturally would have been paid in full. With the second project in plac e, the original unsecured lender has a proportional claim on the unencumbered assets (i.e., the to tal payoff of all the pr ojects net of secured claims).13 The wealth transfer, the source of the asset s ubstitution problem, is the difference between (1) the original lenders claim when the first asset performs well and there is no second asset purchase, D0U, and (2) the value of the or iginal unsecured lenders cl aim when the original asset performs well and a second asset is acquired a nd performs poorly. Th e wealth transfer, WXFER, is: S S U U U U XFERD D Z A X A D D D D W1 0 1 0 1 0 0 0 (3-2) 13 For simplicity of exposition, equation 3-1 assume s that the secured claims on each projects payoff are less than each projects payoff. Vi olating this assumption would not qualitatively impact the results, but would have the unpaid porti on of the secured debt be treated as additional unsecured debt. Because we are interested in case s in which the first project performs well, we know the secured portion of the or iginal debt must be less than the good payoff to the first project. Further, assuming all the debt f unding the second project is unsecured is more conservative with respect to wealth transfers from the first lender. The use of secured debt to fund the second asset would allocat e a higher proportion of the total payoff of the two projects to the second lender(s) before proportionally alloca ting the remainder to the unsecured lenders.
59 The wealth transfer is limited by th e initial amount of secured debt, D0S. The ability to extract wealth from the original lenders by investing in a high risk asset is limited by the amount of secured debt in the initial capital structure. Thus, the asset substitution problem is limited by the amount of secured debt outstanding. The following shows that increasing the amount of secured debt in the initial capital structure increases debt capacity in the sense that by increasing the fractio n of the debt that is secured, the firm can increase to tal debt without increasing th e asset substitution problem. Setting the total differential of th e wealth transfer equal to ze ro results in the amount, R, by which unsecured debt must be reduced when a dolla r of secured debt is a dded to keep the wealth transfer constant: 2 0 11 1U U XFERD D W R (3-3) R is less than one for all plausible values of th e parameters. Therefore, the wealth transfer can be held constant with a reduction in the face value of unsecured debt of less than one dollar when the face value of secured debt against th e first project is increased by one dollar. Replacing unsecured debt with secured debt will always allow more debt for a given level of wealth transfer. Model Discussion and the Equilibrium Leverage and Level of Secured Debt The two key results of the model are that in creasing the proportion of secured debt in a firms liability structure (1) mitigates the asset substitution problem but (2) does not mitigate the underinvestment problem. Thus debt capacity increases with the propor tion of debt that is secured. That is, with greater reliance on secured debt financing the firm can increase leverage without increasing the as set substitution problem. Thus, the model suggests that if a firms
60 leverage decision trades off benefits of leverage (tax deductibility of interest payments, avoidance of issuing undervalued equity or el imination of free cash flow) against risky debt related investment distortions, then firms with greater reliance on secu red debt should have greater financial leverage as the risky debt related investme nt distortions are lower. The model provides a clear benefit of secured de bt financing. However, there is no cost of secured debt that would motivate the use of uns ecured debt. The most obvious limitation of the amount of secured debt is a lack of assets that can be pledged as collat eral. Further, there are other costs of secured debt contracting that are generally not present with unsecured debt discussed in detail in Stulz and Johnson (1985) and Triantis (1992). For example, there are expenses associated with the attachment and perfec tion of the security intere st. There is a loss of flexibility with secured debt financing as debt contracts must be renegotiated before a secured asset is sold. Finally, secured lenders have an in centive to move for liquidation of critical assets, perhaps at fire sale prices, in financial dist ress (for example see Brown, James and Mooradian (1994)). Comparison to Other Models Other theoretical models consider the role of secured debt. Our model is closest to Stulz and Johnson (1985) in that it considers how the seniority of debt impacts wealth transfers associated with new investment. Stulz and Johnson (1985) show that the underinvestment problem is limited when the second asset is fina nced by debt that is s ecured by the second asset or debt that is senior to the original debt. In this model, the extent that the existing debt is secured impacts the wealth transfers asso ciated with new investment. In Triantis (1992) security protects the borro wer from asset substitution with regard to assets in place as the borrowe r cannot sell an asset and buy a higher risk asset without first paying off the secured debt. In Triantis (1992) th e extent of the asset substitution problem with
61 regard to assets in place depends on the proporti on of assets that are encumbered by secured debt. In our model, the asset s ubstitution problem depends on the proportion of the original debt that is secured. Other models, Bester (1985), Besanko and Thakor (1987a), Besanko and Thakor (1987b) and Chan and Thakor (1987), allow for private in formation about the borrowers prospects. In these models, pledging collateral signals favorab le private information about the borrowers prospects. These models do not yield competing expl anations for the predictions of this model. Data and Preliminary Analysis This study uses COMPUSTAT data for the years 1985 through 2004. The sample excludes observations for regulated firms (his torical SIC codes between 4900 and 4939) and for financial firms (historical SIC codes between 6000 and 6999).14 Leverage is defined as total debt divided by book debt plus the market value of eq uity. The sum of book debt and market equity, firm value, is obtained from total assets (dat a6) minus common equity (d ata60) plus the product of fiscal year closing price (data199) and shares outstanding (data25). Tota l debt is the sum of long-term debt (data9) and debt in current liabilities (data34). B ecause debt in current liabilities includes debt due in one year (dat a44), total debt is adjusted downward by the value of debt due in one year to avoid double counting if a footnote flag indicates that debt due in one year is included in the long-term debt value. The fraction of the firms debt that is secure d (Fraction Secured) e quals secured debt and mortgages (data241) divided by total debt. The Fraction Secured variable is decomposed into industry and firm-specific components. Industry Fraction Secured is the annual mean of fraction secured for each industry, where industry is defi ned based on historical SIC code. Residual 14 Although there appear to be a few exceptions us ing the simple SIC code screen, COMPUSTAT indicates it does not collect secured debt information for regulated firms.
62 Fraction Secured is the firm-year deviation from the corresponding industry-year mean. Unsecured Debt is the total debt mi nus secured debt and mortgages. The estimated models of leverage include the variable of interest in this study, Fraction Secured, and the following variable s used in previous studies. The Market-to-book ratio (firm value divided by total assets) is th e standard proxy for growth options. Fixed Assets is the ratio of net property, plant and equipmen t (data8) to total assets. Log Sa les, defined as the log of net sales (data12), measures firm size. Profitabi lity is EBITDA (data172+data14+data15+data16) divided by total assets. Volatility is the standard deviation of the three most recent annual first differences of EBITDA that do no t require the current years da ta to calculate divided by the mean total assets over the four yearends used to calculate the first differences in EBITDA15. Following Faulkender and Petersen (2006) I us e a dummy variable fo r the presence of a rating to proxy for capital market access. The vari able is set to one if the firm has an S&P longterm issuer credit rating (data280). Dummy va riables for net operating loss carry forwards and investment tax credits are set to one if the co rresponding data items (dat a52 and data51) have a nonzero value16. After eliminating observations with missing va lues, the final sample includes 64,791 firm years of data. There are 12,930 observations where leverage is zero. The fraction of the firms debt that is secured is undefined in this case. Thus, much of the analysis focuses on the sample of 51,861 firm-year observations where firms have po sitive leverage. Table 3-1 presents summary statistics for the variab les discussed above. 15These variables are among those found to be si gnificant and widely us ed in recent capital structure papers. For example, see Hovaki mian, Opler and Titman (2001), Johnson (2003), Billett, King and Mauer (2006) a nd Faulkender and Petersen (2006). 16Missing values are set to zero.
