1

Chapter 4
Lender Borrower Relationship

Rules versus Discretion

Heteroscedastic Model
Cerqueiro, Degryse & Ongena (JFI 2011)

Who Makes the Credit Decisions?

Person or Machine Behind the Desk?

Are credit decisions identical?

Imagine
a set of twins that applies for a bank loan
If twins are fully identical in all respects (same sex,
profession, street address, etc.)
The applications are made online through the same
automated system
the twins should receive the same loan rate
as both the inputs in the credit application
and the models processing these inputs
are identical

Now bring in a loan officer

Twins may not get the same loan rate. Why?
Extra input: the loan officers judgment.

experience
bargaining ability
uncertainty regarding the costumers prospects
soft information
eccentricities
the color of the applicants jacket
the loan officers mood
weather conditions and/or many other personal factors

Idiosyncratic in nature and (hence) not verifiable

by third parties = discretion
7

Rules vs. Discretion

Rules

Discretion

Loan Rates
8

Loan Pricing Technologies

Max.

Rules

Standardization
Level
Weight of Objective
Information

Discretion

Min.

Rules versus Discretion

Rules a Computer that uses:
A standardized pricing model
Only objective criteria as inputs
Predictable loan rates
Discretion a Loan Officer who may:
Add subjective judgements as inputs
Combine different inputs in any subjective way
Make pricing mistakes
Hard-to-predict loan rates
10

This Paper
Uses a heteroscedastic regression model to
assess the determinants of the importance
of Rules and Discretion on contracted
loan rates
Loan rates should reflect some latent
combination of objective (rules) and subjective
(discretion) criteria
Heteroscedastic model analyzes how the
predictive power of a linear loan-pricing model
changes with given firm, market and loan
characteristics
11

Main Results
The importance of discretion
decreases with:
Loan and firm size
Prime rate

increases with:
Borrower opaqueness (e.g., poor credit history)
Distance between firm and lender
Age of firms owner

Discretion has decreased over the last 15 years for

small loans to opaque firms

12

I.Why Loan Rates Are So Hard to Predict?

Loan Pricing Models and R2

Study

R2

# Var. # Obs.

Petersen & Rajan (JF 1994)

15%

32

1,389

Berger & Udell (JB 1995)

10%

22

371

Brick & Palia (JFI 2007)

11%

80

766

Degryse & Ongena (JF 2005)

22%

83

15,044
13

I. Well, It Depends....
Heterogeneity in Loan Pricing Models
Sample split regressions (by loan size)
From Degryse & Ongena (JF 2005)

Specification: Loan Rate = Controls + Residual

Loan Size ($)

# Obs.

R2

Small (< 5,000)

5,850

0.01

Large (> 50,000)

1,850

0.67
14

This Crisis?
Lower R on acceptance model for US subprime
versus prime loans
DellAriccia, Igan & Laeven (JMCB Forth)

15

16

II. Methodology
Econometric Model
Heteroscedastic regression model:
Mean equation: yi = 'Xi + ui
Variance equation: i2= exp(Zi)
Extreme cases:
Rules: R2 of mean equation 1
Discretion: R2 of mean equation 0
Model estimated by MLE (normality assumption)
17

18

19

II. What Does Each Equation Tell Us?

Loan Rate

Case 1 No Price Discrimination

Pooling
Equilibrium

Age of the Firm

20

II. What Does Each Equation Tell Us?

Case 2 Price Discrimination

Loan Rate

Loan Officer Increasingly

Dislikes Smile of Owner

Loan Officer Increasingly

Likes Smile of Owner

Age of the Firm

21

II. Mean Equation

Case 2 Perfect Discrimination

Loan Rate

Loan Officer Increasingly

Dislikes Smile of Owner

More Good Smile Owners

in the Pool

Loan Officer Increasingly

Likes Smile of Owner

Age of the Firm

22

II. Variance Equation

Case 2 Perfect Discrimination

Loan Rate

Loan Officer Increasingly

Dislikes Smile of Owner

Loan Officer Increasingly

Likes Smile of Owner

Age of the Firm

23

Loan Rate

II. Hypothetical Example

Loan Size
24

Loan Rate

II. Hypothetical Example

Rules

Loan Size
25

II. Hypothetical Example

Loan Rate

Discretion
Rules

Loan Size
26

II. Hypothetical Example

Loan Rate

Discretion
Rules

Loan Size
27

II. Interpretation of and

<0
Loan Rate

<0

Loan Size
28

II. Technical Comment

Parameters in mean () and variance () equations
are uncorrelated
similar to the consistency of the coefficients
being unaffected in a linear regression model
by heteroscedasticity in the error term

