Informational Frictions and the Credit Crunch

• Author: OLIVIER DARMOUNI
• Date: 01 August 2020
• Published date: 01 August 2020
• DOI: http://doi.org/10.1111/jofi.12900

THE JOURNAL OF FINANCE •VOL. LXXV, NO. 4 •AUGUST 2020

Informational Frictions and the Credit Crunch

OLIVIER DARMOUNI∗

ABSTRACT

In this paper, I estimate the magnitude of an informational friction limiting credit

reallocation to firms during the 2007 to 2009 financial crisis. Because lenders rely on

private information when deciding which relationship to end, borrowers looking for a

new lender are adversely selected. I show how to separately identify private informa-

tion from information common to all lenders but unobservable to the econometrician

by using bank shocks within a discrete choice model of relationships. Quantitatively,

these informational frictions appear to be too small to explain the credit crunch in

the U.S. syndicated corporate loan market.

THE DEFINING FEATURE OF A lending relationship between a bank and a borrower

is its stickiness: switching lenders is rare and costly.1In turn, credit markets

are more vulnerable: a shock forcing a particular bank to cut lending can

have aggregate effects if affected borrowers cannot easily find a new lender.2

Understanding why relationships are sticky is important, as it can guide the

design of institutions or policies to prevent breakdowns in lending markets.

In this paper, I estimate the effects of a key friction behind relationship

stickiness: the information gap between a borrower’s existing lender and

its potential new lenders. Over the course of a relationship, lenders acquire

∗Olivier Darmouni is with Columbia University. I am indebted to David Sraer, Jakub Kastl,

Markus Brunnermeier, Motohiro Yogo, and ValentinHaddad for their continuous support. I would

also like to thank Tobias Berg; Charles Calomiris; Adrien Matray; Atif Mian; Justin Murfin; Hoai-

Luu Nguyen; Tomek Piskorski; Matthew Plosser; multiple reviewers; and seminar participants

at Princeton University, Columbia, Harvard Business School, UT Austin, NYU Stern, Yale SOM,

Wharton, Berkeley Haas, FDIC, HEC Paris, the Toulouse School of Economics, Wharton, Kellogg,

the New York Fed, WFA, Copenhagen Business School, BYU, and UCLA Anderson for many

discussions and comments that improved this paper.Special thanks to Gabriel Chodorow-Reich for

his help and support at the early stages of this project. Previous versions of this paper were entitled

“The Effects of Informational Frictions on Credit Reallocation” and “Estimating Informational

Frictions in Sticky Relationships.” Lira Mota provided outstanding research assistance. I declare

that I have no relevant or material financial interests that relate to the research described in

this paper.

Correspondence: Olivier Darmouni, Assistant Professor,Columbia Business School, 3022 Broad-

way, New York, NY 10027; e-mail: o.darmouni@gmail.com.

1See Srinivasan et al. (2014) for a survey of the extensive literature on banking relationships.

2A large body of works studies the effect of bank shocks on firm borrowing and real outcomes af-

ter the financial crisis, including Ivashina and Scharfstein (2010), Chodorow-Reich (2014), Jim´

enez

et al. (2019), Greenstone, Mas, and Nguyen (2020), and Schwert (2018).

DOI: 10.1111/jofi.12900

C2020 the American Finance Association

2055

2056 The Journal of Finance R

abstract and hard-to-verify private (“soft”) information about their borrowers

that is unobservable to other lenders.3The information gap represents the

informational advantage that stems from relationship lending. The main con-

tribution of this paper is to provide the first direct estimate of the magnitude of

the information gap and its role in explaining the credit crunch that followed

the 2007 to 2009 crisis (Chodorow-Reich (2014)).

The key identification challenge is that, empirically, private information is

difficult to disentangle from common information that all lenders can observe

but that the econometrician cannot. This paper shows that shocks to banks can

be used to separately identify lenders’ private information from information

common to all lenders. Using loan-level data from the U.S. syndicated loan

market, I find that lenders’ private information appears to be too small to

explain why relationships are sticky in this market, and therefore, cannot

quantitatively account for much of the associated drop in lending documented

in prior studies.4

The information gap reduces aggregate lending by creating adverse selection

in the market for borrowers looking for a new relationship. Lenders’ private

information gives them the ability to selectively choose which relationships to

end when scaling down lending after a shock, leaving their worst borrowers

looking for funds elsewhere. This is the predominant view to rationalize re-

lationship stickiness and the credit crunch observed the U.S. syndicated loan

market. However, testing this private information channel directly has proven

elusive. This paper offers a solution to this econometric challenge.