63 Mean leverage ratios for sub-samples of the data sorted by (1) fraction secured and marketto-book ratio and (2) fraction secured and fixed a sset ratio are reported in Tables 3-2 and 3-3. Table 3-2 shows mean leverage for firms two-way sorted by fraction of debt secured and marketto-book ratio. As shown in previous work, high er market-to-book ratios are largely associated with lower leverage. However, sorting the data by market-to-book ratios reveals some interesting non-linear relationships not reporte d in other work. The maximum m ean leverage (34.1%) occurs for the firms with market-to-book ratios between 80% and 110%. Firms with low market-to-book ratios (less then 80%) have less leverage than firms with market-to-book ratios between 80% and 110%. This provides some evidence consistent w ith the idea that the market-to-book ratio captures growth options. Firms in the market-t o-book ratio range between 80% and 110% have no or minimal growth options a ccording to this measure. Leve rage should not increase as the market-to-book ratio falls furthe r. In addition, it is interesti ng to note that as market-to-book ratios increase beyond 200%, the declines in financial leverage are very modest. This provides preliminary evidence that the fracti on of the debt that is secured is positively related to debt capacity. The mean leverage rati o largely increases with fraction secured within market-to-book subsamples (Table 3-2) and within fixed assets subsamples (Table 3-3). The data sorted in this fashion, as predicted by the model, do not suggest th at securing debt is a mechanism for mitigating the under investment problem. The positive relation between leverage and the fraction of the debt that is secured is most pronounced for the less than 80%, 80% to 110%, and 110% to 150% market-to-book cohorts and minimal for the higher market-to-book cohorts. This is in sharp contrast to the fi ndings of Billet, King and Mauer (2007) and Johnson (2003). Billet, King and Mauer (200 7) provide a two-way sort of leverage by market-to-book and debt covenants and find that str onger covenant protection has larg er positive impact on leverage
64 for high market-to-book firms. Johnson (2003) in teracts the market-tobook ratio with the maturity of the debt in a regres sion analysis and finds that shorte ning the maturity of the debt has a greater positive impact on leverage for higher market-to-book ratio firms. Table 3-3 shows mean leverage for firms twoway sorted by fraction of debt secured and the ratio of net fixed assets to total assets. As shown in other work, most notably Titman and Wessels (1988), leverage increases with the extent that the firm s assets are tangible. More importantly for this analysis, leverage ratios increase with the Fraction Secured holding Fixed Assets constant (i.e., moving down the Fixed Asse ts columns). The relationships in Table 3-3 suggest that firms borrow more (1) against hard assets, perhaps because they can be redeployed by the lender and (2) when these hard, collateralizeb le assets, are actually pledged as collateral. Regression Analysis The empirical analysis begins w ith regression models that atte mpt to explain leverage with variables that have been used in other studies and the fraction of the firms debt that is secured. This analysis finds a positive and significant re lation between leverage and the fraction of the debt that is secured. Next I examine how the impact of the fraction of the debt that is secured on leverage varies across different market-to-book sub-samples in order to determine whether securing debt increases debt capacity by mitigat ing the underinvestment problem. The remainder of the empirical analysis (1) addresses the senior ity structure of the debt and (2) provides an instrumental variable estimation of the leverage model to control for potential endogeneity in the original leverage model. Analysis of Leverage and the Fraction of the Debt That is Secured Table 3-4 reports results of a regression of leverage on the fraction of debt secured and other traditional explanatory variab les. The standard errors ar e robust to clustering by firm. Model (1) uses Fraction Secured as an explanat ory variable. The coefficient on the Fraction
65 Secured is positive and significant at the 1%. The results of model (1) imply that a one standard deviation change in th e Fraction Secured will change Leve rage by 2.75% (the sample standard deviation of leverage is about 19%). The significantly pos itive relation between Fraction Secured and Leverage is consistent with the mode l prediction that securing debt increases debt capacity controlling for other vari ables that effect leverage incl uding the fixed assets ratio. Model (2) splits the Fraction Secured into the industry-year mean and the firm-year deviation from the corresponding mean. Both components are strongly economically and statistically significant. Model (2) implies that one standard deviation ch anges in industry-year mean fraction secured and firm-year deviations will change leverage by 1.91% and 2.20%. The finding that the firm-year deviation from the indus try mean fraction secured is positively related to leverage is important because one might argue that the choice to secure debt is driven by the nature of the firms assets and that the positive relation between the fraction of the debt that is secured and leverage is driven by the nature of th e assets rather than that the assets are secured per se. To the extent that the nature of the asse ts within an industry ar e similar, the firm-year deviation from the industry mean, which is positively related to leverage, is less likely to be driven by the nature of the assets and more likely to be the result of the firms desire to minimize the agency cost of debt associated with the chosen higher leverage. Next the relation between fraction secured and leverage is estimated on firms with rated debt and firms without rated debt subsamples. Results of estimated models for the two subsamples are reported in Table 3-5. There is a stronger rela tion between fraction secured and leverage for firms with public debt market acces s. The point estimates on the fraction secured variable, and the industry-mean and firm-speci fic components of the fraction secured, are approximately three times larger for the rated sample.
66 Having established that leverage increases with the fraction of the debt that is secured, the first prediction of the model, we turn to the models second pred iction: secured debt does not increase debt capacity by mitigating the underinvestment problem. The results presented in Table 3-2 suggest that this is not the case. Leverage does not increase more dramatically with the fraction of debt that is secured for firms with high market-to-book ratios. The results presented in Table 3-2 are confirmed by the cl uster-robust regression analysis fo r groups of firms in different market-to-book regions result s presented in Table 3-6. The coefficients on firm-specific component of the faction secured variable are almost identical for the market-to-book less than 0.8, market-to-book between 0.8 and 1.1 sub-samples and then decline monotonically as the sample market-to-book ratio increases. The coefficient on the industry-mean fraction secured is positive and relatively large for the low and intermediate market-to-book ratio sub-samples (the coeffici ents are .109, .130 and .135 for the market-to-book less than 0.8, market-to-book between 0.8 and 1.1 and market-to-book between 1.1 and 1.5 subsamples respectively). Moving to higher market-t o-book ratio sub-samples the point estimate for the industry mean-fraction secu red variable falls. For mark et-to-book between 2.0 and 2.5 and market-to-book between 2.5 and 3.0 sub-samples, the coefficient on the industry-mean fraction secured variable is insignificant. Finally, the in dustry-mean fraction secured variable is negative and significant for the highest market-to-book firms. Analysis of the Seniorit y Structure of Debt This section provides estimates of separate m odels that explain the two components of the firms leverage: the ratio of secured debt to ma rket value of assets a nd the ratio of unsecured debt to market value of assets. The same explan atory variables, other th an fraction of the debt that is secured, are used to explain both compon ents. Table 3-7 presents the estimated model of secured debt and Table 3-8 presents the estimate d models of unsecured de bt as proportions of
