Uncovering Collateral Constraints

JOSÉ MARIA LIBERTI AND JASON STURGESS∗

April 1, 2013

ABSTRACT

Collateral may be used as an ex-ante commitment against agency risk, or for hedging against

ex-post default risk. Using a panel data of 8,820 small and medium firms in 15 countries

with direct measures of ex-ante agency risk and ex-post realized default, we find that the

commitment motive alone explains collateralization. Going from the lowest to highest quartile

of ex-ante agency risk distribution increases initial collateralization by 14 percentage points,

but the same change in expected default risk leads to no change in collateralization. We

also uncover a collateral “pecking order” driven by commitment concerns. While the bank

is willing to accept firm-specific assets susceptible to agency risk (e.g. plant machinery and

inventory) for low agency risk firms, it prefers non-specific assets (e.g. real estate and cash

and bank guarantees) for firms prone to agency risk.

∗DePaul University, Tilburg University and European Banking Center, and McDonough School of Business,

Georgetown University, respectively. We thank Martin Brown, Jim Booth, Gabriella Bucci, Doug Diamond,

Tim Johnson, Vasso Ioannidou, Atif Mian, Steven Ongena, Daniel Paravisini, Lee Pinkowitz, and Amit Seru for

helpful comments and suggestions

Collateral is one of the most common characteristics of loan contracts. Understanding the

role of collateral is important, not only because of its widespread use, but also because of

its implications for monetary policy. Many influential theories use the presence of collateral

to explain a wide variety of phenomena including financing constraints, business cycles and

poverty traps.1 But why does collateral exist?

There is an extensive theoretical literature examining the use of collateral. These theories

of collateral can be conceptually divided into two categories. The commitment view suggests

that collateral provides a credible mechanism for commitment against agency risk such as

moral hazard and asymmetric information. These models argue that collateral mitigates finan-

cial frictions stemming from moral hazard and adverse selection effects.2 The hedging view

on the other hand suggests that collateral provides a convenient hedge against ex-post default,

even in the absence of agency costs. Theories supporting the hedging view of collateral re-

lieves financing constraints by allowing the lender to reduce expected losses upon default.3

The two views on the role of collateral differ substantially on how one should perceive the

role of collateral. While the commitment view credits collateral with preventing agency risk

altogether, the hedging view treats collateral as a passive instrument used only for transferring

default risk from one economic agent to another.

In this paper, we construct new tests that empirically separate these two views and find

commitment to be the primary motive for collateralization. Our empirical design to separate

commitment from hedging is based on the simple observation that if borrowers use collateral1An incomplete list of such work includes collateral constraints as explanations for business cycles, trans-

mission and amplification of macro shocks (Bernanke and Gertler (1989), Kiyotaki and Moore (1997), Aghion,Banerjee and Piketty (1999)); income inequality (Banerjee and Newman (1993)); and poverty traps (Mookherjeeand Ray (2002)).

2See for example Aghion and Bolton (1992), Barro (1976), Chan and Thakor (1987), Johnson and Stulz(1985), Hart and Moore (1994, 1998), Hart (1995), Holmstrom and Tirole (1997), Park (2000), Rajan and Winton(1995), and Stiglitz and Weiss (1981) plus others.

3The arguments in Dewatripont, Lagos, and Matthews (2003), Holmstrom (1999), Inderst and Mueller (2005),Innes (1990), Jensen and Meckling (1976), Lacker (1992), and Zwiebel (1996) all propose that collateral is usedto hedge default risk.

1

to credibly commit themselves against agency risk, then one should not observe that particular

risk in equilibrium ex-post. Specifically, if commitment explains collateralization then default

should be unrelated to agency risk. Suppose there are two types of risks that a bank faces

from a borrower ex-ante: agency risk and production risk. Agency refers to the usual risk

that borrower may be of bad type or that he might misbehave in the future. Production risk

refers to the natural business risk inherent in all projects. For the sake of simplicity, assume

that ex-ante agency and production risks are uncorrelated with each other. If collateral is used

to commit against agency risk then collateral should be positively correlated with measures

of ex-ante agency risk, but uncorrelated with ex-post realized risk. Conversely if collateral

provides a hedge against realized default, then collateral should be positively correlated with

observed default and uncorrelated with ex-ante agency risk.

The data requirements to identify the role of collateral are quite specific. We need ex-ante

measures of both agency and production risk, as well as measures of ex-post realized risk (i.e.

default). In addition one needs information regarding the size of loan and the value and type

of any collateral used to secure the loan. We explore the role of collateral using a novel cross-

country data set containing 8,820 small and business loans issued by a multinational bank in 15

countries. The 15 countries, which range from India, Turkey, and Chile, to Korea, Malaysia,

and Hong Kong, differ widely in their level of institutional and financial development, which

allows us to further study the effects of the commitment view of collateral across economies

where agency risk may differ.

Following Liberti and Mian (2010), we estimate the collateral cost of financing and es-

timate how it varies with agency and business risk. We estimate the collateral using two

measures. The first is the dollar cost of collateral, that is, the value of collateral demanded

for every dollar lent out. Our second measure of collateral cost is the type of asset pledged

as collateral. For example, a firm that is forced to pledge non-firm-specific assets (e.g., land)

2

is more constrained relative to a firm that can also pledge firm-specific assets (e.g., inventory,

account receivables) as collateral.4 This allows is to uncover both the role of collateral and the

pecking order of collateral with respect to asset type.

To take our methodology to the data, we rely on the availability of both ex-ante and ex-

post risk measures. The ex-ante measure of firm risk is computed by the bank. This measure

is derived from an information template that includes the bank loan officer’s assessment of

production risk (such as profitability, ability, etc), as well as agency risk (such as reliability of

information, management character, etc.), but not collateralization. The ex-post risk measure

is observed default. We first purge from the ex-ante measure of overall firm risk the component

that predicts ex-post realized default. Since the original ex-ante measure takes into account

both preventable agency risk and the expected realized risk (i.e. production risk in the example

earlier), focusing on the component orthogonal to realized ex-post risk allows us to isolate ex-

ante agency risk. We can then use the ex-ante agency and ex-post realized risk measures to

test commitment and hedging view predictions.

We employ a within-country estimate of the collateral cost to identify the role of collateral.

These estimations completely absorbs factors influencing the collateral choice and the levels

of agency and production risk in an economy, as well as the demand or supply of collateral

within each industry. Further, this approach ensures that any results are not simply a result of

institutional factors such as economic development that might affect the level of agency risk.

We also ensure that results are not affected by time-varying factors such as business cycles or

growth opportunities.5

4The recent U.S. credit crisis highlights the severe problems in financing that can arise when lenders no longerfeel comfortable accepting a particular class of assets (in this case, mortgage backed securities) as collateral.

5The paper also performs a number of additional robustness checks for alternative explanations and endo-geneity concerns. We postpone this discussion until section V in the interest of brevity.

3

The paper finds that consistent with the commitment view, initial collateralization is strongly

and positively correlated with ex-ante agency risk but completely uncorrelated with production

risk, as instrumented by default. These results reject the hedging motive for collateralization.

The magnitude of the commitment effect is also quite large. Going from the lowest to highest

quartile of ex-ante agency risk distribution increases the rate of initial collateralization by 14

percentage points when the mean rate of collateralization is 54 percent. However, a similar

shift in production risk is associated with only a 1 percentage point increase in collateraliza-

tion.

Next, we exploit differences in financial development across the 15 countries to provide

further support for the commitment view of collateral. If collateral is used as an ex-ante

commitment mechanism to prevent agency risk, and stronger financial development such as

creditor rights protect lenders from agency costs, then if the commitment view view explains

the use of collateral the collateral spread should be lower in economies with stronger financial

institutions and/or development. We show this to be the case. Going from the lowest to highest

quartile of ex-ante agency risk distribution increases the rate of initial collateralization by as

much as 20% in countries with weak creditor rights, but only 9% in countries with strong

creditor rights. The results confirm that stronger financial institutions lower collateral spreads

by improving financial development in a country, and that the channel through which financial

development lowers collateral spread is by protecting lenders from agency risk.

We also uncover an interesting “pecking order” of collateralized assets that lends further

support to the commitment view that collateral limits agency risk. We find that the bank is

more likely to accept firm-specific assets that are prone to agency concerns from firms with

low agency risk. Examples of agency prone assets include inventory and machinery since

their value is susceptible to bad actions such as stealing or neglect by firm management. On

the other hand the bank only accepts non-specific assets not susceptible to agency concerns

4

from firms with high ex-ante agency risk. Non-specific assets include land and real estate,

cash, and bank guarantees which are difficult to hide or abscond with, and have valuations

less susceptible to management neglect. We also reveal a middle-ground of assets such as

promissory notes, import and export letters of credit, financial securities that are used to a

lesser extent as collateral but nonetheless provide some protection from agency concerns.

