Zombie Lending Due to the Fear of Fire Sales

March 10, 2021

Abstract

This paper documents a new externality stemming from the fear of fire sales: in-

creased zombie lending during real estate price downturns. Firms use pledgeable

assets such as real estate collateral to borrow. Using firm and syndicated loan data in

the US, we confirm that the sensitivity of firms’ debt to real estate collateral is posi-

tive. However, this sensitivity falls during real estate price declines due to an increase

in lending to low-quality firms despite a fall in real estate collateral value. Zombie

credit to high collateral firms increases as lenders internalize the price externalities

of liquidating real estate collateral. Zombie presence depresses investment and prof-

itability of healthier firms. Our paper highlights a new mechanism for zombie lending

resulting from reduced collateral liquidation in markets prone to fire sales.

JEL Classification: G21, G33, L25

Keywords: Fire sales, zombie lending, collateral, real estate, crises

 Fire sales of assets occur during downturns, when financial institutions are forced to
sell at prices below their fundamental values, triggering an even further deterioration in
prices (Kiyotaki and Moore, 1997; Brunnermeier and Pedersen, 2009). Such fire sales can
erode the balance sheets of financial institutions during periods of crisis (Carlson et al.,
2009; French, 2010; Caballero and Simsek, 2010) and can have significant negative exter-
nalities. For instance, Benmelch and Bergman (2011) document the negative externalities
on the value of other firms’ assets. Lenders may also internalize such fire-sale externali-
ties, for instance Giannetti and Saidi (2019) find that lenders continue to provide liquidity
to distressed industries to avoid such externalities.
We conjecture that if such support goes to otherwise insolvent or “zombie" firms, it
may end up creating another externality by hindering creative destruction and diverting
resources away from healthier firms (Caballero et al., 2008).This paper thus examines a
new fire-sales externality: increased zombie lending during downturns. We do this us-
ing the real estate market. Firms use real estate as collateral to alleviate agency frictions,
allowing them to raise capital (Chaney et al., 2012; Gan, 2007; Cvijanovic, 2014). Real
estate markets, however, are also illiquid and prone to fire-sale externalities during eco-
nomic downturns (Harding et al., 2009; Campbell et al., 2011; Anenberg and Kung, 2014;
Hartley, 2014). This paper examines whether lenders keep credit flowing to otherwise
insolvent borrowers (zombie firms) during real estate downturns to avoid fire-sale exter-
nalities in real estate markets.
This paper first builds a theoretical model that motivates lender incentives in extend-
ing zombie credit, due to the fear of fire sales during downturns. Using firm- and loan-
level data from the US, we empirically show that during real estate price declines, lenders
do not liquidate failing firms as they fear further deterioration in real estate prices due to
fire sale externalities. Lenders instead extend zombie credit to keep these failed firms
alive and arrest further declines in real estate prices. Lenders with higher exposure to the
local real estate markets are more likely to extend zombie credit as they are more exposed

1
to the negative price spillovers from liquidation. Firms with larger real estate collateral
are also more likely to receive zombie loans as liquidation can create greater price exter-
nalities. Keeping inefficient zombie firms alive gives rise to another externality. It hinders
creative destruction and depresses investment and profitability of healthy firms in indus-
tries congested by zombie firms.
We begin our analysis by first building a theoretical model to illustrate this new ex-
ternality of fire sales. Firms borrow from banks using their real estate assets as collat-
eral. Some of these firms eventually default when faced with a negative shock and turn
into low productivity firms, which we call zombie firms. Banks can then either liqui-
date these firms and sell their real estate collateral, or extend zombie credit and keep
them alive while bearing the cost of their negative NPV zombie loans. During the nor-
mal state, very few firms default, whereas during the adverse state, several firms default.
If only a few firms default, as in the normal state, banks can liquidate these firms and
sell their collateral at a fair price as there are enough buyers, and the market is liquid
(Brunnermeier and Pedersen, 2009; Shleifer and Vishny, 1992). However, in the adverse
state, when many firms default, buyers do not have enough liquidity to buy the assets
of defaulted firms. Hence, there will be cash-in-the-market pricing and assets are sold at
fire-sale prices, resulting in a decline in real estate prices. Banks realize that at the margin,
selling the collateral when the local real estate prices are declining can further deteriorate
prices. These price declines can reduce the collateral value of nearby firms that can result
in a tightening of financial constraints and a lowering of investment.
Banks, thus have an incentive to extend zombie loans to failed firms instead of liq-
uidating them. Banks care about the value of the portfolio of loans that did not default,
which in turn depends on the health of the local economy for which real estate price is
a proxy. Banks also care about the price at which they sell the collateral of some of the
firms they choose to liquidate. We show that during the adverse state, banks extend zom-
bie loans to defaulted firms due to potential fire sale discounts, while during the normal

2
state they liquidate all defaulted firms as they are able to sell their collateral at a fair
price. In addition, in the adverse state, banks with higher market shares will internalize
the effect of the price decline more compared to banks with lower market shares.
We conduct our empirical analysis and test the model’s implications in several steps.
In the model we argue that when real estate prices are declining, banks may give zombie
loans to firms which have defaulted merely to keep them alive. Such lending may break
down the relationship between collateral value and lending. To test this we first estimate
how the value of real estate assets affects debt issuance by firms using a specification
similar to Chaney et al. (2012). To test the assymetry in the relationship during good and
bad times, we separately estimate the effect for periods when local real estate prices are
increasing and when local real estate prices are declining. We show that the elasticity
between debt and real estate prices is positive overall. However, the elasticity is 0.6 times
lower when real estate prices are declining (analogous to the adverse state in our model)
as compared to when real estate prices are increasing (normal state in the model). That
is, the rate of deleveraging when real estate prices are declining is lower than the rate of
leveraging when real estate prices are increasing. This provides suggestive evidence that
banks may not be demanding as much collateral in the adverse state and may be using
liberal credit policies to support firms facing negative shocks.
We test this idea further and determine whether the “excess” borrowing is by healthy
or financially distressed firms. We divide firms into distressed and healthy based on their
ability to service their debt. Distressed firms are defined as firms which are in the bottom
tercile of interest coverage ratio (ICR), which is the ratio of profits to interest expenses.
We show that in the adverse state, distressed firms raise 40% more debt compared to
non-distressed firms. Potentially, banks were extending credit to financially distressed
firms.
We next examine whether banks were extending credit to distressed firms at subsi-
dized rates, also known as zombie lending. Following Caballero, Hoshi and Kashyap

3
(2008), we identify zombie firms based on whether they receive credit at interest rates
lower than the most creditworthy firms in the economy. Indeed, banks extend credit to
zombie firms, and in particular to firms with high real estate collateral. Firms with above-
median real estate holdings are 2.3 times more likely to receive a zombie loan as compared
to firms with below-median real estate holdings. This result is in line with the predictions
of our model wherein banks are more likely to extend zombie credit to firms with higher
collateral as liquidating them would result in larger price declines.
To further test the model and to pin down the channel, we estimate how the market
share of banks in a metropolitan area (MSA) affects the probability of zombie lending.
A bank with higher market share will internalize the effect of price declines more than
a bank with a smaller market share. We show that a one percent increase in a bank’s
market share increases the probability of extending zombie credit by 0.042% whereas a
1% increase in the share of collateral in an MSA held by the bank increases the probability
of extending zombie credit by 0.047%.
Finally, we show that the presence of zombie firms has a negative impact on the
healthier non-zombie firms. As pointed out by Caballero, Hoshi and Kashyap (2008),
the presence of zombie firms in an industry can divert resources which would otherwise
have been available to healthier, more efficient non-zombie firms. A 1% increase in the
share of zombie firms in an industry leads to non-zombie firms reducing their investment
by 1.16%. Subsequently, the profitability of non-zombie firms also declines by 0.95%.
Overall, our paper also has implications for the ongoing COVID-19 crisis, which is a
systemic shock affecting firms and financial institutions. Broadly, the crisis will render
two kinds of firms on the verge of bankruptcy: those with good fundamentals that be-
come bankrupt because of the COVID-19 shock and those with poor fundamentals that
would have become bankrupt even without the shock. Banks should ideally liquidate
firms with poor fundamentals, but since the shock is systemic there will be few buyers in
the market that can lead to fire sales of assets. Hence, banks may extend zombie loans to

4
firms with poor future prospects so as to arrest the further deterioration of prices. This
can further weaken recovery as the presence of zombie firms hinders the natural process
of creative destruction and keeps otherwise insolvent firms alive. Hence, regulators needs
to ensure that credit does not flow to the worse firms in the economy.
The key contribution of this paper is that it shows that as lenders internalize the neg-
ative externality of a decline in collateral prices by extending zombie credit, they create
another externality by keeping unproductive firms alive which in turn hinders the pro-
cess of creative destruction. Thus the actions that banks take may be optimal privately
and even locally, but may be inefficient for the economy as a whole.
Our paper contributes to a number of strands of literature. Prior literature has docu-
mented that fire sales exist and are costly and hence lenders and borrowers take actions
to avoid fire sale of assets (Shleifer and Vishny, 2011). For example, Asquith et al. (1994)
show that distressed firms are more likely to restructure debt than liquidate their assets
when the industry is facing a down turn. Schlingemann et al. (2002) provide evidence
that firms divest business units from industries which have more liquid markets, rather
than liquidating the worst performing units. Banks with high market share in an industry
are also more likely to provide liquidity to firms in such industry during times of distress
due to the fear of fire sale externalities (Giannetti and Saidi, 2019).
We show that banks with a higher market-share in the local market are more likely to
internalize the effect of real estate price declines and extend zombie credit. Our result is
similar to Favara and Giannetti (2017), who show that banks with a higher market share
are less likely to trigger foreclosures. However, our paper highlights the negative effects
of keeping zombie firms alive. Our findings add to the large literature on the effects of
bank concentration on different aspects of lending activity such as the quantity of credit
provision (Garmaise and Moskowitz, 2006) and bank-firm relationships (Petersen and
Rajan, 1995). We highlight that bank concentration can affect the ex-post decision of a
bank to either liquidate a firm or to extend zombie credit to it.

5
 This paper also contributes to the literature on zombie loans, made even more rele-
vant with the ongoing crisis. In their seminal paper, Caballero et al. (2008) show that the
presence of zombie loans can make industries unproductive as it prevents the process of
creative destruction. The literature on the causes of zombie lending has shown that, in the
presence of limited liability, undercapitalized banks have an incentive to engage in zom-
bie lending (Giannetti and Simonov (2013), Acharya et al. (2019), Blattner et al. (2018))
as they are reluctant to liquidate firms and recognize losses. Our paper provides a novel
channel through zombie lending arises: to prevent the liquidation of assets in illiquid
markets prone to fire sales.
Our paper also adds to the literature on the role of collateral in credit provision. The-
oretical models starting with Besanko and Thakor (1987) and Hart and Moore (1994) have
revealed the importance of collateral in alleviating agency frictions and increasing firms’
access to credit. Companies that have access to more redeployable collateral receive larger
loans with longer maturity and at lower interest rates (Benmelech et al., 2005). Chaney
et al. (2012) show that real estate is a major source of collateral for firms and that increas-
ing collateral value increases investments. Cvijanovic (2014), on the other hand, shows
that increasing real estate prices lead to an increase in firm leverage. Our paper shows
that higher levels of real estate assets can help firm secure loans, but for very different
reasons. Liquidating firms with larger real estate will result in larger price externalities
and as a result banks are prepared to extend zombie loans to such firms.
The rest of the paper is organized as follows. Section 1 presents a model of bank
lending where banks choose between liquidating firms or giving zombie loans. Section 2
discusses the data. Section 3 contains our empirical strategy. Section 4 presents the results
and section 5 concludes.