67 firm value. The estimated coefficients on nearly all of the variables are of the same sign in both models. The coefficients are often very similar. For example, the coefficient on the dummy variable for net operating loss ca rry-forwards is 0.011 in the mode l of unsecured debt to total assets and 0.013 in the model of secured debt to total assets. Th e most interesting exception is that the profitability variable is positively relate d to unsecured debt to tota l assets (the result for total leverage found in this and ot her papers) and negatively related to secured debt to total assets (a finding more consistent with proposed theories). The most important result for this analysis is that the ratio of fixed assets to total assets variable is positive and significant at the 1% in both the unsecured debt to total assets and secured debt to total assets models. The finding th at the ratio of unsecured debt to total assets increases with the proportion of ha rd assets indicates that firms borrow more against hard assets even when they are not pl edged as collateral. A second model of unsecured debt to total asse ts adds the proportion of secured debt to total assets (the proportion of assets that are encumbered) as an explanatory variable. The coefficient on secured debt to total assets vari able, is negative and signi ficant: secured borrowing crowds out unsecured borrowing. The coefficient on th e secured debt to tota l assets variable is far greater than -1 (-.08), which is predicted by the model and thus leverage increases with the proportion of the debt that is secured. More im portantly, the ratio of uns ecured debt to total assets remains significantly positively related to the ratio of fixed assets controlling for the proportion of fixed assets that ar e encumbered by secured debt. The finding that the public debt dummy is positive and significan t in the model of unsecured debt to total assets a nd negative and significant in the m odel of secured debt to total assets coupled with the earlier finding that the point estimate on the fraction of debt secured
68 variable is almost three times as large for the sub-sample of firms with access to public debt markets shed some light on the findings of Faulke nder and Petersen (2006) that firms with access to public debt markets are more levered. These results imply that firms with public debt access rely less on secured debt financi ng but secured debt financing has a greater impact on leverage. Secured debt does not crowd out unsecured debt financing as much for firms with public debt access. This suggests public lenders are either more willing than private lenders to provide debt capital on subordinated basis or public lenders rely on more on th e presence of secured debt to protect them against the asset substitution problem. Instrumental Variables Analysis The relationship between a firms leverage and the fraction of that firms debt that is secured could be jointly determined by an unobser ved factor. The most obvious candidate for such a relationship is the nature of the firms assets. That is there are unmeasured characteristics of assets that make good collateral for secured le nding also increase debt capacity. This concern motivates the analysis where the fraction secure d is decomposed into an industry average and deviation from the industry aver age (firm specific) component. Th e idea discussed above is that the nature of the assets is relatively constant across firms in an industry. The positive relation between leverage and industry average secured debt could very well be rela ted to the nature of the assets rather than the fact that the assets are secured. Ho wever, the industry specific component, which is positively related to leverage, seems much less likely to be driven by within industry differences in the nature of the hard assets. I further address this issue by performing a two-stage instru mental variables regression. The model gives us some guidance as the cost of perfecting the collateral influence the attractiveness of secured debt financing. Thus I use components of property, plant and equipment (PP&E) as instruments for the fraction of debt secured. The PP&E components are only
69 available from COMPUSTAT for select years ending around 1997. In the (unreported) first stage, both instruments are str ongly statistically significant. The model explains leverage using the explanat ory variables used in the earlier analysis, including the fraction of debt that is secured. If the fraction of de bt secured is jointly determined with a firms leverage level, the measured e ffects of changing the use of secured debt may include or be biased by firm characteristics which simultaneously influence both the dependent variable, leverage, and one of the explanatory va riables, fraction secured. Using two-stage least squares estimation, predicted levels of fracti on secured are generated using all the exogenous explanatory variables and two inst ruments: the fraction of assets represented by the depreciated value of buildings and the fraction of assets re presented by the depreciate d value of capitalized leases and leasehold improvements. The results shown in Table 3-10 show that (1) the coefficient on the instrument for secured debt financing is positive and significant and (2) the coefficients on the control variab les are qualitatively the same as in the OLS regressions reported earlier. Conclusion This paper provides a model of the seniority st ructure of debt where the asset substitution problem decreases with the proportion of a firms debt that is secured. Sp ecifically, the firm is less likely to purchase a high risk asset in order expropriate wea lth from lenders when the debt structure has considerable secured debt. This id ea is extended to show that debt capacity is increasing in the fraction of the debt that is secured. However, in contrast to shortening the maturity of the debt, increasing the fraction of the debt that is secured does not mitigate the underinvestment problem. An empirical analysis of COMPUSTAT firms supports the two predictions of the model. First, I find a positive and signi ficant relation between leverage and the fraction of the firms
70 debt that is secured. Second, the re lation between leverage and the fr action of the firms debt that is secured is not related to th e borrowers market to book ratio. Thus, as predicted by the model, secured debt does not mitigate the underinvestme nt problem. Secured debt is unlike protective covenants (Billett, Mauer and King (2007)) and de bt maturity (Johnson (2003)) which appear to enhance debt capacity by mitigating the underinvestment problem.
71 Table 3-1. Summary statistics. Sample of firms Variable Obs Mean Std. Min Max Positive leverage Leverage 51,8610.2260.1870.000 0.740 Positive leverage Fraction secured 51,861 0.3310.3670.000 1.000 Positive leverage Industry fraction 51,861 0.3310.1500.000 1.000 Positive leverage Residual fraction 51,861 0.0000.335-0.805 0.913 Positive leverage Market-to-book 51,861 2.0862.9890.528 34.231 Positive leverage Fixed assets 51,861 0.3240.2360.000 0.925 Positive leverage Log sales 51,861 4.8622.486-6.908 12.564 Positive leverage Profitability 51,861 0.0040.494-4.932 0.581 Positive leverage Volatility 51,861 0.1630.2730.004 2.297 All firms Leverage 64,791 0.2050.1900.000 0.740 All firms Market-to-book 64,791 2.1523.0520.528 34.231 All firms Fixed assets 64,791 0.3130.2360.000 0.925 All firms Log sales 64,791 4.7982.556-6.908 12.564 All firms Profitability 64,791 0.0060.484-4.932 0.581 All firms Volatility 64,791 0.1630.2710.004 2.297 Sample is all firms with data available from COMPUSTAT between 1985 and 2004 excluding regulated firms (SIC codes between 4900 and 4939) and financial firms (SIC codes between 6000 and 6999). Leverage is the ratio of book debt to book debt plus the market value of equity. The fraction secured is secured debt divided by total debt. The fraction secured variable is decomposed into industry fraction secured, the a nnual mean fraction secured for all firms in the industry, and residual fraction s ecured, the difference between the firms fraction secured and industry fraction secured. The cont rol variables in the analysis of leverage include market-tobook, the ratio of firm market value to firm book va lue, fixed assets is th e ratio of net property, plant and equipment to total book as sets, log sales is the log of firm sales, profitability is EBITDA divided by total assets, and volatility is the standard deviat ion of the three most recent first differences in EBITDA. Variables using to tal debt in the denominator are undefined for firms with no debt.
