Upload
others
View
9
Download
0
Embed Size (px)
Citation preview
Debt Restructuring Costs and Bankruptcy Risk:Evidence from CDS Spreads
Abstract
We examine the effects of a recent change to the US tax code that reduced the costs lenders incurwhen restructuring loans out of court (Regulation TD9599). The regulation contemplated particulartypes of loans, allowing us to use a triple-differences approach to identify the degree to which borrow-ers are differentially affected by the shift in restructuring costs. We model these changes and showhow CDS market responses allow us to disentangle the relative costs of in-court versus out-of-courtrestructuring. Our evidence is consistent with markets anticipating more out-of-court workouts afterthe passage of TD9599. CDS spreads decline by record-low figures on the regulation’s announcementand the drop is concentrated among distressed firms with high syndicated loans-to-total debt ratios— those borrowers most affected by the reduction in renegotiation costs. Stock returns of thesedistressed firms as well as of their syndicate lenders outperform the market on the announcement ofTD9599. As a result of the reduction in bankruptcy likelihood, distressed firms’ access to syndicatedloans expands and their financing costs decline. Our analysis provides insights into how alteringregulatory constraints can improve welfare in financial distress.
Key words: Debt renegotiation, bankruptcy, credit default swaps, corporate taxes, credit access.
JEL classification: G33, G32.
1 Introduction
Firms unable to meet their financial obligations often attempt to negotiate with their cred-
itors out of court (Gertner and Scharfstein (1991) and Franks and Torous (1994)). These
negotiations are thought to be beneficial because early, mutually-agreed restructurings avoid
the deadweight costs of drawn-out court battles (Gilson (1997)). Yet, these negotiations are
costly, fraught with coordination problems, and do not always succeed (Bolton and Scharfstein
(1996)). Arguably, the relative costs of in- and out-of-court restructurings are likely to deter-
mine whether distressed firms file for bankruptcy. But disentangling these costs is difficult,
and there exists only limited evidence on the extent they truly affect distress resolution.
In this paper, we examine a statutory-induced shift to debt renegotiation costs, showing a
well-identified connection between the relative costs of in-court versus out-of-court restructur-
ings and the likelihood that renegotiation of distressed debt takes place. Our analysis exploits
a number of features of a recent piece of regulation adopted in the U.S. that substantially
reduced the tax payments that certain lenders owe upon restructuring debt out of court. Reg-
ulation TD9599 (which we describe in detail shortly) was adopted on September 12, 2012 and
significantly reduced restructuring costs for syndicated loans, but not for other types of debt.
Notably, TD9599 has unusual legal powers over loans that were issued in the past — years
before the regulation was passed — but restructured after its implementation.
Taxes present an important obstacle to out-of-court renegotiation and are a major deter-
minant of the choice between bankruptcy and restructurings (Gilson (1997)). Lenders incur
large tax costs on restructured loans, and those costs erode the continuation value upon which
lenders and borrowers can bargain out of court. Prior to TD9599 the tax treatment on renego-
tiated syndicated loans was highly punitive, as lenders would owe taxes on so-called “phantom
gains,” which far exceeded any income earned from renegotiations.1
To identify how this reduction in out-of-court costs affected debt renegotiation, we look
at the market price reaction of an instrument directly linked to renegotiation: credit default
swaps (CDS). In a standard CDS contract, a buyer and a seller write an agreement that ref-
erences either a specific debt issue or firm. The buyer pays the seller a periodic fee (the CDS
spread), and the seller makes a lump-sum payment if the underlying reference experiences a
1Prior to TD 9599, creditors who acquired syndicated debt and restructured it out of court would owe taxesbased on the difference between the secondary market purchase price and the loan’s par value — which is higherthan the distressed debt’s market value. The creditor would therefore owe taxes on large unrealized gains. Assuch, the more deeply distressed the borrower, the higher the lender’s tax burden on a loan restructuring.
1
credit event.2 CDS spreads thus directly relate to bankruptcy risk: they represent the amount
buyers are willing to pay to insure against corporate default (see Longstaff et al. (2005)).
Examining changes in CDS spreads around TD9599’s announcement gives unique insights into
the effects of reducing out-of-court renegotiation costs on bankruptcy risk. Our identification
strategy is further strengthened by the fact that TD9599 only affected syndicated loans. As
such, one can measure the relative impact of that regulation on firm bankruptcy risk according
to the weight of those specific types of loans on the firm’s overall debt profile.
To guide our examination, we analytically model the effect of changes in renegotiation costs
on CDS spreads as a way to generate testable predictions about the 2012 IRS ruling. The model
endogenizes CDS insurance and shows how the IRS ruling facilitated out-of-court renegotia-
tion for firms that are distressed. In the model, a borrower raises financing from competitive
lenders, and upon experiencing distress can either declare bankruptcy or try to renegotiate out
of court. If the borrower files for bankruptcy, lenders that bought CDS insurance receive the
full value of debt, while uninsured lenders are exposed to default and to additional losses. If
out-of-court renegotiation occurs, lenders owe taxes. The tax is based on the difference between
the debt purchase price and either (1) the par value of the renegotiated issue, or (2) the market
value of the renegotiated payment. Our modeling is predicated on IRS tax standards. In par-
ticular, prior to TD9599, the IRS classified syndicated loans as non-publicly traded debt with
tax value determined by the loan’s par. The new regulation, however, reclassified syndicated
loans as publicly-traded debt, with tax value based on market prices at time of renegotiation.
One key insight of the model is that the effect of taxes on renegotiation depends on whether
lenders purchase CDS insurance. If lenders are uninsured, the borrower will offer a payment
equal to the lenders’ recovery in bankruptcy. The market value of this payment is lower than
the debt’s face value, so taxable income is lower and renegotiation is easier when taxes are
levied on market values, such as is the case of taxes under TD9599. If lenders are insured,
however, the borrower’s offer must exceed the face value of debt; else lenders trigger their CDS
insurance. In this case, the tax burden is larger when taxes apply to market values, as in the
case of TD9599, making renegotiation harder.
Our model is unique in demonstrating how CDS markets, taxes, and alternative distress
resolution mechanisms are connected in the real economy. When the borrower’s fundamentals
2A credit event is defined by the International Swaps and Derivatives Association (ISDA) as a default onthe underlying debt issue, debt acceleration, failure to pay, repudiation, or bankruptcy filing. Since April2009, the standard CDS contract does not recognize out-of-court restructuring as a credit event.
2
are strong, bankruptcy risk is low regardless of the tax rule and lenders’ insurance decision.
However, when the borrower’s financial situation is weak, bankruptcy is avoided when the
fraction of uninsured lenders is sufficiently large. In this case, taxes based on market val-
ues can significantly decrease the size of the out-of-court payment and reduce the probability
of bankruptcy. This, in turn, lowers CDS spreads. Our model goes further in generating
predictions for the impact of changes in renegotiation costs on credit access.
We test our model’s predictions using a triple-differences strategy focusing on how CDS
spreads change around the announcement of TD9599. We do so using a sample of non-financial
firms for which trading of CDSs is liquid. Since TD9599 only affects syndicated loans, we com-
pare spread changes for firms with high versus low ratios of syndicated loans to total debt.
To better identify our effects, we add another layer of contrasts to our tests and verify our
model’s prediction that the regulation’s effect on spreads is stronger for firms with weaker
fundamentals. We do so by comparing distressed and non-distressed firms; where we classify
firms according to metrics such as Altman’s Z-Score and distance-to-default (Merton (1974).
We show that CDS spreads drop by 27 basis points in the two-week window around the
announcement of TD9599. This is the single largest two-week drop in spreads since the depths
of the Financial Crisis in 2009. Furthermore, this drop is exactly concentrated among firms
for which renegotiation costs presumably decrease the most: spreads decline by 36 basis points
for distressed firms at the top tercile of the syndicated loans–debt ratio, but only by 15 points
for distressed firms at the bottom of that ratio.3 At the other hand of the spectrum, for non-
distressed firms the difference in spread changes across high versus low syndicated loans–debt
ratio is close to zero. Indeed, our tests show that the spread difference between high versus
low loans–debt ratio increases monotonically with measures of financial distress around the
inception of TD9599.
We look for additional externalities of TD9599 and find that the increase in renegotiation
likelihood created shareholder value gains for both borrowers and lenders. In particular, dis-
tressed firms with high syndicated loans–debt ratios experienced a 1.6% abnormal return in the
3-day window around the regulation’s announcement. The tax change also benefits syndicated
lenders: they outperform the market by 4.3%.
Next, we analyze how the market for new financing responds to a reduction in renegotia-
tion costs. Consistent with our model’s predictions, we find that markups on new syndicated
loans issued to distressed firms drop by 28 basis points (13% of the sample mean) relative
3Notably, distressed and undistressed firms have similar syndicated loans–debt ratio distributions.
3
to non-distressed firms following TD9599’s announcement. We also find that distressed firms
are 12% more likely to obtain a new loan after TD9599’s announcement. Notably, access and
financing terms improve primarily for long-term loans, which most expose lenders to deterio-
ration in borrower financial conditions and hence likely benefitted most from the tax change.
Our results indicate that reducing out-of-court renegotiation costs leads distressed borrowers
to obtain greater access to the syndicated loan market and cheaper funding.
One potential concern is that CDS spreads of distressed/high-loan firms could have been
decreasing relative to spreads of other firms even before TD9599, due to market trends and dy-
namics that we do not observe; in other words, our underlying “parallel trends assumption” may
be violated. To investigate this possibility, we re-estimate our specification over a large num-
ber of experimental windows from January 2010 through December 2012, assigning a “placebo
event” to each window. We find that among distressed firms, the drop in spreads following the
actual TD9599 announcement is larger than all but one of the 50 placebo events we simulate.
Notably, spread changes for 44 of the 50 placebo events are either statistically insignificant
or positive. Furthermore, our triple-differences tests present a high threshold for other unob-
servable variables that could bias our results. To contaminate our inference, bankruptcy risk
would have to decrease for non-tax related reasons on the TD9599 announcement, precisely
for firms that are distressed and have high loans–debt ratios. Broad macroeconomic shocks,
for example, are unlikely to produce such a nuanced confounder.
Our paper is related to recent literature showing how CDS trading affects firms’ credit risk.
Saretto and Tookes (2013), Ashcraft and Santos (2009), and Hirtle (2009) show that firms
with CDSs written on their debt observe lower interest rates on their bank loans and increased
corporate leverage and debt maturity. Our study also relates to theoretical and empirical anal-
yses showing that CDS trading increases distressed firms’ probability of bankruptcy through
the “empty creditor problem” (Bolton and Oehmke (2011), Campello and Matta (2012), and
Subrahmanyam et al. (2014)). Our work adds to this new literature by using CDS markets to
show how debt renegotiation costs affect bankruptcy risk and financing costs.
Our paper is also related to empirical work investigating how the relative costs of in- versus
out-of-court renegotiation affect debt restructuring and financing.4 This work suggests that
out-of-court renegotiation likelihood is decreasing in the number and dispersion of lenders,
borrowers’ asset tangibility, and secured debt, among other factors. Precisely identifying such
4Papers that contribute to this literature include Gilson et al. (1990), Asquith et al. (1994), Gilson (1997),Benmelech and Bergman (2008), Roberts and Sufi (2009), and Morellec et al. (2013).
4
linkages is, nonetheless, complicated by lack of variation in renegotiation costs. Some recent
papers attempt to overcome this problem by examining the effect of changes to bankruptcy
codes and procedures on financing costs (see Giambona et al. (2013) and Benmelech et al.
(2005)). We contribute to this literature by showing how a reduction to out-of-court renego-
tiation costs increases the odds of renegotiation, subsequently increasing distressed borrowers’
financing conditions, shareholder value, and access to credit.
Lastly, our study contains important implications for economic policymakers. Crucially, all
of our findings are derived from an arbitrary relaxation of regulatory tax constraints. We show
that both borrowers and lenders benefited from TD9599, providing an estimate of the dead-
weight loss from taxing renegotiated debt at par values (which is customary worldwide). Our
results imply that policies that reduce renegotiation costs can improve contracting efficiency
and the availability of credit at lower costs.
The rest of the paper is organized as follows. Section 2 describes the tax treatment of
renegotiated debt, explaining the statutory changes introduced by Regulation TD9599. Sec-
tion 3 contains our model. Section 4 discusses our data and presents our main results on CDS
spreads, equity values, and access to credit. Section 5 contains robustness checks. Section 6
concludes. All proofs are in the Appendix.
2 Background on the IRS Redefinition of Public Debt
In this section, we briefly describe the tax treatment of out-of-court debt restructuring. We
then discuss the critical features of Regulation TD9599 redefining public debt.