Our work is the first to empirically separate commitment and hedging motives of collat-

eralization. A number of theoretical papers have highlighted the role of collateral as com-

mitment against agency risk. Our empirical results are consistent with papers such as Barro

(1976), Stiglitz and Weiss (1981), and Chan and Thakor (1987) which argue that the threat

of agency risk in the form of unobserved borrower attribute or action leads to greater use of

collateral as a commitment device.6 Although papers such as Bester (1985) make the opposite

prediction by suggesting that low risk firms might sort on high collateral - low interest rate

contracts, our empirical results do not favor such explanations.

Our results are also in line with the standard principal-agent result that it is inefficient for

principal (banks) to try and transfer realized risk to the agent (borrower). However, a priori

there might have been other reasons for banks to transfer at least some of the realized risk to

borrowers through collateral. These include risk-shifting as a mechanism for inducing greater

effort by the borrower (see e.g. Jensen and Meckling (1976), Innes (1990), and Dewatripont,

Legros, and Matthews (2003)). Another reason could be a conscious effort by loan officers to

be excessively risk-averse due to intra-organizational agency and career concerns (e.g. Holm-

strom (1999) and Zwiebel (1996)).

This paper is closely related to recent empirical work that highlights the role of collat-

eral in financial contracts. Berger and Udell (1995) and Jimenez, Salas, and Saurina (2006)

6Rajan and Winton (1995), and Park (2000) suggest that collateral may also be used as commitment by thelender to provide monitoring effort. However, these models are written in the context of institutional loans withother public bond-holders, an environment different from the one our firms belong to.

5

show that creditors require firms with poor repayment histories or firms with greater default

risk to secure their loans with collateral. Benmelech and Bergman (2009) construct industry-

specific measures of redeployability and show that more redeployable collateral leads to lower

credit spreads, higher credit ratings, and higher loan-to-value ratios. Benmelech and Bergman

(2011) propose that a firms bankruptcy reduces collateral values of other industry participants,

thereby increasing the cost of external debt finance industry wide. Similar to our results,

pledging more liquid non-specific collateral eases financing constraints. Chaney, Sraer and

Thesmar (2011) show that investment is sensitive to collateral value by examining US real

estate pledged as collateral. Finally, perhaps the closest work to ours are Berger, Frame and

Ioannidou (2011, 2012), who examine a credit registry to test theories of collateral. Their

results conclude that collateral is explained by commitment against moral hazard such as risk

shifting and/or hedging default (loss mitigation). While they too examine the type of asset

pledged as collateral, they are unable to observe the collateralization rate across asset type and

therefore are are unable to provide evidence on the collateral pecking order.

The rest of the paper proceeds as follows. Section I describes the data of our paper. Sec-

tion II analyzes the theoretical framework and discusses the identification strategy for the

empirical tests. Section III examines the dual role of collateral and tests the commitment and

hedging views of collateral. Section IV uncovers a pecking order for collateral requirements

and provides a calibration exercise in terms of real economic measures. Section V discusses

identification concerns. Section VI concludes raising some important questions for future

research suggested by our findings.

I. Data Description

Our data comes from the small and medium-sized lending division of a large multinational

bank that operates in 15 emerging market economies. The data contain every loan issued by

6

the bank and follow each loan over a 2-year period (on average) from 2002 to 2004, with

information updated every six months. Although the original data set has 12,591 firms we

are left with a cross-sectional sample of 8,820 firms after applying several screening rules.

First, we drop 766 firms that are already in default at the beginning of our sample period.

These firms are not actively borrowing during our sample period, and as such we do not know

their ex ante risk assessment, nor the initial level of collateralization demanded by the bank.

Second, another 1,599 firms are excluded as they are missing the ex ante firm risk variable,

and without this variable we cannot calculate collateral spreads. Finally, 1,406 firms do not

draw any loan from the bank during our sample period and hence are dropped because there

is no collateral information on these firms.7

The range of countries in our final sample of 8,820 firms is diverse in terms of geograph-

ical location, financial development, and per capita income (Table I). The number of loans

is not uniform across countries, varying from 1,440 in the Czech Republic to 96 in Pakistan.

This potentially raises the concern that our results might be driven by one or two countries

with a large number of observations. Accordingly, we carefully test for this in the analysis

section below. There are a total of 87 (finely defined) industries in our sample. The full list of

industries, and the number of firms belonging to each industry, is reported in the Appendix.

For every loan we observe the borrowers identity, industry, and country. We also observe

the total approved loan, loan outstanding, loan default status, the firms size and risk as deter-

mined by the bank, and both the type and liquidation value of the collateral used to secure

the loan. We use the first observation for each loan in our sample to represent the initial loan

characteristics at the time of origination. We then determine for each loan its end-of-sample

period default status. This variable is one if a firm goes into default by the end of the sample

period (i.e., within 2 years), and zero otherwise. Additionally, we determine for each loan7The bank has approved some loan amount for these firms, but as these firms chose not to withdraw against

the approved amount, they are not required to put up any collateral.

7

its end-of-sample period risk status, which captures whether the loan has been downgraded.

This variable is one if a firm’s risk classification worsens by the end of the sample period (i.e.,

within 2 years), and zero otherwise. Table II provides summary statistics for all the variables

in our data set. Because our empirical methodology uses country and country-industry fixed

effects, we report country and country-industry demeaned standard deviations as well.

A key variable in our analysis is the ex-ante risk grade of a borrower. The grade varies

from “A” (best) to “D” (worst) and represents the riskiness of the borrower at the time of loan

origination as determined by the banks loan officer. Additionally, we transform the risk grade

into a numerical variable by assigning “A”=1, “B”=2, “C”=3, and “D”=4. The risk grade is

based upon two sets of information, which take into account measures of production as well as

agency risk. The first includes objective measures of firm performance related to production

risk based on firm and industry fundamentals such as profitability, sales growth, and past credit

history. The second set includes subjective measures of firm performance related to agency

risk such as assessment of the quality and reliance of information, management interviews,

and site visits.8 The firm risk grade is an ex ante assessment of the firm, before any decision

is made about how much to lend to the firm and on what terms. Thus, risk grade does not

include information on ultimate loan terms such as collateral, interest rates, and maturity.

This is important because otherwise firms with a high level of collateral may be given a safe

grade due to the collateral, and not because the firms cash flows are less risky. Table II shows

that all four grades are fairly well represented in the data and that there is significant variation

in grades not only across countries but also within country and country-industry categories.

The bank also constructs a variable on firm size using firm sales. Specifically, the bank

categorizes firms into four sales size groups, where a grade of “0” corresponds to smaller8For example, before coming up with the final ex ante risk grade for a firm, a loan officer responds to questions

such as: How reliable is the information provided by the management? Does the firm have good governancemechanisms? Does the firm have professional management? and other questions related to management and firmperformance that are subjective in nature.

8

firms and a grade of “3” corresponds to larger firms. We find that firms in our sample are

skewed towards smaller-sized firms, which is consistent with the focus of the lending program.

Further, there is significant variation in firm size, not only across countries, but also within

country and country-industry. Our data also includes information on loans and risk. The mean

outstanding loan amount is $386,650, and 5.17% of the firms enter into default and 6.61% of

firms experience a risk downgrade by the end of our sample period.

An important dimension of our data is its information on loan collateralization. For each

loan, the bank records the liquidation value of collateral pledged for the loan. This reflects the

banks assessment of the market value of the collateral in the event of bankruptcy, assuming the

lender receives full ownership of the collateral. We divide the liquidation value of collateral

(in the beginning of the sample period) by the approved loan amount to construct the collater-

alization rate for a loan. The average collateralization rate is 54% with a standard deviation of

45%.

In addition to the value of collateral, our data also include the type of asset pledged as col-

lateral. Asset types correspond to one of eleven categories: (i) firm-specific collateral which

captures collateral that is specific to the operational business of the firm under consideration9,

(ii) industry-specific collateral, such as inventory and machinery, that is specific to the opera-

tional business of the firm’s industry, (iii) assets, which includes equipment that is non-specific

to the industry such as vehicles, (iv) accounts receivable including receivables, contract orders,

and post-dated checks, (v) promissory notes, (vi) import and export letters of credit, which are

used as a method to facilitate payment of international trade transactions, (vii) leases such

as asset-backed-financing, (viii) guarantees including bank guarantees and standby letters of

9We discussed with loan officers the reason to categorize this collateral as firm-specifc. Loan officers indicatesthat this category captures collateral that does not merit classification in any of the other categories but is specificto the operational business of the firm under consideration.