6
1 Model
In our model economy, there are four kinds of agents - firms, banks, depositors and out-
side investors and two dates t = 0 and 1. At t = 0, there are a continuum of atomistic
identical firms, each owns C units of real estate (or land) which can be used a collateral
and has a positive NPV project which requires one unit of investment. There are a con-
tinuum of atomistic banks of mass one which raise funds from insured depositors. Using
these deposits, each bank finances a portfolio of 1 unit of a continuum of firms taking the
real estate assets of the firm as collateral. The face value of each loan is denoted by F,
which needs to be paid by the firms at t = 1.
The project owned by firms can either succeed or fail. There are two aggregate states
of nature - normal (N) and adverse (A). The probability of success is denoted by q ∈
{α, β}, where α ( β) is the probability of success in normal (adverse) state. We assume
α > β. The project fails with the complementary probability. If the project fails, then the
firm defaults and pays nothing. Now the bank has two options. First, it can liquidate
the firm and sell the collateral C at prevailing market prices (to be determined later). The
section option is to roll over the loan and provide the required financing to keep the firm
alive. This rolled over loan is essentially a zombie loan which has very low probability
of success in the future. The total cost to the bank of giving a zombie loan to a firm and
keeping it alive is L. We will assume that L is small positive number very close to zero.1
If a firm’s project succeeds, then it pays F to its bank. The successful firm is then
endowed with another project which is financed by the same bank, and the process can
repeat in future. The continuation value of each firm to the bank is denoted by V. So
we are assuming that the banks are not competitive and earn positive profits. Banks
may be earning this profit because of some monopoly power or information rent from
relationship lending.
1 This assumption is not necessary but considerably simplifies the proof of proposition 1. The results will
still hold as long as L is small enough.

7
 As discussed above, if a firm fails at t = 1, then the bank can liquidate it and sell
the collateral at market price denoted by p. We assume that the intrinsic or fair value
of one unit land is given by Z which is independent of the aggregate state.2 This value
can be interpreted as the net present value that can be generated by the investors who
buy the land. This means that an investor will never pay a price larger than Z. The
price is determined as following. We assume that at t = 1, there are outside investors
who are ready to buy the land. These outside investors have a total wealth of W.3 The
investors can be interpreted as experts who understand the local economy and the real
estate markets. Their wealth characterizes the demand for the real estate assets.
If a bank decides to give a zombie loan to a failed firm with probability λ and liq-
uidate it with probability 1 − λ, then the supply in the real estate market in the normal
state is given by (1 − α)(1 − λ)C. We assume that α is high enough (hence supply is low
enough) such that even if banks liquidate all their loans (λ = 0), land is sold at fair price Z.

Assumption 1. W > Z (1 − α)C.

As the readers might have guessed, we will assume that in the adverse state if all
banks liquidate with probability one, then land will not be sold at fair price and there will
be cash-in-the-market pricing.

Assumption 2. W < Z (1 − β)C.

If all banks liquidate with probability 1 − λ in the adverse state then the price is given
by
2 At
the cost of some notational complexity, we can relax this assumption, and instead assume that the
fair value of the land at t = 1 depends on the state (normal or adverse), without changing the results.
3 Again, the wealth of the investors can be taken to be state dependent without changing the results.

8
 

 Z,
 if W ≥ (1 − β)(1 − λ)CZ.
p(λ) = (1)
W



(1− β)(1−λ)C
, if W < (1 − β)(1 − λ)CZ.

This price is weakly decreasing in 1 − λ, or equivalently weakly increasing in the number

of zombie loans.
We assume that the continuation value of the successful firms to the bank, V, will be
a function price. There are many justifications for this assumption. First, the profitabil-
ity of a firm or the probability of its success depends on how number and scale of other
firms functioning in the economy. This is because firms use goods produced by other
firms as inputs or the workers in one firm use their wage to consume goods produced by
other firms. Thus, as the number of firms operating in the economy and their scale in-
creases, the positive feedback loops on each other also increases which further increases
the profitability and the probability of success. We assume that this will be true even if the
firms are low productive zombie firms as even they employ people and use inputs (see
Bebchuk and Goldstein (2011)). Thus the banks have an incentive to make zombie loans
which also keeps the prices higher. Further, as pointed out by Benmelch and Bergman
(2011), higher price of collateral due to prevention of bankruptcy can reduce the cost of
capital and increase investment by neighboring firms. The larger scale of operation will
hive higher positive spillover effect of firms in the locality. Finally, as the price of real es-
tate falls, it will reduce the home equity value of the residents in the area, who may in turn
reduce their consumption (Mian et al. (2015)). We capture these ideas in a reduced form
by assuming V depends on p and is denoted by V ( p). Also V (.) is increasing (V 0 (.) > 0),
concave (V 00 (.) < 0) and reaches its maximum at p = Z (V ( Z ) = V̄).
The state contingent utility function of a bank is given by

q( F + V ( p(λ))) + (1 − q)(1 − λ)Cp(λ) − (1 − q)λL; q ∈ {α, β}. (2)

9
The first term is the current revenue plus the continuation value of the successful firms.
The second term is the revenue from the liquidation of collateral and the final term is the
loss from zombie loans. There are two reasons a bank may want to give a zombie loan.
The first reason is to increase the liquidation price of collateral and the second reason
is to increase the continuation value which depends on the liquidation price. Next we
determine the equilibrium when all the banks are atomistic as assumed so far.

1.1 Equilibrium with atomistic banks

If all the banks are atomistic, they all take price as given and will choose their λ to max-
imize their utility. It is clear that irrespective of the state, banks will choose λ = 1. So
by assumptions 1 and 2, in good state the price will be equal to the fair value and in the
bad state there will be cash-in-the-market pricing. The equilibrium is characterized by
the following lemma.

Lemma 1. When banks are atomistic, then in both states they choose λ = 0. In the normal
state the price is given by Z and in the bad state the price is given by W/(1 − β)C.

The more interesting scenario is one where all banks are not atomistic, which we
analyse next.

1.2 Equilibrium when banks are not atomistic

Now let us assume that one of the banks ("large bank") has a higher fraction of market
share denoted by f < 1 and the others are still atomistic. As before in the normal state, all
banks will continue to liquidate all defaulted loans (no zombie lending), and the market
price is given by Z. But in the bad state the large bank will internalize the effect of its
liquidation strategy on the selling price of collateral as well as on the continuation value
of the successful loans. The atomistic banks will continue to liquidate all loans in the bad
state since they will take price as given. We assume that in the bad state the price is below

10
fair value.

Assumption 20 . W < Z (1 − f )(1 − β)C.

Assumption 20 is stronger than assumption 2. It implies that even if the large bank
gives zombie loans to all failed firms, i.e. chooses λ = 1, the supply in the adverse state
is high enough that land will not be sold at fair price. Now if the large bank chooses the
probability zombie loan as λ, then the price, p(λ), is given by

W
p(λ) = . (3)
(1 − β ) C [1 − λ f ]

The utility function of the large bank in the adverse state is given by

β( F + V ( p(λ))) + (1 − β)(1 − λ)Cp(λ) − (1 − β)λL. (4)

This is the same as (2), where q takes value β. But now the price is given by (3) rather than
(1). The large bank chooses λ to maximize its utility. We denote the equilibrium value of
λ by λ∗ .
As discussed above, there are two benefits of giving the zombie loan. First, it in-
creases the price of collateral. Second, the increased price increases the value of V. The
large bank will internalize these effects and give zombie loans with positive probability
(λ > 0). More interestingly, it can be shown that as the market share increases, it gives
zombie loans to higher fraction of failed firms.

Proposition 1. If assumptions 1 and 20 hold true and

dV ( p(0)) (1 − β)(1 − f )C dV ( p(1))

> > , (5)
d( p) βf d( p)

11
then there exists a unique λ∗ ∈ (0, 1) which maximizes the large bank’s utility. Also, λ∗
increases as as f increases.
Proof: See appendix.

The intuition for the result is simple. More liquidation results in lower selling price
which further results in lower continuation value of the second round of loans given to
firms. As the market share of the large bank increases, it internalizes these costs more
and gives zombie loans with higher probability. Condition (5) simply gives the boundary
conditions required for interior solution. It says that V 0 (.) should be large enough at p(0)
and small enough at p(1) for an interior solution to exist.
We have so far assumed that all firms have the same collateral. But in an economy,
firms have different levels of collateral. So the next question is how does the level of
collateral affect the likelihood of receiving a zombie loan. We turn to this issue next.

1.3 Collateral level and the probability of zombie lending

We now assume that there are two types of firms. Half of the firms have high collateral
denoted by CH and remaining half firms have low collateral denoted by CL < CH . Aver-
age size of the collateral is still C.4 The other characteristics of the firm, V and L, remain
the same.5 As before there is a large bank with market share f and atomistic banks with
a combined market share of (1 − f ). Each banks’ portfolio is equally distributed between
the two types of firms and the face value of loans remain the same.6
In the normal state, by assumption 1, all banks will continue to liquidate all firms. In
4 This assumption is not necessary, and merely reduces the effort of refining assumptions 1 and 20
5 Itmay seem unreasonable to assume the other characteristics of a firm do not change with the size of
the collateral. But in the empirical part of the paper, we will be comparing firms with different ratios of
real estate collateral as a fraction of their total assets. Hence we can assume that the total assets of all firms
is same, but some firms have more real estate assets than the others. The other characteristics of the firms,
i.e. V and L, depend on total assets and not the real estate collateral. We are basically abstracting from
modelling the market for not real estate assets of the firm when it is liquidated.
6 We can assume that the face value changes with collateral level without changing any result. Here the

face value can be interpreted as the average face value of the loans. Since liquidation probability does not
affect the current pay off, the face value is irrelevant to our calculations.

12
the adverse state, when a firm goes bankrupt, the larger bank chooses to give a zombie
loan to high (low) collateral firm with probability λ H (λ L ). The total collateral liquidated
by large firms is denoted by τ and is given by

τ = f (1 − β)((1 − λ H )CH + (1 − λ L )CL )/2 (6)

where τ. The equilibrium values are denoted by λ∗H , λ∗L and τ ∗ . The atomistic banks
will continue to liquidate all firms in the adverse state.
The price in the adverse state is given by

W
p(λ H , λ L ) = . (7)
τ + C (1 − f )

Utility function of the large bank is given by

1−β
β( F + V ( p(λ H , λ L ))) + τ p(λ H , λ L ) − (λ H + λ L ) L. (8)
2

Given this set up, it can be shown that the high collateral firms are more likely to get
a zombie loan than the low collateral firms.