2.1 Tax Treatment of Debt Restructuring
When a debt issue is significantly modified outside of the bankruptcy procedure, the IRS treats
the modification as an exchange of the old debt issue for a new one. An extension of the issue’s
maturity or a change in the interest payments are examples of significant modifications. A debt
exchange qualifies as a taxable event, with implications for lenders, borrowers, and investors
who purchase debt on the secondary market.
Debtholders must report income to the IRS when restructuring debt, and their tax obliga-
tions depend on whether the IRS classifies the debt as publicly or privately traded. For private
debt, tax is based on the spread between the par value of the original debt contract and the
debt’s secondary market price when the debtholder purchased it; alternatively, on the issue’s
5
Figure 1. Tax Treatment of Debt Restructuring
Panel A. Debt is privately traded
Scenario 1 Scenario 2
• Original debt issue has par of 100 • Same as 1, except lender retains• Investor buys debt on secondarymarket for 40
debt
• Renegotiated debt has marketvalue of 50 and par of 100
Debtholder’staxes:
τ × (100− 40)→ tax on unrealized“phantom gains”
τ × (100− 100)→ no tax owed
Panel B. Debt is publicly traded
Scenario 1 Scenario 2
• Original debt issue has par of 100 • Same as 1, except lender retains• Investor buys debt on secondarymarket for 40
debt
• Renegotiated debt has marketvalue of 50 and par of 100
Debtholder’staxes:
τ×(50−40)→ lower tax obligation τ × (50− 100)→ receive tax credit
original purchase price if the debtholder is the first (original) lender. Debt restructurings often
modify the maturity date or coupon rate while leaving the par value unchanged. For distressed
firms, the original par value is generally far higher than the loan’s market price. Accordingly, a
distressed-debt investor may owe taxes on “phantom gains” that exceed actual profits earned
from trading the debt.
For public debt, however, debtholders owe tax on the spread between the fair market value
of the modified debt and the purchase price of the original issue. In this case, investors owe tax
on just the profits that they can realize by selling the restructured debt. Furthermore, when
the first lender retains the debt and renegotiates it, the lender receives a tax credit reflecting
the loss they have incurred. This makes debt restructuring even more attractive.
Figure 1 displays an example of the tax obligations that debtholders face when restructur-
ing debt. In Panel A, a privately traded debt issue is restructured so that its new market value
is 50, while its par value remains 100. If a distressed-debt investor purchases the debt before
it is restructured for 40, they will owe tax of τ(100− 40). Because the debt is private, the IRS
6
regards the event as if the investor bought a debt issue for 40 and exchanged it for debt worth
100. The investor owes a tax payment that may be substantially larger than any gains they
obtain from trading the firm’s debt. The figure also shows that if the primary lender retains
the debt, it will receive no tax credit, but it will also not owe taxes.
Panel B shows that the debtholder is better off when the issue is public. In this case, an
investor would owe tax of just τ(50−40). The gain of 10 is the actual profit the investor would
receive from selling the modified debt at its market price. If the primary lender retains the
debt, it receives a tax credit based on the write-down from the restructuring.
Debt restructuring is a taxable event also for borrowers and the taxes owed depend on how
the IRS classifies the debt. Borrowers owe no tax when a private debt issue is restructured so
that the par value does not change.5 For public debt, borrowers incur “cancellation of debt
income” equal to the spread between the issue’s original par value and the market price of the
modified issue. This leads to a tax payment in the typical case that the modified debt is worth
less than the original issue. However, in this case the borrower’s tax payments are offset by an
equal-sized tax credit, called an “original issue discount.” The main difference is in the timing
of the tax payments: the income tax is immediately recognized, while the credit is spread out
over the years to maturity of the modified debt issue.6
2.2 Change in Debt Classification under TD9599
In parallel to developments in the debt markets during the Financial Crisis, government offi-
cials announced in various forums their plans to update the tax definition of public debt. At a
Practicing Law Institute Conference held on October, 2009, Treasury Deputy Tax Legislative
Counsel Jeffrey Van Hove and Treasury Deputy Assistant for Tax Policy Emily McMahon gave
presentations discussing the update. The public debt definition update was also listed as one
item in the department’s 2009-10 Priority Guidance Plan, released the following month. On
September 12, 2012 the IRS announced Regulation TD9599 redefining public debt, with an
effective date of November 13, 2012.
Before TD9599’s adoption, taxes were based on a 1994 regulation that classified debt as
publicly traded if it satisfied one of three conditions:
5Specifically, for private debt restructuring borrowers owe tax on the spread between the par value of theoriginal issue and the par value of the modified issue. When the par value does not change, this spread is zero.
6The 2009 American Reinvestment and Recovery Act delayed this tax by allowing firms that restructureddebt in 2009 or 2010 to spread out the cancellation of debt income from 2014 to 2018. This provision, however,expired in 2010 and does not affect our results.
7
1. The issue is listed on a securities exchange or traded in a market such as the interbank
market.
2. The issue’s price appears in a quotation medium.
3. A price quote can be obtained from dealers or traders.
TD9599 added to these conditions that debt would be classified as public also if a “soft”
price quote could be obtained from at least one broker, dealer, or pricing service. Most syn-
dicated loans satisfied this new condition and were reclassified from private to public debt.7
Notably, TD9599 also specifies a size threshold: debt can only be considered public if the
original issue amount exceeds $100 million. Issues smaller than this threshold were reclassified
as private, even if traded on an exchange.
The IRS applies the tax treatment for public debt to a restructuring if either the original or
modified debt issue meets the conditions outlined in TD9599. Therefore, a syndicated loan that
is issued before November 2012, but restructured afterward, is considered public for tax pur-
poses. This feature of the tax laws mitigates selection bias in our analysis, as the tax treatment
under TD9599 effects loans that were issued well before the regulation was even proposed.
3 The Model
The economy lasts for three periods t = 0, 1, 2. There is a borrower and a continuum of lenders
indexed by i ∈ I. The borrower needs financing and issues a measure 1 of debt securities in
t = 0. Each security promises to pay 1 unit of funds in t = 2 in exchange for a price p paid
in t = 0. Each lender has a unit demand for securities and the borrower raises funds from a
subset of mass 1 of lenders. All players are risk neutral and there is no discounting.
With probability λ the borrower is “sound” and generates a verifiable cash flow of y > 1
in t = 2. With probability 1 − λ the borrower is in “distress” in which case the value of
the borrower’s assets depends on whether he renegotiates the debt out of court or files for
bankruptcy (in-court restructuring). If the debt is restructured in court, the assets have a
verifiable recovery ratio of r < 1. In the event of an out-of-court renegotiation, the assets are
worth a positive amount v (θ), which is continuous and strictly increasing in θ. The borrower’s
7At the time, Cleary Gottlieb, a leading international law firm stated: “The final regulations are likely tocause most syndicated loans to be treated as publicly traded, especially as a result of the fact that indicativequotes — a term that is very broadly defined — may cause a loan to be publicly traded.”
8
Figure 2. Model Timing
t = 0 t = 1 t = 21. Borrower issues debt of 1. Lenders receive signals xi 1. If cash flow y is realized1 in exchange for p about the fundamental θ the borrower is sound
2. Lenders choose whether 2. Otherwise there isto buy CDS protection renegotiation:
(i) if renegotiation succeedsassets are worth v (θ)(ii) otherwise they yield r
continuation value θ is unknown to all participants until t = 2 and is drawn from a continuously
differentiable and strictly positive density k with support on the real line. We assume that
v (θ) is nonverifiable in t = 0 and 1, but becomes verifiable to the borrower and participating
lenders in t = 2. This opens room for out-of-court renegotiation between the parties in t = 2.
We also assume that v (θ) is non-verifiable to outside lenders, so the borrower cannot pledge
existing assets to obtain outside financing.
Renegotiation proceeds as follows. The borrower offers to each lender i an amount qi. If a
lender rejects the borrower’s offer, then renegotiation fails. In this case each lender i is entitled
to receive r in court. If all lenders accept the borrower’s offer, assets are worth v (θ) and each
lender i receives qi. The borrower receives v (θ)−∫qidi.
We also assume lenders are exposed to the borrower’s risk of bankruptcy, which creates
the need for insurance. In particular, we assume that lenders suffer an additional loss ` > 0
if they are not insured when the borrower files for bankruptcy. This loss can be thought as
regulatory penalty for the bank’s capital going below a required threshold (Thompson (2010))
or a deadweight loss that results from the lender approaching insolvency (Duffee and Zhou
(2001)). This is a simple way to model the bank’ exposure without having to model its capital
structure.
Each lender i’s insurance decision is made in t = 1, after receiving a noisy signal about the
borrower’s fundamental given by
xi = θ + σηi, (1)
where σ > 0 and the noisy terms are i.i.d. according to continuous and integrable density h
with support on the real line. Given their signals, lenders chose whether to buy CDS insurance
from a CDS provider. There is continuum of providers that act as price takers, and each of
9
them offers a CDS contract that pays 1 unit if there is a credit event in t = 2, in exchange
for a fee of f or “spread” paid up front. A credit event occurs only if the project fails and the
borrower declares bankruptcy. The CDS market is populated by liquidity traders that always
buy CDS and that CDS providers do not know whether the buyers of CDS are debt holders.
As we will show, the probability of bankruptcy depends only on lenders’ CDS positions. This
implies that CDS providers cannot learn about the probability of a credit event from CDS
demand. We assume that although lenders’ insurance positions are known to all participants
in t = 2, they cannot be contracted upon.
Finally, and importantly for our purposes, we model the real-world feature that lenders pay
a tax rate of τ on the gains from renegotiations that occur out of court. Following TD9599,
the amount of taxes paid depends on whether debt securities are classified as private or public.
If private, taxes are levied on the difference between the par value and the purchase price:
τ (1− p). If public, the tax rate applies to the difference between the debt’s value upon
renegotiation and the purchase price: τ (qi − p).
3.1 Equilibrium and Results
Our analysis proceeds as follows. We focus on the case in which the borrower is in distress in
t = 2. We first solve for the borrower’s offer to lenders that induces renegotiation, separately
for each tax classification of debt. Next, we show how the relationship between taxes and
renegotiation likelihood depends on the fraction of lenders that purchase insurance. We then
apply “global games” analysis to study lenders’ decisions about whether to purchase insurance,
and the resulting equilibrium CDS spread f and debt price p.8
Lenders agree to renegotiate if the borrower offers a stake in the continuation firm that
exceeds their outside option, which is 1 if the lender is insured and r otherwise. The borrower
optimally offers each lender i a stake that just meets this outside option. Let qpari and qmkti be
the offers made to lender i when the tax rate applies to the par (1 − p) and market (qi − p)taxable incomes, respectively. Then
qpari − τ (1− p) = qmkti − τ(qmkti − p
)= max r, 1− si , (2)
where si = 1 if lender i does not have a CDS and 0 if otherwise. Rearranging Eq. (2) yields
qmkti = (max r, 1− si − τp) (1− τ)−1 and qpari = qmkti + τ(1− qmkti
).
8See Carlsson and van Damme (1993a,b) and Morris and Shin (2003) for a comprehensive discussion onglobal games.
10
These expressions show that when qmkti < 1, the borrower must make a higher offer to
induce renegotiation when taxes apply to par values (where debt is classified as private) than
when taxes apply to market values (debt is classified as public). Note that qmkti < 1 leads to
renegotiation only when lenders are not insured. When lenders are insured, the borrower must
offer a net-of-tax payment qmkti > 1 — otherwise the lender will force bankruptcy and claim the
CDS payout of 1. However, in this case lenders owe more tax when debt is public (τ(qmkti − p
))
than when debt is private (τ (1− p)). This outcome can occur when the borrower does not
have sufficient cash to repay the debt, but can offer an equity stake in the going concern v (θ)
that exceeds debt owed.9
The impact of TD9599 on out-of-court renegotiation depends on lenders’ decisions to pur-
chase CDS insurance on the borrower. We now solve for the relationship between renegotia-
tion outcome and the fraction of insured lenders. Out-of-court renegotiation under tax rule
j = par,mkt fails if and only if:
v (θ) < Qj ≡∫qji di. (3)
Let l be the fraction of lenders that do not insure. When l = 1, renegotiation can succeed
for offers below 1, and Qmkt ≤ Qpar. Conversely, when all lenders insure (l = 0) the borrower
must offer more when taxes are levied on market values, so Qmkt ≥ Qpar. Since taxes are the
same when the out-of-court offer equals the face value of debt, there exists a critical level of
uninsured lenders such that Qmkt = Qpar = 1.