72 Table 3-2. Mean leverage and number of firm -year observations conditional on fraction secured and market-to-book. Market-to-Book Fraction secured < 0.8 0.8 1.1 1.1 1.5 1.5 2.0 2.0 2.5 2.5 3.0 > 3.0 All firms Exactly 0 0.224 0.256 0.204 0.147 0.114 0.093 0.086 0.160 1,366 3,569 5,246 3,967 2,174 1,402 5,542 23,266 < 0.1 0.327 0.351 0.264 0.190 0.147 0.117 0.102 0.234 632 3,064 4,528 2,650 1,301 664 1,872 14,711 0.1 0.2 0.284 0.344 0.261 0.179 0.121 0.103 0.071 0.223 307 1,308 1,666 1,016 442 300 743 5,782 0.2 0.3 0.289 0.361 0.275 0.193 0.124 0.106 0.067 0.235 259 1,151 1,332 747 391 215 658 4,753 0.3 0.4 0.313 0.355 0.271 0.184 0.123 0.106 0.064 0.233 245 1,047 1,137 691 341 185 558 4,204 0.4 0.5 0.279 0.358 0.275 0.181 0.123 0.099 0.046 0.221 246 912 1,042 662 321 190 641 4,014 0.5 0.6 0.284 0.358 0.286 0.186 0.120 0.087 0.046 0.222 240 965 1,064 655 328 228 703 4,183 0.6 0.7 0.281 0.356 0.276 0.181 0.129 0.096 0.044 0.218 244 949 1,092 647 357 247 712 4,248 0.7 0.8 0.296 0.345 0.266 0.176 0.126 0.094 0.046 0.218 311 985 1,020 636 366 208 657 4,183 0.8 0.9 0.306 0.365 0.283 0.179 0.122 0.090 0.064 0.243 363 1,231 1,186 700 376 233 525 4,614 > 0.9 0.347 0.368 0.285 0.189 0.143 0.121 0.080 0.257 656 2,112 2,173 1,182 651 366 868 8,008 Exactly 1.0 0.282 0.310 0.232 0.149 0.101 0.086 0.058 0.201 296 805 762 547 268 155 426 3,259 All firms 0.285 0.335 0.255 0.173 0.126 0.100 0.075 5,165 18,098 22,248 14,100 7,316 4,393 13,905
73 Table 3-3. Mean leverage and number of firm-y ear observations conditional on fraction secured and fixed assets ratio. Fixed assets ratio Fraction secured < 0.1 0.1 0.2 0.2 0.3 0.3 0.4 0.4 0.5 0.5 0.6 0.6 0.7 0.7 0.8 0.8 0.9 > 0.9 All firms Exactly 0 0.146 0.140 0.154 0.161 0.170 0.188 0.200 0.228 0.222 0.214 0.160 6,949 4,650 3,388 2,460 1,507 1,155 918 806 835 598 23,266 < 0.1 0.212 0.230 0.226 0.235 0.240 0.248 0.269 0.272 0.303 0.308 0.234 2,942 3,334 2,731 1,875 1,204 859 736 581 345 104 14,711 0.1 0.2 0.159 0.207 0.219 0.246 0.253 0.239 0.259 0.299 0.304 0.241 0.223 900 1,305 1,118 761 529 391 317 253 173 35 5,782 0.2 0.3 0.167 0.194 0.230 0.254 0.273 0.281 0.305 0.316 0.315 0.264 0.235 742 964 899 683 457 365 264 193 142 44 4,753 0.3 0.4 0.147 0.193 0.226 0.254 0.271 0.294 0.317 0.339 0.338 0.327 0.233 668 942 728 600 409 256 218 181 152 50 4,204 0.4 0.5 0.143 0.169 0.208 0.260 0.264 0.304 0.308 0.328 0.323 0.332 0.221 724 861 698 553 378 257 208 157 140 38 4,014 0.5 0.6 0.135 0.162 0.209 0.251 0.281 0.323 0.323 0.318 0.359 0.316 0.222 796 871 678 563 386 296 233 192 123 45 4,183 0.6 0.7 0.136 0.155 0.197 0.227 0.284 0.291 0.341 0.325 0.337 0.299 0.218 720 892 710 564 375 313 233 211 173 57 4,248 0.7 0.8 0.148 0.151 0.181 0.211 0.264 0.276 0.309 0.350 0.325 0.348 0.218 631 834 616 549 414 382 256 252 187 62 4,183 0.8 0.9 0.203 0.175 0.188 0.223 0.252 0.284 0.330 0.371 0.347 0.392 0.243 610 784 744 600 445 402 396 298 248 87 4,614 > 0.9 0.227 0.194 0.201 0.227 0.256 0.284 0.355 0.381 0.374 0.364 0.257 1,023 1,394 1,285 1,098 753 623 597 594 476 165 8,008 Exactly 1.0 0.146 0.132 0.152 0.190 0.227 0.269 0.275 0.289 0.330 0.319 0.201 500 626 513 363 271 210 215 185 247 129 3,259 All firms 0.165 0.177 0.196 0.218 0.239 0.257 0.286 0.307 0.306 0.281 17,205 17,457 14,108 10,669 7,128 5,509 4,591 3,903 3,241 1,414
74 Table 3-4. Determinants of market levera ge using cluster-robust OLS regressions. Leverage (1) Leverage (2) Fraction secured 0.0749*** (17.1) Market-to-book -0.00228*** -0.00215*** (-8.45) (-7.91) Fixed assets 0.102*** 0.0968*** (16.1) (15.3) Log sales 0.00227*** 0.00263*** (4.90) (5.68) Profitability -0.007 04** -0.00831*** (-2.57) (-3.02) Loss carry forwards 0.0211*** 0.0219*** (6.59) (6.79) Investment tax credits -0.0623*** -0.0630*** (-13.2) (-13.4) Rated debt 0.0712*** 0.0714*** (16.4) (16.4) Volatility -0.0267*** -0.0297*** (-5.06) (-5.45) Industry fraction secured 0.128*** (12.2) Residual fraction secured 0.0652*** (14.4) Constant 0.147*** 0.130*** (37.2) (25.4) Observations 43,259 42,735 R-squared 0.08 0.08 Fraction Secured is secured debt divided by total debt. Market-to-book is the ratio of the firm value to the firm book value. Fixed assets is the ratio of fixed assets to total assets. Log sales is the natural log of net sales. Profitability is EBITDA divided by sales. Dummy variables are used for contemporaneous net operating loss carry forwards, investment tax credits and having a long-term issuer credit rating. Volatility is th e volatility of earnings, defined as the standard deviation of changes in EBITDA over the four-y ear period preceding estimation divided by firmmean total assets over that period. Industry fraction secured and re sidual fraction secured represent the decomposition of fraction secured into industry-year means and firm-year deviations from those means. Robust t-statistics are provided in parentheses with *** indicating significance at the 1% level, ** significance at th e 5% level, and signifi cance at the 10% level.
75 Table 3-5. Determinants of leverage by market access subsamples usi ng cluster-robust OLS regressions. (1) (2) (3) (4) Unrated Firms Rated Firms Leverage Leverage Leverage Leverage Fraction Secured 0.0568*** 0.183*** (12.4) (15.1) Market-to-book -0.00210*** -0.00201*** -0.00261*** -0.00245*** (-7.12) (-6.71) (-4.18) (-3.94) Fixed Assets 0.108*** 0. 105*** 0.0822* ** 0.0753*** (15.0) (14.5) (7.14) (6.56) Log Sales 0.00379*** 0.00398*** -0.00245*** -0.00169* (7.37) (7.66) (-2.70) (-1.86) Profitability -0.00243 -0 .00285 -0.242*** -0.242*** (-0.88) (-1.04) (-8.32) (-8.31) Loss Carry Forwards 0.0217*** 0.0228*** 0.0114* 0.0114* (6.15) (6.40) (1.75) (1.74) Investment Tax Credits -0.0643*** -0.0648*** -0.0527*** -0.0541*** (-12.1) (-12.2) (-5.83) (-5.94) Volatility -0.0265*** -0. 0294*** -0.0182 -0.0207 (-4.91) (-5.26) (-0.64) (-0.73) Industry Fraction Secured 0.0987*** 0.237*** (8.48) (11.1) Residual Fraction Secured 0.0493*** 0.171*** (10.3) (13.6) 0.145*** 0.131*** 0.262*** 0.243*** (34.0) (23.6) (28.3) (22.4) Observations 33,324 32,819 9,935 9,916 R-squared 0.06 0.06 0.14 0.15 Fraction Secured is secured debt divided by total debt. Market-to-book is the ratio of the firm value to the firm book value. Fixed assets is the ratio of fixed assets to total assets. Log Sales is the natural log of net sales. Profitability is EBITDA divided by sales. Dummy variables are used for contemporaneous net operating loss carry forwards, investment tax credits and having a long-term issuer credit rating. Volatility is th e volatility of earnings, defined as the standard deviation of changes in EBITDA over the four-y ear period preceding estimation divided by firmmean total assets over that period. Industry fraction secured and re sidual fraction secured represent the decomposition of Fraction secured into industry-year means and firm-year deviations from those means. Models (1) and (2 ) use the sample of firms without public debt ratings. Models (3) and (4) use the sample of firms with public debt ratings. Robust t-statistics are provided in parentheses w ith *** indicating significance at th e 1% level, ** significance at the 5% level, and signi ficance at the 10% level.