Proposition 2. Given assumption 1 and 20 ,

τ∗
i. If τ ∗ ≤ f (1 − β)CL /2, then λ∗H = 1 and λ∗L = CL f (1− β)/2
.

τ ∗ −CL f (1− β)/2

ii. If τ ∗ > f (1 − β)CL /2, then λ∗L = 1 and λ∗H = CH f (1− β)/2
.

Proof: See appendix.

The proposition says that the large bank prefers to first liquidate the low collateral
firms and then the high collateral firms. Part i. of the proposition says that if the total
collateral sold by the large bank in equilibrium is less than the total collateral of the low

13
collateral firms (this is the inequality in the if condition of part i.), then it will only liqui-
date the low collateral firms and all the high collateral firms get a zombie loan. But if the
total collateral sold by the large bank in equilibrium is more than the total collateral of the
low collateral firms (this is the inequality in the if condition of part ii.), then the bank will
first liquidate all the low collateral firms and the remaining collateral will come from the
high collateral firms.
The intuition is as following. The cost of giving a zombie loan, L, is fixed and is
independent of the collateral level. So, for a given amount of collateral that the large
bank sells (which determines the effect on price and V (.)), it wants to liquidate as many
firms as possible to minimize the loss from zombie lending.

2 Data
To test our hypotheses, we need information on the value of collateral available to firms
and their borrowings. We also need information on the exposure of various lenders to
these firms. We use accounting data for listed US firms from Standard & Poor’s COM-
PUSTAT database. Details on syndicated loans are from the Thompson Reuters Dealscan
database. We access the House Price Index from the Office of the Federal Housing Enter-
prise Oversight and the Consumer Price Index from the Bureau of Labour Statistics.

2.1 Firm Data

We use firms with non-missing real estate assets, headquartered in the United States. We
then exclude firms operating in finance, insurance, real-estate, construction, and mining.
We restrict our sample period from 1993 to 2015 and to firms that have data for at least
three years in this period. This leaves us with 6,804 firms and 73,126 firm-year observa-
tions.

14
2.1.1 Real Estate Assets:

We classify real estate as total Buildings, Land, and Improvement and Construction in Progress.
Real estate assets are not marked to market but are held on the balance sheet at histori-
cal cost. To impute the market value of real estate, we need to calculate the average age
of assets. The ratio of accumulated depreciation in 1993 to the gross book value of the
real estate measures the proportion of the cost claimed as depreciation. Assuming a de-
preciable life of 40 years and straight-line depreciation, we calculate the average age of a
firm’s real estate. We inflate real estate assets using state-level and MSA-level real estate
inflation. For real estate purchased before 1975, we use CPI to inflate its value.

Real Estate Prices: We get the House Price Index (HPI) from the Office of the Federal
Housing Enterprise Oversight. The HPI is available at the state and MSA levels. We
match each firm’s ZIP and FIPS code to the respective MSA code. The HPI is then nor-
malized to 1 in 2006. Previous work by Gyourko (2009) has shown that residential and
commercial real estate prices are highly correlated so we assume that residential real es-
tate price change reflects all real estate prices changes in an area. Since we don’t know
the exact location of all real estate holding of a firm, we assume that a firm’s real estate is
at the same location as its headquarters.
We restrict our sample to firms active in 1993 as accumulated depreciation is not avail-
able in COMPUSTAT after 1993. Our final sample has 3,430 firms and 36,831 firm-year
observations. Chaney, Thesmar and Sraer (2012) show that for the median land-holding
firm in COMPUSTAT, the market value of real estate is a sizable fraction of tangible as-
sets of the firm and so is a good proxy for the collateral available to a firm. We define
RealEstate as the market value of real estate standardized by lagged PPE.

15
2.1.2 Dependent Variables

We aim to explain a channel of subsidized lending so our main variable of interest is

the probability of zombie lending. The other variables of interest are long-term debt, debt
issuance, investment, productivity, return-on-assets, and interest coverage ratio. Debt variables
are normalized by lagged PPE.
We define the Investment rate as the ratio of capital expenditures (Capex) to the previ-
ous year’s PPE. The Interest Coverage Ratio is the ratio of a firm’s earnings before interest
and taxes (EBIT) and its interest expense while Return-On-Assets is calculated as the ratio
of operating income minus depreciation and amortization to assets. We also calculate the
operating return on assets as the ratio of operating cash-flows to assets. And finally, leverage
is the ratio of a firm’s total debt to assets.
We calculate a firm f ’s productivity A at the two-digit SIC code (sector) level i using a
α β
Cobb-Douglas production function. We estimate the function Yf ,t = A f ,t K f ,ti −1 L f ,ti , where
Y is the value-added, K is fixed assets and L is the employment.

2.1.3 Controls

For our analysis, we need to control for factors that affect a firm’s borrowing decision.
We use Tobin’s-q to proxy for investment opportunities and the Kaplan-Zingales Index (KZ)
(Kaplan and Zingales, 1997) and firm age to proxy for financing constraints. We cut-off
firm age (in 1993) at 33 so that no firm’s birth year is before 1960. Tobin’s q (market-to-book
ratio) is calculated as the ratio of a firm’s market value to its book value. The KZ-Index is
calculated as

KZ-Index = − 1.002 ∗ (ib − dp)/att−1 − 39.368 ∗ (dvc − dvp)/att−1 − 1.315 ∗ che/att−1

+ 3.139 ∗ (dltt + dlc)/(dltt + dlc + seq) + 0.283 ∗ q

where ib is the income before extraordinary items, dp is depreciation and amortization,

16
at is the book value of a firms assets, dvc and dvp are the common and preferred dividends,
che is the cash and short-term investment, dltt and dlc are the long-term and short-term
debt, and seq is the shareholders’ equity.
To ensure that our results are robust, we winsorize all ratio variables at the 1% and
99% levels.

2.2 Zombie Lending using Accounting Data

Zombie firms are companies that are unprofitable and can go bankrupt in the future. To
identify zombie lending, we follow the approach of Caballero, Hoshi and Kashyap (2008)
and Acharya, Eisert, Eufinger and Hirsch (2019). The classification of a firm as a zombie
hinges on the firm receiving subsidized credit from banks. A firm receives subsidized
credit if, its interest rate on borrowing is lower than the rate paid by the most creditwor-
thy firms in the economy. We calculate this interest rate paid by the most creditworthy
firms in two ways. First, we calculate the median of the average interest rate (total interest
expense/total debt) paid by firms with an AAA rating in any given year. Second, we cal-
culate the median of the average interest rate paid by the top decile of firms by interest
coverage ratio (ICR). The interest coverage ratio is a good proxy for the S&P rating of a
firm and thus of the highest rated firms.
To be conservative, we take the lower of the two interest rates as the rate paid by the
most creditworthy firms in the economy. Given this interest rate benchmark (r top ) and the
total debt of a firm (Dit ), we calculate the minimum required interest payment of a firm
(Rmin ),
top
Ritmin = rt ∗ Dit

Next, we calculate the excess interest paid by the firm. Excess interest is the difference
between the actual interest expense of a firm (Rit ) and the minimum required interest
payment.
xit = Rit − Ritmin

17
Given xit , a firm is classified as a zombie if it meets the following criteria: (i) xit is negative
i.e. the excess interest paid by the firm is negative which implies that its interest cost is
less than that of the most creditworthy firms (ii) it is in the bottom tercile of firms when
classified by the 3-year average interest coverage ratio. For small firms (<$5bn in market
cap), ICR = 3 corresponds to a rating of BB while for larger firms, ICR = 2 is equivalent
to a BB rating (Damodaran). When using the bottom tercile of ICR as a proxy for S&P
rating, only 1 datapoint has an ICR above 3 and 1.2% of the data points have an ICR
greater than 2. Hence, selecting the bottom tercile is a good proxy for a firm’s credit
rating.

2.3 Loan Level Data

To test our hypothesis, we need to calculate the market share of lenders in each loca-
tion. Thompson Reuter’s Dealscan database contains detailed information on syndicated
lending facilities to a firm. We merge Dealscan with COMPUSTAT using the link table
provided by Chava and Roberts (2008). We can match 13,185 facilities (loans) to firms
in our dataset, 36,549 observations of a loan from a bank to a firm and a total of 244,763
bank-firm-year data-points. Using this merged dataset, we can calculate the outstanding
loans of a bank to a firm in each year. Assuming that the majority of a firm’s real estate is
in the same MSA as its headquarter, we can calculate the outstanding loans of each lender
in each MSA. With the exact interest rate of each loan, we can identify specific loans as
zombie loans. With the COMPUSTAT data, we classified an entire firm as a zombie.

2.3.1 Zombie Lending using Lending Data

To identify zombie firms from the COMPUSTAT data, we relied on an average rate of
interest paid by a firm. Dealscan allows us to identify the specific loans which are "subsi-
dized". The variable "AllInDrawn" is a composite way of reporting the pricing of facilities.
We can thus compare the rate across facilities regardless of the underlying fee and spread

18
structure. The AllInDrawn rates are quoted as a spread over LIBOR. The spread is not
calculated for fixed-rate loans, letters of credit, or subordinated debt. We ignore these.
Since secured loans are cheaper than unsecured loans, we divide facilities into secured
and unsecured. As before, we classify loans as zombie if the interest rate on the loan is
lower than that of the highest-rated firm.

2.3.2 Bank Market Shares

Syndicated loans are only a fraction of banks’ total lending. However, they consist of the
largest loans to the largest firms and thus account for a sizeable portion of total lending.
Previous studies (Giannetti and Saidi, 2019; Chodorow-Reich, 2014) have used syndicated
loans to evaluate bank lending policies. In our setting, negative spillover effects are larger
for large borrowers whose loans have good coverage in Dealscan.
We consider the bank holding company as the ultimate provider of credit and aggre-
gate all loans accordingly. 30% of the loans in our sample have only one lender. For
the remaining loans, we ascribe a loan to a lender if its role is either Lead Arranger, Agent,
Bookrunner, Manager, Underwriter or Sole Lender. Though both the arranger and the partic-
ipant commit capital, the average lead arranger share is four times as large as the average
participant share. Participants in a syndicated loan are more likely to sell their loans in
the secondary market (Irani, Iyer, Meisenzahl and Peydro, 2018). For these reasons, the
lead arranger is considered the "lender" in literature (Ivashina and Scharfstein, 2010).
If the arranger share data is missing, we attribute the median allocation of arrangers
in our sample to the arranger. In the case of more than one arranger, we assign each
arranger an equal fraction of the lead arranger’s total share. We assume that a bank that
arranges a loan retains it on its balance sheet. So, to calculate the loans retained by a bank,
we add all loans that have not matured. In our setting, lenders with a large market share
in a location have stronger incentives to avoid price-default spirals. This is because their
decision to liquidate loans has a larger price impact. We assume that all creditors have

19
the same seniority.
To evaluate this mechanism, we identify loans that are still outstanding. We define
MarketShare as the dollar amount of loans arranged by a bank that have not matured,
divided by the dollar amount of all loans issued in an MSA. Large banks have a more
geographically diversified portfolio while most small banks tend to lend locally. Thus
for small banks, loans to firms in an MSA is large compared to their total lending. Our
model shows that liquidation by smaller banks does not have a price externality. So, a
large portfolio share of loans to an MSA will not lead to zombie lending incentives. We
define PortfolioShare as the total outstanding loans of a bank in an MSA divided by the
total outstanding loans of a bank. Finally, CollateralShare is the share of collateral in an
MSA that a bank has access to divided by the total collateral available in an MSA. To
calculate CollateralShare, we ascribe the collateral of a firm to a bank in proportion to the
bank’s share of outstanding loans to the firm.

