What Do One Million Credit Line Observations Tell Us about Exposure at Default?

A Study of Credit Line Usage by Spanish Firms

Gabriel Jimnez (Banco de Espaa) Jose A. Lopez (Federal Reserve Bank of San Francisco) Jess Saurina (Banco de Espaa)*

6th Annual Bank Research Conference September 24, 2006

* The views expressed here are those of the authors and not necessarily those of the Banco de Espaa, the
Federal Reserve Bank of San Francisco or the Federal Reserve System.
Filename

Outline
Introduction-Motivation
Contributions of the paper

Database

Empirical evidence
Event study Econometric modeling Data description Estimation results

Policy analysis

FINANCIAL STABILITY DEPARTMENT

Filename

1
1

Overview

Corporate credit lines present a variety of interesting research questions:

- Firms: Why establish? Why and when draw? - Banks: Why underwrite? How monitor? - Risk managers: What will exposure at default be?
Race to Default
Source: BofA LEQ doc

Commitment

Exposure

Outstanding Today Deteriorating Credit Quality Default

FINANCIAL STABILITY DEPARTMENT

Filename

2
2

Overview (continued)

Academic research on corporate credit lines has been relatively limited, most probably due to a lack of data.
Melnik and Plaut (1986): 101 survey respondents
Ham and Melnik (1987): 90 survey respondents Berger & Udell (1995): 89 Nat. Survey of Small Bus. Fin.

Agarwal et al. (2004): bank proprietary dataset of 712 firms

Sufi (2006): handcrafted dataset of 300 publicly-traded firms based on SEC 10-K annual filings

Gatev and Strahan (2006): Dealscan for CP back-up lines

Our dataset based on the Banco de Espaas credit register allows us to construct a comprehensive sample of firms (more than 200,000) over a long period of time (1985-2005)
FINANCIAL STABILITY DEPARTMENT
Filename

3
3

Overview (continued) Contributions Contributions to the literature:

1. Corporate finance findings:

Five years prior to default, credit line usage is higher (60%) for these firms than for non-defaulting firms (50%). The usage rate increases monotonically as default approaches, up to almost 90% at default.

Borrowers identified ex-ante as riskier, get less access to credit lines; analogous to Sufis profitability result.
Borrowers that have previously defaulted on loans, access their credit lines less, suggesting bank monitoring. Credit line usage has cyclical characteristics; i.e., use increases in recessions and declines in expansions.
4
4

FINANCIAL STABILITY DEPARTMENT

Filename

Overview (continued) Contributions

2. Public policy findings:

Exposure at default (EAD) exhibits procyclical behavior. To our knowledge, this is the first evidence on this point. Various loan characteristics lower EAD: Larger commitment size Longer maturities Collateral requirements With respect to the Basel II framework, EAD parameterization in the standard approach may be too low For the foundation approach, it seems appropriate. Not sufficient recognition of the procyclicality of EAD that could augment the expected procyclicality of capital
FINANCIAL STABILITY DEPARTMENT
Filename

5
5

Database Banco de Espaas credit register

Banco de Espaa maintains a credit register known as the Central de Informacin de Riesgos or CIR. The CIR contains monthly loan-level information on all credits above a threshold of 6,000 granted by Spanish banking institutions (i.e., commercial banks, not-for-profit savings banks, credit cooperatives, etc.) since 1984 - hence, full census of Spanish corporate borrowing

The CIR contains detailed information on loan details: borrower name, industry, province of headquarters instrument type (i.e., commercial loan, etc.) maturity use of collateral total commitment & amount drawn default status
6
6

FINANCIAL STABILITY DEPARTMENT

Filename

Database Banco de Espaas credit register (cont.)

Definition of default status in the CIR database: the borrower has loan payments overdue by more than 90 days, the legal definition of default in Spain, or it has been classified as a doubtful borrower by the originating bank (i.e., the lender itself believes there is a high probability of non-payment). Not a terminal state, so new loans can be granted. Useful data transformations are available: length of a banking relationship number of loans outstanding percent of a firms credit lines provided by a bank Important shortcomings: no pricing data (i.e., interest rates, fees, etc.) no covenant information
FINANCIAL STABILITY DEPARTMENT
Filename