Expression (3) can be re-written as:
v (θ) < l qji (si = 0) + (1− l) qji (si = 1) (4)
Re-arranging (4) yields the threshold fraction of insured lenders P j (θ) that cause renegotiation
to fail under each tax classification:
Pmkt (θ) = [1− v (θ) + τ (v (θ)− p)] (1− r)−1 , (5)
and
P par (θ) = Pmkt (θ) + τ (1− v (θ)) (1− r)−1 . (6)
When l < P j (θ), the borrowers’ asset value v (θ) is too low to generate net-of-tax offers that
match insured lenders’ CDS payout. One implication is that when the borrower’s continuation
9In this case, lenders’ tax obligations would continue to be determined by the trading status of debt. TheIRS considers any property received in lieu of debt payments to be subject to tax if the property’s valueexceeds the purchase price of the debt.
11
value in distress is low (v (θ) < 1), the threshold value P par (θ) is higher than Pmkt (θ). The
opposite holds when the borrower’s continuation value in distress is high (v (θ) > 1). Expres-
sions (5) and (6) therefore show that the effect of TD9599 on renegotiation likelihood depends
on lenders’ decisions whether to purchase CDS. We discuss this link in greater detail after
solving for the full equilibrium below.
To generate testable predictions, we need to endogenize the CDS spread f and price of
debt p. This is a challenging problem, because each individual lender’s demand for insurance
(and hence the CDS spread) depends on both the borrower’s continuation value and other
lenders’ decisions whether to buy CDS. To see why, consider the payoffs to any lender i. If the
lender purchases insurance, they receive a constant payout π ≡ 1 − f whether a credit event
occurs or not. If the lender remains uninsured, the payoff is π = λ+ (1− λ) r when P j (θ) ≤ l
(renegotiation succeeds) and π = λ+ (1− λ) (r − `) when P j (θ) > l (renegotiation fails).
We use techniques developed by the global games literature to solve for the equilibrium. We
briefly describe the process here, and present the full details of the analysis in the Appendix.
The steps we follow to solve for equilibrium are:
1. We focus on the situation when lenders’ signals about θ become nearly precise. This
allows the equilibrium to depend only on strategic uncertainty about other lenders’ de-
cisions, and not fundamental uncertainty about the borrowers’ continuation value.
2. For the threshold lender who is indifferent between insuring and not insuring given beliefs
about l, the expected payoff from not insuring equals the constant payoff from buying
CDS. We use this fact to compute the equilibrium cutoff θ∗j for which the threshold lender
buys insurance: ∫ P j(θ∗j )
0
π dl +
∫ 1
P j(θ∗j )π dl = 1− f , (7)
3. We use expression (A.2) in the Appendix to solve for the equilibrium demand for CDS.
To solve for equilibrium CDS supply, we use the fact that CDS providers are com-
petitive so the fee they charge allows them to break even in expectation. This yields
fj = (1− λ)K(θ∗j), where K
(θ∗j)
is the quantity demanded of CDS.10
4. We substitute the above expression for fj into expression (A.2) to derive the equilibrium
CDS spread.
10Note that demand in expression (A.2) depends on f , which is endogenous. We first take f as givenand solve for the equilibrium demand and supply of CDS. Then, we show in the full characterization of theequilibrium with endogenous CDS fees that there is a unique cutoff satisfying (A.2).
12
This process leads to the following proposition on the equilibrium CDS spread, and its
relationship to tax classification of debt:
Proposition 1 Suppose lenders’ exposure to bankruptcy risk ` and recovery ratio r are suf-
ficiently large. In the limit σ → 0, the unique equilibrium of the game starting in t = 1 is
characterized as follows: (i) lenders follow monotone strategies with cutoff θ∗∗j such that no
lender insures if θ > θ∗∗j and all lenders insure if θ < θ∗∗j and (ii) CDS providers charge a CDS
fee given by f ∗j , where
θ∗∗j = P j−1 ([K(θ∗∗j)− (1− r)
]`−1)
(8)
f ∗j = (1− λ)K(θ∗∗j). (9)
In addition, θ∗∗par ≥ θ∗∗mkt if the probability that v (θ) ≤ 1 is sufficiently high, with reverse
inequality if otherwise.
This result shows that equilibrium effect of regulation TD9599 on the probability of bankruptcy
occurs through P j (defined in (5) and (6)), which determines the difficulty to renegotiate the
debt out of court. As discussed earlier, P par ≥ Pmkt when v (θ) ≤ 1 and P par > Pmkt other-
wise. If the borrower is likely to be in a critical situation in the event of distress (v (θ) ≤ 1),
then renegotiation only succeeds when a large fraction of lenders do not insure. Yet when
lenders do not insure, the tax burden of renegotiating debt is lower when taxes are based on
market prices, reducing the offer the borrower must make to induce renegotiation. Proposition
1 shows that these two effects interact so that lenders coordinate to not insure when v (θ) ≤ 1
and taxes are based on market values.
To complete the equilibrium, our last step is to determine the price of debt p∗j . Since lenders
are competitive, the equilibrium p∗j satisfies the following breakeven condition in t = 0:
p∗j = K(θ∗∗j (p∗)
) [1− f ∗j
]+(1−K
(θ∗∗j(p∗j)))
[λ+ (1− λ) r] , (10)
The right-hand side of (10) is positive and less than 1, so there exists p∗j in the unit interval that
satisfies the equality. Moreover, the equilibrium price of debt p∗j is decreasing in the probability
of bankruptcy given distress. Therefore, the borrower’s cost of financing is affected by the tax
rule through θ∗∗j (pj). From the results in Proposition 1, we conclude that the equilibrium fi-
nancing cost is lower when taxes apply to market values and the borrower’s continuation value
upon distress is likely to be low (v (θ) ≤ 1). Conversely, if the borrower’s continuation value
in distress is likely to be high, then taxes levied on par values result in a relatively lower cost
of debt.
13
We now characterize the full equilibrium of the game starting in t = 0 through the following
proposition:
Proposition 2 Suppose lenders’ exposure to bankruptcy risk ` and recovery ratio r are suf-
ficiently large. In the limit σ → 0, the unique equilibrium of the game starting in t = 0 is
characterized as follows: (i) the borrower issues debt at price p∗j , (ii) lenders follow monotone
strategies with cutoff θ∗∗j(p∗j)
such that no lender insures if θ > θ∗∗j(p∗j)
and all lenders insure
if θ < θ∗∗j(p∗j), and (iii) CDS providers charge a CDS fee given by f ∗j
(p∗j), where
θ∗∗j(p∗j)
= P j−1 ([K(θ∗∗j(p∗j))− (1− r)
]`−1)
,
f ∗j(p∗j)
= (1− λ)K(θ∗∗j(p∗j))
, (11)
p∗j = λ+ (1− λ) r − (1− λ)K(θ∗∗j(p∗j)) [
K(θ∗∗j(p∗j))− (1− r)
]. (12)
In addition, p∗par ≤ p∗mkt, θ∗∗par
(p∗par
)≥ θ∗∗mkt (p∗mkt), and f ∗par
(p∗par
)≥ f ∗mkt (p∗mkt) if the probabil-
ity that v (θ) ≤ 1 is sufficiently high, with reverse inequalities if otherwise.
Before we finish the model analysis, we derive an important result on the effect of taxes
on the borrower’s equity value. Since the borrower is the residual claimant, his payoff in
equilibrium (equal to v (θ)−Qj) is equivalent to equity value in our economy. If the borrower
is in distress in t = 2, he gets v (θ) − Qj(l, p∗j
), which equals v (θ) − Qj
(1, p∗j
)if θ > θ∗∗j
(p∗j)
and equals v (θ)−Qj(0, p∗j
)if otherwise. Therefore, his ex-ante payoff in t = 0 is
λ (y − 1)+(1− λ)
[∫ θ∗∗j (p∗j)
−∞
[v (θ)−Qj
(0, p∗j
)]k (θ) dθ +
∫ ∞θ∗∗j (p∗j)
[v (θ)−Qj
(1, p∗j
)]k (θ) dθ
],
(13)
which can be rewritten as
λ (y − 1) + (1− λ)[E (v (θ))−
(K(θ∗∗j(p∗j))Qj(0, p∗j
)+(1−K
(θ∗∗j(p∗j)))
Qj(1, p∗j
))].
(14)
Expression (14) shows that bankruptcy risk reduces the borrower’s equity value. Which
tax rule affects equity most depends on the expected fraction of lenders that do not insure
in equilibrium under each regime. If the probability that lenders do not insure is high when
taxes apply to par, then equity is likely to increase following a change to a tax regime based on
market values. The reason is that the average tax burden faced by lenders is lower for market
values. This lead us to the following proposition:
14
Proposition 3 The following holds for the borrower’s ex-ante payoff under different tax rules:
1. Suppose the probability that v (θ) ≤ 1 is sufficiently high. If the equilibrium expected
fraction of lenders that do not insure when taxes apply to par values is large enough(1−K
(θ∗∗par
(p∗par
))≥ l∗
(p∗par
)), the borrower’s equilibrium payoff in t = 0 is higher if
taxes apply to market values vs par values.
2. Suppose the probability that v (θ) ≤ 1 is sufficiently low. If the equilibrium expected
fraction of lenders that do not insure when taxes apply to market values is small enough
(1−K (θ∗∗mkt (p∗mkt)) ≤ l∗ (p∗mkt)), the borrower’s payoff in t = 0 is higher if taxes apply to
par values vs market values.
3.2 Testable Hypotheses
One can derive several results based on our model. They are summarized in the following set
of hypotheses.
Hypothesis 1: The CDS spread associated with the debt of a distressed borrower is lower than
that of an non-distressed borrower when taxes apply to market values. The converse is true
when taxes apply to par values.
Hypothesis 2: The equity value of a distressed borrower is lower than that of an non-distressed
borrower when taxes apply to market values. The converse is true when taxes apply to par val-
ues.
Hypothesis 3: The cost of financing of a distressed borrower is lower than that of an non-
distressed borrower when taxes apply to market values. The converse is true when taxes apply
to par values.
These results motivate the empirical tests of the next section, where we use the adoption
of TD9599 as a surrogate for a change from a system in which debt restructing taxes are based
on par values to a system in which those taxes are based on market values.
4 Empirical Results
In this section, we present our main results establishing that TD9599 made renegotiation more
likely, and thereby increased firm value. We first describe our data. We then present our
empirical model and identification strategy. Next, we show empirical evidence confirming our
15
model’s predictions that the tax change led distressed firms’ CDS spreads to decrease, their
equity values to increase, and their financing costs to decline.
4.1 Data
The starting point for our sample is a set of firms for which the most CDS contracts are
written. We identify these firms by collecting data from the Depository Trust & Clearing
Corporation (DTCC) on the thousand firms with the most outstanding CDS contracts. The
DTCC has published this list of firms on a weekly basis since October 2008, along with gross
and net notional CDS positions on each firm.11 We restrict our analysis to these firms because
secondary trading of their CDS contracts is substantially more liquid than for smaller firms
(Oehmke and Zawadowski (2013)).
We merge the DTCC sample with several other databases. We collect CDS spreads for
standard, five-year CDS contracts from Thomson Reuters’s Datastream. Data on syndicated
loans are from LPC–Dealscan. These data include loan signing and maturity dates, principal,
pricing, and other terms such as covenants. We use Compustat for firm fundamental data and
CRSP for stock price data.
Our sample starts with 1,215 firms that appear in the DTCC database between October
2008 and March 2013 with CDS spreads available from Datastream. We exclude 445 firms that
are either foreign-based or state-owned, and 116 financial institutions. We drop 152 with no
information in LPC–Dealscan and 242 firms with missing Compustat data. We are left with
a sample of 260 unique firms.
4.2 Empirical Specification and Identification Strategy
Our model predicts that redefining the status of debt from private to public under the US tax
law could lead to a decline in CDS spreads. TD9599 works as a surrogate for such change. The
regulatory piece, however, only affected the tax status of syndicated loans. As such, its effects
should be larger for firms whose overall debt obligations contained more syndicated loans. This
observation allows us to build our baseline difference-in-differences specification for firm i in
11Gross notional is the sum of all CDS contracts written on the reference entity. Net notional positions arecalculated after canceling out offsetting positions, such as when a party buys CDS contracts on a particularfirm and then later sells contracts on the same firm.
16
week t:
CDSSpreadi,t = α + βHighSyndicatedi,t + µPostTDt
+ γ(HighSyndicatedi,t × PostTDt) + δXi,t + εi,t (15)
CDSSpreadi,t is the CDS spread on the firm’s debt. HighSyndicatedi,t is an indicator that
equals 1 if the firm’s syndicated loans-to-total debt ratio at the start of fiscal year 2012 is in
the highest tercile of the sample, and 0 if it is in the lowest tercile (we omit firms in the middle
tercile). The indicator PostTDt equals 1 for weeks after the announcement of TD9599, and 0
for weeks before, where we show results for alternative time windows.