76 Table 3-6. Determinants of leverage by mark et-to-book ratio subsamples using cluster-robust OLS regressions. Market-to-book ratio < 0.8 0.8 1.1 1.1 1.5 1.5 2.0 2.0 2.5 2.5 3.0 > 3.0 Industry fraction 0.109*** 0.130*** 0.135*** 0.0893***0.0222 0.0346 -0.0341* secured (3.22) (7.06) (9.97) (5.68) (1.11) (1.53) (-1.93) Residual fraction 0.0639*** 0.0640*** 0.0494***0.0365***0.0146* 0.0224* -0.0140* secured (4.91) (8.01) (7.69) (5.12) (1.77) (1.94) (-1.80) Market-to-book 0.000207 -0.00183*** -0.000823 -0.000479 -0.00129**0.0000225 -0.000543 (0.19) (-2.88) (-1.62) (-0.91) (-2.02) (0.029) (-1.39) Fixed assets 0.0547*** 0.0870*** 0.0763***0.0604***0.0485***0.0427*** 0.0418*** (2.81) (7.57) (9.15) (6.52) (4.07) (2.84) (3.43) Log sales 0.00449*** 0.000944 -0.000115 0.000426 -0.00115 -0.000477 0.00296*** (2.67) (1.01) (-0.17) (0.58) (-1.28) (-0.41) (3.22) Profitability 0.00903 -0.00767 -0.0512*** -0.0391***-0.0430***-0.0263*** -0.0135*** (0.67) (-0.70) (-4.89) (-4.32) (-4.35) (-2.63) (-4.44) Loss carry forwards 0.00572 0.0205*** 0.0231***0.0230***0.0195***0.0263*** 0.00994* (0.59) (3.58) (4.98) (4.49) (3.40) (3.34) (1.82) Investment tax -0.0754*** -0.0547*** -0.0503***-0.0385***-0.0402***-0.0412*** -0.0384*** credits (-3.34) (-5.37) (-8.28) (-6.54) (-7.25) (-5.23) (-8.40) Rated debt 0.174*** 0.107*** 0.0724***0.0700***0.0561***0.0478*** 0.0277*** (9.42) (15.4) (14.1) (11.1) (7.95) (5.19) (3.93) Volatility -0.0491* -0.0710*** -0. 00519 0.0526***0.0523*** 0.0516*** 0.0421*** (-1.70) (-4.31) (-0.44) (4.01) (4.15) (2.75) (6.54) Constant 0.208*** 0.243*** 0.164*** 0.0973***0.0906***0.0595*** 0.0504*** (12.0) (24.2) (22.6) (12.2) (8.89) (6.06) (6.62) Observations 2,876 9,965 11,783 7,216 3,542 1,980 5,373 R-squared 0.08 0.10 0.08 0.09 0.08 0.07 0.08 Industry fraction secured and residual fraction secured represent the decomposition of fraction secured into industry-year means and firm-year de viations from those means. Market-to-book is the ratio of the firm value to the firm book value. Fixed assets is the ratio of fixed assets to total assets. Log sales is the natural log of net sale s. Profitability is EBITDA divided by sales. Dummy variables are used for contemporaneous net operating loss carry forwards, investment tax credits and having a long-term issuer credit ra ting. Volatility is the volatility of earnings, defined as the standard deviation of change s in EBITDA over the four-year period preceding estimation divided by firm-mean total assets over th at period. Robust t-sta tistics are provided in parentheses with *** indicating sign ificance at the 1% level, ** signi ficance at the 5% level, and significance at the 10% level.
77 Table 3-7. Cluster-robust OLS model e xplaining the fraction of debt secured. Fraction Secured Leverage 0.286*** (17.7) Fixed assets 0.0973*** (8.08) Log sales -0.00289*** (-3.48) Profitability 0.0608*** (16.4) Rated debt -0.234*** (-35.4) Constant 0.300*** (45.3) Observations 43,259 R-squared 0.09 Leverage is firm leverage, measured as total debt divided by firm (m arket) value. Fixed assets is the ratio of fixed assets to total assets. Log sales is the natural log of net sales. Profitability is EBITDA divided by sales. A dummy variable is used for the presence of a long-term issuer credit rating. Robust t-statistic s are provided in parentheses w ith *** indicating significance at the 1% level, ** significance at the 5% level, and signi ficance at the 10% level.
78 Table 3-8. Model of unsecured debt to firm value using cluster-robust OLS regressions. Unsecured-to-firm-value Unsecured-to-firm-value Fixed assets 0.0349*** 0.0411*** (7.91) (9.22) Market-to-book -0.00128*** -0.00136*** (-7.05) (-7.46) Log sales 0.00221*** 0.00227*** (6.71) (6.86) Profitability -0.0113*** -0.0104*** (-5.52) (-5.07) Loss carry forwards 0.0110*** 0.0122*** (4.83) (5.29) Investment tax credits -0.0342*** -0.0364*** (-10.4) (-11.0) Rated debt 0.104*** 0.102*** (31.0) (30.2) Volatility -0.0148*** -0.0163*** (-4.10) (-4.48) Secured to firm value -0.0838*** (-11.8) Constant 0.0867*** 0.0911*** (32.8) (33.3) Observations 49,313 49,313 R-squared 0.10 0.11 Fixed assets is the ratio of fixe d assets to total assets. Market -to-book is the ratio of the firm value to the firm book value. Log sales is the na tural log of net sales. Profitability is EBITDA divided by sales. Dummy variables are used for contemporaneous net operating loss carry forwards, investment tax credits and having a long -term issuer credit rating. Volatility is the volatility of earnings, defined as the standard de viation of changes in EBITDA over the four-year period preceding estimation divided by firm-mean to tal assets over that pe riod. Secured-to-firmvalue is the ratio of secured de bt to book debt plus the market value of equity. Unsecured-tofirm-value is the ratio of unsecured debt to book de bt plus the market valu e of equity. Robust tstatistics are provided in pare ntheses with *** indicating sign ificance at the 1% level, ** significance at the 5% level, and significance at the 10% level.
79 Table 3-9. Model of secured debt to firm value using cluster-robust OLS regressions. Secured-to-firm-value Market-to-book -0.00103*** (-6.74) Fixed assets 0.0727*** (15.5) Log sales 0.000640** (2.12) Profitability 0.0110*** (9.06) Loss carry forwards 0.0133*** (6.58) Investment tax credits -0.0262*** (-8.96) Rated debt -0.0209*** (-7.86) Volatility -0.0174*** (-5.98) Constant 0.0539*** (23.2) Observations 49,679 R-squared 0.03 Fixed assets is the ratio of fixe d assets to total assets. Market -to-book is the ratio of the firm value to the firm book value. Log sales is the na tural log of net sales. Profitability is EBITDA divided by sales. Dummy variables are used for contemporaneous net operating loss carry forwards, investment tax credits and having a long -term issuer credit rating. Volatility is the volatility of earnings, defined as the standard de viation of changes in EBITDA over the four-year period preceding estimation divided by firm-mean to tal assets over that pe riod. Secured-to-firmvalue is the ratio of secured debt to book debt plus the market va lue of equity. Robust t-statistics are provided in parentheses w ith *** indicating significance at th e 1% level, ** significance at the 5% level, and signi ficance at the 10% level.
80 Table 3-10. Two-stage least square s model of leverage using net components of property, plant and equipment as instrume nts for fraction secured. Leverage Fraction secured 0.458*** (4.74) Market-to-book -0.00257 (-1.00) Fixed assets 0.0897*** (3.51) Log sales 0.000926 (0.41) Loss carry forwards 0.0441*** (3.51) Investment tax credits -0.0568*** (-2.61) Rated debt 0.147*** (5.20) Volatility -0.0390 (-1.45) Constant 0.00301 (0.076) Observations 2,745 R-squared Standard errors are robust to clustering by firm. The compon ents of property, plant and equipment used are buildings (data155) and leases (data159). Fixed assets is the ratio of fixed assets to total assets. Market-to-book is the rati o of the firm value to the firm book value. Log sales is the natural log of net sales. Profitabi lity is EBITDA divided by sales. Dummy variables are used for contemporaneous net operating loss carry forwards, investment tax credits and having a long-term issuer credit rating. Volatility is the volatility of earnings, defined as the standard deviation of changes in EBITDA over the four-year period preceding estimation divided by firm-mean total assets over that period. Robust t-sta tistics are provided in parentheses with *** indicating significance at the 1% level, ** significance at the 5% level, and significance at the 10% level.
