19. Capital Structure: Empirical Financial Contracting

The financial-contracting literature answers capital-structure questions (e.g. why does debt even exist) from a different angle than traditional trade-off theory. Its incentive-conflict issues are often more convincing than the tax-benefit argument when deriving the optimal capital structure. Theory papers explaining capital structure with incentive conflicts: management compensation and risk shifting (Jensen-Meckling 1976), management effort (Innes 1990), truthful disclosure/CSV (Townsend 1979), incomplete contract and corporate control rights (Aghion-Bolton 1992; managers receive private benefit and so have an excess continuation bias). The crucial idea: debt financing implements contingent control allocation — the external financier can take control when the manager makes sub-optimal decisions due to incentive conflicts.

Empirically, financial contracts have many aspects worth describing and measuring. Kaplan and Strömberg (2003) is seminal in empirical venture-capital (VC) contracting, finding that contracts with VCs give them a lot of control rights (cash-flow, board, voting, liquidation rights), and interpreting the findings relative to theory, showing how theory and empirics disclose truth together. Below are selected papers in the field.

Study of 153 Chapter 11 bankruptcies of private and public firms in 2001.

19.1.1 Institutional details. Liability structure: claimants ranked by priority from high to low — secured loan creditor, unsecured bond creditor, subprime bond creditor, equity holder. Chapter 11 bankruptcy: filed when insolvent, initiating the Automatic Stay (temporarily suspending creditors' right to take assets, giving the firm time to restructure), under court supervision, ending after the firm emerges (restructures) or liquidates; unlike Chapter 7, Chapter 11 aims to maintain operation and minimize social impact. Absolute priority violation: if the equity holder gets any payment before all debt is paid. Section 363 Sale: the debtor can tentatively market assets, negotiate with buyers, and file a motion for a bankruptcy auction; secured creditors can make a credit bid (use their value claim as the bid). Remark 19.1: a Section 363 Sale isn't necessarily a traditional liquidation (a firm may sell a non-core unit to focus on its core, even without bankruptcy). Debtor in possession (DIP): in bankruptcy can still operate but is a legally different firm replacing the original; can raise DIP financing to support operations, and since secured loan creditors can object to higher-seniority new creditors they are typically the DIP lenders (very profitable, rates ~30% APR).

19.1.2 Descriptive findings. Creditors have pervasive control before/in bankruptcy, managers and equity holders little (contrary to the traditional view that Chapter 11 gives them more control): 70% of CEOs replaced within two years before filing; absolute priority violation only 8%; 75% use DIP financing (giving creditors enormous control) — 95% of DIP lenders can reject reorganization plans unless fully paid, 92% have a security interest in all assets.

19.1.3 Hypothesis testing. Core hypothesis: whether and to what extent creditor conflicts (liability structure) affect final asset allocation in bankruptcy (in an MM 1958 world, parties bargain perfectly without friction, so liability structure shouldn't matter, only fundamentals). By priority order, the secured creditor (higher), unsecured creditor (lower) and equity holder (residual claimant) have payoff functions as in Figure 19.1.

Without frictions, liquidation value transforms one-to-one into claim value (the 45° line); the secured creditor is fully paid right at the kink \(L^\star\). With estimated liquidation value \(L_0\) (near \(L^\star\)): the secured creditor controls and dislikes risk (restructuring's extra variance only adds a positive probability of not being paid in full, no upside) → more likely to liquidate → violates MM. With value \(L_1\): the secured creditor is far out of the money, acts like an equity holder, always wants to restructure. With value \(L_2\): the secured creditor doesn't care much, and the unsecured creditor can bargain with it to not liquidate. The story is supported by Figure 19.2 (a Lowess curve of reorganization probability, x-axis \(\log(\text{secured debt}/\text{assets})\), $=0$ corresponding to \(L^\star\)) and Figure 19.3 (a probit of reorganization probability): the "secured debt $>0$" coefficient $+0.380$ (prefer restructuring when far out of money), and "secured debt 25%–100% / $>\(100% of assets" coefficients \)-0.352,-0.354$ (prefer liquidating near \(L^\star\)). Comments: the small sample limits statistical power; the story missed selection bias (a large secured-debt overhang may signal worse fundamentals → more likely to liquidate); segment \(AB\) in Figure 19.1 may not be flat (the unsecured creditor may bribe the secured one not to liquidate); an alternative not ruled out — only parties "in the money" can vote on the plan, so secured creditors have a strong incentive to under-evaluate liquidation value to push others "out of money" and reap a huge premium from the value hike after the case closes (so they liquidate less at \(L_1\) and more at \(L^\star\)).