Xi,t is a vector of control firm-level variables that affect the level of CDS spreads (see Tang
and Yan (2013)), as well as time-varying macroeconomic variables that could influence spread
changes. Firm-level controls include: Firm Leverage, measured as current liabilities and long-
term debt over total assets; Firm Volatility, the standard deviation of daily stock returns from
the past year; Tobin’s Q, the ratio between the market and the book value of the firm; Log
Assets ; Return on Assets, measured as net income before interest on debt over total assets;
Tangibility, measured as property, plant and equipment over total assets; Liquidity Needs, the
ratio of current to total debt; and Credit Rating, the S&P rating on the firm’s long-term debt.
Market-level controls include: Market Volatility, the weekly level of the VIX index; and Term
Slope, the weekly difference between the yield on a 10-year U.S. Treasury bond and 2-year
Treasury Note. Complete definitions for each variable are in the Data Appendix.
The primary coefficient of interest in Eq. ((15)) is γ. A negative γ indicates that CDS
spreads decline more for firms with high syndicated loans–debt ratios than firms with low
loans–debt ratios after TD9599 is announced. Also of interest is µ, which estimates the av-
erage change across all firms in CDS spreads after that same announcement. Unobservable
shocks could affect spread levels for several weeks, so we cluster standard errors at the firm
level to account for possible serial correlation.12
According to our analysis, the tax-induced changes in debt restructuring costs will be more
relevant for borrowers with weaker fundamentals — borrowers on the verge of going bankrupt,
triggering CDSs. We test this prediction empirically by adding yet another layer of contrasts
to Eq. (15), which is based on measures of financial distress. In doing so, we essentially pur-
sue a triple-difference estimation approach, but we do so under a more parsimonious model
12Our results do not change when we estimate our base tests on spreads averaged across the weeks beforeand after the announcement.
17
representation. To wit, we estimate Eq. ((15)) separately for distressed and non-distressed
firms, contrasting the interaction terms of interest, γ. In these estimations financial distress is
measured by both the Altman Z-Score and distance-to-default (based on the Merton (1974)).
Table 1 presents summary statistics for the variables used in our tests. Data on firm funda-
mentals are from the start of fiscal year 2012. Panel A separates the sample into distressed and
non-distressed firms, based on whether a firm’s 2012 Altman Z-Score is above or below 3. Un-
surprisingly, distressed firms have higher CDS spreads and leverage ratios. They also have lower
growth opportunities (as measured by Tobin’s Q) and a lower return on assets. Importantly
for our identification strategy, distressed and non-distressed firms have very similar summary
statistics for the syndicated loans–debt distribution. This suggests that these treatment as-
signment variables are largely orthogonal and unconfounded with each other. As such, whether
a firm has a high or low loans–debt ratio says nothing about whether or not it is in distress.
Table 1 About Here
Panel B compares firms with syndicated loans–debt ratios in the highest and lowest terciles
of the sample distribution (firms with HighSyndicated equal to 1 or 0). High- and low-loan
firms are similar along many key firm-level characteristics, including Z-Score, leverage, growth
opportunities, and recent performance. However, firms that use more syndicated loans are
smaller than firms using other types of debt. This is consistent with prior literature and
underscores the importance of controlling for firm size in our regressions.
In order for Type I error to affect our estimation, an omitted variable must cause CDS
spreads to drop exactly on TD9599’s announcement date. Further, in light of our triple-
difference approach, the omitted variable would bias our interpretation of the interaction co-
efficient γ only if it was correlated with firms’ financial conditions and syndicated loans–debt
ratios. This is a high bar for a plausible omitted variables case.
4.3 Changes in CDS Spreads
4.3.1 Graphical Evidence
We start out by testing whether CDS spreads generally respond to the announcement of
TD9599 by the IRS on September 12, 2012. Our model predicts that spreads should de-
cline for at least some firms in the market. Figure 3 shows that the sample average CDS
spreads dropped by 27 basis points when the IRS announced the new regulation. This is the
single largest drop for our sample firms since mid-2009.
18
Figure 3. Effect of TD9599 on Aggregate CDS Spreads
This figure compares the drop in CDS spreads after the Sept. 12, 2012 TD9599announcement to the average change in CDS spreads across other 2-week blockssince 2010. Spread changes are calculated for the entire sample of firms in theintersection of the LPC–Dealscan, CRSP, and DTCC databases.
Following our theoretical priors, we partition our sample firms according to their level
of financial distress (based on Z-scores) and their use of syndicated loans. Specifically, we
use a 2 × 2 partition where we classify firms into 4 buckets: Distressed–High Syndicated
Loans, Distressed–Low Syndicated Loans, Non-Distressed–High Syndicated Loans, and Non-
Distressed–Low Syndicated Loans. For each of these four groups, Figure 4 illustrates the
behavior of characteristic-adjusted spreads from 15 weeks prior through 10 weeks after the
IRS announcement. To be precise, for each group we regress weekly CDS spreads on Firm
Leverage, Log Assets, and week-fixed effects. We then plot the coefficients of the weekly effects,
with each group starting from a common base in week -15 to emphasize how spreads diverged
19
after that point.
Figure 4 shows that CDS spreads drop by 48 basis points at the announcement for Distressed–
High Syndicated Loans. Spreads drop by a much smaller 18 basis points for Distressed–Low
Syndicated Loans firms, 20 basis points for Non-Distressed–High Syndicated Loans firms, and
by just 13 basis points for Non-Distressed–Low Syndicated Loans firms. These are rough es-
timates of the impact of Regulation TD9599 on spreads of various sample firm partitions, but
the patterns depicted are striking.
4.3.2 Regression Results
In Table 2 we estimate regressions based on Eq. ((15)), separately for distressed and non-
distressed firms, where distress is measured by Z-Scores and firms are deemed as financially
distressed when Z-Score is less than 3. We use 7-week and 2-week windows around the IRS
announcement (alternative windows yield consistent results and are available upon request).
The main variable of interest is the interaction term HighSyndicated×PostTD. The negative
coefficient in columns (1) and (2) indicates that in the 7-week window spreads decreased about
30 basis points for those firms at the highest tercile of syndicated loans–debt ratio relative to
firms at the lowest tercile. These coefficients are statistically significant at the 1% test level.
In contrast, comparable coefficients for non-distressed firms are much smaller and statistically
indistinguishable from 0. Notably, the coefficient on PostTD is negative and highly significant
only for distressed firms. The results in the narrower 2-week window around the rule change
are similar (columns (5) through (8)). Finally, the coefficients on our control variables are
consistent with the findings of other work on CDS spreads. Firms with higher leverage and
firm-specific volatility have significantly higher spreads. Increases in the term slope, which
may indicate improving macroeconomic conditions, lead spreads to drop.
Table 3 repeats the analysis using distance-to-default to measure firm distress. In the
Merton (1974) model, a firm defaults when the value of its assets falls below the book value
of debt. Distance-to-default is the number of standard deviations by which the log of (asset
market value)/(debt book value) must fall in order for default to occur. In Table 3, distressed
firms are those with a below-median distance-to-default measure, and non-distressed firms are
those with an above-median measure.
Table 3 About Here
The results are largely similar to those in Table 2. Spreads decrease more for distressed than
20
Figure 4. CDS Spreads Sorted by Distress and Loans–Debt Ratio
This figure plots CDS spreads in the weeks surrounding the Sept. 12, 2012 TD9599announcement. All spreads are adjusted for leverage and firm size. Sample firmsare sorted into four buckets based on firm distress and syndicated loans–debt ratio.Distressed firms have 2012 Altman Z-Score < 3, and Non-distressed firms have Z-Score ≥ 3. HighSyndicated firms have a syndicated loans–debt ratio in the highesttercile of the sample distribution, and LowSyndicated firms have a loans–debt ratioin the lowest tercile. The loans–debt ratio is principal for all syndicated loansoutstanding in 2011 divided by total debt in 2011. The sample is all firms in theintersection of the LPC–Dealscan, CRSP, and DTCC databases.
for non-distressed firms, and the decrease is larger for high-loan–debt firms than low-loan–debt
firms. In the narrower 2-week window, spreads decrease by 24 basis points for distressed firms,
compared to just 6.7 basis points for non-distressed firms. Distressed/high-loan firms experi-
ence a further 19.6 basis point-drop in spreads relative to distressed/low-loan firms. As in Table
2, among non-distressed firms spreads do not change for high-loan–debt versus low-loan–debt
firms.
21
The results in both Table 2 and 3 support the predictions of our model, and indicate that
the change in tax status of syndicated loans increased out-of-court renegotiation likelihood.
Our theoretical model predicts that lenders’ tax burdens decrease when syndicated loans are
reclassified from private to public, and that this reduces the minimum offer that distressed
borrowers must pledge for renegotiation to succeed. The observed drop in CDS spreads on
TD9599’s announcement is consistent with lenders demanding less insurance after the tax
change. Notably, spreads decrease relatively more for distressed firms, as our model predicts.13
Because we obtain the same results using Z-Score and distance-to-default, throughout the
rest of the paper we report results for distress measured using just Z-Score. This conserves
space, yet results for distance-to-default are available from the authors upon request.
4.4 Stock Returns
Next, we examine how TD9599 affected the market value of borrower and lender firms. Hy-
pothesis 2 predicts that the tax change increases the equity value of borrowers with weak
fundamentals, relative to other borrowers. This should obtain because bankruptcy filings typ-
ically wipe out shareholders’ equity in the firm, while shareholders usually retain a stake in
the going concern after out-of-court renegotiation. We test this empirically by examining
borrowers’ stock returns surrounding TD9599’s announcement date.
Panel A of Table 4 presents borrower firms’ cumulative abnormal returns (CARs) from 1
day before to 2 days after the announcement, and also from a wider 3-day window around
the announcement. We keep only firms that have at least one syndicated loan outstanding at
the start of fiscal year 2012, sorting firms into four portfolios based on financial condition and
syndicated loans–debt ratio. Daily abnormal returns are calculated based on the Fama-French
three-factor model (Fama and French (1993)). CARs are the sum of returns over each event
window. We account for the fact that all firms share a common event date by using the Brown
and Warner (1980) portfolio method to calculate t-statistics.
Table 4 About Here
The table shows that only distressed/high-loan firms outperform the market around the
13We further observe a monotonic relationship between firm distress and the decrease in CDS spreads,which supports our model’s predication that the effect of the tax change is related to firm fundamentals.In unreported results, we re-estimate column (6) of Table 2 for highly-distressed firms (Z-Score below 1.23)versus “gray zone” firms (Z-Score between 1.23 and 3). The coefficient on HighSyndicated × PostTD issignificant in both subsamples, but it is twice as large for highly-distressed firms. The coefficient on PostTDalso monotonically decreases with firm distress.
22
regulation announcement. These firms earned a 1.16% abnormal return in the shorter window,
and a 1.57% abnormal return in the 3-day window. Both CARs are statistically significant.
Over the same period, distressed/low-loan firms underperformed the market by close to 1%,
and non-distressed firms did not outperform the market. These results suggest that TD9599
led to a significant increase in the market value of the borrower firms most affected by the
regulation. It is unlikely that an unobserved contemporaneous shock led to this value increase,
as the value of other distressed borrowers simultaneously decreased.
We also examine whether TD9599 created value for major syndicated lenders. Gains for
these lenders would stem from regulation-induced increases in their distressed-debt recoveries
and a drop in their tax obligations on renegotiated loans. In this analysis, we consider a port-
folio of 21 syndicated lenders that each were lead arrangers for at least 50 syndicated loans
outstanding in 2012. Together, these lenders arranged 95% of outstanding loan principal in
our sample. We find that a size-weighted portfolio of these lenders earned a 4.3% CAR in the
3-day window around the announcement of TD 9599 (t-stat of 2).
Our results indicate that taxing renegotiated debt based on market instead of par values
generates a substantial increase in value for lenders. Not only did TD9599 affect the CDS
market by decreasing bankruptcy likelihood, but it also benefitted shareholders of both lender
and borrower firms. Furthermore, this result shows that a completely different group of firms
also benefitted from the regulation in a way that is consistent with our model. An alternative
explanation for our results would need to account for the decrease in CDS spreads for borrowers
and the simultaneous increase in stock returns for lenders.
4.5 Financing Access and Costs
Our model predicts that borrowers’ financing costs decrease following a reduction in the tax
burden of renegotiating debt. This is because the increase in renegotiation likelihood leads to
higher payoffs from distressed borrowers to lenders, and in response lenders reduce the price
of new financing. In this section, we test this prediction empirically by examining whether
borrowers affected by the rule change subsequently gain access to cheaper funding.