7
7

Database Our credit line database

After applying our filtering procedures, We have a sample of 915,563 credit line-year observations corresponding to 352,328 new credit lines granted to 258,532 non-financial firms by 444 banks time period: Decembers from 1985 to 2005 Roughly 85% of the observations are individual credit lines held by a firm with a single bank, and the remaining 15% of the observations correspond to firms that hold more than one credit line with a bank. We examine the period from 1985 to 2005, which includes a deep recession around 1993, and two expansionary periods around the late 1980s - early 1990s and from 1997 onward.
FINANCIAL STABILITY DEPARTMENT
Filename

9
9

Empirical evidence Event study based on default

Transform calendar time data into event time data For each of the 17 years in the sample for which it is possible, define it as year zero and trace usage ratios back to year -5 For defaulted firms, shows usage up to default point For non-defaulted firms, just shows usage history
% DRAWN Credit Lines
100

80

%
60 40 -5 -4 -3 -2 -1 0 1 2

No. years from default

Non defaulted Defaulted

FINANCIAL STABILITY DEPARTMENT

Filename

11
11

Empirical evidence Econometric modeling (continued)

Ln _ RDRAWN ijkt 1it 2 Firm jt 3 Bankkt 1GDPG t 2 RIR t i it
Where i denotes the credit line, j the firm, k the bank and t the time.

it is the time to default for credit lines that default at t+, >0 two specifications discrete, dummy variables; i.e., it(-5) = 1 & rest = 0 continuous variable; i.e., it(-5) = -5 Firmjt is variables that controls for firm characteristics Bankkt is variables that control for bank characteristics GDPGt is real, annual growth rate in Spanish GDP RIRt is 3m real interbank interest rate (measure of funding) i is a fixed effect for the credit line, and it is an error term
FINANCIAL STABILITY DEPARTMENT
Filename

15
15

Empirical evidence Econometric modeling (continued)

Why include i ? - All credit line time-invariant characteristics, such as its maturity or collateral requirements, are included here. - The CIR has limited information on firm characteristics, and those effects not captured elsewhere go here.

What is Firmjt? - It is a measure of firm risk based just on the CIR, which is an indicator variable for whether the firm has ever defaulted on any CIR loan.
What is Bankkt? - A measure of bank risk based just on the CIR, which is non-performing loan ratio of the bank. - A measure of the size of the bank using the total share of the bank each year as a proxy.
FINANCIAL STABILITY DEPARTMENT
Filename

16
16

Empirical evidence Baseline estimation results (Table 4)

Within-Groups No. observatios 904,542 Sample period 1985-2005 Dependant variable Ln_RDRAWNijkt Model 2 Coefficient t-ratio No. years from defaultit Firm riskjt Bank NPL ratiokt Bank Sharekt GDPGt RIRt Constant F-test (p-value) 2.761 -0.685 0.015 0.598 -0.043 0.231 -0.524 0.00 41.48 *** -8.13 *** 3.24 *** 4.79 *** -5.23 *** 43.07 *** -16.41 ***

FINANCIAL STABILITY DEPARTMENT

Filename

18
18

Empirical evidence Baseline estimation results (Table 4)

Panel data estimation technique: Within-groups estimation to address possible correlation between i and RHS variables Dependent variable is ln_RDRAWNijkt
RHS variables Firm risk Sign (-) Riskier firms draw down less; suggests bank monitoring Large banks more confident(?) Riskier banks are more lenient

Bank share Bank NPL ratio

(+) (+)

GDP growth Interest rates

(-) (+)

Increase use in bad times High funding costs lead to increased use of credit lines
19
19

FINANCIAL STABILITY DEPARTMENT

Filename

Empirical evidence Baseline estimation results (continued)

RHS variables Years to default

Sign (+)

As a firm approaches default, it drawns down more heavily on its existing credit lines, even after controlling for CIR firm and lender characteristics as well as macroeconomic conditions. Recall that years to default is a negative variable.

Provides confirmation & context for our event study diagram.

FINANCIAL STABILITY DEPARTMENT

Filename

20
20

Empirical evidence Econometric modeling (continued)

Extension to the baseline model:

Ln _ RDRAWN ijkt ( 1 X it )it 2 Firm jt 3 Bank kt 1GDPG t 2 RIR t i it

Where Xit is a variable of interest that may have a differential impact on the usage rate depending on a firms time to default. We use: maturity indicator equal to one if greater than one year collateral indicator commitment size (with the top 5% tail winsorized) Bank type since there was variation in non-performing loans at different points in the sample; see Salas and Saurina (2002, JFSR)