19.2 Roberts-Sufi (2009a). Aghion-Bolton (1992) model control shift via an abstract performance-related "state \(s\)," but \(s\) isn't necessarily payment default. Roberts-Sufi shift to financial-covenant violation (not payment default) in private credit agreements — a much larger sample (39% of COMPUSTAT public firms violate during 1997–2007). Financial covenants are typically written on cash-flow- and leverage-based variables (most common: debt/cash flow with a 3.5 cutoff; interest coverage = operating income/interest expense), giving creditors the right to terminate or accelerate payment, but always nebulous (saying what can happen, not what must); violation is cyclical (Figure 19.4). Core question: the effect of creditor intervention (implied by violation \(viol_{it}=1\)) on capital structure (leverage \(lev_{it}\))? The selection bias (17.1):

$$\mathbb{E}[lev_{1it}\mid viol{=}1]-\mathbb{E}[lev_{0it}\mid viol{=}0]=\underbrace{\mathbb{E}[lev_{1it}\mid viol{=}1]-\mathbb{E}[lev_{0it}\mid viol{=}1]}_{\kappa}+\underbrace{\mathbb{E}[lev_{0it}\mid viol{=}1]-\mathbb{E}[lev_{0it}\mid viol{=}0]}_{\text{selection bias}<0} \tag{19.1}$$

is likely negative (violating firms probably have too much debt, would reduce leverage anyway). With no precise covenant-variable data for an exact RD, they use a quasi-RD: omitted variables' effect on leverage is likely continuous, while the violation indicator is certainly discontinuous at 1; throw in usual covenant variables to high orders to control fundamental variation. Results: a large within-firm effect on net debt issuance, creditors reduce leverage; by 10-K/10-Q filings, on violation creditors typically cut the credit facility and raise the rate. Drawback: no evidence on how ex-post control affects ex-ante financing.

19.3 Akins et al. (2019) change-of-management restrictions (CMR): study private syndicated-loan agreements; CMR appears in 9% of all firms, 17% of small firms. Findings: CMR is more likely with small, risky borrowers, especially when the CEO is also the founder (interpreted as creditors wanting to reduce "CEO turnover risk"); CMR is binding — firms with CMR have lower CEO turnover.

Study the effect of private credit agreements between public-firm borrowers and banks on investment decisions. Findings: 32% of agreements explicitly restrict capital expenditure; poor performance (deteriorating credit quality) → more likely to face an investment restriction (Figure 19.5), so creditors intervene to restrict new investment and reduce the volatility of firm value; the restriction is binding (Figure 19.6: firms asymmetrically cluster just below the capital-expenditure limit) and indeed reduces capital expenditure (Figure 19.8); newly restricted firms' market values and operating performance rise after the restriction (Figures 19.7a, 19.7b) — but interpreting this too aggressively as efficiency improvement is questionable (the restriction may kill the worst-but-still-positive-NPV project, which can't be efficient, yet average performance rises as the lower end is removed from the average).

The selection bias (\(Rest_i=1\) if restricted, \(I_{1i}/I_{0i}\) investment under restriction/none):

$$\mathbb{E}[I_{1i}\mid Rest{=}1]-\mathbb{E}[I_{0i}\mid Rest{=}0]=\underbrace{\mathbb{E}[I_{1i}\mid Rest{=}1]-\mathbb{E}[I_{0i}\mid Rest{=}1]}_{\kappa}+\underbrace{\mathbb{E}[I_{0i}\mid Rest{=}1]-\mathbb{E}[I_{0i}\mid Rest{=}0]}_{\text{selection bias}<0}$$

likely negative (restricted firms may already over-invest, would cut anyway). Graphical resolution: Figure 19.8 shows the pre-trend of "always-restricted" and "newly-restricted" firms is almost the same before the new restriction → selection bias is not a big issue.