3 Empirical Strategy
We conduct our empirical analysis in three stages. We first document the effect of col-
lateral value on firms’ borrowing, investment, and productivity during positive and neg-
ative shocks to real estate collateral value. Second, we link this to the probability of re-
ceiving zombie loans during real estate downturns.Third, we exploit heterogeneity across
bank-exposures to local real estate markets to establish the mechanisms driving the zom-
bie lending.

3.1 Motivating specifications: Firm borrowing and real estate collateral

We motivate our empirical analysis by separating the effect of collateral value on debt
during positive and negative real estate price shocks. We use a specification similar to
Chaney et al. (2012), who relate increases in real estate collateral value to firm investment.

20
We use the specification:

yi,t =αi + δt + β 1 × Real Estatei,t + β 2 × Negative Shock MSA,t

(9)
+ β 3 × Real Estatei,t × Negative Shock MSA,t + Controlsi,t−1 + ei,t

for firm i, in year t. The outcome variable, yi,t is the change in debt from t-1 to t. αi and
δt are the firm and year fixed effects and control for the time-invariant firm-level factors
and the macroeconomic shocks affecting all firms, respectively. Real Estatei,t is the market
value of real estate held by a firm normalized by lagged plant, property and equipment. It
is calculated by multiplying the change in the Housing Price Index (HPI) by the real estate
holding of each firm in 1993. Negative Shock MSA,t is 1 if real estate price growth in an
MSA is negative in year t. Control variables included are Tobin’s q, KZ index, age and age
squared. These variables control for firms’ investment opportunity, financial constraints,
and age. Standard errors are clustered at the firm level. β 1 measures the sensitivity of
yi,t to real estate prices. The coefficient of interest, β 3 measures the differential effect on
sensitivity to real estate prices during real estate price declines. β 1 + β 3 , measures the
sensitivity to real estate prices during real estate price declines. β 2 controls for the overall
decline in the outcome variable.
If β 3 equals zero, this would imply that there are no differential effects on collateral-
based lending during real estate price declines. If β 3 is greater than zero, this would imply
that sensitivity to real estate collateral increases during downturns, say if, lender risk
aversion increases during periods of stress and lenders switch to more secured lending
(that is, lending secured by real estate as collateral). If β 3 is less than zero, this would
imply that sensitivity to real estate collateral declines during real estate price declines.
Sensitivity may decline if banks continue lending to borrowers whose real estate collateral
values decline (due to the real estate price declines).7
7 Another reason for the decline in sensitivity could be, if say, firms switch to borrowing from other
sources of financing and start relying less on real estate collateral-based borrowing.

21
 One reason for such continued support during real estate prices declines is if lenders
believe that the shock is temporary and firms are likely to recover in the future. We
examine whether debt issuance is driven by the healthier firms relative to the distressed
firms using the specification:

yi,t = β 1 ∗ Low ICRi,t + β 2 ∗ Negative Shock MSA,t

+ β 3 ∗ Low ICRi,t ∗ Negative Shock MSA,t (10)

+ Controlsi,t−1 + αi + δt + ei,t

for firm i in time t. αi and δt are the firm and year fixed effects. yi,t is the debt issuance
in year t. Low ICRi,t is 1 for firms in the bottom tercile of firms the 3-year moving aver-
age of the interest coverage ratio (ICR). ICR is the ratio of profit to interest expenses and
measures a firm’s ability to service its debt. Low ICRi,t captures the relatively distressed
or low quality borrowers in the economy. Control variables included are Tobin’s q, KZ
index, age and age squared. These variables control for firms’ investment opportunity,
financial constraints, and age. Standard errors are clustered at the firm level. The co-
efficient of interest, β 3 , estimates the change in debt issuance for the low-quality firms
relative to the high-quality firms during downturns. A positive β 3 would indicate that
the increase in lending is higher for the low-quality firms. A negative β 3 would indicate a
decline in lending to distressed borrowers consistent with say if banks become risk-averse
during downturns and continue lending to only the safer borrowers.

22
3.2 Main specification: Zombie lending and real estate collateral
To estimate the probability of zombie lending to high collateral vs low collateral firms, we
run a logit regression as below.

Zombiei,t = β 1 ∗ High Real Estatei,t + β 2 ∗ Negative Shock MSA,t

+ β 3 ∗ High Real Estatei,t ∗ Negative Shock MSA,t (11)

+ Controlsi,t−1 + αi + δt + ei,t

for firm i in year t. High Real Estatei,t is an indicator for firms with above-median real
estate value, measured as the market value of real estate held by a firm normalized by
lagged plant, property and equipment. Negative Shock MSA,t is an indicator for an MSA-
year with negative real estate price change as measured by the house price index. αi and
δt are the firm and year fixed effects and control for the time-invariant firm-level unob-
servable factors and the annual shocks affecting all firms uniformly. Control variables
included are Tobin’s q, KZ index, age and age squared. These variables control for firms’
investment opportunity, financial constraints, and age. Tobin’s Q and the KZ-Index to
control for the demand and supply of credit respectively. The coefficient of interest, β 3 ,
estimates the probability of a high collateral firm receiving a zombie loan compared to a
low collateral firm during a negative shock. β 1 estimates the probability of receiving a
zombie loan for high real estate firms during normal times.

3.3 Specifications to examine underlying mechanisms

If the existence of zombie firms has negative spillover effects on an industry (Caballero,
Hoshi and Kashyap, 2008), banks should be wary of lending to zombie firms. ideally, they
would force these firms to restructure and help the creative destruction process. Banks
may, on the other hand, continue lending to these firms if they believe that fire sale of
assets can depresses the value of other firms, which can then negatively affects the bank’s

23
assets. Our model (proposition 3) predicts that banks with a large market share of loans
bear a proportionate cost of the fire sale of land, depressing prices, and would likely
trade-off the cost of zombie lending with zombie spillover on their own balance sheets.
To see if this mechanism is in play, we test if banks with large market shares in an MSA
are more likely to engage in zombie lending to firms with high real estate collateral.
??? We estimate a bank k’s propensity to provide a zombie loan to a firm with high
real estate holdings (because these firms would have a larger impact on real estate prices
in case of a fire sale) following a shock to the MSA depends on its market share in an
MSA. Our specification is as follows:

Zombie Relni,k,t = β 1 ∗ High Real Estatei,t + β 2 ∗ Negative Shock MSA,t + β 3 ∗ Lender Sharek,t−1

+ β 4 ∗ High Real Estatei,t ∗ Negative Shock MSA,t ∗ Lender Sharek,t−1

+ αk + δt + ei,k,t
(12)
here, the outcome Zombie Relni,k,t is a subsidized loans by lender k to a firm i in an MSA.
NegShock is an indicator variable indicating declining real estate prices in the MSA. High-
RealEstate indicates firms with above-median real estate holdings and Lendershare is the
market share of the lender. LenderShare can be defined in two ways. MarketShare is the
share of loans outstanding held by the bank in an MSA and CollateralShare is the share of
collateral in an MSA that a bank has claims on divided by the total available collateral in
an MSA. δt and αk are year and lender fixed-effects respectively. This allows us to exclude
possible alternative explanations that weakly capitalized banks may want to lend to their
clients so that they don’t have to recognize bad loans and hence raise more equity capital
(Caballero, Hoshi and Kashyap, 2008 and Acharya, Eisert, Eufinger and Hirsch, 2019).
We cluster standard errors at the lender level to allow for bank policies to be correlated
across time and MSAs.
The coefficient of interest, β 4 if positive will indicate that during distress, banks with

24
higher market share are more likely to provide a subsidized loan to firms with large real
estate holdings.

3.4 Identification
Endogeneity in our estimation can arise if decline in real estate prices across MSAs are
correlated and a fall in real estate prices led to lax monetary policy and subsequent in-
crease in lending by banks. Our identification strategy thus exploits the variation across
MSAs in the size and timing of the housing bust in the 2000s. Different MSAs in the
United States had different house price appreciation during the boom. MSAs differ in the
size of changes in local housing demand (Davidoff, 2016 and Ferreira and Gyourko, 2012)
and in local housing supply elasticity (Mian and Sufi, 2011). There is a consensus that
the variation in housing prices during the the boom and bust of the 2000’s was not the
result of changes in traditional fundamentals like productivity, income, or population,
but rather was the result of factors specific to the housing market. These explanations
include irrational exuberance, the introduction of products like interest-only-mortgages,
and changes in lending standards.
We use the size and timing of structural breaks in housing prices as an event study
which allows us to assess whether sharp changes in local housing demand lead to a
change in trend in zombie lending or borrowing by firms. We interpret the estimate as
the reduced-form effect of a structural break in local house prices, which is valid whether
the break is caused by speculative forces or from a combination of these forces and other
economic shocks.
To create our event, we search for sharp changes in housing prices between 2001 and
2010. We rely on the assumption that underlying economic fundamentals do not change
abruptly and that sharp breaks from the trend in a market’s housing price reflects varia-
tion due to exogenous or housing-related shocks and not due to abrupt changes in bank
lending. Figure 1 shows the housing price index from 2001 to 2010 for three MSAs which

25
illustrate the identification and timing of the structural break. Using quarterly house
prices for each MSA between 2001 and 2010, we estimate an OLS regression (Bai and Per-
ron, 1998) with a single structural break and search for a break that maximizes the fit of
the equation

HPI MSA,t = αmsa + τmsa ∗ t + λmsa (t − t∗msa )1[t > t∗ ] + e MSA,t (13)

here, HPI is the log house price index in an MSA in quarter-year t, and t∗ is the time of the
structural break. τ is the MSA’s linear price trend before the break and λ is the size of the
structural break. This estimation is similar to Ferreira and Gyourko (2012) and Charles,
Hurst and Notowidigdo (2018).
Finally, we estimate the event study regression

yi,t = β 1 ∗ break t−1 + β 2 ∗ HighRealEstatei + β 3 ∗ break t−1 ∗ HighRealEstatei + ei,t (14)

here, break is an indicator for the year which maximizes the R2 of equation (??), and
HighRealEstate is an indicator for high real estate holding firms. β 3 is the coefficient of
interest that allows us to assess the instantaneous impact of a structural break in prices
on lending and probability of zombie lending (yi,t ).