9

credit, (ix) financial securities such as bonds and shares, (x) cash, both foreign and domestic,

and (xi) real estate, including land and building.

Table II shows the composition of collateral by summarizing the percentage of collateral

value that belongs to each of the seven collateral categories. Land and real estate are the most

common types of collateral, followed closely by firm-specific assets, (non-specific) assets,

accounts receivable, and cash. The type of collateral varies significantly in its “specificity”

to the firms operation and performance. For example, whereas firm specific assets, and to a

lesser degree industry-specific assets, are highly specific to the state of a firm, real estate, bank

guarantees, and cash are not.

We want to emphasize that country bank managers are free to lend to whoever they want

and have complete discretion in terms of the value and type of collateral they want to demand

from each borrower. Headquarters approves and allocates overall lending limits for each coun-

try but each country-based lending division has complete discretion to implement, execute

and monitor the lending program locally. Pricing, credit process, procedures and delinquency

management are managed by each division at the local-country level, although lending terms

and requirements are standardized and consisted across countries. Product offerings are sim-

ilar across countries.10 The central objective given to each country manager is to maximize

the return on lending assets while minimizing defaults. Thus, none of our findings on the

relationship between collateralization rates and firm risk are hard wired by bank rules.

10The rationale of these programs is to offer a small and mid-market segment of borrowers almost all creditand non-credit products available to large corporate borrowers. The strategy of these programs is designed toachieve a robust and consistent growth in the small and medium-sized borrower segment targeting well-managedcompanies with typically entrepreneurial management style, growth prospects and leveraging cross-selling op-portunities. The attractiveness of this segment lies in its large and well-established base, which is typically theengine of growth for developing economies, revenue characteristics, high cross-sell opportunities, and capacityfor self-funding through marketing of liability products. Generally, this segment, although competitively banked,provides an untapped market for traditional products offered to large-sized borrowers.

10

One downside of the cross-country data set described above is that it does not have infor-

mation on firm financials or loan interest rates. However, we were able to gather firm financial

and loan interest rate data from the same lending program for Argentina for 587 firms from

1995 to 2001.11 Although our primary cross-country data set comes from the central computer

archives of the bank, this second database is hand-collected from credit dossiers in Argentina.

The hand-collected data include information on a firm’s ex-ante risk grade, annual balance

sheet, income statement, and interest rates. However, the credit files made available to us did

not contain information on collateralization. We therefore use this second data set not for

computing collateral spreads, but for estimating how other firm attributes such as interest rate,

profitability, and supply of collateralizable assets vary with firm risk.

II. Empirical Methodology

A. Conceptual Framework

We present a simple model of lending to illustrate the dual role of collateral in lending con-

tracts. Collateral is used to commit borrowers against agency risk and/or hedge against pro-

duction risk. Since banks hold debt claims on their borrowers’ assets, default is their primary

measure of risk. There are two fundamental sources of potential default risk facing a bank.

We define one as ex-ante preventable agency risk, and the other as ex-post production risk.

Agency risk may take the form of information asymmetry and/or moral hazard. However,

the assumption is that the bank and lender can write some contract ex-ante that mitigates or

prevents this agency risk. On the other hand, production risk cannot be prevented ex-ante, but

instead the loan contract describes how the bank protects itself from this risk ex-post.

11The number of firms in the pre-2000 sample from Argentina is much larger than the number of firms in ourprimary sample (601 versus 120) because the Argentine crisis of 2000 to 2001 forced many firms out of business.

11

Consider an economy with a continuum of firms each wanting to invest $1 by borrowing

from a bank. The loan contract between a bank and firm takes the form of an interest rate

r > 1 and a collateral amount w < 1 pledged by the firm. Firms vary by their risk attributes

denoted by the pair (α, β), where α and β are both between 0 and 1. β captures the production

risk inherent in a firm’s technology. The firm can produce Y > 1 with probability (1 − β)

and 0 otherwise. α on the other hand captures the degree of agency risk inherent in a firm.

If a firm produces an output Y, it can choose to repay the promised interest rate r < Y to

the bank or declare default strategically. In case of default, the firm looses part of its future

productivity due to a loss of reputation in financial markets given by (1 − α)Y . It also loses

its initial collateral worth w < 1 in case of default. Hence the firm chooses not to default only

if the following IC condition holds:

Y − r ≥ αY − w (1)

Our measure of agency risk, α, in (1) effectively represents the fraction of firm assets that

the borrower can abscond with. Figure I summarizes the above set up.

In the above framework, a bank can use collateral w either to receive commitment against

agency risk (α), or to hedge against realized risk (β). Papers such as Barro (1976), Stiglitz

and Weiss (1981), and Chan and Thakor (1987) argue that in the face of agency risk of the sort

captured by α, a bank may impose collateral requirements to minimize risk. This can be seen

from (1) where collateral worth w = r − (1− α)Y guarantees no agency risk in equilibrium.

A second potential use of collateral comes as a hedge against production risk. It is common

among practitioners to think of collateral as a hedge against actual default by a firm. Theory

also provides a number of rationales for why a banker may want to do so. For example, work

by Jensen and Meckling (1976), Innes (1990), and Dewatripont, Legros, and Matthews (2003)

12

suggests that banks may want to transfer more of the risk towards the firm for incentivizing

managers. Similarly, organizational literature such as Holmstrom (1999) and Zwiebel (1996)

suggests that loan officers within a bank hierarchy might be excessively risk averse due to

intra-firm agency issues and career concerns. Consequently the higher β is, the more attractive

collateralization appears to the bank under the hedging hypothesis.

The commitment and hedging views differ dramatically in their perception of collateral.

The commitment view describes collateral as an effective tool for minimizing agency risk

in the economy. If the commitment view of collateral is at play then agency risk should be

preventable (through collateral) and thus realized default must be a result of production risk

only. The hedging view treats collateral as a passive tool used only for sharing existing risk

across agents.

Combining the commitment and hedging view implies that collateral can be used both to

prevent agency risk and risk-share upon default, and further, that default should be a result of

production risk only if collateral is used as commitment against agency risk. These predictions

are summarized in the proposition below.

Proposition 1 Under the pure commitment view, collateralization is uncorrelated with pro-

duction risk β, but positively correlated with agency risk α (i.e. w = r − (1 − α)Y ). Under

the pure hedging view, collateralization is positively correlated with production risk β, but

uncorrelated with ex-ante agency risk α. Combining the commitment and hedging views, col-

lateralization ex-ante commits against agency risk, and therefore realized default is a result of

production risk only.

13

B. Regression Specification

To investigate the dual role of collateral in lending contracts, one needs to be able to separate

agency risk (α) from production risk (β). Unfortunately, risk classifications, such as ours,

generally contain both agency risk and expected production risk, which combined might ex-

plain predicted default. However, Proposition 1 implies that in the presence of both agency

and production risk, default will be unrelated to agency risk and that production risk alone will

predict default.

Therefore, we start by assuming that both the commitment and hedging views of collateral

are at play and that default is explained as a function of production risk only. Next, we can

decompose the ex-ante risk grade into its components that reflect agency risk and production

risk by purging risk grade of the component that predicts future default. In particular, let R0i

be the initial ex-ante risk grade and ZTi be the final ex-post realized default for firm i in our

sample. Then by projecting R0i on ZTi ,

R0i = α + β1ZTi + εi (2)

one can separate R0i into the component R̂0i that predicts Z

Ti , and the orthogonal residual

component RZ0i that contains only ex-ante agency risk information. RZ0i thus becomes our

firm-level measure of agency risk, and R̂0i the expected production risk measure. Proposition

1 can then be tested by running the initial rate of collateralization Y 0i on agency and production

risk measures:

Y 0i = αcj + β1RZ0i + β2R̂

0i + εi (3)

14

According to proposition 1, if only the commitment view of collateral is correct, then

β1 > 0 and β2 = 0. Alternatively, β1 = 0 and β2 > 0 if only the hedging view is correct.