Follow-ups. Nini et al. (2012) use 1996–2008 SEC filings to identify all violations (10%–20% of firms), finding firms after violation typically: lower growth in assets and PP&E, lower capital expenditures and cash acquisitions (Figure 19.9); lower new debt issuance, total debt, shareholder payouts, higher cash (Figure 19.10); higher CEO turnover (Figure 19.11, creditors have informal governance influence); operating performance and stock price increase after violation (efficiency improves, possibly from creditor intervention). Ershahin et al. (2018) use establishment-level U.S. Census data on within-firm resource allocation of violating firms (near the threshold, controlling covenant variables), finding within violating firms employment/investment fall, non-core/less-productive establishments close more often; effects stronger when management agency cost is higher; creditor intervention improves efficiency by relocating resources more productively.

Using supervisory data of syndicated bank loans during the 2008 crisis, study how a bank's (lead lender's) health transmits to the real sector through corporate borrowing. Empirical design: identification assumption — covenant violators of healthier and less-healthy lenders have similar characteristics (supported by a balance test, Figure 19.12, showing similar leverage, credit ratings, and propensities to violate). DiD:

$$Y_{b,f,t}=\beta_0+\beta_1[\text{Bad Lender}_b]+\beta_2[\text{Bind}_{f,t-1}]+\beta_3[\text{Bad Lender}_b\times\text{Bind}_{f,t-1}]+\boldsymbol\gamma'\mathbf X_{f,f,t}+\varepsilon_{b,f,t}$$

\(Y\) = outcome of firm \(f\) (with loan \(l\) from lead bank \(b\)) in period \(t\) (one specific \(Y\) is Credit Reduced, $=1$ if the loan is terminated before maturity or reduced); Bad Lender\(_b\) = the lead bank's health rank normalized to $[0,1]$ (1 worst); Bind\(_{f,t-1}\) = an indicator $=1$ if the firm violates in \(t-1\) or \(t\); \(\mathbf X\) controls; \(\beta_3\) the parameter of interest. Results (Figure 19.13): \(\beta_3\) significantly positive in all specifications → credit is reduced following violation. Findings: corporate borrowers borrow long-term; ~1/3 of firms breach covenants (enabling banks to force renegotiation); an unhealthy bank (in crisis) less often grants a waiver and requires a lower loan volume in renegotiation; covenant violation accounts for an 11% decline in loan volumes during the 2008–2009 crisis (~the total credit decrease then); so financial-sector health transmits to the real sector largely through covenant violation (banks use violations to reduce credit); in smaller/more concentrated syndicates the lead lender matters more and the outcome is more sensitive to its health. Contribution: previously people believed financial-sector troubles shouldn't transmit to the real sector since loans are long-term; this paper shows it's wrong — the pressure transmits via forced renegotiation following violation, at high frequency.

19.6.1 Theoretical relevance. Liquidation value = the value of the second-best use of assets in bad cash-flow states, with real effects on firm behavior. Perspectives: Williamson (1988) (transaction cost) — concern for liquidation value may reduce ex-ante operating efficiency; higher asset specificity → higher liquidation transaction cost and worse debt contracts, so a firm may inefficiently pursue less-specific, more-redeployable assets (redeployability = alternative use; illiquidity = difficulty of sale; salability = the combination that determines liquidation value). Hart-Moore (1994) (human capital) — human capital cannot be committed/perfectly contracted, so liquidation value is crucial to debt terms. Berglöf-Von Thadden (1994) (incomplete-contract renegotiation) — a small liquidation value + a short-term senior creditor reduces the entrepreneur's agency problem (the short-term creditor is tougher, choosing to liquidate more often if the liquidation value is so small it would take a loss). Shleifer-Vishny (1992) — liquidation value is endogenous (depends on the set of buyers and their financial conditions). The basic testable implication: "how does liquidation value affect capital structure?" Traditional research regresses leverage on PP&E/Total Assets, but tangible assets need not have high liquidation value and are a poor proxy.