29

II. If Loan Size Not in Mean Equation?

=0
Loan Rate

<0

Loan Size
30

III. Data
Datasets used:
1993, 1998 and 2003 SSBF
Allows for a temporal analysis of loan pricing
practices

31

III. Variables in Mean Equation

Number of predictors: 62, including:
Cost of capital
Loan characteristics
Firm / Owner characteristics
Accounting information
Credit history

Relationship characteristics
Competition / Location measures
Other controls: SIC codes, regions, lender type
R2 of mean equation: 25%
Not particularly low & not due to low number of regressors

32

33

34

III. Variables in Mean Equation

Number of predictors: 62, including:
Cost of capital
Loan characteristics
Firm / Owner characteristics
Accounting information
Credit history

Relationship characteristics
Competition / Location measures
Other controls: SIC codes, regions, lender type
R2 of mean equation: 25%
Not particularly low & not due to low number of regressors

35

III. Omitted Variables (in Mean Equation)?

Does Variance Equation Capture Discretion ?
SSBF: include all available variables in mean
equation
Objective: Predict loan rates as accurately as
possible
Relevance of information: Information predicts 88% of
accept/reject outcome

Disregard collinearity problems

Results very robust to exclusion / inclusion of
variables
Admittedly surveyor knows less than bank itself
36

III. Omitted Variables (in Mean Equation)?

Does Variance Equation Capture Discretion ?
Belgian sample
Used in Degryse & van Cayseele (JFI 2000), Degryse &
Ongena (JF 2005, JFI 2007), Degryse, Laeven &
Ongena (RoF 2009)
Loan portfolio of a large bank (over 15,000 loans)
Contains all information observable to the bank
Ensures that discretion relates to pricing decisions
made by the loan officer!
Results are surprisingly similar to SSBF

37

III. Robustness Tests Mean Equation

Model specification
Non-linear terms: squared, cubed
Add linearly predicted loan rates squared and
cubed ( la Ramsey RESET)
Regression Equation Specification Error Test

Splines: all continuous variables in ten

equally spaced splines
38

III. Robustness Tests Mean Equation

Add Dun and Bradstreet Credit Score percentile
Not publicly available hence loan officer may
not have it
Sample selection bias?
Probit selection model of granting loan
Drop all loan characteristics, add Loan amount
requested

39

III. Robustness Tests Mean Equation

Bank heterogeneity
Confidential bank information obtained from Federal
Reserve Board!
100 bank size dummies (fixed effects not possible)

Branch heterogeneity in Belgian sample

One bank
(Somewhat) different institutional setting

Jointness of loan terms (Brick & Palia, JFI 2007)

Drop loan amount, collateral, maturity and type
Fully jointly with variance equation: not so straightforward

40

III. Variables in Variance Equation

Parsimonious specifications
Variables motivated by theoretical ideas
as much as possible
admittedly no structural testing

Robust to many additions / removals of control

variables in variance equation
Remember robustness to all alterations in mean equation
41

III. Variables in Variance Equation

Discretion a product of market imperfections
Information search costs Stigler (JPE 1961)
Information asymmetries von Thadden (FRL 2004), Stigler (JPE 1961),

Reinganum (JPE 1979), MacMinn (JPE 1980), Varian (AER 1980), Baye and Morgan
(AER 2001)

Firm opaqueness Petersen & Rajan (QJE 1995)

Strength of firm-bank relationship Petersen & Rajan (JF 1994), Berger
& Udell (JB 1995)

Firm-bank distance

Hauswald & Marquez (RFS, 2005)

Competitive structure of banking markets

Market concentration
(QJE 1995)

Hannan (JBF 1991, RIO 1997), Petersen & Rajan

42

III. Information Search Costs

Borrowing firm wants more information and to
price shop if loan is large for example
More competition among possible lenders
possibly makes loan pricing more
homogeneous
ln(Loan Amount)