In (3), β̂ is an unbiased estimate of β only if the error term is uncorrelated with our

measures of agency and production risk. The concern, however, is that country-specific, or

country-industry-specific, factors, may be spuriously correlated with expected firm risk. For

example, the average level of collateralization in a country or country-industry may depend

on macro factors (such as the industry mix of investments), and these factors may in turn be

correlated with the average agency or production risk as well. In such circumstances, β will be

biased. Similarly, the measurement of ex ante loan risk may not be comparable across coun-

tries. For example, a risk grade of A in one country may not be comparable to a grade of A

in another. We address the concern of country-specific spurious factors by including country

(c) - industry(j) fixed effects in equation (3). The country-industry fixed effects account for

aggregate changes in the demand or supply of collateral within each country-industry. Further,

because the cross-sectional data are constructed around the same time period for all countries,

country-industry fixed effects also absorb any contemporaneous or expected shocks hitting

various economies. Thus, our coefficient of interest is not affected by time-varying factors

such as business cycles or growth opportunities.

Similarly, in equation (2) we are concerned that country-specific, or country-industry-

specific, factors, may be spuriously correlated with default. Once again, we address this con-

cern by also including country fixed effects in equation (2) to ensure that comparisons are

made within a country-industry, and average differences in default risk across countries due

to macro factors, as well as differences in grading schemes across countries, are factored out.

However, this introduces a further concern. Our estimation of agency and production risk in

equation (2) relies on using firm-level default as an instrument that is related to production

risk but not related to agency risk. However, it is likely that both agency and production risk

15

vary with country-industry and therefore the inclusion of country-industry fixed effects results

in our estimation of production risk R̂0i being biased by agency risk. However, so long as we

(i) include the same fixed effects and explanatory variables in equations (2) and (3); and (ii)

include estimates of both agency and production risk in equations (3) the coefficients β1 and

β2 should remain unbiased.

Once we have estimated the effects of agency and production risk on collateralization

we test for effects of total, agency and production risk on the level of collateral type (the

pecking order of collateral), and also for how the effects of agency and production risk on

collateralization and the pecking order of collateral vary with financial development across

countries.

III. The Dual Role of Collateral

A. Separating Agency Risk and Production Risk

Table III estimates equation (3) using collateralization rate as the dependent variable. How-

ever, instead of using agency and business risk on the right-hand side, we first use the banks

total risk assessment of a loan applicant, which reproduces the results in Liberti and Mian

(2010). The purpose is to show the total correlation between collateralization and ex-ante

subjective risk assessment. The risk assessment varies from A to D, with A being the omit-

ted category in the estimations. Coefficients on other grade dummies therefore represent the

average difference from grade A firms within a given country-industry. Throughout, all esti-

mations include 782 country-industry fixed effects and we present standard errors computed

after allowing for correlation across observations in a given country-industry. This is our

default methodology throughout the paper.

16

Column (1) shows a positive collateral spread on average as collateralization increases

with firm risk. The largest increase in collateralization occurs for firms with the worst risk

assessment (23% of firms). The rate of collateralization is 11.05% points higher for grade

D firms compared to grade A firms. This jump is all the more striking given that the mean

collateralization rate is already 54%.

The results in column (1) are strengthened once we add sales-size and loan-size size con-

trols in columns (2) and (3). Size controls include sales size indicators and approved loan

amount decile fixed effects. The approved loan amount decile corresponds to the decile that a

loan falls into in the approved amount distribution

In column (4) we replace the loan-size fixed effects with a continuous measure of loan size.

As Liberti and Mian (2010) point out, all else equal, the bank demands greater collateralization

for larger loans, possibly reflecting the increased moral hazard concerns with greater leverage.

The results in columns (3) and (4) reveal that the rate of collateralization is 5.4% points higher

for grade C firms, and 14.8% points higher for grade D firms, compared to grade A firms,

when comparing across firms that belong to the same industry in the same country and also

controlling for differences in firm- and loan-size.

The strengthening of results as we include more controls is related to the issue of whether

the coefficient in risk grade is an underestimate of the true elasticity of demand coefficient.

We will discuss in detail this concern in Section V. For example, columns (3) and (4) suggests

that riskier firms have smaller outstanding loans, and since all else equal smaller loans require

lower collateralization, including size controls increases the coefficient on risk grade. In par-

ticular, the results in column (4) suggest that banks are risk averse and try to limit the exposed

risk by demanding more collateralization as the level of exposure to a single party increases,

even controlling for risk. This evidence also suggests that the marginal cost of collateralization

increases with loan size, leading to convex cost of borrowing for firms.

17

Although collateral spread is robust to controls such as country-industry, and size fixed

effects, there may be a concern that the estimate is primarily driven by one or two countries.

Table I shows that the distribution of loans across countries is highly skewed, with countries

such as the Czech Republic having 1,440 loans whereas others such as Pakistan have only 96.

The regressions in Table III weigh each loan equally, in effect giving a lot more importance

to the Czech Republic relative to Pakistan. We test whether the estimated collateral spread is

primarily driven by a couple of countries by giving each country equal weight in the regression

regardless of the number of loans from that country. To do so, we estimate the country-specific

collateral spread for grade “D” borrowers in equation (3). We only present the country spe-

cific coefficient for grade “D” firms because results are driven by this coefficient. We then use

this country-specific collateral spread as the dependent variable in column (5), which is run

at the country level. The equal country-weighted collateral spread for grade “D” borrowers

is almost identical to earlier estimates, and significant at the 5% level. Additionally, Figure 2

plots the collateral spread for grade “D” borrowers separately for each country. Countries in

Figure 2 are ordered by their number of firms in sample. In addition to the estimated demand

elasticity coefficient we also present the 95% confidence interval around it. Although the

country specific estimates are sometimes imprecise because of the small number of observa-

tions, generally speaking the estimated demand coefficients are not drastically different from

the average demand coefficient in column (4) of Table III. For example, 11 of the 15 country

estimates either include or are significantly greater than the average elasticity of demand es-

timate, represented by the horizontal line, within their 95% confidence. The four remaining

countries include Sri Lanka, for which we have only 102 borrowers and South Africa, which

has only 0.6% of firms classified as grade “D”.

In Table IV we present estimation results of equation (3) using collateralization rate as the

dependent variable and further decompose ex-ante risk assessment into agency and production

18

risk, as described in Section II.B. In panel B of Table IV we present estimates from equation

(2) and in panel A we present estimates of equation (3), where we use results from panel B to

estimate agency and production risk. Column 1 first repeats the estimation of collateralization

on total risk presented in column (4) of Table III, but instead uses the parametric estimation of

risk grade. The parametric estimation of risk grade is used in the decomposition of risk in panel

B. Comparing the results in column (1) with those in column (4) of Table III reveals that the

non-parametric and parametric estimations results are similar both in terms of coefficient on

risk grade and also explanatory power. In the remainder of our results we employ a parametric

estimation of risk grade.

Columns (2) and (3) examine how agency and production risk affect collateralization. We

first use observed default to instrument for production risk in equation (2), and then include the

instrumented production risk and estimated agency risk in equation (3). In column (2) we omit

country-industry fixed effects from the estimation of equation (2), while we include them in

column (3). The results in columns (2) and (3) reveal that the increase in collateralization with

risk grade is solely explained by agency risk. The coefficients on production risk and agency

risk describe the collateral spread with respect to an increase in each of these risks. The

results in column (2) reveal that an increase in risk grade from “A” to “D” (an increase in the

parametric risk grade of 3.0 points) associated with agency risk increases collateralization by

14.25%, and increase in risk grade from “A” to “D” associated with production risk increases

collateralization by 0.78%. The results also imply that collateralization rate is completely

uncorrelated with the ex-post default rate measure. This suggests that, consistent with the

commitment view of collateral, collateralization is employed to mitigate agency risk, but not

to hedge against expected production risk.

Columns (4) and (5) presents estimates that employ our risk downgrade variable, risk

grade decrease, instead of default in the first stage instrumentation of production risk. A valid

19

concern is that default captures only the handful of borrowers that ultimately default, and

therefore may underestimate production risk. To address this concern we use an indicator

variable equal to one when a borrower experienced a risk downgrade over the period of the

sample, and zero otherwise. The indicator variable, risk grade decrease, captures events when

the borrower’s risk classification was downgraded as well as default events. The results are

very similar to those presented in columns (2) and (3).

Once again, although the results on how agency and production risk affects collateral-

ization is robust to controls such as country industry, and size fixed effects, there may be a

concern that the estimate is primarily driven by one or two countries. The regressions in Ta-

ble IV weigh each loan equally, in effect giving a lot more importance to countries with a

larger sample size such as the Czech Republic and Korea. This would be a concern, for ex-

ample, if collateral was used to hedge production risk only in countries with a smaller sample

size. We test whether the estimated demand elasticity coefficients for agency and production

risk are primarily driven by a couple of countries by giving each country equal weight in the

regression regardless of the number of loans from that country. To do so, we estimate the

country-specific collateral spread for each of production and agency risk. We then use these

country-specific collateral spreads as the dependent variable in columns (5) and (6), which are

run at the country level. The equal country-weighted collateral spread results for production

and agency are almost identical to earlier estimates: collateralization appears to entirely driven

by commitment against agency risk.