3.5 Spillover
Most studies of zombie lending have focused on Southern Europe post the sovereign debt
crisis or Japan during the lost decades of 1990-2010. Since we are studying a different
country, we similarly explore the spillover effects of zombie lending on healthy firms.
Our regression follows Caballero, Hoshi and Kashyap (2008)

yi,t = β 1 ∗ NonZombiei,t + β 2 ∗ IndZombiePct j,t + β 3 ∗ NonZombiei,t ∗ IndZombiePct j,t

+ αi/j + δt + ei,t
(15)

26
 where NonZombie is an indicator for firms not classified as zombie in the year. Ind-
ZombiePct is the percentage of zombie firms in an industry. The unit of observation is
the firm-year where i denotes the firm and t denotes the year. αi/j is either the firm or
industry fixed effect accounting for either firm-specific or industry-specific shocks. δt is
the year fixed effect controlling for the annual shocks affecting all firms uniformly.
The dependent variables are investment, profit, productivity, change in employment,
and return on assets. We expect the existence of zombie firms in an industry to depress
all the above variables for healthy firms.

4 Results
We now present the results of our analysis. First we identify the effect of collateral value
on borrowing during and adverse shocks. We then test to see if lending during adverse
shocks is dominated by high-quality borrowers or is some of it misallocated. Next, we
estimate the probability of zombie loans for borrowers with high collateral and banks
with larger market shares. We conclude by documenting the spillover effects of zombie
lending on healthy firms.

4.1 Sensitivity of Debt to Real Estate Value

Increases in collateral values provides firms access to more and cheaper credit for longer
maturities (Benmelech, Garmaise and Moskowitz, 2005; Benmelech and Bergman, 2009).
This is the collateral channel of lending. We begin by separating the sensitivity of debt to
collateral during normal and adverse times. Table 2 estimates (9) where the dependent
variable is the change in long-term debt. During normal times, the as real estate values
rise, firms are able to raise more debt. A 1% increase in real estate value results in an
increase in debt by 0.046%. During adverse periods the elasticity of debt issuance to
collateral value decreases to 0.017. That is, the deleveraging during a downturn slower
than the leveraging during good times.

27
4.2 Inefficient Lending
A possible explanation of firms increasing their borrowing and investment during neg-
ative shocks is that interest rates fall during a recession which makes it a good time to
borrow cheaply for firms that have profitable investment opportunities and are not finan-
cially constrained.
Firms that have a low interest coverage ratio have a lower ability to cover their in-
terest expense from their operating income, are more susceptible to downturns, and are
therefore considered riskier. Generally, we expect banks to lend less to firms with a low
interest coverage ratio or lend at a higher interest rate. We divide firms into terciles by
interest coverage ratio. A comparison of our sorting with S&P ratings shows that the bot-
tom terciles of firms by interest coverage ration corresponds to firms with ratings of BB
or lower as discussed previously. These firms should in general receive less credit than
other firms.
We test this intuition via (10) in Table 3. We regress debt issuance on an indicator for
the lowest moving average interest coverage ratio tercile and negative shock. Our results
(Table 3, Row 1) indicate that low interest coverage ratio firms indeed issue 12.5% less
debt than other firms during normal times. During a negative shock (Table 3, Row 3), we
find that low ICR firms surprisingly issue 45% more debt than firms with higher interest
coverage ratios. These results are robust to controls of firm demand for credit as well as
financial constraints as measured by Tobin’s Q and the KZ-Index respectively.
This is not a clear indication of inefficient lending. If these firms borrow at a higher
interest rate than more creditworthy firms, we would not classify these as zombie loans.
Having established an asymmetric borrowing pattern where safe firms issue more debt
during normal times while riskier firms issue more debt during negative shocks, we pro-
ceed to identify instances of zombie lending, i.e. we want to check if the additional bor-
rowing by risky firms during negative shocks is at market rates or at subsidized rates
intended to keep them afloat.

28
4.3 Zombie Lending and the Collateral Channel

4.3.1 Estimating Zombie Lending

We follow the specification of Caballero, Hoshi and Kashyap (2008) and Acharya, Eisert,
Eufinger and Hirsch (2019) to identify zombie firms. We then estimate (11) to verify our
hypothesis that collateral is a channel for zombie lending. The results are presented in Ta-
ble 4, Panel A. We successively add controls to the baseline specification and find that the
coefficient of interest (β 3 ) is always significant. In our full specification (column 4), dur-
ing a negative shock, firms with above-median real estate holdings (High Real Estate) are
2.3 times more likely to receive a subsidized loan compared to firms with below-median
real estate holding. To control for variation between industries where certain sectors may
get preferential treatment due to government policy intervention or any industry-specific
news, we run the regression with industry fixed effects (Column 5) as well. In this case,
firms with above-median real estate are 1.5 times more likely to receive zombie loans
compared to firms with below-median real estate holdings. This is our main result and
validates our hypothesis that collateral is an important channel for evergreening of loans
viz. zombie lending.

4.3.2 Survivorship Bias and Robustness

Survivorship Bias: Our method to calculate the market value of real estate holdings of a
firm requires that the sample firms exist in 1993 (the last year when accumulated depreci-
ation is available in COMPUSTAT). This introduces a survivorship bias in our sample. To
establish the robustness of our result, we assume that tangible assets i.e. property, plant,
and equipment are a good proxy for collateral value and divide our sample into firms
with above and below the median value of PPE in the first year of the appearance of the
firm in the sample. We then re-estimate equation (11) in Table 4B. Here, high PPE firms
are 1.48 times more likely than low PPE firms to receive zombie loans. The results are

29
similar to those in Panel A which alleviates some concern about survivorship bias in our
baseline result.

Robustness In Table 5, we reestimate (11) with an indicator for zombie loans identified
from the dealscan on an indicator for firms with above-median collateral in 1993 (High
Real Estate) and separately, firms with above-median book value of collateral in the year
of their first appearance in the dataset (High Real Estate1) which controls for survivorship
bias as above. We use lender and year fixed effects to control for heterogeneity arising
out of lender characteristics and time-varying heterogeneity. The identification of zombie
using dealscan data is cleaner because we have the exact borrowing cost and are able to
identify specific loans as zombie loans. The results confirm our hypothesis that firms with
above-median collateral value are more likely to receive zombie loans during negative
shocks than firms with below-median collateral value.
These three results robustly confirm our findings and identify collateral as a channel
for zombie lending.

4.4 Mechanism
Table 6 tests the main result of our model viz. the probability of zombie lending depends
on the market share of the bank in an MSA. We explore whether high market-share banks
are more likely to indulge in zombie lending during a downturn and if so, do they pref-
erentially lend to high collateral firms. We estimate (12) by regressing bank market share
and indicators for high real estate and negative shock on an indicator of zombie loans.
Column 1 shows that High Real Estate firms are more likely to get a zombie loan from a
bank if the bank has a higher share of loans in the MSA. A 1% increase in a bank’s mar-
ket share increases its probability of giving a zombie loan by 0.042%. This is the gist of
proposition 3 that a bank which has a larger presence in an MSA internalizes the negative
spillover of the bankruptcy of a firm with large real estate and does not recognize bad
loans. Instead it provides the firm with subsidized loans hoping for a recovery in the

30
future and preventing a spillover of a fire sale to its other assets.
More important to the fire sale mechanism is the share of collateral that a bank has
liens to which it will need to liquidate in case of a default. Column 2 tests if banks with
higher collateral share in an MSA are more likely to give zombie loans. We estimate that
an increase of 1% in collateral share is associated with an increase of 0.047% increase in
the probability of zombie lending to high real estate firms after a fall in real estate prices.
Along with an increase in zombie lending to specific firms, banks should increase
their zombie lending in an MSA in order effectively prevent a collapse in real estate prices.
Table 7 regresses measures of a bank’s share of zombie loans in an MSA on the bank’s
market share and collateral share and finds that banks with a larger market (or collateral)
share are likely to have a larger share of zombie loans during distress in the MSA. In Table
7, BankNewZombie is the ratio of the new zombie loans disbursed by a bank in an MSA
to the total outstanding loans of a bank. An increase in this measure would indicates
a bank subsidizing its borrowers either in a specific MSA. It could also indicate a bank
systematically giving out cheap loans to everyone. MSANewZombie is the ratio of new
zombie loans disbursed by a bank in an MSA to the total loans disbursed by the bank in
the MSA. MSANewZombie specifically measures if a bank is giving out more zombie loans
in the MSA compared to other banks.
During normal times, an increase in market (or collateral) share of a bank reduces the
zombie-proportion of the new loans it originates (Row 1). In column 1 (& 2) we see that a
1% increase in market share during a shock leads to an increase in the proportion of new
zombie loans disbursed by the bank by 0.027% (0.022%) while it increases the new zombie
share of loans by 0.063% (0.053%). The results indicate that high market (collateral) share
banks originate a higher number of new zombie loans in an MSA during a downturn.

31
4.5 Identification
We use the variation between MSAs in the timing and size of structural breaks in house
prices during the 2007-2009 financial crisis as our identification strategy. Figure 2 shows
the distribution of the structural breaks in different MSAs by year. There is sufficient
variation in the year of the structural break to infer that the increase in lending and zombie
lending during a negative shocks as in Table 2 and Table 4 are not due to global factors
affecting lending. Moreover, most of the structural breaks are concentrated in 2005 and
2006 which were years before the Fed reduced interest rates so that any additional lending
was not firms taking advantage of low rates. Furthermore, since the shocks would have
begun in the MSAs identified, we can be certain that the additional lending is not because
of increased collateral value either.
The structural breaks are used to get the instantaneous effect of a change in the value
of collateral to changes in debt and the probability of zombie lending. We estimate (14)
in Table 8. Break 1 indicates a negative structural break between 2001 and 2010 while
Break 2 is a negative and significant structural break. We regress the zombie firms and
∆ Debt on HighRealEstate1 and the structural break indicator to create an event study
for our results in Table 2 and Table 4 & Table 5. We find that the probability of zombie
lending is approximately 50% higher (Columns 1 & 2) for high real estate firms post a
structural break compared to low real estate firm. The magnitude of the estimate is very
similar to the magnitude estimated in Table 4. The two indicators of structural break are
increasingly significant as is expected from their construction. The result confirms our
initial results that during a negative shock, firms with high collateral are more likely to
receive zombie loans and any loans in general as well. Columns 3 & 4 also show that high
real estate are more likely to raise additional debt post a structural break (as in Table 2).

32
4.6 Spillover Effects of Zombie Firms
Finally, we explore the spillover effects of zombies on non-zombie firms in an industry.
Non-zombie firms in an industry can be affected by zombie firms through two channels.
Evergreening of loans shifts the supply of credit to these zombie firms reducing the credit
available to healthier firms. The effect of zombie firms on investment can go both ways.
Healthy firms may know that other firms are unproductive increase their investments
hoping to gain a greater share of the market. On the other hand, the presence of zom-
bie firms may induce banks to reduce their exposure to the sector which would reduce
investments by healthy firms.
The prevalence of zombies also distorts the competitive process in the industry Acharya,
Crosignani, Eisert and Eufinger, 2020. In a market without distortions, impaired firms
would reduce employment and lose market share. This gives more productive (non-
zombie) firms access to a larger talent pool, allowing them to increase market share and
thereby profitability. But, subsidized credit keeps zombie firms artificially alive which
congests the market. Zombie firms also bear lower interest rates than non-zombie firms
and so have lower costs. This reduces product market prices and correspondingly mark-
ups and increases wages in the industry.
Following Caballero, Hoshi and Kashyap (2008), equation (15) allows us to estimate
the effect of the presence of zombie firms in an industry on healthy firms. Table 9 presents
results which show that non-zombie firms have lower investments and return-on-assets
when (β 3 < 0) they operate in industries with many zombie firms compared to firms in
industries with lower zombie percentage. The estimates imply that a 1% rise in zombie
percentage leads to non-zombie firms reducing their investment by 1.16% and their ROA
is reduced by 0.95%.