43

III. Information Asymmetries

Effect of opacity on discretion?
Positive: variance of loan rates increasing function of
uncertainty about quality of pool of borrowers
Model simulations of von Thadden (FRL 2004) /
Rajan (JF 1992) (as in Black 2006)
Variance of loan rates:
decreases in average repayment probability in credit market
increases in difference in repayment probability between
good and bad firms
changes non-monotonically in proportion of good firms (1/2)

Uncertainty increasing in distance: Hauswald &

Marquez (RFS 2005)
44

III. Information Asymmetries

Effect of Opacity on Discretion?
Negative: Lenders may try to avoid giving
away information by pricing long-time
borrowers too differentially (Gan & Riddiough,
RFS 2006)

Ex ante: Collateral, Minority, Corporation,

Clean Record, IRS Problem, and ln(Duration)
Ex post: ln(Distance)
45

III. Concentration & Size of Markets

Possibly more discretion in:
more concentrated banking markets (Petersen
and Rajan, QJE 1995)
larger banking markets: greater diversity of
underwriting procedures and by higher search
costs (Varian, AER 1980)

Concentrated and MSA

46

Effect seem ultimately

unaddressed empirical question

47

IV. Results of Variance Equation

(1993 NSSBF)

48

IV. Loan Amount

49

IV. Borrower Opaqueness / Switching Costs

Variable

S.e. ()

Ln(Loan Amount)

-0.27 ***

0.02

Loan is Secured (0/1)

-0.18 **

0.08

Firm is a Corporation (0/1)

-0.24 ***

0.09

Ln(Age of the Firms Owner)

0.39 ***

0.13

Firm not Owned by Minority (0/1)

-0.34 ***

0.13

Firm with Clean Legal Record (0/1)

-0.25 ***

0.09

Firm with no IRS Problem

-0.16 **

0.07

Duration of Firm-Bank Relationship

-0.12 **

0.05

Concentrated Banking Market (0/1)

0.10

0.08

Firm Located in MSA (0/1)

0.18 **

0.09

Ln(Firm-Bank Distance)

0.10 ***

0.02

Number of observations

1,425
50

IV. Public Information about the Firm:

Age of the Owner

51

IV. Private Information about the Firm:

Duration of Firm-Bank Relationship

52

IV. Economic Significance of Results

53

IV. Discontinuity in Discretion?

Decompose Continuous Variables
in Three (Five) Splines
Loan Amount: reflecting search costs?
< 47,000 $: strongly negative
> 325,000 $: moderately negative
47,000 $ < < 325,000 $: no effect

Duration of Firm-Bank Relationship

< 3 years: negative
> 10 years: negative
3 < < 10 years: positive

Ln(Firm-Bank Distance)
< 2 miles: positive

Ln(Age of the Firms Owner)

all positive

54

IV. Asymmetric Deviations?

A quantile (median) regression of Loan Rate on the set of
explanatory variables used in the loan-pricing model
Robust to skewness in the distribution of residuals
For example: if positive deviations from the true pricing model
are more sizeable than the negative ones, then a mean
regression model will underestimate positive deviations and
overestimate the negative ones

Dummies Rip-off and Bargain: estimated disturbance is

larger or smaller than one standard deviation above or
below the sample mean of the residuals (the estimated
value is 17 bp)
Estimate logit regressions of the variables Rip-off and
Bargain on the same set of variables we employ in the
variance analysis
55

56

IV. Has Discretion Varied Over Time?

Empirical Test:
Merged samples 1993, 1998 and 2003 SSBF
Variance equation:

lni = 1Zi + 2t + 3Zit

57

Table 5

58

IV. Discretion Changed Over

Time
Results:
Discretion decreased for small loans to
opaque businesses Berger, Frame & Miller (JMCB 2005)
More discretion in periods of high liquidity
(i.e., low prime rate)
Rajan (EFM 2006), Jimnez, Ongena, Peydro & Saurina (ECMA
Forth), Ioannidou, Ongena & Peydro (2013)

59

V. Conclusions
Heteroscedastic model identifies determinants of
unexplained dispersion of loan rates (discretion)
Discretion increases with...
Borrower opaqueness (Switching costs)
Public information about the firm
And decreases in...
Loan size (Information search costs)
Prime Rate
Discretion has decreased over the last 15 years for
small loans to opaque firms

60

Many Possible Other Applications

Like finding out what loan officers actually do ...

61

62