We use a difference-in-differences specification that compares new financing terms for dis-
tressed vs. non-distressed borrowers, in a 12-month window before and after the TD9599
announcement. Our sample includes new syndicated loans issued to all U.S. non-financial
firms in the intersection of Compustat, LPC–Dealscan, and CRSP. We exclude syndicated
loans with issue-date principal below $100M, because TD9599 did not change the tax status
23
of these loans. We use a 12-month window because firms may not take new loans immediately
after the tax change, and also to account for seasonal patterns in loan issuance (see Murfin
and Peterson (2013)).
Table 5 contains our results. Columns (1) and (2) test whether distressed borrowers were
more likely to obtain a new loan after the TD9599 announcement. These columns present
results from logistic regressions with two observations per firm, one for the 12-month period
before and one for the 12-month period after the announcement (we omit September 2012 in
all tests). For each period, Obtained Loan equals 1 if the firm signed a syndicated loan, and 0
if it did not. This analysis is akin to a simple, characteristic-adjusted cross-tabulation of the
fraction of distressed and non-distressed borrowers that obtained a loan in each 12-month pe-
riod. The key explanatory variable is the interaction term Distressed× PostTD, which tests
whether distressed borrowers are more likely to obtain a new loan after the TD9599 announce-
ment than non-distressed borrowers. Column (1) compares distressed borrowers with Z-Score
< 3 to non-distressed borrowers with Z-Score ≥ 3, and column (2) compares highly-distressed
borrowers with Z-Score < 1.86 to non-distressed borrowers.14 Using different Z-Score cutoffs
allows us to examine whether financing terms improved for borrowers with different levels of
distress. The regressions also control for the same firm and macroeconomic characteristics as
in the previous section.
Table 5 About Here
The results show that distressed borrowers are more likely to obtain a new loan after
TD9599, relative to non-distressed borrowers. Coefficients in columns (1) and (2) represent
the marginal effects of each variable. The coefficient on Distressed× PostTD in column (1)
indicates that distressed borrowers are 11% more likely to obtain a new loan after TD9599.
Column (2) shows that the likelihood of obtaining a new loan is even greater for highly dis-
tressed borrowers, and statistically significant at the 1% level. The negative coefficient on
PostTD indicates an overall tightening in the loan market after late 2012.
Next, columns (3) and (4) show that markups on new syndicated loans decreased for dis-
tressed borrowers after TD9599. The table shows results from OLS regressions, with each
observation corresponding to a new loan in the 12-month window around the announcement.
The dependent variable New Loan Markup is the percentage-point all-in drawn spread. This
14Z-Score cutoffs of 1.86 and 3 are widely used in the literature to define firms as highly distressed, “grayzone”, or non-distressed (see e.g. Altman (2000)).
24
variable equals the spread that borrowers pay on top of a floating base rate, and thus directly
measures borrowers’ financing costs. PostTD equals 1 for loans signed after September 2012
and 0 for loans signed before. The regressions also control for market volatility and the term
slope of interest rates from the loan’s signing week.
The results show that markups decreased by 26 to 30 basis points for distressed firms after
TD9599, relative to non-distressed firms. This drop represents a 14% decrease over the mean
pre-announcement markup of 2.2%. These estimates are statistically significant at the 5%
level, with p-value below .02 in column (3). The decrease is similar for all distressed firms and
highly distressed firms.
One potential concern is that the decrease in markups could be due to more high-quality
borrowers obtained loans after TD9599, instead of decreases in out-of-court renegotiation
costs. Our results indicate that instead loans were increasingly issued to riskier borrowers
after TD9599, making the observed decrease in markups even more striking. Furthermore, we
re-estimate columns (3) and (4) using a sample of borrowers that obtained loans both before
and after TD9599. In unreported results, we find that loan markups decrease for the same
distressed borrower after the tax change.
The evidence presented in this section supports Hypothesis 3 of our model. An increase in
renegotiation likelihood allows more distressed firms to enter the syndicated loan market, and
to borrow at lower cost. These results are consistent with our model’s prediction that higher
renegotiation likelihood increases lenders’ payoffs, allowing them to increase their up-front
lending commitments to distressed borrowers.
5 Robustness Checks
We conclude our analysis by testing whether unobserved heterogeneity across firms, and not
reduced out-of-court renegotiation costs, could explain the decrease in CDS spreads around
TD9599’s announcement. One potential concern with our specification in tables 2 and 3 is
that the parallel trends assumption may not hold: prior to TD9599 spreads may have been
decreasing relatively more for distressed/high-loan firms, due to macroeconomic or other ef-
fects unrelated to debt restructuring costs. It is also possible that distressed/high-loan firms
have more volatile CDS spreads in general, so that any positive shock coinciding with TD9599
could produce larger declines in the spreads of these firms. It is therefore important to rule
out the possibility that structural differences alone could lead to the results we observe.
25
Figure 5. CDS Spread Changes, TD9599 Announcement vs.Placebo Events
This figure compares the decrease in CDS spreads around the Sept. 12, 2012TD9599 announcement to the change in spreads around a series of placebo events.Placebo estimates are constructed by splitting the sample into non-contiguous 4-week blocks ranging from January 2010 to December 2012. Each placebo eventestimates the regression from Column 6 of Table 2, with PostTD redefined to equal1 on the third and fourth week of each block. Bars are the coefficients on HighSyn-dicated x PostTD for each block, and are sorted from largest to smallest coefficient.The sample comprises all distressed firms (2012 Altman Z-Score < 3) in the inter-section of the LPC–Dealscan, CRSP, and DTCC databases.
We do so by conducting a placebo analysis. We split the period from January 2010 to
December 2012 into non-contiguous 4-week blocks. In each block we repeat the regression
from Table 2 Column 6, however we re-define PostTD to equal 1 in the third and fourth week
of each block. We focus our test on just distressed firms. Figure 5 plots the distribution of
the resulting coefficients on the interaction term HighSyndicated×PostTD from each block,
26
sorted from largest to smallest coefficient. We find that the change in CDS spreads around
TD9599’s announcement is the second-largest decrease out of all 50 blocks. Furthermore, in
only 6 of the blocks is the interaction term coefficient negative and statistically significant at
conventional levels.
These results indicate that the set of firms most affected by TD9599 experienced an un-
precedentedly large decrease in CDS spreads right when the regulation was announced. This
provides further support that bankruptcy risk decreased for these firms due to the reduction
in taxes owed upon out-of-court renegotiation. Any alternative explanation must include an
unobserved variable that coincides with TD9599 and produces an abnormally large decrease
in bankruptcy risk for distressed/high-loan firms.
6 Conclusion
Firms experiencing distress usually attempt to renegotiate debt out of court (Gertner and
Scharfstein (1991) and Franks and Touros (1994)). These negotiations are thought to be value-
enhancing as mutually-agreed resolutions should preserve more value than costly restructuring
in bankruptcy under court supervision. Out-of-court debt modifications contemplated by dis-
tressed firms are significantly large and they must compensate creditors for both their outside
options and taxes associated with the exchange.
In this paper, we model and test the effect of changes in debt restructuring costs on
bankruptcy risk, financing costs, and shareholder value of borrowers and lenders. We do so
examining a change in the US tax code that reduces the taxes lenders pay when restructuring
syndicated loans out of court, but does not affect other types of debt (Regulation TD9599).
This modification allows us to disentangle the relative costs of in-court and out-of-court re-
structuring. Using data from CDS contracts, we show that renegotiation becomes more likely
following a relative decrease in out-of-court costs, and this in turn increases the market value
of both firms and banks. We further show that the capital markets respond by granting firms
access to cheaper bank credit and by increasing up-front commitments to long-term funding.
Our paper presents well-identified evidence that taxes are a major transaction cost in debt
renegotiation of distressed firms. In all, the analysis provides insights into how altering regu-
latory constraints can improve welfare in financial distress.
27
Appendix A
We first explain some technical details regarding the procedure we follow to solve our propo-
sitions, and then present proofs.
Lender i’s net payoff of not insuring over insuring is
Πi =
π − π, if P j (θ) ≤ lπ − π, if P j (θ) > l
. (A.1)
The net payoff Πi is increasing in the fraction of lenders that do not insure (strategic com-
plementarity) and in the fundamental θ (state monotonicity), which determines the value of
the borrower’s assets out of court. Specifically, if π > π and π < π, lenders face an enormous
coordination problem as their decision to insure or not crucially depends on their beliefs about
both the fundamental θ and the fraction l of lenders that do not insure.
Suppose lenders follow a monotone strategy with a cutoff k, that is, they do not insure
if their signal is above k and insure otherwise. Lender i’s expectation about the fraction of
lenders that do not insure conditional on θ is simply the probability that any lender observes
a signal above k, that is, 1−H(k−θσ
). This proportion is less than z if θ ≤ k − σH−1 (1− z).
Each lender i calculates this probability using the estimated distribution of θ conditional on
his signal xi. A well known result in the literature of global games is that as σ → 0 this
probability equals z for xi = k. That is, the threshold type believes that the proportion of
lenders that do not insure follows the uniform distribution on the unit interval.
By focusing on the situation when signals become nearly precise, we focus on strategic
uncertainty rather than on fundamental uncertainty. The equilibrium cutoff can then be com-
puted by using the fact that the threshold type must be indifferent between insuring and not
insuring given his beliefs about l. Let θ∗j be the cutoff under tax regime j .Then θ∗j is such that∫ P j(θ∗j )
0
(π − π) dl +
∫ 1
P j(θ∗j )(π − π) dl = 0, (A.2)
which yields the condition
P j(θ∗j)
= (π − π) [(1− λ) `]−1 . (A.3)
The threshold in (A.3) exists and is unique for π > π and π < π. However, these payoff
depend on the CDS fee f , which is endogenous. Our strategy is to assume these conditions
hold and solve for the equilibrium demand for CDS taking f as given. Then we show in the
full characterization of the equilibrium with endogenous CDS fees that there is a unique cutoff
satisfying (A.2). These results for a fixed CDS fee f are formalized in Lemma 1 below.
Lemma 1 Suppose there exists ε such that π−π > ε > 0 > −ε ≥ π−π. In the limit σ → 0,
the unique equilibrium of the game starting in t = 1 for a given f is in monotone strategies
with cutoff θ∗j and has all lenders not insuring if θ > θ∗j and insuring if θ < θ∗j , where
θ∗j = P j−1 ((π − π) [(1− λ) `]−1) (A.4)
28
Proof of Lemma 1. Morris and Shin (2003) prove this result for a general class of global
games that satisfies the following conditions: (i) Πi increasing in θ, (ii) Πi increasing in l, (iii)
there exists a unique θ∗ that satisfies∫ 1
0Πidl = 0, (iv) there exists θ, θ, and ε > 0 such that
Πi ≤ −ε for all l ∈ [0, 1] and θ ≤ θ and Πi > ε for all l ∈ [0, 1] and θ ≥ θ, (v) continuity of∫ 1
0g (l) Πidl with respect signal xi and density g, and (vi) the expected value of ηi is finite. The
assumption π − π > ε > 0 > −ε ≥ π − π stated in the proposition implies (i), (ii), (iii), (iv).
Condition (v) is clearly satisfied. Condition (vi) is assumed in the description of the model.
The results above characterized lenders’ demand for insurance for a given CDS fee f . In
order to find the equilibrium demand for CDS we need to determine its supply. CDS providers
are competitive and the fee they charge is such that the break even in expectation. This
requirement implies that
fj = (1− λ)K(θ∗j)
. (A.5)
Therefore, the CDS fee fj charged by CDS providers is increasing in “quantity demanded”
K (θ∗), that is, the probability of a credit event conditional on distress. The equilibrium CDS
fee is the one that simultaneously satisfies supply and demand. Plugging (A.5) into (A.2) gives
the following condition for the equilibrium cutoff θ∗∗j :∫ P j(θ∗∗j )
0
(1− λ)[K(θ∗∗j)− (1− r)− `
]dl +
∫ 1
P j(θ∗∗j )(1− λ)
[K(θ∗∗j)− (1− r)
]dl = 0.
(A.6)
This equation simplifies to
K(θ∗∗j)− (1− r)− P j
(θ∗∗j)` = 0. (A.7)
This leads to the proof of Proposition 1.
Proof of Proposition 1. With endogenous f (θ) we no longer need the assumption
π−π > ε > 0 > −ε ≥ π−π made in Lemma 1. Now, Πi clearly satisfies (i), (ii), (iv), and (v).
Condition (vi) is assumed in the model. Therefore, it suffices to show (iii). We show that this
condition holds for ` and r sufficiently large.
First, let us derive some properties of the aggregate offer Qj (0). We have that Qmkt (0) ≥Qpar (0) ≥ 1 if all lenders buy CDS and Qmkt (1) ≤ Qpar (1) if no lender insures. Thus, there
exists l∗ such that Qmkt (l∗) = Qpar (l∗) = 1. Because the aggregate offer when taxes apply to
market values is more sensitive to the fraction of lenders that do not insure, Qmkt ≥ Qpar for
l ≤ l∗ and Qmkt ≤ Qpar for l ≥ l∗.