19.6.2 Evidence from commercial real estate: Benmelech et al. (2005). Using 14,159 U.S. commercial real estate transactions Jan 1, 1992–Mar 30, 1999 (COMPS.com) to study the effect of liquidation value on debt contracting. A good industry (purchases highly levered, LTV 0.68–0.77). Focus on how each property's leverage is affected by commercial zoning regulation — zoning captures the flexibility of permitted uses, used as a proxy for redeployability (the second-best-use/liquidation value). The selection bias (17.1) (\(z_i=1\) more flexible zoning, \(z_i=0\) stricter):

$$\mathbb{E}[lev_{1i}\mid z{=}1]-\mathbb{E}[lev_{0i}\mid z{=}0]=\underbrace{\mathbb{E}[lev_{1i}\mid z{=}1]-\mathbb{E}[lev_{0i}\mid z{=}1]}_{\kappa}+\underbrace{\mathbb{E}[lev_{0i}\mid z{=}1]-\mathbb{E}[lev_{0i}\mid z{=}0]}_{\text{selection bias}>0} \tag{19.2}$$

likely positive (flexible-zone properties are typically large businesses like supermarkets, high leverage by nature; strict zones are small businesses like dry cleaners, low leverage by nature). The specification:

$$\text{loan characteristic}_i=F(\text{redeployability}_i,\ \text{price}_i,\ \text{cap rate}_i,\ \text{controls}_i)+\varepsilon_i \tag{19.3}$$

with census-tract fixed effects as partial controls to reduce the bias. Results (Figure 19.14): higher redeployability (higher liquidation value) → higher leverage. Comments: census-tract FE could deteriorate the bias (the municipal government intentionally creates zoning-flexibility variation within tracts; very different firms by nature may fall into different zoning lands within a tract → FE doesn't solve it and makes the comparison less comparable); the variable \(\text{price}_i\) should not be on the RHS of (19.3) (price is a channel through which redeployability affects leverage: higher redeployability → higher price → more debt; estimating a combination of channels, we shouldn't shut down this major economic channel).

19.6.3 Evidence from 19th-century railroads: Benmelech (2009). Using data on 221 railroads in 1868/1873/1877/1882 to study the effect of asset salability on leverage and debt maturity. A good dataset (railroad salability depends mainly on gauge — the horizontal distance between two rails; the more nearby railroads using the same gauge, the higher the salability). Findings: railroads with higher salability use longer-maturity debt; leverage is unrelated to salability. Comment: a personal-competence issue (railroads then were much affected by founders' talents; a talented leader could both use longer-maturity debt and unify nearby gauges → salability and debt maturity correlated through personal competence, not a direct effect).

The inevitability of ex-post renegotiation always changes ex-ante incentives. Roberts-Sufi (2009b) provide a purely descriptive study of renegotiation in private credit agreements (1,000 U.S. public-firm private credit agreements, 1996–2005). Findings: renegotiation is frequent — 90%+ of long-term agreements have major terms (principal, interest, maturity, collateral) renegotiated before stated maturity; it mostly happens early in the agreement's life without financial stress or covenant violation; it is cyclical (Figure 19.15a); favorable renegotiation (favorable to borrowers: (1) more amount, constant spread; (2) lower spread, constant amount; (3) both) happens in good times when firms voluntarily choose to (Figure 19.15b, since only renegotiation with both amount and spread changing favorably must be proposed by firms); unfavorable renegotiation happens in bad times when firms violate covenants (Figure 19.15c).

Puzzle and conjecture: the puzzle — firms can renegotiate terms in their favor in good conditions, but banks cannot (unless firms violate covenants), so banks insure firms against cyclicality (good times have upside, bad times no downside as long as no violation); why? More puzzling — banks write down all contingency terms they know are very likely to be renegotiated; why? The authors conjecture banks write down contingency terms to assign bargaining power to each party in later renegotiation. Per Figures 19.16a–19.16d: in 19.16a/b firms with a covenant are more likely to get a favorable outcome when performing well → good-performing firms with covenants have stronger bargaining power than without; in 19.16c/d firms with a pricing grid (interest payment based on which performance grid the borrower falls into) are more likely to get an unfavorable outcome when performing badly → bad-performing firms with covenants have weaker bargaining power than without.