FINANCIAL STABILITY DEPARTMENT

Filename

21
21

Empirical evidence Further estimation results

Ln _ RDRAWN ijkt ( 1 X it )it 2 Firm jt 3 Bank kt 1GDPG t 2 RIR t i it

Interacted variable Firm characteristics:

Sign

Firm risk (-) -for a given time to default, lower quality borrowers draw less, suggesting banks monitor these firms more closely

FINANCIAL STABILITY DEPARTMENT

Filename

22
22

Empirical evidence Further estimation results (continued) Interacted variable Line characteristics: Sign

Commitment size (-) -for a given time to default, larger credit lines are drawn less, suggesting more bank monitoring Maturity (-) -for a given time to default, credit lines with longer maturities are drawn less, suggesting more bank monitoring Collateral indicator (-) -for a given time to default, credit lines with collateral are drawn less, suggesting more bank monitoring
23
23

FINANCIAL STABILITY DEPARTMENT

Filename

Empirical evidence Further estimation results (continued) Interacted variable Bank characteristics: Sign

Bank NPL ratio (+) -for a given time to default, credit lines by riskier banks are drawn more, suggesting less bank monitoring Bank share (+) -for a given time to default, credit lines by larger banks are drawn more, suggesting less bank monitoring Bank type -not statistically significant

FINANCIAL STABILITY DEPARTMENT

Filename

24
24

Empirical evidence Further estimation results (continued) Interacted variable Sign

Macroeconomic conditions:
GDP growth (-) -in recessions, credit line usage increases at any given time to default Our two sets of macroeconomic results are the first empirical evidence of the procyclicality of credit line usage (and related to the Gatev & Strahan (2006) work on credit line origination and pricing)

FINANCIAL STABILITY DEPARTMENT

Filename

25
25

Empirical evidence Summary of empirical findings

Credit line usage is related to: macroeconomic conditions firm characteristics bank characteristics For firms that eventually default, credit line usage increases as the default approaches. These increases are influenced upward by bank riskiness and recessions downward by firm riskiness, loan terms and expansions

FINANCIAL STABILITY DEPARTMENT

Filename

26
26

Policy analysis EAD in Basel II

Basel II capital calculations:

Regulatory capital = f(PD, M) * LGD * EAD,

Much work has been done on PD quantification and validation. Much less work has been done on LGD.

Virtually no work has been done on EAD, even though it has a one-for-one effect on capital requirements.

FINANCIAL STABILITY DEPARTMENT

Filename

28
28

Policy analysis Prior research on EAD Prior public research into EAD has focused on other forms.

Loan equivalent amounts (LEQ):

EADit() = DRAWNit + LEQit() * UNDRAWNit -This is the more common form, since one is typically more interested in how much more of a line is at risk.

Credit conversion factors (CCF):

EADit() = CCFit() * (DRAWNit + UNDRAWNit)
Obviously related algebraically to LEQ, but perhaps a bit more intuitive for capital calculations based on total commitment amounts.

FINANCIAL STABILITY DEPARTMENT

Filename

29
29

Policy analysis Prior research on EAD (continued)

Asarnow and Marker (1995, the Citi study): presented (in an appendix) a set of LEQ estimates based on corporate loan data from 1988 to 1993 Araten and Jacobs (2001, the Chase study): based on 408 credit lines to 399 defaulted borrowers (i.e., no longer able to draw down) from 1994 to 2000
Unconditional LEQ of 43.4% i.e., borrowers that default draw down almost half of their undrawn commitment on their way down LEQit() is a decreasing function of i.e., borrowers draw down more as they get closer to default

FINANCIAL STABILITY DEPARTMENT

Filename

30
30

Policy analysis Our empirical EAD results

Based on 8,384 defaulted credit line-year observations corresponding to 2,883 credit lines LEQit() declines from 73% at = -5 to 36% at = -1 with an unconditional mean of 48%
100

80

60

%
40 20 0 -5 -4 -3 -2 -1

No. years from default

FINANCIAL STABILITY DEPARTMENT

Filename

31
31

Policy analysis Our empirical EAD results (continued)

Further analysis based on different characteristics: Large credit lines have lower LEQs One interpretation is that larger lines are more actively monitored by lenders Another is that larger lines are granted to larger firms that dont need bank financing as much, even in default Credit lines with shorter maturities have higher LEQs One interpretation is that longer maturities allow for better monitoring of the borrower Collateralized credit lines have lower LEQs Not surprising since a drawdown puts more collateral at risk to be seized by the lender
32
32

FINANCIAL STABILITY DEPARTMENT

Filename