33
5 Conclusion
Theory predicts a positive relationship between debt and collateral. We show that firms
with collateral can raise more debt during a shock than firms with less collateral. Some
of this lending results from increased risk aversion and move towards safer lending by
banks. But, we observe that some is directed to low-quality firms.
We next show that firms with high collateral have a higher probability of receiving
zombie loans. This identifies collateral as a channel for zombie lending. Previous work
by Peek and Rosengren (2005) and Acharya, Eisert, Eufinger and Hirsch (2019) show
that weak banks (banks close to their regulatory capital requirements) are responsible
for zombie lending. Since equity capital is expensive, banks close to regulatory capital
requirements prefer to evergreen their loans. This allows a bank to not recognize non-
performing assets (or defaults) which would force the bank to set aside capital buffers for
the impending default.
Our model and empirical tests show that banks behave rationally in evergreening
some of these loans. Banks with large market share are aware that since they hold a lien
on a large proportion of real estate in an area, a fire sale will hurt their portfolio. By
evergreening some loans, they minimize the spillover on their balance sheet.
However, zombie lending has negative spillovers on healthy firms in an economy.
The credit misallocation reduces return-on-assets and employment of healthy firms in
industries dominated by zombie firms. Firms in an industry dominated by zombies face
a congested market, have lower mark-ups, and higher labour costs. This reduces future
investment as well.
The presence of collateral can lead banks into believing that some loans are safe while
ignoring the underlying financials of a firm. Banks are averse to calling in collateral dur-
ing recessions because of possible contagion effects of selling collateral in a bad economy.
Zombie lending in some sense subverts regulatory oversight by dressing up the bank’s

34
and firm’s balance sheets. Besides it also has negative spillover effects on healthy firms.
Regulators would, thus, need to be especially wary of it.

35
References
Acharya, V., Crosignani, M., Eisert, T. and Eufinger, C. (2020), ‘Zombie credit and (dis-)
inflation: Evidence from europe’, Available at SSRN
Acharya, V., Eisert, T., Eufinger, C. and Hirsch, C. (2019), ‘Whatever it takes: The real
effects of unconventional monetary policy’, The Review of Financial Studies, 32(9), 3366–
3411
Almeida, H. and Campello, M. (2007), ‘Financial constraints, asset tangibility and corpo-
rate investment’, Review of Financial Studies
Anenberg, E. and Kung, E. (2014), ‘Estimates of the size and source of price declines due
to nearby foreclosures’, American Economic Review
Asquith, P., Gertner, R. and Scharfstein, D. (1994), ‘Anatomy of financial distress: An
examination of junk-bond issuers’, The Quarterly Journal of Economics
Bai, J. and Perron, P. (1998), ‘Estimating and testing linear models with multiple structural
changes’, Econometrica
Banerjee, R. and Hofmann, B. (2018), ‘Rise of zombies: Causes and consequences’, BIS
Quarterly
Bebchuk, L. A. and Goldstein, I. (2011), ‘Self-fulfilling credit market freezes’, The Review
of Financial Studies
Benmelch, E. and Bergman, M. K. (2011), ‘Bankrupcy and the collateral channel’, Journal
of Finance
Benmelech, E. and Bergman, N. (2009), ‘Collateral pricing’, Journal of Financial Economics
Benmelech, E., Garmaise, M. and Moskowitz, T. (2005), ‘Do liquidation values affect fi-
nancial contracts? evidence from commercial loan contracts and zoning regulation’,
Quarterly Journal of Economics
Besanko, D. and Thakor, A. (1987), ‘Collateral and rationing: Sorting equilibria in monop-
olistic and competitive credit markets’, International Economic Review

36
Blattner, L., Farinha, L. and Rebelo, F. (2018), ‘When losses turn into loans: The cost of
undercapitalized banks’, Working Paper
Brunnermeier, M. K. and Pedersen, L. H. (2009), ‘Market liquidity and funding liquidity’,
The review of financial studies, 22(6), 2201–2238
Caballero, R., Hoshi, T. and Kashyap, A. (2008), ‘Zombie lending and depressed restruc-
turing in japan’, American Economic Review
Caballero, R. J. and Simsek, A. (2010), ‘Fire sales in a model of complexity’, MIT Depart-
ment of Economics Working Paper
Campbell, J. Y., Giglio, S. and Pathak, P. (2011), ‘Forced sales and house prices’, American
Economic Review
Carlson, J., Haubrick, J. G., Cherney, K. and Wakefield, S. (2009), ‘Credit easing: A policy
for a time of financial crisis’, Federal Reserve Bank of Cleveland
Chaney, T., Thesmar, D. and Sraer, D. (2012), ‘The collateral channel: How real estate
shocks affect corporate investment’, American Economic Review
Charles, K., Hurst, E. and Notowidigdo, M. (2018), ‘Housing booms and busts, labour
market opportunities, and college attendance’, American Economic Review
Chava, S. and Roberts, M. (2008), ‘How does financing impact investment? the role of
debt covenants’, Journal of Finance
Chodorow-Reich, G. (2014), ‘The employment effects of credit market disruptions: Firm-
level evidence from the 2008-09 financial crisis’, Quarterly Journal of Economics
Cvijanovic, D. (2014), ‘Real estate prices and firm capital structure’, Review of Financial
Studies
Davidoff, T. (2016), ‘Supply constraints are not valid instrumental variables for home
prices because they are correlated with many demand factors’, Critical Finance Review
Favara, G. and Giannetti, M. (2017), ‘Forced asset sales and the concentration of outstand-
ing debt: evidence from the mortgage market’, The Journal of Finance
Ferreira, F. and Gyourko, J. (2012), ‘Hetrogeneity in neighbourhood-level price growth in

37
 the united states 1993-2009’, American Economic Review
French, Kenneth R., e. a. (2010), ‘The squam lake report’, Princeton University Press: Prince-
ton, NJ
Gan, J. (2007), ‘The real effects of asset market bubbles: Loan- and firm-level evidence of
a lending channel’, Columbia University Academic Commons
Garmaise, M. J. and Moskowitz, T. J. (2006), ‘Bank mergers and crime: The real and social
effects of credit market competition’, the Journal of Finance
Giannetti, M. and Saidi, F. (2019), ‘Shock propagation and banking structure’, Review of
Financial Studies
Giannetti, M. and Simonov, A. (2013), ‘On the real effects of bank bailouts: Micro evidence
from japan’, American Economic Journal: Macroeconomics
Gyourko, J. (2009), ‘Understanding commercial real estate: Just how different from hous-
ing is it?’, NBER Working Paper No.14708
Harding, J. P., Rosenblatt, E. and Yao, V. W. (2009), ‘The contagion effect of foreclosed
properties’, Journal of Urban Economics
Hart, O. and Moore, J. (1994), ‘A theory of debt based on the inalienability of human
capital’, Quarterly Journal of Economics
Hartley, D. (2014), ‘The effect of foreclosures on nearby housing prices: Supply or dis-
amenity?’, Regional Science and Urban Economics
Irani, R., Iyer, R., Meisenzahl, R. and Peydro, J. L. (2018), ‘The rise of shadow banking:
Evidence from capital regulation’, Working Paper
Ivashina, V. and Scharfstein, D. (2010), ‘Bank lending during the financial crisis of 2008’,
Journal of Financial Economics
Kaplan, S. and Zingales, L. (1997), ‘Do investment cash-flow sensitivities provide useful
measures of financing constraints’, Quarterly Journal of Economics
Kiyotaki, N. and Moore, J. (1997), ‘Credit cycles’, Journal of Political Economy
Lamont, O., Polk, C. and Saa-Requejo, J. (2001), ‘Financial constraints and stock returns’,

38
 Review of Financial Studies
Mian, A. and Sufi, A. (2011), ‘House prices, home equity-based borrowing, and the us
household leverage crisis’, American Economic Review
Mian, A., Sufi, A. and Trebbi, F. (2015), ‘Foreclosures, house prices, and the real economy’,
The Journal of Finance
Peek, J. and Rosengren, E. (2005), ‘Unnatural selection: Perverse incentives and the mis-
allocation of credit in japan’, American Economic Review
Petersen, M. A. and Rajan, R. G. (1995), ‘The effect of credit market competition on lending
relationships’, The Quarterly Journal of Economics
Schlingemann, F. P., Stulz, R. M. and Walkling, R. A. (2002), ‘Divestitures and the liquidity
of the market for corporate assets’, Journal of financial Economics
Shleifer, A. and Vishny, R. (2011), ‘Fire sales in finance and macroeconomics’, Journal of
Economic Perspectives
Shleifer, A. and Vishny, R. W. (1992), ‘Liquidation values and debt capacity: A market
equilibrium approach’, The Journal of Finance

39
 Table 1: Summary Statistics in 1993
Mean SD Min Max
Real Estate 0.861 1.726 0.000 22.686
Leverage 0.289 0.377 0.000 3.852
Investment 0.440 0.768 0.000 5.605
Tobin’s Q 2.334 2.778 0.534 62.878
KZ Index 0.743 2.741 -18.260 22.764
Age 11.433 10.876 0.000 34.000
ROA -0.028 0.442 -6.187 0.369
Interest Coverage Ratio 5.532 107.767 -701.667 598.400
Land Supply Elasticity 1.440 0.866 0.595 6.396
Firms 6803
Table 1 reports the summary statistics of the firms in the sample in 1993. Real estate value is
calculated by estimating average age and historical cost of the real estate holdings of a firm in
1993. The real estate index which is normalized to 1 in 2006 is used to inflate real estate holdings
to market value which is then normalized by lagged PPE. Tobin’s Q, KZ-Index, Age, Age2 and
interest coverage ratio are used as controls for credit demand and/or supply. Tobin’s Q is calcu-
lated as the ratio of enterprise value of a firm to its book value following Almeida and Campello
(2007). The Kaplan-Zingales Index is calculated based on the 5-factor model of Lamont, Polk and
Saa-Requejo (2001)). Interest coverage ratio is the ratio of a firms’ EBIT to interest expense. Firm
age is calculated using the first appearance of a firm in the COMPUSTAT dataset with a cut-off
at 1960. Leverage is the ratio of total debt to assets. Return on assets is calculated as the ratio
of operating income minus depreciation and amortization to assets. Investment is measured as
capex normalized by lagged fixed assets.