B. Collateral Effects of Financial Development on Agency Risk and Production Risk

Table IV reveals that collateral is used as an ex-ante commitment mechanism to prevent agency

risk. We now examine how the collateral spread with respect to agency risk varies with finan-

20

cial development. In more financially developed economies, Liberti and Mian (2010) docu-

ment that stronger financial development protects lenders from agency costs because lenders

can employ contracts such as loan covenants to restrict borrowers from risk-shifting. There-

fore we should expect that the positive collateral spread on agency risk shown in Table IV is

lower in economies with stronger financial institutions and/or development.

Table V estimates equation (4) to test how collateral spread varies with financial develop-

ment.

Y 0i = αcj + β1RZ0i + β2R̂

0i + β3R

Z0i ∗ Fc + β4R̂0i ∗ Fc + εi (4)

where Fc is financial development measured as the country-level creditor rights index.

The creditor rights index measures the ease with which creditors secure assets in the event

of bankruptcy, and is taken from Djankov, McLiesh, and Shleifer (2007).12 The index is

the sum of four variables that capture the relative power of secured creditors in the event of

bankruptcy: (i) the requirement of creditor consent when a debtor files for reorganization, (ii)

the ability of a creditor to seize collateral once petition for reorganization is approved, (iii)

whether secured creditors are paid first under liquidation, and (iv) whether an administrator,

and not management, is responsible for running the business during the reorganization. A

value of one is added to the index for each of these creditors protections afforded under a

countrys laws and regulations. Thus a score of 0 suggests very poor creditor rights whereas 4

suggests strong creditor rights. We classify countries as having strong creditor rights (a score

≥ 3) or weak creditor rights (a score ≤ 2), and our creditor rights variable is an indicator

variable equal to one in the countries with high creditor rights. We use the creditor rights

index for 2003 reported in the DMS data set. Given the very high level of persistence in

12The creditor rights index is downloaded from the DMS data at www.andrei-shleifer.com.

21

creditor rights for a country over time, our results do not change if we use the average creditor

rights index over a different time period.13

The results in columns (1) - (4) of Table V estimate equation (4) using the same decom-

position of agency and production risk presented in columns (2) - (5) of Table IV. The results

show that the collateral spread for agency risk is much smaller in economies with stronger

creditor rights. Because all regressions include country fixed effects, there is no need to in-

clude the level of country-specific variables such as the creditor rights index. The results in

column (1) reveal that an increase in risk grade from “A” to ”D” (an increase in the parametric

risk grade of 3.0 points) associated with agency risk increases collateralization by as much as

20.13% in countries with weak creditor rights, but only 8.79% in countries with strong credi-

tor rights. This reflects a drop of 56% in collateral spread associated with agency risk moving

from low to high creditor rights economies. We find no such result for production risk and

collateralization rates. The results confirm that better institutions lower collateral spreads by

improving financial development in a country, and that the channel through which financial

development lowers collateral spread is by protecting lenders from agency risk.

IV. The Collateral Pecking Order

The previous section shows that the value of collateral per dollar lent increases with borrower

agency risk. However, in addition banks may restrict their preferences in terms of the type

of marginal asset accepted as collateral as borrower agency risk increases. For example, a

bank might accept firm-specific assets such as inventory that are more susceptible to concerns

regarding a borrowers agency risk for low risk firms, but as agency risk increases the bank may

demand non-specific assets such as cash, land and real estate, and guarantees as collateral. A

13Our results are robust to using alternate measures of financial development such as the ratio of private creditto GDP.

22

key feature of our data set is that it permits us to look at how the composition of collateral

varies with firm risk.

We examine the collateral pecking order by studying the composition of eleven categories

of collateral demanded by banks. The asset types correspond to one of eleven categories: (i)

firm-specific collateral which captures collateral that is specific to the operational business of

the firm under consideration, (ii) industry-specific collateral, such as inventory and machinery,

that is specific to the operational business of the firm’s industry, (iii) assets, which includes

equipment that is non-specific to the industry such as vehicles, (iv) accounts receivable in-

cluding receivables, contract orders, and post-dated checks, (v) promissory notes, (vi) import

and export letters of credit, which are used as a method to facilitate payment of international

trade transactions, (vii) leases such as asset-backed-financing, (viii) guarantees including bank

guarantees and standby letters of credit, (ix) financial securities such as bonds and shares, (x)

cash, both foreign and domestic, and (xi) real estate, including land and building.

Given the results on the commitment view of collateral, a natural pecking order one might

expect to observe is that banks will accept firm or industry specific assets prone to agency

concerns (e.g. inventory, machinery and account receivables) from borrowers with low agency

risk, and demand non-specific assets from borrowers with high agency risk. These type of

non-specific assets include cash, land and real estate mortgage, and guarantees and liquid

securities. For example, collateral that is mobile and hence not perfectly secured (such as

inventory and firm equipment) may be stolen by managers in bad states of the world, while,

on the contrary, it is more difficult to abscond with land and bank guarantees. One rationale

for such a pecking order could be that firm specific assets are subject to suffer from moral

hazard and hence largely dependent on the quality of firm. Similarly the value of firm-specific

assets might fall faster for lower quality firms. Non-specific assets do not share these concerns

as their value is independent of firm performance and future behavior.

23

It is important to note that, since we estimate collateralization rates using the liquidation

value of the asset, we are not simply studying how different types of asset might have different

liquidation values. If, instead, we were using the market value then any results might be

driven by differences in liquidation value, as described by Schleifer and Vishny (1992) and

documented by Acharya, Bharath and Srinivasan (2007) plus others. Since our estimations

already control for liquidation value effects we can be bolder in suggesting that the pecking

order is driven by agency concerns as well.

We uncover the pecking order by estimating the collateralization rate by asset type. This

allows us to estimate the collateral spread by asset type. Importantly, this methodology esti-

mates the collateral spread by asset type within country-industry which reveals the elasticity

of collateral to risk while controlling for aggregate factors that might otherwise explain the

supply and/or demand of collateral. For example, Table II shows that, unconditional on bor-

rower risk and not controlling for differences across countries that might explain the supply

and demand of collateral, land and real estate, firm-specific assets, equipment, and cash are

most likely to be employed as collateral. But such preliminary analysis is unlikely to uncover

the true collateral pecking order.

Table VI examines how total borrower risk effects the collateral pecking order. For com-

pleteness, in panel A we present results using the non-parametric risk grade and in panel B

we use the parametric risk grade. The results are almost identical. In column (1) we present

the collateral spread results for total risk, which was presented in column (4) of Table III.

In columns (2) - (12) we estimation equation (2) by asset type. The results in columns (1)

through (5) of Table VIII show an interesting and systematic pattern. As firm quality deterio-

rates, the bank is less likely to accept firm- and industry-specific collateral. On the other hand

with deteriorating firm quality, the bank is more likely to demand non-specific collateral types

such as land and real estate, cash, and bank guarantees. In addition, we find that accounts re-

24

ceivable are demanded as collateral more as firm quality deteriorates. This might be explained

by the fact that although accounts receivable tend to be firm-specific, they are also provided

by a third party in the form of receivables and post-dated checks that may alleviate agency

concerns. Additionally, the results in Table VI reveal that loan size plays a role in the type of

collateral pledged. In Table III we showed that larger loans resulted in higher collateralization

rates. Further, larger loans tend to use non-specific assets as collateral.

The collateral pecking order can be more explicitly seen when we plot the marginal impact

of each one grade downgrade on the probability of an asset class being used as collateral.

Figure 3 shows that for initial grade downgrades (i.e. from “A” to “B”, and “B” to “C”),

the bank reduces the percentage of firm-specific collateral allowed. However, the demand for

industry-specific assets is largely unchanged by grade downgrades from “A” to “B”, and “B”

to “C”, but the bank demands less industry-specific assets for downgrades from “C” to “D”.