As θ → v−1 (Qj (1)), P j (θ)→ 1 and the left-hand side of (A.7) becomes Ωj ≡ K (v−1 (Qj (1)))−(1− r)− `, which is negative for ` sufficiently large. If θ → v−1 (Qj (0)), P j (θ)→ 0, and it ap-
29
proaches Ωj ≡ K (v−1 (Qj (0)))− (1− r), which is positive for r large enough. This establishes
(iii).
Finally, we show the comparative statics relative to the cutoffs θ∗∗mkt and θ∗∗par. For θ →v−1 (Qj (l∗)), the left-hand side of (A.7) converges to Ω ≡ K (v−1 (1))− (1− r)− l∗`.
If Ω ≥ 0, then the unique θ∗∗j that satisfies (A.7) is such that v−1 (Qj (1)) ≤ θ∗∗j ≤ v−1 (1).
Since Qpar (1) ≥ Qmkt (1), if v (θ∗∗mkt) < Qpar (1) it directly follows that θ∗∗mkt < θ∗∗par. If v (θ∗∗mkt) ≥Qpar (1), then for all θ such that v−1 (Qpar (1)) ≤ θ ≤ v−1 (1) we have that P par (θ) ≥ Pmkt (θ).
Along with (A.7), this implies that θ∗∗mkt ≤ θ∗∗par.
If Ω < 0, then the unique θ∗∗j that satisfies (A.7) is such that v−1 (1) < θ∗∗j ≤ v−1 (Qj (0)).
Since Qmkt (0) ≥ Qpar (0), if v (θ∗∗mkt) > Qpar (0) it directly follows that θ∗∗mkt > θ∗∗par. If v (θ∗∗mkt) ≤Qpar (0), then for all θ such that v−1 (1) < θ ≤ v−1 (Qpar (0)) we have that P par (θ) ≤ Pmkt (θ).
Along with (A.7), this implies that θ∗∗mkt ≥ θ∗∗par.
Proof of Proposition 2. This follows straight from Proposition 1 and from the expression
for the price of debt provided in (10).
Proof of Proposition 3. Remember that the aggregate offer Qj (l, p) is a weighted
average of the extreme offers when no lender insures and when all lenders insure. That is,
Qj (l, p) = (1− l)Qj (0, p) + lQj (1, p). Therefore, the negative term in brackets in (14) can be
written as
Qj(1−K
(θ∗∗j(p∗j)), p∗j)
=(K(θ∗∗j(p∗j))Qj(0, p∗j
)+(1−K
(θ∗∗j(p∗j)))
Qj(1, p∗j
)).
We also have that
Qpar(1−K
(θ∗∗par
(p∗par
)), p)≥ Qmkt
(K(θ∗∗par
(p∗par
)), p)
if 1−K(θ∗∗par
(p∗par
))≥ l∗ (p) and
Qpar(1−K
(θ∗∗par
(p∗par
)), p)≤ Qmkt
(K(θ∗∗par
(p∗par
)), p)
if 1−K(θ∗∗par
(p∗par
))≤ l∗ (p).
Therefore, if K (v−1 (1)) is sufficiently high and 1−K(θ∗∗par
(p∗par
))≥ l∗
(p∗par
),
Qpar(1−K
(θ∗∗par
(p∗par
)), p∗par
)≥ Qmkt
(1−K
(θ∗∗par
(p∗par
)), p∗par
)≥ Qmkt (1−K (θ∗∗mkt (p∗mkt)) , p
∗mkt) ,
where the last inequality follows from Proposition 2.
If K (v−1 (1)) is sufficiently low and 1−K (θ∗∗mkt (p∗mkt)) ≤ l∗ (p∗mkt), then
Qmkt (1−K (θ∗∗mkt (p∗mkt)) , p∗mkt) ≥ Qpar (1−K (θ∗∗mkt (p∗mkt)) , p
∗mkt)
≥ Qpar(1−K
(θ∗∗par
(p∗par
)), p∗par
),
where the last inequality follows from Proposition 2.
30
References
[1] Altman, E., 1968, “Financial Ratios, Discriminant Analysis and the Prediction of Corpo-rate Bankruptcy,” Journal of Finance 23, 589–609.
[2] Altman, E., 2000, “Predicting Financial Distress of Companies: Revisiting the Z-Scoreand ZETA Models,” NYU Working Paper.
[3] Ashcraft, A.B. and J. Santos, 2009, “Has the CDS Market Lowered the Cost of CorporateDebt?,” Journal of Monetary Economics 56, 514-523.
[4] Asquith, P., Gertner, R. and D. Scharfstein, 1994, “Anatomy of Financial Distress: AnExamination of Junk-Bond Issuers,” Quarterly Journal of Economics 109, 625-658.
[5] Benmelech, E., Garmaise, M.J., and T. Moskowitz, 2005, “Do Liquidation Values AffectFinancial Contracts? Evidence from Commercial Loan Contracts and Zoning Regulation,”Quarterly Journal of Economics 120, 1121–1154.
[6] Benmelech, E. and N. Bergman, 2008, “Liquidation Values and the Credibility of FinancialContract Renegotiation: Evidence from U.S. Airlines,” Quarterly Journal of Economics123, 1635–1677.
[7] Blanco, R., Brennan, S. and I.W. Marsh, 2005, “An Empirical Analysis of the DynamicRelation between Investment-Grade Bonds and Credit Default Swaps,” Journal of Finance60, 2255–2281.
[8] Bolton, P. and M. Oehmke, 2011, “Credit Default Swaps and the Empty Creditor Prob-lem,” Review of Financial Studies 24, 2617–2655.
[9] Bolton, P. and D. Scharfstein, 1996, “Optimal Debt Structure and the Number of Credi-tors,” Journal of Political Economy 104, 1–25.
[10] Campello, M., and R. Matta, 2012, “Credit default swaps and risk-shifting,”EconomicsLetters 117, 639–641.
[11] Campello, M., and R. Matta, 2013, “Credit Default Swaps, Firm Financing and theEconomy,” Working Paper, Cornell University and University of Amsterdam.
[12] Carhart, M.M., 1997, “On Persistence in Mutual Fund Performance,” Journal of Finance52, 57–82.
[13] Carlsson, H., and E. van Damme, 1993, “Global Games and Equilibrium Selection,”Econometrica 61, 989–1018.
[14] Diamond, D., 2004, “Committing to Commit: Short-Term Debt When Enforcement isCostly,” Journal of Finance 59, 1447–1479.
[15] Duffee, G. R., and C. Zhou, 2001, “Credit Derivatives in Banking: Useful Tools forManaging Risk?,” Journal of Monetary Economics 48, 25–54.
[16] Fama, E. and K. French, 1993, “Common risk factors in the returns on stocks and bonds,”Journal of Financial Economics 33, 3–56.
[17] Franks, J. and W. Torous, 1994, “A comparison of financial recontracting in distressedexchanges and Chapter 11 reorganizations,” Journal of Financial Economics 35, 349–370.
31
[18] Gertner, R. and D. Scharfstein, 1991, “A Theory of Workouts and the Effects of Reorga-nization Law,” Journal of Finance 46, 1189–1222.
[19] Giambona, E., Lopez-de-Silanes, F., and R. Matta, 2013, “Improved Creditor Protectionand Verifiability in the U.S.,” University of Amsterdam Working Paper.
[20] Gilson, S., John, K. and L.H.P. Lang, 1990, “Troubled debt restructurings: An empiricalstudy of private reorganization of firms in default,” Journal of Financial Economics 27,315–353.
[21] Gilson, S., 1997, “Transaction Costs and Capital Structure Choice: Evidence from Finan-cially Distressed Firms,” Journal of Finance 52, 161–196.
[22] Hirtle, B., 2009, “Credit Derivatives and Bank Credit Supply,” Journal of Financial In-termediation 18, 125–150.
[23] Kolari, J.W. and S. Pynnonen, 2010, “Event Study Testing with Cross-sectional Correla-tion of Abnormal Returns,” Review of Financial Studies 23, 3996–4025.
[24] Longstaff, F., Mithal, S. and E. Neis, 2005, “Corporate Yield Spreads: Default Risk orLiquidity? New Evidence from the Credit Default Swap Market,” Journal of Finance 60,2213–2253.
[25] Merton, R. C., 1974, “On the pricing of corporate debt: The risk structure of interestrates,” Journal of Finance 29, 449–470.
[26] Morellec, E., 2013, “Financing investment: the choice between bonds and bank loans,”Swiss Finance Istitute Working Paper.
[27] Morris, S., and H. Shin, 2003, “Global Games: Theory and Applications,” in Dewatripont,M., Hansen, M., Turnovsky, S. (Eds), Advances in Economics and Econometrics (Proceed-ings of the Eighth World Congress of the Econometric Society), Cambridge UniversityPress.
[28] Murfin, J., and M. Peterson, 2013, “Loans on sale: Credit market seasonality, borrowerneed, and lender rent seeking,” Yale University and Northwestern University WorkingPaper.
[29] Oehmke, M. and A. Zawadowski, 2013, “The Anatomy of the CDS Market,” ColumbiaUniversity Working Paper.
[30] Parlour, C. A., and G. Plantin, 2008, “Loan Sales and Relationship Banking,” Journal ofFinance 63, 1291–1314.
[31] Parlour, C., and A. Winton, 2008, “Laying off Credit Risk: Loan Sales versus CreditDefault Swaps,” Working Paper, University of California at Berkeley.
[32] Roberts, M.R. and A. Sufi, 2009, “Renegotiation of financial contracts: Evidence fromprivate credit agreements,” Journal of Financial Economics 93, 159–184.
[33] Saretto, A. and H. Tookes, 2013, “Corporate Leverage, Debt Maturity and Credit Supply:The Role of Credit Default Swaps,” Review of Financial Studies 26, 1190–1247.
[34] Subrahmanyam, M., Tang, D. Y., and S.Q. Wang, 2014, “Does the Tail Wag the Dog? TheEffect of Credit Default Swaps on Credit Risk,” Review of Financial Studies, Forthcoming.
32
[35] Tang, D.Y. and H. Yan, 2013, “What Moves CDS Spreads,” University of Hong KongWorking Paper.
[36] Thompson, J.R., 2010, “Counterparty Risk in Financial Contracts: Should the InsuredWorry About the Insurer?,” Quarterly Journal of Economics 90 , 1195–1252.
[37] Vassalou, M. and Y. Xing, 2004, “Default Risk in Equity Returns,” Journal of Finance59, 831–868.
33
Data Appendix: Variable Definitions and Sources
This appendix contains definitions of the variables used in this paper. Programs to derive the dataset are available on request.
Variable Name Definition Data Source
1. Dependent Variables
CDS Spread The weekly spread on a firm’s five-year CDS contracts. This vari-able is winsorized at the 1-99 level.
Thomson ReutersDatastream
Obtained Loan An indicator equal to 1 if a firm obtained a new syndicated loanin the corresponding 12-month period.
LPC–Dealscan
New Loan Markup The markup on new syndicated loan issues, measured in percent-age points. This variable is LPC data item ALLINDRAWN.
LPC–Dealscan
2. Key Independent Variables
HighSyndicated Equals 1 if a firm’s syndicated loans–debt ratio at the start of fis-cal year 2012 is in the highest tercile of the sample distribution,and 0 if the loans–debt ratio is in the lowest tercile. This variableis set to missing for firms with syndicated loans–debt ratio in themiddle tercile. The syndicated loans–debt ratio is the issue-dateprincipal of all syndicated loans oustanding at the start of fiscalyear 2012 divided by total debt in 2012. Issue-date principal isLPC data item FACILITYAMT summed over all observations foreach individual loan. Total debt is measured as total liabilities(Compustat data item LT ). A loan is outstanding if its issue dateis before the end of fiscal year 2011, and its maturity date is afterthe end of fiscal year 2011. Our sample excludes loans that aresyndicated outside the United States or that are rumored, sus-pended, or cancelled. The loans/debt ratio is bounded from aboveby 1.
LPC–Dealscan,Compustat
LowSyndicated Equals 1 if a firm’s syndicated loans–debt ratio at the start of fiscalyear 2012 is in the lowest tercile of the sample distribution, and0 if the syndicated loans–debt ratio is in the lowest tercile. Thisvariable is set to missing for firms with syndicated loans–debt ratioin the middle tercile. The syndicated loans–debt ratio is measuredin the same way as for HighSyndicated.
LPC–Dealscan,Compustat
PostTD Equals 1 for weekly observations after September 12, 2012, and 0otherwise.