40
 Table 2: Sensitivity of Debt to Real Estate Value
(1) (2) (3) (4) (5)
∆ Long-Term Debt
Real Estate 0.066*** 0.059*** 0.061*** 0.051*** 0.046***
(0.014) (0.015) (0.015) (0.012) (0.014)

Negative Shock 0.061 0.067 0.067 0.060 0.059

(0.050) (0.052) (0.053) (0.054) (0.056)

Negative Shock × Real Estate -0.040*** -0.040*** -0.038*** -0.030** -0.029**

(0.013) (0.014) (0.014) (0.012) (0.014)

Tobin’s Q 0.022** 0.031***

(0.009) (0.010)

KZ Index -0.001 -0.009

(0.009) (0.011)

Age 0.005 0.012

(0.008) (0.008)

Age2 -0.000 -0.000

(0.000) (0.000)
Observations 35035 31503 31308 31634 29039
R2 0.11 0.12 0.12 0.12 0.14
Year FE Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 2 estimates the sensitivity of debt to collateral value by regressing change in debt on change
in value of real estate holdings from 1994 to 2014. The sensitivity of debt to real estate holdings is
positive during periods of rising real estate prices indicating that as the value of real estate holding
increases, long term borrowing increases. However, this sensitivity is lower during periods of
falling real estate prices indicating that firms do not reduce their borrowings when the value of
their collateral falls. Real estate value is calculated by estimating average age and historical cost
of the real estate holdings of a firm in 1993. The real estate index which is normalized to 1 in
2006 is used to inflate real estate holdings to market value which is then normalized by lagged
PPE. NegShock is an indicator for years which saw a fall in the Residential Price Index in a MSA.
Tobin’s Q, KZ-Index, Age and Age2 are used as controls for credit demand and/or supply. Tobin’s
Q is calculated as the ratio of enterprise value of a firm to its book value following Almeida and
Campello (2007). The Kaplan-Zingales Index is calculated based on the 5-factor model of Lamont,
Polk and Saa-Requejo (2001)). Firm age is calculated using the first appearance of a firm in the
COMPUSTAT dataset with a cut-off at 1960. Standard errors are clustered at the firm-level

41
 Table 3: Debt Issuance and Interest Coverage Ratio
(1) (2) (3) (4) (5)
Debt Issuance
Low ICR -0.111 -0.111 -0.146 -0.152 -0.125
(0.092) (0.093) (0.096) (0.099) (0.098)

Negative Shock -0.142* -0.092 -0.150 -0.109 -0.061

(0.085) (0.091) (0.095) (0.085) (0.092)

Low ICR × Negative Shock 0.309* 0.412** 0.453** 0.326* 0.454**

(0.182) (0.189) (0.201) (0.188) (0.197)

Tobin’s Q 0.198*** 0.199***

(0.028) (0.028)

KZ Index 0.014 -0.030

(0.023) (0.021)

Age 0.057*** 0.063***

(0.018) (0.020)

Age2 -0.000* -0.000

(0.000) (0.000)
Observations 53707 46436 46237 49679 42872
R2 0.40 0.42 0.40 0.40 0.42
Year FE Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 3 explores lending to low interest coverage ratio firms to high ICR firsm during normal
times and during a shock. The results indicate that during normal times, low ICR firms are less
likely to get debt whereas during a shock, low ICR firms are able to issue more debt compared to
high ICR firms. The interest coverage ratio is measured as the EBIT by the interest expense and
we define low ICR firms as the bottom tercile of firms by moving average ICR. NegShock is an
indicator for years which saw a fall in the Residential Price Index in a MSA. Tobin’s Q, KZ-Index,
Age and Age2 are used as controls for credit demand and/or supply. Tobin’s Q is calculated as the
ratio of enterprise value of a firm to its book value following Almeida and Campello (2007). The
Kaplan-Zingales Index is calculated based on the 5-factor model of Lamont, Polk and Saa-Requejo
(2001). Firm age is calculated using the first appearance of a firm in the COMPUSTAT dataset with
a cut-off at 1960. Standard errors are clustered at the firm-level

42
 Table 4: Probability of Zombie Lending
Panel A: High Real Estate Firms
(1) (2) (3) (4)
Zombie
High Real Estate -0.456 -0.392 -0.379 -0.380
(0.280) (0.300) (0.281) (0.262)

Negative Shock -0.277∗∗ -0.243∗ -0.241 -0.242∗

(0.124) (0.135) (0.150) (0.137)

High Real Estate × Negative Shock 0.837∗∗∗ 0.864∗∗∗ 0.834∗∗∗ 0.835∗∗∗

(0.238) (0.226) (0.281) (0.189)

Tobin’s Q -0.005 -0.002

(0.008) (0.008)

KZ Index -0.014 -0.014

(0.009) (0.009)
Observations 9658 8519 8416 8416
Year FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes

Panel B: High PPE Firms

(1) (2) (3) (4)
Zombie
Negative Shock 0.179∗ 0.028 0.023 0.025
(0.096) (0.095) (0.082) (0.091)

High PPE × Negative Shock 0.320∗∗ 0.403∗∗ 0.398∗∗ 0.397∗∗∗

(0.158) (0.179) (0.182) (0.135)

Tobin’s Q 0.002 0.006∗

(0.003) (0.003)

KZ Index -0.008 -0.010∗∗

(0.005) (0.005)
Observations 26473 22002 21455 21455
Year FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 4 estimates the probability of zombie lending from 1994 to 2014 by estimating a logit model on an
indicator for zombie firms on firms with above median real estate holdings and negative real estate shocks.
Following Caballero, Hoshi and Kashyap (2008), a firms is defined to have received a zombie loan if its
interest expense is lower than the highest rated firms in the year and it is rated BB or lower. Tobin’s Q
and KZ-Index are used as controls for credit demand and/or supply. Tobin’s Q is calculated as the ratio of
enterprise value of a firm to its book value following Almeida and Campello (2007). The Kaplan-Zingales
Index is calculated based on the 5-factor model of43Lamont, Polk and Saa-Requejo (2001). Panel B uses an
unbalanced dataset and proxies for firms with high collateral using firms with above median PPE. Standard
errors are clustered at the firm-level
 Table 5: Probability of Zombie Lending - Dealscan Data
(1) (2) (3) (4)
Zombie Dealscan
High Real Estate 1 -0.022*** -0.026*** -0.028*** -0.027***
(0.005) (0.005) (0.006) (0.005)

Negative Shock -0.016** -0.011 -0.010 -0.010

(0.007) (0.008) (0.008) (0.008)

High Real Estate 1 × NegShock 0.024*** 0.028*** 0.027*** 0.027***

(0.007) (0.008) (0.008) (0.008)

Tobin’s Q 0.002 0.002

(0.002) (0.002)

KZ Index 0.003*** 0.003***

(0.001) (0.001)
Observations 40062 34969 40063 34969
Year FE Yes Yes Yes Yes
Lender FE Yes Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 5 replicates the result from Table 4 using the dealscan dataset. It estimates the probability of zombie
lending from 1994 to 2014 by estimating a linear probability model on an indicator for zombie firms on
firms with above median real estate holdings and negative real estate shocks. High Real Estate1 are firms
with above median real estate holding in 1993. Following Caballero, Hoshi and Kashyap (2008), a firms is
defined to have received a zombie loan if its interest expense is lower than the highest rated firms in the
year and it is rated BB or lower. Tobin’s Q and KZ-Index are used as controls for credit demand and/or
supply. Tobin’s Q is calculated as the ratio of enterprise value of a firm to its book value following Almeida
and Campello (2007). The Kaplan-Zingales Index is calculated based on the 5-factor model of Lamont, Polk
and Saa-Requejo (2001).Standard errors are clustered at the firm-level

44
 Table 6: Probability of Zombie Lending - Bank Market Share
(1) (2) (3)
Zombie Dealscan
High Real Estate 0.005*** 0.005*** 0.008***
(0.001) (0.002) (0.003)

High Real Estate × Negative Shock -0.006*** -0.007*** -0.009***

(0.003) (0.002) (0.003)

High Real Estate × Market Share -0.009 -0.046***

(0.007) (0.017)

Negative Shock × Market Share -0.001 -0.010*

(0.007) (0.005)

High Real Estate × Negative Shock × Market Share 0.042* 0.049***

(0.025) (0.014)

High Real Estate × Collateral Share -0.014*

(0.008)

Negative Shock × Collateral Share -0.007

(0.009)

High Real Estate × Negative Shock × Collateral Share 0.047***

(0.022)

High Real Estate × High Capital -0.011***

(0.004)

High Real Estate × Negative Shock × High Capital 0.010***

(0.003)

High Real Estate × Negative Shock -0.076**

× High Capital × Market Share (0.031)
Observations 13323 13258 8615
Year FE Yes Yes Yes
Lender FE Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 6 estimates a linear probability model to estimate how the probability of zombie lending depends on
bank market share. The dependent variable is an indicator for zombie lending to a firm by a specific lender
in a year. The independent variables include indicators for above median real estate firms, a negative real
estate shock and an estimate of a bank’s market share. Market Share is the percentage of outstanding loans in
an MSA arranged by a bank. Portfolio Share is the percentage of outstanding loans arranged by a bank in an
MSA divided by the total outstanding loans arranged by the bank. Collateral Share is the share of collateral
in an MSA that a bank has access to which is calculated as the share of outstanding loans arranged for a
firm multiplied by the value of collateral of the firm divided by the total collateral available in an MSA.
45 top tercile by Tier 1 capital in a year. Standard errors
High Capital is an indicator variable for banks in the
are clustered at the firm-level
 Table 7: New Zombie Loans in MSA
(1) (2) (3) (4)
BankNewZombie MSANewZombie
Market Share -0.008*** -0.018***
(0.003) (0.005)

Collateral Share -0.007** -0.016***

(0.003) (0.006)

Negative Shock -0.004* -0.003* -0.012*** -0.011***

(0.002) (0.002) (0.004) (0.004)

Negative Shock × Market Share 0.027** 0.063**

(0.013) (0.031)

Negative Shock × Collateral Share 0.022* 0.053*

(0.012) (0.029)
Observations 8865 8802 8865 8802
Year FE Yes Yes Yes Yes
Lender FE Yes Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 7 tests whether high market share banks internalize the cost of spillover in their lending decisions.
The independent variable is the market share of a lender in an MSA in a year. The dependent variables
are the BankNewZombie which is the ratio of new zombie loans of a bank in an MSA to the total loans of
the bank. MSANewZombie is the ratio of the new zombie loans disbursed by a bank in an MSA to the total
value of new loans disbursed by the bank in the MSA. Negative Shock is an indicator for negative real estate
shock. The results indicate that high market share banks originated a higher fraction of zombie loans in an
MSA by number and dollar amount thus confirming our hypothesis. Standard errors are clustered at the
firm-level

46
 Table 8: Structural Break - Probability of Zombie Lending
(1) (2) (3) (4)
Zombie ∆ Long-Term Debt
Break 1 0.129 -0.322***
(0.374) (0.119)

High Real Estate1 × Break 1 0.487* 0.268***

(0.252) (0.078)

Break 2 0.331 -0.293**

(0.365) (0.118)