Further, as the grade deteriorates from “C” to “D”, the bank only accepts non-specific and

hard types of collateral at the margin such as land and real estate, cash, guarantees and to

a lesser extent third party assets such as financial securities, promissory notes and accounts

receivables. Figure 4 presents the cumulative effect of the bank’s pecking order by aggregating

the marginal effects in Figure 3. Thus we get an interesting “collateral pecking order” in terms

of which assets the bank is willing to accept as collateral. While various forms of collateral

are acceptable for the very best firms, as firm risk increases, the bank stops accepting certain

forms of specific collateral. For very high risk firms, the bank only accepts hard and non-

specific forms of collateral such as cash and land. In summary, there is a sharp shift in the

composition of collateral towards non-specific assets - and away from specific assets - as firm

risk increases.

A valid concern is that collateral type is endogenous to borrower risk. Perhaps worse

quality firms have less supply of firm- or industry-specific assets, which would explain the

25

finding that these types of assets are used less by the riskiest borrowers. However, it is worth

emphasizing that the country-industry fixed effects rule out that results are driven by certain

industries exhibiting higher borrower risk and a lower supply of firm- or industry-specific

assets. We address this issue directly in Section V.

Table VII tests whether the collateral pecking is explained by production or agency risk.

The results in Table IV show that the value of collateral per dollar lent varies is explained by

the commitment view of collateral. However, the commitment view might also explain the

collateral pecking order. For example, the asset types most likely to be used as collateral by

the riskiest of firms in Table VI are also the assets least susceptible to agency concerns. We

estimate equation (3) by asset type and present the estimation results in columns (2) - (11). We

find strong support for the commitment view of collateral. The demand for non-specific assets,

which offer lenders better protection from agency risk and borrowers stronger commitment

against agency risk, increases as borrower agency risk increases. However, specific assets that

offer little protection against borrower agency risk are used less when agency risk is high.

The collateral pecking order also reveals some interesting results on the production risk. Our

earlier results concluded that banks did not use collateral to hedge production risk. However,

the pecking order results reveal that production risk does play some role in the composition

of collateral. Land and real estate is demanded when both production and agency risk is high.

Conversely, firm-specific collateral is used only when both production and agency risk is low.

Finally, assets and equipment that is non-specific such as vehicles are employed as collateral

when production risk is high.

Finally we examine how financial development affects the pecking order of collateral.

If the pecking order commits against agency risk and financial development better protects

lenders from agency concerns, one should expect the results in Table VII to be strongest for

borrowers in economies with weak creditor rights. On the other hand the pecking order should

26

be less pronounced in economies with strong creditor rights. Turning to Table VIII we find

this to be the case. The stronger demand for specific assets as collateral when faced with high

agency costs is dampened by strong creditor rights. For example the demand elasticity for

cash is 3.13% per grade of agency risk for weak creditor rights economies but only 0.39%

(and insignificant) for strong creditor rights economies. Interestingly, strong creditor rights

also opens the door for firms to use a wider range of asset type as collateral. The negative

and significant coefficient on CreditorRights x AgencyRisk for industry-specific collateral

implies that banks reduce demand for this specific asset from borrowers with high agency risk

in economies with weak creditor rights, but continue to accept this type of collateral from

borrowers of similar riskiness in economies with strong creditor rights. There is, however, a

notable exception. The demand for land and real estate as commitment against agency risk

is high in all economies. Combined with the results for land and real estate in Table VII this

suggests that land and real estate might be the most sought collateral by lenders.

It is worth reiterating the new findings in Table VIII. We already know from Table IV that

collateral commits against agency risk and that the collateral spread with respect to agency

risk declines with financial development. Therefore, if the coefficient on CreditorRights x

AgencyRisk were negative for all asset types, this would not be a big surprise - all that it

would have meant is that as agency risk and thus collateral spread decreases in financially

developed economies, both specific and non-specific types of collateral are equally likely to

be reduced. However, the coefficients on CreditorRights x AgencyRisk in columns (2) -

(11) of Table VIII paint a different picture. The fact that the the coefficient tends to be neg-

ative for non-specific assets (cash, guarantees), but positive for some specific assets reveals

that not only does financial development mitigate concerns related to agency, as implied by

the lower collateral spread, but the composition of collateral also shifts towards specific as-

sets. As Liberti and Mian (2010) point out this suggests that financial development not only

27

reduces the reliance on collateral, but also enables banks to accept firm-specific forms of as-

sets as collateral. However, our results go further and show that the channel through financial

development reduces demand for non-specific collateral is by better protecting lenders from

agency concerns

V. Identification Issues

The results presented in this paper explore the role of collateral and further how this varies

with financial development. The identification in equations (2), (3), and (4) all hinge on there

not being some alternative explanation for the relation between collateral, both level and type,

and borrower risk, both agency and production. In this section we consider and rule out

identification concerns. The ideal experiment to test equation (2), (3), and (4) is to pick a set

of firms, randomly assign each firm a risk type and grade and then measure how the demand

for collateralization differs across low and high risk firms. Unfortunately nature is seldom

this accommodating. We must therefore pay particular attention to factors that might affect a

firm’s risk rating and its equilibrium rate of collateralization at the same time. We discuss four

such factors and then test how they actually co-vary with firm risk and collateralization in the

data.

First, there may be an incidental correlation between firm grade and collateralization

driven by endogenous movements in a firm’s loan amount. Suppose all firms start off with

the same risk rating and degree of collateralization. Over time, firms experience different id-

iosyncratic shocks and their risk ratings adjust accordingly. This can automatically lead to

changes in the value of collateralization, as measured in our data, even if there is no change

in the demand for collateralization by the bank. For example, suppose that when the riskiness

of a firm deteriorates its bank borrowing declines but the initial collateral remains in place.

Then looking at the cross section of firms at any future point in time it would appear that bank

28

demands higher rates of collateralization from riskier firms, while in reality the correlation is

driven by endogenous movements in firm’s loan amount.

Second, the equilibrium rate of collateralization may be affected by the supply of a firm’s

collateralizable assets as well. For example, firms with greater (or cheaper) supply of collat-

eralizable assets may be willing to put up more collateral per dollar borrowed in exchange for

lower interest rate charged by the bank. We only need to worry about this concern if the sup-

ply of collateralizable assets is negatively correlated with risk grade, i.e. as grade deteriorates,

firms have more collateralizable assets available to them.

Third, the equilibrium rate of collateralization can also be affected by productivity or latent

demand for loans by firms. Since the average bank worries about its total exposure to a single

client, firms with larger loan demand are likely to face higher rates of collateralization. This

observation is a concern for us only if firm productivity or latent demand for loans is negatively

correlated with firm risk, i.e. firms with worse risk grades are more productive.

The fourth factor of concern in terms of identification may be the substitutability between

interest rate and rate of collateralization. For example, a bank may be willing (to some extent)

to trade off lower rates of collateralization for a higher interest rate. However this potential

trade off only biases our coefficient towards zero and thus should not be of great importance.

The second and third concerns mentioned above are also unlikely to hold in practice. For

example, as firm risk deteriorates, its supply of collateralizable assets is likely to go down as

opposed to increase. Nevertheless, since one may come up with specialized stories to raise

these concerns, we shall explicitly test for these in the empirical section.

29

A. Testing for Endogenous Time-Series Fluctuations

First we test for the concern that correlation between risk grade and collateralization is inci-

dentally generated due to time-series endogenous fluctuations in loan amount and collateral

value. Since we have up to four six-monthly observations per firm we can directly test for

this concern. In particular, instead of using our cross sectional sample that picked the first

available observation for each firm, we now use the entire available time series for firms. Us-

ing firm fixed effects we then test how collateral value and loan size varies over time when a

firm’s risk grade fluctuates. The standard deviation of firm grades after de-meaning at the firm

level is 0.36, while the cross-sectional variation in country demeaned firm grades is 0.92. This

result indicates that there is significant variation in firm grades over time and we can conduct

a meaningful time series test. We also control for possible macro shocks hitting an economy

by including country interacted with six monthly period fixed affects (i.e. country-date fixed

effects).

The empirical specification for measuring the endogenous fluctuations in loan amount and

collateral value with risk grade over time thus becomes:

Yict = αi + αct +∑k

βkαkict + εict (5)

where Yict is a variable such as log of loan amount and outstanding for firm i in country c

and date t. αk represents the k risk grade dummies, αi represents firm fixed effects, while αct

country-date fixed effects.

Column (1) in Table IX shows that deterioration in firm grade is negatively correlated

with approved loan amount. In particular relative to “A” grade firms, “B” grade firms have

6.1% less amount of loans approved. For “C” grade firms, the same difference goes up to

30

11.1%. The largest differences is for firms classified as “D”. These firms have 28.0% less

loans approved compared to “A” grade firms. Thus credit approval declines over time if firm

grade deteriorates. Column (2) shows that the outstanding loan amount also decreases with

firm quality. Worse grade firms experience the highest decrease. This result is consistent with

a firm becoming more financially constrained as firm grade deteriorates.