Distressed Equals 1 if the firm’s 2012 Altman Z-Score < 3, and 0 otherwise. Compustat
Altman Z-Score This variable equals 1.2*(Current Assets - Current Liabili-ties)/Total Assets + 1.4*Retained Earnings/Total Assets +3.3*EBIT/Total Assets + 0.6*Market Capitalization/Total Lia-bilities + Total Sales/Total Assets. Current Assets is Compustatdata item ACT, Current Liabilities is LCT, Total Assets is AT,Retained Earnings is RE, EBIT is data item EBIT, Total Lia-bilities is LT, and Total Sales is SALE. Market Capitalization isdefined as for Market Leverage. All Z-Score component variablesare measured in the most recent quarter ending before September2012, and are winsorized at the 1-99 level. The Z-Score is set tomissing if any of the component variables is missing.
Compustat
3. Control Variables
Firm Leverage The sum of debt in current liabilities (Compustat data item DLC ),long-term debt (DLTT ), and preferred stock (PSTK ), dividedby this number plus market capitalization. We measure mar-ket capitalization by matching monthly stock data from CRSPto Compustat fiscal-year-end datadate. Market capitalization isthe firm’s month-end stock price (CRSP data item PRC ) multi-plied by month-end shares outstanding (SHROUT ), and expressedin millions of dollars. This variable is measured at the end of thefirm’s fiscal year, and is winsorized at the 1-99 level.
Compustat, CRSP
Firm Volatility The annualized standard deviation of daily stock returns from theprevious fiscal year. This variable is missing for firms with fewerthan 50 trading days, and is winsorized at the 1-99 level.
CRSP
Market Volatility The weekly closing value of the VIX index. Yahoo! Finance
Tobin’s Q Market capitalization plus total assets minus book value of equity,all divided by total assets. Market capitalization is defined in thesame way as for the variable Market Leverage. Total assets isCompustat data item AT, and book value of equity is Compustatdata item SEQ. This variable is measured at the end of the firm’sfiscal year, and is winsorized at the 1-99 level.
Compustat, CRSP
Log Assets The natural logarithm of 1 plus the firm’s total assets (Compustatdata item AT ). This variable is measured at the end of the firm’sfiscal year, and is winsorized at the 1-99 level.
Compustat
Return on Assets The ratio of net income before interest expenses to total assets.Net income before interest expenses is the sum of Compustat dataitems NI and XINT. Total assets is data item AT. This variableis measured at the end of the firm’s fiscal year, and is winsorizedat the 1-99 level.
Compustat
Tangibility The ratio of plant, property and equipment (Compustat data itemPPENT ) to total assets (AT ). This variable is measured at theend of the firm’s fiscal year, and is winsorized at the 1-99 level.
Compustat
Liquidity Needs The ratio of debt in current liabilities (Compustat data item DLC)to this number plus long-term debt (DLTT). This variable is mea-sured at the end of the firm’s fiscal year, and is winsorized at the1-99 level.
Compustat
Term Slope The weekly difference between the yield on a 10-year U.S. Treasurybond and a 2-year U.S. Treasury Note.
U.S. TreasuryDept.
Credit Rating A numeric index of Standard & Poor’s credit rating for the firm’slong-term debt (Compustat data item SPLTICRM ). The indexranges from 1 for a “AAA” rating to 21 for a “D” or “SD” rating.
Compustat
Tab
le1.
Su
mm
ary
Sta
tist
ics
Pan
elA
.Sum
mar
ySta
tist
ics
by
Fir
mF
inan
cial
Con
dit
ions
Dis
tres
sed
Fir
ms
Non
-dis
tres
sed
Fir
ms
Dis
tres
sed
–t-
stat
NM
ean
St.
Dev
.25
thP
erc.
Med
ian
75th
Per
c.N
Mea
nSt.
Dev
.25
thP
erc.
Med
ian
75th
Per
c.N
on-d
istr
esse
d
Avg
.C
DS
Spr
ead
160
217.
6924
0.21
77.8
313
5.97
249.
6047
105.
8689
.51
56.3
467
.82
114.
0011
1.84
3.12
***
Syn
dica
ted
loan
s–de
bt18
00.
300.
280.
110.
210.
3847
0.31
0.22
0.15
0.22
0.47
0.00
0.05
Z-S
core
180
1.46
0.74
0.98
1.45
1.91
474.
081.
013.
343.
744.
63-2
.61
-19.
85**
*F
irm
Lev
erag
e17
00.
390.
190.
230.
360.
5047
0.17
0.10
0.11
0.15
0.20
0.21
7.25
***
Fir
mV
olat
ilit
y17
036
.42
16.5
925
.46
34.2
744
.23
4730
.45
10.9
123
.93
28.5
736
.71
5.97
2.33
**T
obin
’sQ
170
1.32
0.41
1.08
1.23
1.51
472.
070.
891.
371.
972.
46-0
.75
-8.3
4***
Ass
ets
180
35,6
3546
,763
8,25
519
,678
40,3
0047
20,7
9916
,899
7,42
114
,983
31,6
1814
,836
2.14
**R
etu
rnon
Ass
ets
180
0.06
0.05
0.04
0.06
0.09
470.
110.
060.
080.
110.
14-0
.04
-5.5
8***
Tan
gibi
lity
180
0.40
0.25
0.17
0.39
0.62
470.
310.
210.
140.
250.
480.
092.
24**
Liq
uid
ity
Nee
ds17
90.
100.
110.
010.
060.
1447
0.12
0.11
0.02
0.10
0.16
-0.0
2-1
.01
Cre
dit
Rat
ing
178
9.56
3.17
89
1247
7.21
2.37
66
92.
35-4
.74*
**
Pan
elB
.Sum
mar
ySta
tist
ics
by
Syndic
ated
Loa
ns–
Deb
tR
atio
Hig
hSyn
dica
ted
Low
Syn
dica
ted
Hig
hSyn
dica
ted
–t-
stat
NM
ean
St.
Dev
.25
thP
erc.
Med
ian
75th
Per
c.N
Mea
nSt.
Dev
.25
thP
erc.
Med
ian
75th
Per
c.L
owS
yndi
cate
d
Avg
.C
DS
Spr
ead
7627
3.97
280.
6510
6.70
169.
5739
7.28
7015
3.57
204.
0559
.53
88.7
114
9.37
120.
402.
94**
*S
yndi
cate
dlo
ans–
debt
840.
600.
240.
400.
530.
7879
0.07
0.05
0.00
0.08
0.11
0.53
19.5
4***
Z-S
core
782.
061.
361.
161.
782.
9873
1.74
1.19
0.94
1.42
2.35
0.32
1.56
Fir
mL
ever
age
790.
380.
220.
220.
340.
5674
0.36
0.23
0.17
0.32
0.47
0.03
0.70
Fir
mV
olat
ilit
y79
41.8
420
.09
30.2
538
.87
49.2
274
32.1
711
.72
24.2
030
.22
38.4
19.
673.
61**
*T
obin
’sQ
791.
440.
611.
071.
221.
6074
1.36
0.50
1.08
1.29
1.51
0.08
0.85
Ass
ets
8412
,245
10,6
495,
128
8,56
116
,589
7949
,199
55,3
5714
,382
32,9
9355
,099
-36,
954
-6**
*R
etu
rnon
Ass
ets
840.
060.
060.
040.
070.
0979
0.07
0.04
0.05
0.07
0.09
-0.0
1-0
.61
Tan
gibi
lity
830.
370.
250.
180.
330.
5979
0.39
0.28
0.12
0.38
0.65
-0.0
2-0
.49
Liq
uid
ity
Nee
ds84
0.08
0.11
0.01
0.05
0.13
780.
100.
120.
020.
060.
13-0
.02
-1.1
9C
redi
tR
atin
g83
10.6
73.
329
1014
788.
362.
987
810
2.32
-4.6
5***
Th
ista
ble
pre
sents
sum
mary
stati
stic
sfr
om
fisc
al
yea
r2012
for
ou
rsa
mp
le,
wh
ich
conta
ins
257
non
-fin
an
cial
firm
sin
the
inte
rsec
tion
of
the
Com
pu
stat,
LP
C–D
eals
can
,an
dD
TC
Cd
ata
base
s.O
bse
rvati
on
sin
this
tab
leare
at
the
firm
level
.In
Pan
elA
,“D
istr
esse
dF
irm
s”are
those
wit
h2012
Alt
man
Z-S
core<
3,
an
d“N
on
-dis
tres
sed
Fir
ms”
are
those
wit
hZ
-Sco
re≥
3.
InP
an
elB
,HighSyndicated
isfi
rms
wit
hsy
nd
icate
dlo
an
s–d
ebt
rati
oin
the
hig
hes
tte
rcile,
an
dLowSyndicated
isfi
rms
wit
hlo
an
s–d
ebt
rati
oin
the
low
est
terc
ile.
Avg.CDS
Spread
isth
efi
rm’s
wee
kly
CD
Ssp
read
,aver
aged
over
2011
an
d2012.Syndicatedloans–debt
isth
eis
sue-
date
pri
nci
pal
of
all
syn
dic
ate
dlo
an
son
afi
rm’s
books
at
the
start
of
fisc
al
yea
r2012,
div
ided
by
tota
ld
ebt
in2012.
Firm
Leverage
isth
esu
mof
deb
tin
curr
ent
liab
ilit
ies,
lon
g-t
erm
deb
t,and
pre
ferr
edst
ock
,d
ivid
edby
this
nu
mb
erp
lus
mark
etca
pit
aliza
tion
.Firm
Volatility
isth
eannu
alize
dst
an
dard
dev
iati
on
of
daily
stock
retu
rns
from
the
pre
vio
us
fisc
al
yea
r.Tobin’sQ
ism
ark
etca
pit
aliza
tion
plu
sb
ook
valu
eof
ass
ets
min
us
book
valu
eof
equ
ity,
all
div
ided
by
book
valu
eof
ass
ets.
Return
on
Assets
isn
etin
com
eb
efore
inte
rest
exp
ense
s,d
ivid
edby
tota
lass
ets.
Tangibility
isth
era
tio
of
pro
per
ty,
pla
nt,
an
deq
uip
men
tto
tota
lass
ets.
LiquidityNeeds
isth
era
tio
of
deb
tin
curr
ent
liab
ilit
ies
toto
tal
deb
t.Credit
Rating
isa
nu
mer
icin
dex
of
the
firm
’slo
ng-t
erm
cred
itra
tin
gfr
om
Sta
nd
ard
&P
oor’
s,w
ith
1re
pre
senti
ng
“A
AA
”,
2re
pre
senti
ng
“A
A+
”,
etc.
All
vari
ab
les
are
win
sori
zed
at
the
1-9
9le
vel
.*,
**,
an
d***
rep
rese
nt
stati
stic
al
sign
ifica
nce
at
the
10,
5,
an
d1%
level
s.