81 CHAPTER 4 ANATOMY OF A RATINGS CHANGE Introduction The financial press is reporting widespread concern by fixed income investors that shareholder-friendly activities wi ll hurt credit quality and bo ndholders. Bloomberg and Moodys Investors Service calculated that the bond market has lost nearly $5 billion in the first eight months of 2006 to share repurchases, special di vidends and cash financed acquisitions (Salas 2006). Numerous structural models of credit ri sk beginning with Bl ack-Scholes (1973) and Merton (1974) apply op tion pricing models to value risky debt17. This application assumes that management does not modify the capital struct ure and thus only changes in the operating environment change credit risk18. There is evidence that gover nance and ownership structures that can affect management incentives, impact bond yield spreads. Klock et al. (2004) find better corporate governance has an economically and sta tistically significant favo rable effect on yield spreads. Particularly, their measure of governan ce improves with increased takeover defenses. Yield spreads appear to reflect protection from the potential w ealth effects of acquisitions. Bhojraj and Sengupta (2003) find that after controlling for other f actors, (1) grea ter aggregate institutional ownership and more outside directors are associated with higher ratings and lower yields for new issues and (2) greater concentration of institutional holdings adversely impacts ratings and yields. With more concentrated sh areholdings, owners may be more likely to take actions which expropriate bondholder wealth. 17Acharya and Carpenter (2002) provides an extensiv e review of the models in this tradition. 18Collin-Dufresne and Goldstein (2001, p. 1930) note that the Merton model does not allow the firm to issue additional debt and demonstrate th at precluding this option generates a downwardsloping term structure of credit spreads for specu lative-grade debt, in conflict with the empirical findings of Helwege and Turner (1999).
82 This study provides direct evidence on the extent that the assumptions of the Merton model do not hold. The analysis examines 418 downgrades and 186 upgrades of industrial firms listed in COMPUSTAT that occurred between 1987 a nd 2004. Most of the rating changes were collected from the period 2002 through 2004. Among downgrades, 18% (75 observations) were primarily driven by management choices and acti ons. Management was a significant contributor for another 6% (27 observations). Among upgrad es, 25% (47 observations) were primarily driven by management choices and actions. Ma nagement was a significant contributor for another 16% (30 observations). Data and Methodology For each rating change, we would like to know the extent to which it was caused by management choices rather than by economic or operating factors. Management choices are high-level financial decisions such as acquisiti ons, repurchases and tender offers. Economic factors include competition, demand and sovereign risks. Operating factors include general operating difficulties as well as restructuring costs and risks. Rating change observations were collected fr om COMPUSTAT. The rating changed if the Standard and Poors long-term issu er credit rating at the end of the year was different than the rating at the end of the previous year. From a total of 808 y ear-over-year rati ng changes, the cause of the change could be determined for 75% or 604 changes. Tabl e 4-1 shows the yearly and directional distribution of all 808 year-over-year rating changes and of the 604 changes to which cause could be assigned. I searched Factiva for relevant stories on each rating change from the sample described above. The ultimate search goal was to find one or more news stories that explained why the firm had been upgraded or downgraded. Because the initial database of rating changes was compiled from COMPUSTAT which uses Standard and Poors credit ratings, the ideal article
83 specifically explained, often quo ting S&P analysts, why Standard and Poors had changed the credit rating. When the date of the S&P rating change c ould be identified but there was no explicit attribution from S&P, there were three alternative sources of attribution. If there was information about the rating having been put on watch by S&P and the resolution of the watch was a change in the rating, any di rect explanation of the cause fo r the watch was presumed to be the cause of the change. Whenever possible, this was verified using the additional sources discussed below. If a firms Moodys credit rating changed in th e same direction and at approximately the same time as the S&P rating and it was generally reasonable to expect that the causes of the ratings changes were the same, the attribution for the Moodys rating change was used when it was not possible to draw inferences directly from S&P information. Absent direct information from S&P and Moody s, the final sources of information were other stories about the company a nd transcripts of earnings confer ence calls. These last sources of information were only used if, based on date s and other cues, there was a clear basis for presuming the cause of the rating change could be correctly inferred from the additional sources. The search for articles was adaptive to minimi ze the probability of missing relevant articles without having to read through hundr eds of articles for each rating change. The initial search expression included all or part of the company na me plus restrictions to only return articles mentioning Standard and Poors or Moodys and ratings.19 Using information from the COMPUSTAT quart erly database, the wi ndow of time for each rating change was narrowed to three months The publication date window for the search 19Specifically, the general form of the initi al search expression was CompanyName and ((Standard and Poors) or S&P or Moodys) a nd (rating or ratings or rtgs or rtg).
84 started the 15th of the month before the fiscal quarter and ended the 15th of the month after the fiscal quarter. In most cases, it was also possible to further pinpoi nt the date of the rating change by searching for articles of less than 20 words that contained the company name or a fragment of the company name. These extremely short arti cles are essentially headlines without any additional text. Almost every rating change in the initial da taset of 808 has one or more headlines for ratings and watch changes. When the search failed to produce any results, the requirement that the articles contain references to ratings was dropped. If the search continued to fail, it wa s rerun with only the company name. Finally, a special effort was made to address concerns that the CONAME (company name) field in COMPUSTAT is currently different than the name that would have appeared in the press at the time of the rating cha nge. This concern applies in two directions: (1) the failure to find any articles about a compa ny could reflect a differe nce between the current company name (CONAME) and the company name at the time of the news event and (2) the search could have found information about th e wrong company in cases such as now-merged companies. Reviewing company websites and also searching for articles using a wider date window and only the company name was used when necessary to address the first concern. The second concern was addressed by comparing deta ils such as the star ting and ending credit ratings. Of the 808 changes, sufficient information wa s available to identify the causes for 604. Most of the 204 changes with uni dentified causes are not missing completely from the article database. Rather, the available articles do not contain sufficient detail to confidently identify the cause of the change. Typical of the 204 is a case where the only direct reference to a credit
85 rating change is in one of the extremely short headline stories discussed above as a means to precisely identify the date of the change. The goal of this analysis is to document and understand why firms credit ratings change. The ultimate question is whether management cau sed the change through a deliberate high-level financial decision or whether the change came from economic or operating factors. The simplest examples of deliberate high-leve l financial decisions that can drive credit rating changes are direct capital structure changing tran sactions. If management uses cash on hand or issues debt to repurchase common stock, the firms credit quali ty will fall as its leverage increases. Conversely, management could issue common stock to redeem debt. It follows that the firms credit quality will improve. While these two ex amples are among the most direct and strongest ways in which management can alter the firms credit risk, there are numerous management actions which can directly change the firms credit quality holding the economy and the underlying business opera tions constant. At the opposite end of the sp ectrum from management-induced credit rating changes are credit rating changes which are caused by outsi de economic factors. Changes in industry fundamentals, either positive or negative, change the credit quality of the firm in fairly obvious ways. For a company with intern ational operations, changes in s overeign risk levels directly change the riskiness of the firms debt. Intern ationally or domestically, a firm operating in a regulated market might see an exogenous shock to firm value and credit risk with a changing regulatory climate. Operating performance and risks are grouped with economic factors. Examples of operating factors are earnings or lo sses, as well as the predictability of those results. While some operating factors such as costs or risks associated with the restructuring of operations could be