B. Testing for Endogenous Collateral Supply

The second identification concern suggests that perhaps our elasticity of demand coefficient is

being spuriously affected by the correlation of unobserved supply of collateral with risk grade.

This is a valid concern if the supply of collateralizable assets increases with deteriorating

quality; an unlikely occurrence.

Nonetheless using data from one of the countries in the sample, Argentina, with detailed

firm level financials (described in Section I), we test how measures of the supply of collat-

eralizable assets are correlated with firm risk grade. We present descriptive statistics for the

Argentine data in Table X. We construct different measures of the supply of collateralizable

assets. A number of firm balance sheet assets such as cash, marketable securities, accounts

receivables, inventory and property plant and equipment are collateralizable. However, net

collateralizable assets available for borrowing also depend on senior liabilities already held by

a firm. As such we construct two alternate measures of the supply of collateralizable assets.

Our first measure (Net Worth) is the total book net worth of the firm (i.e. total assets mi-

nus total non-equity liabilities). Our second measure (Net Collateral) is computed by adding

the primary collateralizable assets of the firm and subtracting total collateralized liabilities

issued by the firm. Primary collateralizable assets include cash, marketable securities, ac-

counts receivables, inventory and net fixed assets. Collateralized liabilities include senior and

31

subordinated short and long term debt.14 We also normalize each of the two measures of col-

lateralizable assets by total assets and sales in order to get a sense of the supply of collateral

per borrowing need of the firm.

Table XI estimates regressions for each of the six measures of collateral supply against

firm risk grades, industry fixed effects, and location fixed effects. Since the number of firms is

limited in this sample, we exploit full sample variation by running our regressions on pooled

data but cluster standard errors at the firm level. It can be seen that regardless of the exact

definition used, collateral supply is positively correlated with firm grade, i.e. supply increases

as grade quality improves. Given this positive correlation or at least no evidence for any

negative correlation, endogenous collateral supply cannot explain our elasticity coefficients.

C. Testing For Endogenous Firm Productivity

The third identification concern is that firm productivity, or latent demand, can bias our elastic-

ity of demand estimates upward if firm productivity is negatively correlated with firm quality

and larger loans require higher rates of collateralization. We have already seen in column (4)

of Table III that larger loans attract higher rates of collateralization. Further, column (1) of

Table XII presents results for the main sample of borrowers in 15 countries showing that worse

quality firms borrow around 7% less than the best quality firms. In addition we use Argentine

data to address concerns related to firm productivity. In columns (1) and (2) in Table XII show

that firm productivity whether measured through ROA or EBITDA/Sales is positively corre-

lated with firm grade. Consequently as with collateral supply, endogenous firm productivity

or latent demand for loans cannot explain our elasticity of demand coefficients.

14Excluding subordinated debt does not change our results significantly.

32

D. Testing for Endogenous Interest Rates

Table XII tests for the extent of substitutability between interest rates and rates of collater-

alization. Since we have detailed information on the type of revenue generated from a firm,

we construct two measures of interest rates. Our first measure is computed by dividing the

total lending revenue generated from a firm by the average lending given to that firm during a

year. Our second measure is computed by dividing the total lending and non-lending revenue

generated from a firm during a year by its average borrowing in that year.

Columns (3) and (4) in XII show that there is little substitutability between interest rates

and collateralization. The coefficients are estimated with reasonable precision as the standard

errors are small in terms of economic magnitude. Intuitively higher interest rates are not very

useful in guarding a bank against the threat of ex-post repudiation, while collateralization is.

This finding can also serve as a possible explanation for why earlier work such as Petersen

and Rajan (1994) and Mian (2006) find negligible elasticity of interest rates with respect to

measures of firm risk.

Based on the above evidence the results in Tables III - VIII provides accurate estimates for

the elasticity of collateralization demand with respect to agency risk.

VI. Concluding Remarks

The use of collateral predominantly as a commitment device to prevent agency risk raises a

number of interesting questions for further inquiry. At one end of the spectrum, existing macro

literature treats collateral as one of the main causes for frictions that lead to volatility, conta-

gion, and poverty traps. At the other end however, micro theory coupled with the evidence

presented in this paper perceives collateral as a critical factor in limiting agency risk. It may

33

not be unreasonable therefore to think of collateral as the “necessary evil” needed to sustain

financing in a less than perfect world.

This view of collateral raises a number of interesting research questions regarding alterna-

tive mechanisms available to an economy for limiting agency concerns. Even a cursory look

across economies suggests that there are other potential avenues for dealing with agency risk.

Such alternatives include more efficient enforcement of laws, market discipline through the

use of credit registries, social or venture networks with better enforcement and information

tools, and better social norms. At least some of these alternative mechanisms for dealing with

agency risk are likely to be more efficient than collateral from a macro perspective. There is

also apparent variation across economies in the availability and use of these alternative mecha-

nisms. What explains such variation? Why do some economies adopt different and potentially

superior mechanisms for dealing with agency risk than others? We hope that future work will

guide us towards the answers.

34

References

[1] Acharya, Viral V., Sreedhar Bharath and Anand Srinivasan, “Does Industry-wide Dis-

tress Affect Defaulted Firms? - Evidence from Creditor Recoveries”, Journal of Finan-

cial Economics, v. 85(3), (2007), 787-821.

[2] Aghion, Philippe, Abhijit Banerjee, and Thomas Piketty, “Dualism and Macroeconomic

Volatility”, Quarterly Journal of Finance, v. 114(4), (1999), 1359-97.

[3] Barro, Robert J., ”The Loan Market, Collateral and Rates of Interest”, Journal of Money,

Credit and Banking, v. 8 (November 1976), 439-56.

[4] Banerjee, Abhijit, and Andrew F. Newman, “Occupational Choice and the Process of

Development”, Journal of Political Economy, v. 101(2), (1993), 274-98.

[5] Benmelech, Efraim and Nittai Bergman, “Collateral Pricing”, Journal of Financial Eco-

nomics, v. 91(3), (2009), 339-360.

[6] Benmelech, Efraim and Nittai Bergman, “Bankruptcy and the Collateral Channel,”,

Journal of Finance, v. 66, (2011), 337-378.

[7] Berger, Allen and Scott Frame and Vasso Ioannidou “Tests of Ex Ante versus Ex Post

Theories of Collateral using Private and Public Information”, Journal of Financial Eco-

nomics, v. 100, (2011), 85-97.

[8] Berger, Allen and Scott Frame and Vasso Ioannidou “Reexamining the Empirical Rela-

tion between Loan Risk and Collateral: The Role of Collateral Types and Characteris-

tics”, University of South Carolina WP, (2012)

[9] Berger, Allen and Greg Udell “Relationship Lending and Lines of Credit in Small Firm

Finance”, Journal of Business, , v. 68 (1995), 351-382.

35

[10] Besanko, David and Anjan V. Thakor, “Collateral and Rationing: Sorting Equilibria

In Monopolistic And Competitive Credit Markets”, International Economic Review, v.

28(3), (1987), 671-89.

[11] Bester, Helmut, “Screening versus Rationing in Credit Markets with Imperfect Informa-

tion”, American Economic Review, v. 75(4), (1985), 850-55.

[12] Bernanke, Ben S., and Mark Gertler, “Agency Costs, Net Worth and Business Fluctua-

tions”, American Economic Review, v.79, (1989), 14-31.

[13] Chan, Y., and Thakor, A., “Collateral and Competitive Equilibria with Normal Hazard

and Private Information”, Journal of Finance, v. 42, 345-64.

[14] Chaney, Thomas, David Sraer and David Thesmar, 2011, “The Collerateral Channel:

How Real Estate Shocks affect Corporate Investment”, American Economic Review,

(2011)

[15] Holmstrom, Bengt, “Managerial Incentive Problems: A Dynamic Perspective”, Review

of Economic Studies, v. 66, no. 1, special issue: contracts, (Jan., 1999), 169-182.

[16] Holmstrom, Bengt; and Jean Tirole, “Financial Intermediation, Loanable Funds, and the

Real Sector”, Quarterly Journal of Economics, v.112(3) (1997), 665-691.

[17] Jimenez, Gabriel and Vicente Salas, and Jesus Saurina, “Determinants of Collateral”,

Journal of Financial Economics, v.81 (2006), 255-281.

[18] Kiyotaki, Nobuhiro, and John Moore, “Credit Cycles”, Journal of Political Economy, v.