Table 2. CDS Spreads Around TD9599, Firms Sorted by Z-Score
Dependent Variable CDS SpreadsWindow around Announcement 7 weeks 2 weeks
Type of Firm Distressed Non-distressed Distressed Non-distressed
HighSyndicated 235.37*** 96.00*** 146.22** 36.50 224.65*** 90.26*** 140.55** 26.08(5.13) (2.86) (2.51) (1.10) (5.00) (2.66) (2.32) (0.57)
PostTD -15.23*** -11.22*** -2.58 3.65 -10.16*** -15.46*** -6.95*** -10.88***(-5.96) (-4.43) (-0.46) (0.42) (-5.59) (-6.59) (-2.98) (-3.52)
HighSyndicated x PostTD -31.25*** -29.52*** -0.04 -15.17 -23.01*** -20.99*** -1.97 -1.24(-3.30) (-2.87) (-0.00) (-1.41) (-3.25) (-2.72) (-0.66) (-0.42)
Firm Leverage 280.69*** 633.15*** 280.09*** 578.12**(3.07) (4.07) (2.95) (2.71)
Firm Volatility 8.04*** 5.08*** 8.10*** 5.52**(5.90) (3.02) (5.73) (2.18)
Market Volatility -0.59 1.16 -0.50 1.39(-1.19) (1.28) (-0.59) (1.20)
Tobin’s Q 91.19*** 2.26 93.92*** -10.02(2.91) (0.18) (2.90) (-0.60)
Log Assets 54.68** -2.42 52.97** -2.69(2.46) (-0.18) (2.34) (-0.18)
Return on Assets -600.98** 55.52 -613.01* 174.73(-2.10) (0.16) (-1.94) (0.42)
Tangibility 6.73 -28.06 1.48 -18.66(0.12) (-0.28) (0.03) (-0.18)
Liquidity Needs 82.08 116.89 82.75 142.80(0.59) (1.12) (0.57) (1.11)
Term Slope -76.18*** -27.26* -98.23*** -61.07***(-6.25) (-1.88) (-5.53) (-3.23)
Credit Rating 28.54*** -5.86 26.66** -12.28(2.87) (-0.66) (2.55) (-1.18)
Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesObservations 1,545 1,410 457 457 412 376 123 123Adjusted R2 0.282 0.827 0.409 0.806 0.255 0.820 0.360 0.769
This table examines the changes in CDS spreads around the Sept. 12, 2012 TD9599 announcement. The sample contains 257 non-financial firms in the intersection of the Compustat, LPC–Dealscan, and DTCC databases. Each regression is at the firm-week levelfor 7- and 2-week windows around the announcement. “Distressed” firms have 2012 Altman Z-Score< 3, and “Non-distressed” firmshave Z-Score ≥ 3. The dependent variable in each regression is weekly CDS spreads, winsorized at the 1-99 level. HighSyndicatedequals 1 if the firm’s syndicated loans–debt ratio is in the highest tercile of the sample distribution, and 0 if the ratio is in thelowest tercile. (Firms in the middle tercile are omitted.) PostTD is an indicator equal to 1 for weeks after Sept. 12, 2012, and 0otherwise. Firm Leverage is the sum of debt in current liabilities, long-term debt, and preferred stock, divided by this number plusmarket capitalization. Firm Volatility is the annualized standard deviation of daily stock returns from the previous fiscal year.Market Volatility is the weekly closing value of the VIX index. Tobin’s Q is market capitalization plus book value of assets minusbook value of equity, all divided by book value of assets. Log Assets is the natural logarithm of total assets. Return on Assetsis net income plus interest expense, divided by total assets. Tangibility is the ratio of property, plant, and equipment to totalassets. Liquidity Needs is the ratio of debt in current liabilities to total debt. All control variables are winsorized at the 1-99 level.Term Slope is the weekly difference between the yield on a 10-year U.S. Treasury bond and 2-year U.S. Treasury note. CreditRating is a numeric index of the firm’s long-term credit rating from Standard & Poor’s, with 1 representing “AAA”, 2 representing“AA+”, etc. All regressions include industry fixed effects based on the Fama-French 12-industry classification. t-statistics are inparentheses. Standard errors are clustered at the firm level. *, **, and *** represent statistical significance at the 10, 5, and 1%levels.
Table 3. CDS Spreads Around TD9599, Firms Sorted by Distance-to-Default
Dependent Variable CDS SpreadsWindow around Announcement 7 weeks 2 weeks
Type of Firm Distressed Non-distressed Distressed Non-distressed
HighSyndicated 289.24*** 148.07*** 11.04 1.87 271.72*** 132.23*** 10.23 1.25(4.09) (3.70) (0.87) (0.19) (3.84) (3.07) (0.81) (0.13)
PostTD -11.95** -7.62 -10.03*** -8.26*** -15.20*** -24.02*** -5.84*** -6.70***(-2.32) (-1.50) (-4.56) (-4.14) (-3.84) (-5.32) (-4.76) (-5.21)
HighSyndicated x PostTD -36.28*** -41.55*** 0.33 0.33 -19.41** -19.60** -0.05 0.03(-2.73) (-3.21) (0.10) (0.10) (-2.14) (-2.07) (-0.03) (0.02)
Firm Leverage 309.92** 118.23* 313.12** 124.00*(2.54) (1.88) (2.41) (1.91)
Firm Volatility 7.49*** 0.41 7.49*** 0.32(4.74) (0.58) (4.34) (0.43)
Market Volatility 0.54 -0.47* 1.23 -1.02**(0.61) (-1.74) (0.86) (-2.33)
Tobin’s Q -21.26 -1.62 -33.13 -0.63(-0.36) (-0.16) (-0.52) (-0.06)
Log Assets 58.33* 1.39 60.05* 1.43(1.92) (0.16) (1.86) (0.16)
Return on Assets -525.12* 321.37* -472.54 328.64*(-1.86) (1.78) (-1.52) (1.75)
Tangibility -10.56 -48.49* -24.54 -50.50*(-0.12) (-1.72) (-0.27) (-1.77)
Liquidity Needs 67.98 41.14 68.95 34.15(0.35) (1.61) (0.33) (1.26)
Term Slope -107.63*** -25.98*** -151.17*** -23.26***(-5.25) (-5.09) (-5.47) (-6.30)
Credit Rating 35.89*** 10.48*** 34.75*** 9.91***(3.03) (3.06) (2.68) (2.77)
Industry Fixed Effects Yes Yes Yes Yes Yes Yes Yes YesObservations 894 879 793 793 240 236 211 211Adjusted R2 0.199 0.834 0.270 0.572 0.156 0.816 0.226 0.537
This table examines the changes in CDS spreads around the Sept. 12, 2012 TD9599 announcement. In this table, “Distressed”firms are those with below-median distance-to-default, and ”Non-distressed” firms are those with above-median distance-to-default.The sample contains 257 non-financial firms in the intersection of the Compustat, LPC–Dealscan, and DTCC databases. Eachregression is at the firm-week level for 7- and 2-week windows around the announcement. The dependent variable in each regressionis weekly CDS spreads, winsorized at the 1-99 level. HighSyndicated equals 1 if the firm’s syndicated loans–debt ratio is in thehighest tercile of the sample distribution, and 0 if the ratio is in the lowest tercile. (Firms in the middle tercile are omitted.)PostTD is an indicator equal to 1 for weeks after Sept. 12, 2012, and 0 otherwise. Firm Leverage is the sum of debt in currentliabilities, long-term debt, and preferred stock, divided by this number plus market capitalization. Firm Volatility is the annualizedstandard deviation of daily stock returns from the previous fiscal year. Market Volatility is the weekly closing value of the VIXindex. Tobin’s Q is market capitalization plus book value of assets minus book value of equity, all divided by book value of assets.Log Assets is the natural logarithm of total assets. Return on Assets is net income plus interest expense, divided by total assets.Tangibility is the ratio of property, plant, and equipment to total assets. Liquidity Needs is the ratio of debt in current liabilitiesto total debt. All control variables are winsorized at the 1-99 level. Term Slope is the weekly difference between the yield on a10-year U.S. Treasury bond and 2-year U.S. Treasury note. Credit Rating is a numeric index of the firm’s long-term credit ratingfrom Standard & Poor’s, with 1 representing “AAA”, 2 representing “AA+”, etc. All regressions include industry fixed effectsbased on the Fama-French 12-industry classification. t-statistics are in parentheses. Standard errors are clustered at the firm level.*, **, and *** represent statistical significance at the 10, 5, and 1% levels.
Table 4. Shareholder Effects of TD9599 Announcement
Type of Firm Distressed Non-distressed
Event Window (-1, +2) (-3, +3) (-1, +2) (-3, +3)
HighSyndicated 1.16%** 1.57%** -0.51% 0.11%(2.11) (2.15) (-1.23) (0.20)
LowSyndicated -0.86%*** -0.98%** -0.12% -0.05%(-2.64) (-2.27) (-0.10) (-0.32)
Difference 2.02%*** 2.55%*** -0.39% 0.16%(4.79) (2.77) (-0.80) (0.19)
This table examines borrower stock returns around the Sept. 12, 2012 TD9599 announcement. The sample contains all firms in theintersection of Compustat, LPC–Dealscan, and CRSP databases that have at least one syndicated loan outstanding at the start offiscal year 2012. “Distressed” firms have 2012 Altman Z-Score < 3, and “Non-distressed” firms have Z-Score ≥ 3. HighSyndicatedis a portfolio of firms that have a syndicated loans–debt ratio in the highest tercile of the sample distribution, and LowSyndicatedis a portfolio of firms that have a loans–debt ratio in the lowest tercile. The loans–debt ratio is principal for all syndicated loansoutstanding in 2011 divided by total debt in 2011. All portfolios are size-weighted based on firms’ total assets in 2012. Abnormalstock returns are based on the Fama-French 3-factor model. Cumulative Abnormal Returns (CARs) are the sum of abnormal dailystock returns over each event window, with Sept. 12, 2012 set as date 0. Below the CARs are t-statistics based on the Brown andWarner (1980) portfolio method for event-date clustering. The estimation window for the benchmark Fama-French model is 7 to1 months prior to the TD9599 announcement. Firms with fewer than 60 days of stock return data in the estimation window aredropped. *, **, and *** represent statistical significance at the 10, 5, and 1% levels, respectively.
Table 5. TD9599 and Financing Terms on New Syndicated Loans
Dependent Variable Obtained Loan New Loan MarkupModel Logistic OLS
Diff-in-Diff ComparisonDistressed Highly Distressed Distressed Highly Distressed
vs. Non-Distressed vs. Non-Distressed vs. Non-Distressed vs. Non-Distressed
Distressed -0.10** -0.12** 0.22** 0.18(-2.11) (-2.22) (2.27) (1.61)
PostTD -0.10** -0.10* 0.18 0.11(-1.97) (-1.96) (1.47) (0.90)
Distressed x PostTD 0.11* 0.16*** -0.30** -0.26**(1.80) (2.71) (-2.43) (-1.98)
Firm Leverage 0.18** 0.18* 0.86*** 0.97***(2.02) (1.74) (3.70) (3.47)
Firm Volatility <-0.01*** <-0.01*** -0.00 -0.00(-3.73) (-3.36) (-0.86) (-0.11)
Market Volatility -0.00 -0.01*(-0.82) (-1.90)
Tobin’s Q 0.00 0.01 0.18*** 0.20***(0.10) (0.19) (2.80) (2.92)
Log Assets 0.02 0.02 -0.03 -0.02(1.46) (1.16) (-1.05) (-0.68)
Return on Assets 0.13 0.15 -0.86 -1.37**(0.54) (0.58) (-1.34) (-2.45)
Tangibility 0.06 0.05 -0.41** -0.24(0.83) (0.56) (-2.31) (-1.22)
Liquidity Needs -0.01 -0.05 0.46** 0.20(-0.08) (-0.48) (1.99) (1.11)
Term Slope -0.18* -0.16(-1.81) (-1.42)
Credit Rating -0.01* -0.01 0.26*** 0.25***(-1.66) (-1.18) (14.28) (11.87)
Industry Fixed Effects Yes Yes Yes YesObservations 1,730 1,292 1,316 1,028Pseudo/Adjusted R2 0.024 0.030 0.460 0.480
This table examines whether firms’ financing terms improved after the Sept. 12, 2012 TD9599 announcement. The samplecontains all firms that are in the intersection of the Compustat, LPC–Dealscan, and CRSP databases. Financial firms are excluded.Columns (1) and (2) are logistic regressions, and coefficients represent the marginal effects of each variable. Each regression has twoobservations per firm, one each for the 12-month period before and 12-month period after the TD9599 announcement (September2012 is excluded in all columns). The dependent variable Obtained Loan equals 1 for firms that obtained a syndicated loan inthe respective 12-month period, and 0 otherwise. Columns (3) and (4) are OLS regressions, with one observation for each newsyndicated loan signed by a firm in the 12-month window around the TD9599 announcement. Syndicated loans with principal lessthan $100M (unaffected by TD9599) are excluded. The dependent variable New Loan Markup is the percentage-point all-in drawnspread on new syndicated loan issues. Columns (1) and (3) show results from a difference-in-differences regression comparingdistressed (2012 Altman Z-Score < 3) and non-distressed (Z-Score ≥ 3) borrowers. Columns (2) and (4) compare highly distressed(Z-Score < 1.86) and non-distressed (Z-Score ≥ 3) borrowers. Distressed equals 1 for firms with Z-Score < 3 (or < 1.86), and 0 forfirms with Z-Score ≥ 3. PostTD equals 1 for observations after September 2012, and 0 for observations before September 2012.Firm Leverage is the sum of debt in current liabilities, long-term debt, and preferred stock, divided by this number plus marketcapitalization. Firm Volatility is the annualized standard deviation of daily stock returns from the previous fiscal year. MarketVolatility is the weekly closing value of the VIX index. Tobin’s Q is market capitalization plus book value of assets minus bookvalue of equity, all divided by book value of assets. Log Assets is the natural logarithm of total assets. Return on Assets is netincome plus interest expense, divided by total assets. Tangibility is the ratio of property, plant, and equipment to total assets.Liquidity Needs is the ratio of debt in current liabilities to total debt. Term Slope is the weekly difference between the yield on a10-year U.S. Treasury bond and 2-year U.S. Treasury note. Credit Rating is a numeric index of the firm’s long-term credit ratingfrom Standard & Poor’s, with 1 representing “AAA”, 2 representing “AA+”, etc. All regressions include industry fixed effectsbased on the Fama-French 12-industry classification. All firm-level control variables are measured at the end of the previous fiscalyear and are winsorized at the 1-99 level. t-statistics are in parentheses. Standard errors are adjusted for heteroskedasticity. *,**, and *** represent statistical significance at the 10, 5, and 1% levels.