86 viewed as management choices, the focus of this analysis is to treat operations as exogenous to credit ratings. Drawing the line between manage ment choices and other causes in this way gives conservative estimates of when firms deviate fr om the assumptions of structural credit risk models. Table 4-2 details potential causes of ratings changes and their assignments to management, economic or operating factors. Using these causes as a guide, an overall scor e of 1 through 5 was given to each year-overyear rating change. Changes attributable entir ely to management are scored 1 while changes attributable entirely to economic s and operations are scored 5. Table 4-3 summarizes the scoring system. The assignment of the score generally re flects an aggregation of the factors from Table 4-2. However, it is sometimes possible to a ssign a score even though th ere is not underlying detail such as that described in Table 4-2. Fo r example, if a firm is downgraded because of a series of shareholder friendly activities in th e absence of any underlying business changes, it would be possible to assign a score of 1 even if there is no further detail about those shareholderfriendly activities. Finally, note that because the basis of the an alysis is year-over-year rating changes, when there were two or more rating changes in a fisc al year, the overall sc ore was based on all the changes. Results Management actions frequently drive credit ra ting changes, both up and down. Table 4-4 shows the causes of downgrades by the quality of the debt. Downgrades resulting primarily from management actions are less common for firms th at have a speculative grade rating before or after the downgrade. However, even for firms losing investment grade ratings, management choices play a role. Fifty-nine firms in the sa mple were downgraded from investment grade to
87 speculative grade. Ten of those downgrades were entirely the result of management actions and four more of those downgrades were primar ily the result of management action. Management-induced downgrades from inve stment grade to speculative grade are particularly significant. First, as stated at the outset of the paper, deliberate credit quality changes by management are not contemplated in current credit risk models. Second, moving across the divide between investment grade a nd speculative grade brings (1) a well recognized change in demand as certain clienteles would be forced to sell the newly speculative-grade bonds and (2) an increase in required yield that is much greater than that whic h would be required for a comparable change within the investment grade ranks. Two examples illustrate the most extreme cases where management actions are the sole driver of a firm moving from inve stment grade to speculative grade. Referencing an analyst at Standard and Poors, Reuters News (2004a) repo rted on July 21, 2004 that the rating on Citizens Communications had been lowered to BB+ from BBB because of concern that Citizens [had] shifted toward a more shareholder-friendly fi nancial policy that [w ould] limit further deleveraging and financial flexibil ity. Specifically, the firm pl anned to pay dividends which would lead to a smaller financial cushion (Reu ters News (2004). Zale Corporation announced a large stock buyback on July 1, 2003. The BBBcredit, which was placed on watch for a possible downgrade by S&P following the announ cement, was downgraded to BB+ before the end of the month (Reuters News Service (2003) a nd Dow Jones Capital Markets Report (2003)). Table 4-5 shows the causes of credit rating upg rades by firm credit quality. An interesting contrast can be drawn between downgrades to sp eculative grade debt and upgrades to investment grade debt. From Table 4-4, we know that 41 out of 59 downgrades to speculative grade were entirely caused by economics and operations. Ta ble 4-5 shows that only 7 out of 24 upgrades to
88 investment grade were entirely caused by ec onomics and operations. Management matters everywhere and sometimes, as illustrated by the cases of Citizens and Zale above, proshareholder actions are taken even when the im pact on bondholders is si gnificant (i.e. the debt falls to speculative grade). But, management matters even more in the opposite direction: a firms chance of being upgraded from speculative to investment grade relies substantially on management choices. Table 4-6 recasts the information detailed in Tables 4-4 and 4-5 to show the frequency with which management matters as a percentage of all upgrades or downgrades by credit quality. Management matters most for upgrades to invest ment grade. Thirty-eight percent of firms upgraded to investment grade were driven primarily by management. Another 21% had a significant management influence. As suggest ed earlier, the yield and demand differences between the highest specul ative grade and the lowest investme nt grade are far greater than the differences between other one notch changes in credit quality. One group of ratings changes stands out as being dominated by economic and operating factors: downgrades within speculative grade. For 85% of downgrades within the specula tive ranks, economic or operating factors were the primary cause of the downgrade. This should be expected because speculative grade bonds typically have covena nts precluding the types of mana gement actions that lead to downgrades. Whenever possible, detailed data in the categories outlined in Table 4-2 was collected along with the assignment of each rating change to management versus economic or operating causes. Table 4-7 summarizes the incidence of se lected actions and events as being one of the specific proximate causes of the rating change. Downgrades are dominated by a few particular causes. Among economic and opera ting causes, industry fundamentals were responsible for a
89 total of 112 downgrades. Among management causes, acquisitions led to 67 downgrades, common stock repurchases led to 26 downgrades and financial policies led to 19 downgrades. This last category captures both explicit polic y changes, typically articulated as a proshareholder shift, and policy cha nges inferred from firm actions. Interestingly, financial policies were the leading cause of rating upgrades, follo wed by dispositions and ac quisitions. On April 14, 2004, Yum! Brands was upgraded to BBBby Standard and Poors. Quoting an S&P analyst, the upgrade was attribut ed to consistently high free ca sh flow, and S&Ps expectation that the company [would] maintain a prudent financial policy (Lemos 2004). Here, both management actions and economic and operating factors played a significant role in the upgrade20. Among economic causes of upgrades, industr y fundamentals was most common, although not necessarily as the singular driver of the upgrade. In August 2003, S&P raised the investment-grade BBBcredit rating on Northrop Grumman21 because of the companys position within the generally attractive defense sector, bu t also because the firm was expected to pursue a more moderate financial policies (AFX UK Focus 2003). Conclusion Credit rating changes can be driven by either market conditions or management actions. For downgraded firms, 24% of changes are prim arily or substantially caused by management action. For upgraded firms, 41% of change s are primarily or s ubstantially caused by management action. Structural models of cred it risk, which assume capital structure is not 20 The 2004 rating change for Yum! Brands was scored as a three to reflect the significant impact of both management and fa vorable operating factors. 21 The 2003 rating change for Northrop Grumman was sc ored as a three to reflect the significant impact of both management choices and the favorable economic environment.
90 actively changed, have significant limitations when a substantial fraction of rating changes are coming through management action. This sh eds light on the mechanism through which corporate governance affects credit spreads.
91 Table 4-1. Credit rating change data by year. Year Downgrades Assignable downgr ades Upgrades Assignable upgrades 2004 127 101 103 70 2003 183 136 94 58 2002 142 119 38 27 2001 and before 84 62 37 31 All years 536 418 272 186
92 Table 4-2. Potential causes of credit rating changes. Cause Management, economics or operations Explanation Competition Economics Costs Economics Demand, prices or revenue Economics Economy Economics Fines, lawsuits and criminal law actions Economics Industry fundamentals Economics Market position Economics Market share Economics Public utilities commission and similar regulatory issues Economics Reporting and/or SEC issues Economics Sovereign or country issues Economics Acquisitions Management Acquiring large capital assets, divisions or entire companies. Aggressive growth Management The firm is either undertaking or ceasing a period of aggressive growth. Capital expenditures Management Refers to significant capital expenditures outside ordinary renewal and replacement levels Dispositions Management Dispositions of either large capital assets or divisions. Dividends Management Significant changes in dividend policy or payment of an extraordinary dividend. ESOP Management Establishment or curtailment of an employee stock ownership plan. Financial policy Management Integration risks or benefits Manageme nt Integration of large acquisitions. Refinancing or issuance of preferred stock Management Refinancing preferred stock with regular debt or using preferred stock to reduce regular debt leverage. Repurchase or issuance of common stock Management Using cash on hand or debt issuance to repurchase common stock or issuing common stock to reduce leverage or increase cash on hand. Leveraged buyouts are included here. Family ownership changes Management Operations restructuring costs or benefits Operations Operations restructuring risks Operations Strategic uncertainty Operations Cash flows Operations Earnings or losses Operations Earnings predictability Operations General operations Operations Operating risk changes Operations
93 Table 4-3. Summary of ove rall scoring criteria. Score Underlying factors 1 Management actions were the only or the overwhelming cause of the rating change. 2 Management actions were the primary cause of the rating change, but economic or operating factors contributed. 3 Both management actions and either economic or operating factors were a significant and proximate cause of the rating change. 4 Economic or operating factors were the primary cause of the rating change, but management actions contributed. 5 Economic or operating factors were the only or the overwhelming cause of the rating change.