The above channel makes clear why shocks to banks’ ability to lend can be

useful in identifying the information friction. It implies an “inference hypothe-

sis”: borrowers leaving the most-affected lenders are less adversely selected, as

described in Dell’Ariccia and Marquez (2004). Intuitively, these lenders cannot

continue lending even to relatively good borrowers. Therefore, this inference

hypothesis implies that a firm’s ability to borrow from a new lender after a

breakup depends on the size of the shock faced by its previous lender.

There is evidence consistent with this effect in the U.S. syndicated loan mar-

ket over the period 2004 to 2010. Exploiting the financial crisis that originated

in the real estate sector, I use a lender’s exposure to this shock to measure

its ability to lend in the corporate loan market. Conditional on leaving a re-

lationship, a one-standard-deviation increase in the crisis exposure of a firm’s

existing lender implies a 20% increase in the probability of borrowing from a

new lender.5

However, this evidence does not solve the main identification challenge of

isolating private information. In fact, the same reduced-form correlation would

3See,forexample,Sharpe(1990), Rajan (1992), and Detragiache, Garella, and Guiso (2000).

Examples of soft information acquired during a relationship include the quality of management,

potential future investment projects, as well as information whose public disclosure would hurt

the firm.

4This is not to say that there is no asymmetric information between lenders and borrowers, but

rather that there is no asymmetric information across lenders in this particular market.

5An equivalent finding in labor markets can be found in Gibbons and Katz (1991).

Informational Frictions and the Credit Crunch 2057

emerge if there were only common information that all lenders could observe

but that the econometrician could not. In that case, new lenders would not

learn any additional information from a relationship being ended. Rather, they

simply would prefer lending to better borrowers, which are mechanically more

likely to come from more affected lenders, that is, there is selection on com-

mon information.

The key idea to address this challenge is to exploit a comparison with the

sample of borrowers who renewed their relationships. This comparison is useful

because relationship renewal reflects how informed lenders lend to borrowers

and introduces a benchmark against which new lenders can be compared.

Models with and without private information make different predictions on

the joint pattern of renewal and creation of relationships.

To this end, I introduce a two-stage discrete-choice model of firm borrowing.

In the first stage, firms try to renew their relationship with their existing

lender. Each lender faces a shock impacting its ability to lend. If a borrower

fails to receive a new loan from its existing lender, it can turn to new lenders

in the second stage. The main ingredient of the model is the existence of three

layers of information: (i) all lenders have some information about borrowers,

but (ii) each lender has private information about its existing borrowers, and

(iii) the econometrician observes neither.

Empirically, the approach relies on the assumption that shocks to banks’

ability to lend are unrelated to the unobservable characteristics of its borrow-

ers.6The first stage can be estimated by regressing the probability that a firm

renews its relationship with its existing lender on firm and lender characteris-

tics. The information gap is estimated in the second stage, using the subsample

of firms whose relationship ended in the first stage. In line with the “inference

hypothesis” above, this stage estimates how the probability that a firm finds a

new lender depends on the shock faced by its previous lender. Unlike a purely

reduced-form approach, it is possible to control for the mechanical selection

on common information of firms that did not renew their relationship. Indeed,

the first stage precisely characterizes how renewal depends on shocks to a

firm’s previous lender. The maintained assumptions are that the distribution

of borrower unobservables and the lending rule (as a function of borrower and

lender characteristics) are common across lenders, up to some matching error

orthogonal to the previous lender’s shock.7

In the context of U.S. syndicated loans, I find that the information gap is

small, and thus, this friction is unlikely to explain relationship stickiness

in this market. Quantitatively, the information gap cannot account for any

6Any such credit supply shock would do. I follow Chodorow-Reich (2014) to construct a va-

riety of shocks to banks active in the syndicated loan market that appear to be orthogonal to

borrowers’ characteristics.

7As opposed to the literature that identifies informational frictions by comparing firms with

different degrees of opacity, this approach relies on comparing how lenders with different infor-

mation would treat the same firm. The two-step approach bears some resemblance to econometric

models in the line of Heckman (2013) but is used to account for differences in information among

agents, a feature that is absent from these models.