105(2), (1997), 211-48.

[19] Lacker, Jeffrey, “Collateralized Debt as the Optimal Contract”, mimeo Federal Reserve

Bank of Richmond, (1992).

36

[20] Liberti, José M. and Atif Mian, “Collateral Spread and Financial Development”, Journal

of Finance,v. 105(1), (2010), 147-177.

[21] Mian, Atif, “Distance Constraints. The Limits of Foreign Lending in Poor Economies”,

Journal of Finance, v. 61(3), (2006),1465-1505.

[22] Mookherjee Dilip, and Debraj Ray, “Contractual Structure and Wealth Accumulation”,

American Economic Review, v. 92(4), (2002), 819-49.

[23] Myers, Stewart and Raghuram Rajan, “The Paradox of Liquidity”, Quarterly Journal of

Economics, v.113(3), (1998), 733-771.

[24] Park, Cheol, “Monitoring and Structure of Debt Contracts”, Journal of Finance, v. LV,

no. 5, (2000), 2157-2195.

[25] Petersen, Mitchell and Raghuram Rajan (1994), “The Benefits of Lender Relationships:

Evidence from Small Business Data”, Journal of Finance, v.49, (1994), 3-37.

[26] Rajan, Raghuram and Andrew Winton, “Covenants and Collateral as Incentives to Mon-

itor”, Journal of Finance, v. 50(4) (1995), 1113-42.

[27] Schleifer, Andrei, and Robert W. Vishny, “Liquidation Values and Debt Capacity: A

Market Equilibrium Approach”, Journal of Finance, v.47, (1992), 1343-66.

[28] Stiglitz, Joseph, and Andrew Weiss, “Credit Rationing in Markets with Imperfect Infor-

mation,”American Economic Review, LXXI (1981), 393-410.

[29] Zwiebel, Jeffrey, “Dynamic Capital Structure Under Managerial Entrenchment ”, Amer-

ican Economic Review, (1996).

37

Figure 1. Model Summary: Defining Business and Agency RiskThe figure summarizes the model defining business and agency risk presented in Section I. A firm hasa investment opportunity for $1, which it finances through a (collateralized) bank loan. Firms vary intheir risk attributes denoted by the pair (α, β), where α and β are both between 0 and 1. β captures thebusiness production risk inherent in a firms technology. The investment payoffs Y > 1 with probability(1 − β) and 0 otherwise. α on the other hand captures the degree of agency risk inherent in a firm. Ifa firm produces an output Y , it can choose to repay the promised interest rate on the bank loan r < Yor declare default strategically. In case of default, the firm looses part of its future productivity due to aloss of reputation in financial markets given by (1−α)Y . It also loses its initial collateral worth w < 1in case of default.

38

Figure 2. Collateral Demand Elasticity By CountryThe figure plots the demand elasticity coefficient separately for each country for grade D firms. Coun-tries in Figure 2 are ordered by their number of firms in sample. The point estimate along with the 95%confidence interval is presented for demand elasticity in each country.

-50

050

100

PakistanSri Lanka

ArgentinaRomania

SlovakiaSingapore

South AfricaTaiwan

IndiaMalaysia

TurkeyHong Kong

ChileKorea

Czech

39

Figure 3. Marginal Effect of the Collateral Pecking OrderThe figure plots the marginal effect of the collateral pecking order. The marginal effect on collateraldemand of a one grade downgrade is plotted for each asset class. The one grade downgrade is fromgrade A to B, B to C, or C to D. Assets classes vary in there specificity from firm- and industry-specific(high specificity) to guarantees, cash, and land (low specificity).

-6-4

-20

24

6

AtoB BtoC CtoD

FIRM SPEC IND SPEC

PNOTE IMP EXP L/C

LEASE FIN SEC

AR ASSETS

GUA CASH

LAND

40

Figure 4. Cumulative Effect of the Collateral Pecking OrderThe figure plots the cumulative effect of the collateral pecking order. The cumulative effect on collateraldemand is plotted for each asset class. The cumulative effect measures the effect on collateral frommoving from grade A (1) to grade D (4). Assets classes vary in there specificity from firm- and industry-specific (high specificity) to guarantees, cash, and land (low specificity).

-10

-8-6

-4-2

02

46

810

Cum

ulat

ive C

hang

e in

Col

late

ral (

%)

1 1.5 2 2.5 3 3.5 4Risk Grade

FIRM SPEC IND SPECPNOTE IMP EXP L/CLEASE FIN SECAR ASSETSGUA CASHLAND

41

Table IData Description by Country

This table presents the distribution of data by country along with the country’s financial (creditor rights) andeconomic development (GDP per capita) indicators . The data comes from a sample of 8,820 small and medium-sized firms in 15 different emerging markets borrowing from a large multinational bank. The appendix reportsthe sample distribution of firms by industry. Countries are ordered in alphabetical order.

Avg. Loan Financial Development

Country Number Size No. of % of % of Creditor GDP perof Firms ($000) Industries Total Firms Total Loans Rights Capita

Argentina 120 86 18 1.4 0.3 1 3650Chile 1,248 127 77 14.1 4.7 2 4390Czech 1,440 296 73 16.3 12.5 3 6740Hong Kong 1,169 618 65 13.3 21.2 4 25430India 494 626 49 5.6 9.1 2 530Korea 1,427 94 71 16.2 3.9 3 12020Malaysia 697 339 62 7.9 6.9 3 3780Pakistan 96 599 35 1.1 1.7 1 470Romania 135 191 47 1.5 0.8 1 2310Singapore 237 680 58 2.7 4.7 3 21230Slovakia 140 472 43 1.6 1.9 2 4920South Africa 307 1505 59 3.5 13.5 3 2780Sri Lanka 102 468 17 1.2 1.4 2 930Taiwan 443 723 54 5.0 9.4 2 13320Turkey 765 358 54 8.7 8.0 2 2790

Total/average 8,820 387 87 100.0 100.0 2.3 7,019

42

Table IISummary Statistics: Cross-Country Firm-Level Data

This table presents for all the variables used in the empirical analysis for the sample of 8,820 firms at the begin-ning of the sample (except the Default Status and Risk Grade Decrease Status which are estimated at the sample’send). Standard deviation (SD) within country is computed after demeaning each variable at the 15 country levels,while SD within each country-industry is computed after demeaning each variable at the 782 country-industrycategories. Variable definitions are given in the Appendix.

Variable SD within SD within Obs.Mean SD Country Country-Industry Obs.

Risk Grade 2.65 0.99 0.90 0.81 8,820A 0.15 0.35 8,820B 0.29 0.45 8,820C 0.33 0.47 8,820D 0.23 0.42 8,820

Sales Size Indicator 0.94 1.01 0.84 0.76 8,8200 0.40 0.49 8,8201 0.37 0.48 8,8202 0.14 0.34 8,8203 0.07 0.26 8,8204 0.02 0.15 8,820

Total Loan Approved (in $000) 556.59 973.23 738.24 641.53 8,820Log Total Loan Approved 11.93 1.96 1.46 1.30 8,820Loan Outstanding (in $000) 386.65 909.96 660.81 571.42 8,820Default Status 5.17 22.14 17.21 13.71 8,820Risk Grade Decrease Status 6.61 24.84 21.21 17.78 8,820Collateralization Rate 54.09 44.69 33.12 30.75 8,820

Breakdown of Collateralization Rate by:Firm-specific Assets 9.36 60.64 36.96 31.63 8,820Industry-specific Assets 4.41 19.25 6.62 5.21 8,820Assets 7.42 24.18 14.28 9.69 8,820Accounts Receivable 8.01 24.92 12.34 10.85 8,820Promissory Notes 0.58 7.01 3.46 2.79 8,820Import/Export Letters of Credit 0.10 3.01 1.00 0.67 8,820Leasing 3.57 17.64 5.88 5.13 8,820Guarantees 1.04 9.94 5.83 3.34 8,820Financial Securities 0.02 1.51 0.45 0.21 8,820Cash 7.66 23.83 14.87 12.06 8,820Land/Real Estate 11.92 30.33 21.33 19.01 8,820

43

Table IIICollateralization Rate with Respect to Overall Ex-Ante Firm Risk

This table reports estimates of the collateralization rate with respect to overall ex-ante firm risk grade. The unitof observation is the borrower firm. The dependent variable is collateralization rate. Risk Grade is the borrowergrade assigned by loan officers are the beginning of the sample. Risk Grade takes on values of ”A” (best) to ”D”(worst). Risk Grade ”A