94 Table 4-4. Causes of downgrades by i nvestment versus speculative grade. Cause of downgrade Score All downgrades Investment grade Speculative grade Transition to speculative grade All management 1573512 10 Primarily management with some economics or operations influence 21895 4 Management and economics or operations were significant 3271413 0 Primarily economics or operations with some management influence 4341416 4 All economics or operations 528281160 41 All causes 418153206 59
95 Table 4-5. Causes of upgrades by inve stment versus speculative grade. Cause of upgrade Score All upgrades Investment grade Speculative grade Transition to investment grade All management 1 35 11 17 7 Primarily management with some economics or operations influence 2 12 3 7 2 Management and economics or operations were significant 3 30 7 18 5 Primarily economics or operations with some management influence 4 31 7 21 3 All economics or operations 5 78 31 40 7 All causes 186 59 103 24
96 Table 4-6. Frequency of substant ial management influence by credit quality and direction of ratings change. Management dominates Both influences significant Economic and operating factors dominate Number Percentage NumberPercentage Number Percentage Downgrades within investment grade 44 29 14 9 95 62 Upgrades within investment grade 14 24 7 12 38 64 Downgrades to speculative grade 14 24 0 0 45 76 Upgrades to investment grade 9 38 5 21 10 42 Downgrades within speculative grade 17 8 13 6 176 85 Upgrades within speculative grade 24 23 18 17 61 59 All Downgrades 75 18 27 6 316 76 All Upgrades 47 25 30 16 109 59
97 Table 4-7. Incidence of selected specific causes of downgrades and upgrades. DowngradesUpgrades Fines, lawsuits and criminal law actions 152 Industry fundamentals 11216 Public utilities commission and similar regulatory issues74 Sovereign or country issues 86 Acquisitions 6719 Aggressive growth 50 Dispositions 1120 Dividends 110 ESOP 10 Financial policy 1932 Integration risks or benefits 211 Refinancing or issuance of preferred stock 20 Repurchase or issuance of common stock 2610
98 LIST OF REFERENCES Acharya, Viral V., and Jennifer N. Carpenter, 2002, Corporate bond valuation and hedging with stochastic interest rate s and endogenous bankruptcy, Review of Financial Studies 15, 13551383. AFX UK Focus, 2003, S&P upgrades Northrop Grummans credit ratings, August 18. Berkovitch, Elazar and E. Han Kim, 1990, Financia l contracting and leverage induced overand under-investment, Journal of Finance 45, 765-794. Besanko, David and Anjan Thakor, 1987a, Competi tive equilibria in the credit market under asymmetric information, Journal of Economic Theory 42, 167-182. Besanko, David and Anjan Thakor, 1987b, Collate ral and rationing: Sorting equilibria in monopolistic and competitive credit markets, International Economic Review 28, 671-689. Bester, Helmut, 1985, Screening vs rationing in credit markets with imperfect information, American Economic Review 75, 850-855. Bhojraj, Sanjeev, and Partha Sengupta, 2003, Eff ect of corporate governance on bond ratings and yields: The role of institutional investors and outside directors, Journal of Business 76, 455-475. Billet, Matthew T., Tao-Hsien Dolly King, and David C. Mauer, 2007, Growth opportunities and the choice of leverage, debt maturity, and covenants, Journal of Finance 62, 697-730. Black, Fischer, and Myron Scholes, The pric ing of options and co rporate liabilities, Journal of Political Economy 81, 637-654. Brown, David T., Brian A. Ciochetti and Ti mothy Riddiough, 2006, Theory and evidence on the resolution of financial distress, Review of Financial Studies 19, 1357-1397. Brown, David T., Christopher M. James and Robert M. Mooradian, 1994, Asset sales by financially distressed firms, Journal of Corporate Finance 1, 233-257. Chan, Yuk-Shee and Anjan Thakor, 1987, Collate ral and competitive equilibria with moral hazard and private information, Journal of Finance 42, 345-364. Collin-Dufresne, Pierre, and Robert S. Goldstei n, 2001, Do credit spreads reflect stationary leverage ratios?, Journal of Finance 56, 1929-1957. Dow Jones Capital Markets Report, 2003, S&P cuts Zale Corp credit rtg to BB+, off watch, July 31. Esty, Benjamin, and Aldo Sesia, Jr., 2005, An overview of project finance 2004 update, Harvard Business School, Case 9-205-065.
99 Faulkender, Michael, and Mitchell A. Petersen, 2006, Does the source of capital affect capital structure?, Review of Financial Studies, 19, 45-79. Flannery, Mark J., Joel F. Houston and S ubramanyam Venkataraman, 1993, Financing multiple investment projects, Financial Management, 22, 161-172. Helwege, Jean, and Christopher M. Turner, 1999, The slope of credit yield curve for speculativegrade issuers, Journal of Finance 54, 1869-1884. Hovakimian, Armen, Tim Opler and Sherid an Titman, 2001, The debt-equity choice, Journal of Financial and Quantitative Analysis 36, 1-24. Howe, John and James Schilling, 1988, Capital stru cture theory and REIT security offerings, Journal of Finance, 43, 983-993. John, Teresa and Kose John, 1991, Optimality of project financing: Theory and empirical implications in finance and accounting, Review of Quantitative Finance and Accounting, 1, 51-74. Johnson, Shane, 2003, Debt maturity and the effect s of growth options and liquidity risk on leverage, Review of Financial Studies, 16, 209-236. Kahn, Charles and Andrew Winton, 2004, Moral hazar d and optimal subsidiary structure for financial institutions, Journal of Finance 59, 2537-2575. Klock, Mark S., Sattar A. Mansi, and William F. Maxwell, 2005, Does corporate governance matter to bondholders?, Journal of Financial and Quantitative Analysis 40, 693-718. Kolasinski, Adam, 2006, Subsidiary debt, capital structure and inte rnal capital markets, Working paper, University of Washington. Lemos, Mara, 2004, S&P Upgrade puts Yum! Brands firmly in invest-grade land, Dow Jones Capital Market Reports, April 14. Lewellen, Wilbur, 1971, A pure financial ra tionale for the conglomerate merger, Journal of Finance, 26, 521-537. Ling, David and Wayne Archer, 2005, Real Estate Principals: A Value Approach, (McGraw-Hill Irwin). Merton, Robert C., On the pricing of corporate debt: The risk structure of interest rates, Journal of Finance 29, 449-470. Myers, Stewart, 1978, Determinants of corporate borrowing, Journal of Financial Economics, 5, 147-175. Pulvino, Todd, 1998, Do asset fire sales exist? An em pirical investigation of commercial aircraft transactions, Journal of Finance, 53, 939-978.
100 Riddiough, Timothy and Zhonghua Wu, 2006, Financial constraints, external-source liquidity, and investment, Working paper, University of Wisconsin. Reuters News, 2004, Update 1 S&P cuts Citizen s Communications rating to junk, July 21. Reuters New Service, 2003, Zale plans st ock buyback, debt refinance, July 2. Salas, Caroline, 2006, Bondholders lose in He inz, Sprint stock buybacks, Bloomberg.com, August 15. Stulz, Rene and Herb Johnson, 1985, An analysis of secured debt, Journal of Financial Economics, 14, 501-521. Titman, Sheridan and Roberto Wessels, 1988, The determinants of capital structure choice, Journal of Finance, 43, 1-19. Triantis, George, 1992, Secured debt under conditions of imperfect information, Journal of Legal Studies 21, 225-258. Williams, Joseph, 2001, Agency, ownership and re turns on real assets, Working paper, Professors Capital. Williamson, Oliver, 1988, Corporate finance and corporate governance, Journal of Finance 63, 567-591.
101 BIOGRAPHICAL SKETCH Hugh Marble III grew up in Rhode Island and completed his undergraduate work in economics at the University of Rhode Island. He received a Master of Business Administration from Rollins College and spent slightly over th ree years working in public finance before beginning doctoral studies at the University of Florida.