High Real Estate1 × Break 2 0.517** 0.286***

(0.253) (0.079)
Observations 5157 5157 19753 19753
Year FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table 8 uses a structural break in real estate prices to create an instrument for local negative real
estate shocks following Charles, Hurst and Notowidigdo (2018). We regress and indicator for
zombie lending and change in debt on the structural break IV and an indicator for above median
real estate holding to estimates the probability of zombie lending. Following Caballero, Hoshi
and Kashyap (2008), a firms is defined to have received a zombie loan if its interest expense
is lower than the highest rated firms in the year and it is rated BB or lower. Break1 indicates
when the structural break is negative and break2 is an indicator for a significant negative break.
Columns 1-2 estimate the probability of zombie lending following a negative structural break
while columns 3-4 estimate the change in debt following a negative structural break which tie in
with the results of Table 4,5 and 2 respectively. Standard errors are clustered at the firm-level

47
 Table 9: Spillover Effects of Zombie Firms
(1) (2) (3) (4) (5) (6) (7) (8)
Investment ROA Productivity ∆ Emp Investment ROA Productivity ∆ Emp
Non Zombie 0.019 0.079*** 0.530*** -0.017** -0.063* 0.304*** 0.869*** -0.004
(0.030) (0.025) (0.044) (0.007) (0.033) (0.031) (0.050) (0.008)

IndustryZombiePct 1.045** 0.592* -0.296 -0.039 0.906 1.761** 0.864* -0.147

(0.463) (0.328) (0.480) (0.105) (0.624) (0.661) (0.461) (0.144)

Non Zombie × IndustryZombiePct -1.165** -0.953*** 0.419 0.057 -0.838 -2.086*** -0.296 -0.056
(0.473) (0.345) (0.503) (0.107) (0.613) (0.599) (0.478) (0.134)
Observations 107400 117029 76159 95448 107400 117029 76159 125300
R2 0.33 0.65 0.54 0.18 0.06 0.05 0.02 0.02
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes No No No No
Industry FE No No No No Yes Yes Yes Yes
48

Standard errors in parentheses

* p < 0.10, ** p < 0.05, *** p < 0.01
Table 9 explores the spillover effects of zombie lending from 1994 to 2014. The specification follows Caballero, Hoshi and Kashyap (2008). The
regressors are an indicator for non-zombie firms, the percentage of zombie firms in an industry and an interaction term. Our variables of interest
are investment, return on assets, productivity and employment growth. The results show that a greater percentage of zombie firms in an industry
depresses investment and returns on assets of non-zombie firms in the industry. Investment is measured as capex normalized by the lagged PPE, and
return on asset as EBIT by lagged assets. Productivity is calculated using a Cobb-Douglas function for firms with the capital-labour ratio determined
at the 2-digit SIC level. We control for year and firm fixed effects in the first four columns and industry and year fixed effects in the last four.Standard
errors are clustered at the firm-level
 Figure 1: Structural Breaks
(A) Positive Structural Break - Midland,TX

(B) Non-Significant Structural Break - Waco,TX

(C) Negative Structural Break - Merced,CA

Note: Figure 1 This figure shows graphs of quarterly house price data for three MSAs. The house
price index for each city is normalized so that 2001:I = 100 . The dotted lines report the house
price series, while the solid lines reports the estimated house price and also the structural breaks.
Figure 1A shows a positive and significant structural break while Figure 1C shows a negative and
significant structural break.

49
 Figure 2: Distribution of Structural Breaks

Note: Figure 2 shows the percentage of MSA’s that have a structural break in a year. Equation 13
identifies the year between 2001 and 2010 in which a structural break in housing prices occurs in
an MSA. The heterogeneity in the timing of this structural break is used as an instrument for our
identification strategy in equation 14.

50
Appendix A: Proofs
A.1 Proof of Proposition 1
Taking the first order condition for equation 4 and substituting L = 0 (since it has been
assumed to be very small), we get

dV ( p(λ)) (1 − β)(1 − f )C
− = 0. (A.1)
d( p(λ)) βf

λ∗ is given by the solution to this equation. The first term is decreasing in λ and last term
is independent of λ. Hence we get a unique solution λ∗ if condition (5) holds.
Next we analyse how λ∗ changes with f . The derivative of the first term in (A.1) w.r.t
is f equals V 00 ( p(λ)) p f (λ), where p f is partial derivative of p w.r.t. f . Both V 00 ( p(λ)) and
p f (λ) are negative, hence product is positive. Thus the first term is increasing in f . The
second term in (A.1) is clearly decreasing in f . Hence as f increases λ∗ must increase.

A.2 Proof of Proposition 2

We prove only the first part of the proposition. The second part can be proved analo-
gously. The proof is by contradiction. Suppose τ ∗ ≤ f (1 − β)CL /2 and in the optimal
solution all high collateral firms have not been given zombie loan (λ∗H < 1). Now sup-
pose if we increase λ∗H by e and reduce λ∗L by η = e CCHL > e, then the total collateral
liquidated remains the same but the number of zombie firms reduces. Since total collat-
eral liquidated is same, the price will remain same and the first two terms in (8) will be
same. But the number of zombie loans will reduce increasing the utility. Thus λ∗ < 1
cannot be true in equilibrium. Finally, we get λ∗L by diving total collateral liquidated by
all the collateral of the low collateral firms which failed.

51
 Table B1: Sensitivity of Debt to Instrumented Real Estate Value
(1) (2) (3) (4) (5)
∆ Long-Term Debt
Real Estate I 0.068*** 0.061*** 0.064*** 0.053*** 0.049***
(0.014) (0.015) (0.015) (0.012) (0.014)

Negative Shock 0.079 0.070 0.073 0.068 0.063

(0.052) (0.054) (0.054) (0.056) (0.058)

Negative Shock × Real Estate I -0.047*** -0.047*** -0.046*** -0.037*** -0.037***

(0.011) (0.012) (0.011) (0.010) (0.011)

Tobin’s Q 0.021** 0.029***

(0.010) (0.011)

KZ Index 0.005 -0.002

(0.010) (0.012)

Age 0.009 0.014*

(0.008) (0.008)

Age2 -0.000 -0.000

(0.000) (0.000)
Observations 30874 27982 27838 28007 25683
R2 0.10 0.12 0.12 0.12 0.14
Year FE Yes Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes Yes
Standard errors in parentheses
* p < 0.10, ** p < 0.05, *** p < 0.01
Table B1 replicates the result from Table 2 using instrumented real estate values. We re-estimate
the sensitivity of debt to collateral value. The sensitivity of debt to real estate holdings is posi-
tive during periods of rising real estate prices indicating that as the value of real estate holding
increases, long term borrowing increases. However, this sensitivity is lower during periods of
falling real estate prices indicating that firms do not reduce their borrowings when the value of
their collateral falls. Local house price elasticities are from Saiz (2010). Real estate value is cal-
culated by estimating average age and historical cost of the real estate holdings of a firm in 1993.
Negative Shock is an indicator for years which saw a fall in the Residential Price Index in a MSA.
Tobin’s Q, KZ-Index, Age and Age2 are used as controls for credit demand and/or supply. Tobin’s
Q is calculated as the ratio of enterprise value of a firm to its book value following Almeida and
Campello (2007). The Kaplan-Zingales Index is calculated based on the 5-factor model of Lamont,
Polk and Saa-Requejo (2001)). Firm age is calculated using the first appearance of a firm in the
COMPUSTAT dataset with a cut-off at 1960. Standard errprs are clustered at the firm-level

52
 Table B2: Probability of Zombie Lending Instrumented
(1) (2) (3) (4)
Zombie
High Real Estate I 0.281 0.312 0.306 0.305
(0.250) (0.262) (0.262) (0.262)

Negative Shock -0.204 -0.196 -0.195 -0.194

(0.137) (0.148) (0.148) (0.148)

High Real Estate I × Negative Shock 0.560∗∗∗ 0.573∗∗∗ 0.556∗∗∗ 0.556∗∗∗

(0.185) (0.194) (0.196) (0.196)

Tobin’s Q -0.003 0.001

(0.008) (0.008)

KZ Index -0.018∗∗ -0.018∗∗

(0.009) (0.009)
Observations 8419 7461 7367 7367
Year FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table B2 replicates the result of Table 4 using instrumented real estate prices. We estimate the probability
of zombie lending from 1994 to 2014 by fitting a logit model on an indicator for zombie firms on firms
with above median real estate holdings and negative real estate shocks. Following Caballero, Hoshi and
Kashyap (2008), a firms is defined to have received a zombie loan if its interest expense is lower than the
highest rated firms in the year and it is rated BB or lower. Tobin’s Q and KZ-Index are used as controls
for credit demand and/or supply. Tobin’s Q is calculated as the ratio of enterprise value of a firm to its
book value following Almeida and Campello (2007). The Kaplan-Zingales Index is calculated based on the
5-factor model of Lamont, Polk and Saa-Requejo (2001). Panel B uses an unbalanced dataset and proxies
for firms with high collateral using firms with above median PPE. Standard errprs are clustered at the
firm-level

53
 Table B3: Alternate Zombie Definitions
(1) (2) (3) (4)
Zombie Zombie 1 Zombie 2 Zombie 3
HighRealEstate -0.380 -0.504∗∗ 0.230 -0.347
(0.270) (0.212) (0.222) (0.232)

Neg Shock -0.242 -0.176 0.032 -0.137

(0.175) (0.153) (0.118) (0.145)

High Real Estate × Neg Shock 0.835∗∗∗ 0.720∗∗∗ 0.341∗ 0.522∗∗

(0.210) (0.268) (0.184) (0.252)

Tobin’s Q -0.002 -0.002 -0.010 -0.954∗∗∗

(0.007) (0.010) (0.008) (0.183)

KZ Index -0.014 -0.013 0.063∗∗∗ 0.026

(0.010) (0.010) (0.010) (0.021)
Observations 8416 8174 14555 8442
Year FE Yes Yes Yes Yes
Firm FE Yes Yes Yes Yes
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table B3 replicates the result of Table 4 using alternate definitions of zombie firms. instrumented real estate
prices. Zombie is the same definition of subsidized lending used in the main text following Caballero et al.
(2008). Zombie 1 is a similar definition of zombie firms which identifies them as firms with excessinterest <
0, 3yrMovAvgICR < 1 and the firm issuing debt in the year. Zombie 2 follows from the definition used
in Banerjee and Hofmann (2018) where firms have 3yrMovAvgICR < 1 and age >= 10. Lastly, Zombie3
is a firm in the bottom tercile of ICR and Tobin0 sQ <= 1. We estimate the probability of zombie lending
from 1994 to 2014 by fitting a logit model on an indicator for zombie firms on firms with above median
real estate holdings and negative real estate shocks. Tobin’s Q and KZ-Index are used as controls for credit
demand and/or supply. Tobin’s Q is calculated as the ratio of enterprise value of a firm to its book value
following Almeida and Campello (2007). The Kaplan-Zingales Index is calculated based on the 5-factor
model of Lamont, Polk and Saa-Requejo (2001). Standard errprs are clustered at the firm-level

54