Transcript Forum for the Free Flow of Information Organizational Launch Meeting

INFORMATION POLICY
INSTITUTE
Summary of Research Findings: Full
File Credit Reporting in Latin America
By Michael Turner, Ph.D.
January 16, 2006
What is Known and Unknown
About Credit Reporting I
Without payment information, lenders face three
problems:



“asymmetric information”: lenders are relatively ignorant of
risk profiles and make bad decisions
“adverse selection”, lend excessively to high-risk
consumers
“moral hazard”, have a hard time dissuading defaults
because of difficulty of punishing defaulters
Benefits of a well-structured system of information
sharing:



consumers: lower prices for credit, wider access
lenders: increased profits, larger markets
economy: a more stable financial sector.
2
What is Known and Unknown
About Credit Reporting II
Research has established that credit reporting is better for private
sector lending and loan performance than no reporting but also:


reporting full-file (positive and negative information) is better than
reporting only negatives, and
in a full-file setting, private bureau information leads to better
performance of consumer and small-business loan portfolios than
public bureau information.
Unknown: the degree to which differences in participation in a fullfile reporting affect the financial sector.


But how much participation is needed?
In other words, how much more accurate are assessments of risk
when lenders have more information?
Measure effect of participation to convince
furnishers to provide information
3
Why These Objectives?
No legal hurdles to full-file reporting in Latin America, but
substantial differences in participation rates
Research can resonate in Latin America because:




Economic development depends generating savings, allocating
capital, and transforming risk, especially with end of state-led
models of development.
Can assist in improving the efficiency of the financial sector,
which has been relatively inefficient
Can expand private sector lending, which has been relatively
stagnant
Can help to reduce the chances of financial crises, which has
been chronic
4
Methodology:
Two Ways to Show Benefits
1.
Statistically compare the private lending sector in economies with
different reporting systems (private full-file, private negative-only,
public full-file, public negative-only) and different participation rates
2.
Simulations using 5+ million complete files from “close” or similar
economy (Colombia) to:



Generate 4 scenarios

75% provide positive and negative information, 25% only negative

50% provide positive and negative information, 50% only negative

25% provide positive and negative information, 75% only negative

100% provide only negative information
Using generic commercial scoring model (ACIERTA) and reoptimzed
research-grade model, test impact of changes in participation on

market size,

loan performance, and

the distribution of credit among genders and age groups
Compare positive payment information vs. rich socio-demographic
information (as in Costa Rica)
5
Estimations
(Tests of the Impact of Participation)
No direct data on participation rates. Coverage (%) of adults in public
and private credit registries (World Bank data) as proxy.
Controls for estimations:




GDP per capita at purchasing power parity
Economic growth rates
Index for legal rights of borrowers and lenders*

Measures how much collateral and bankruptcy laws facilitate
lending

Based on studies of collateral and insolvency laws

Includes 3 aspects related to legal rights in bankruptcy and 7
aspects found in collateral law.
Index measuring depth of credit reporting† derived based on scope,
accessibility and quality of credit information, based on 6 factor
* (i) Secured creditors are able to seize their collateral when a debtor enter reorganization (ii) Secured creditors, rather than other parties such as government or
workers, are paid first out of the proceeds from liquidating a bankrupt firm. (iii) An administrator, not old management, is responsible during reorganization. (iv)
General, rather than specific, description of assets is permitted in collateral agreements. (v)General, rather than specific, description of debt is permitted in
collateral agreements. (vi) Any legal or natural person may grant or take security in the property. (vii) A unified registry that includes charges over movable
property operates.(viii) Secured creditors have priority outside of bankruptcy. (ix)Parties may agree on enforcement procedures by contract. (x) Creditors may
both seize and sell collateral out of court.
† (i)
Full-file; (ii) information on individuals and firms, (iii) contains information from retailers, trade creditors and financial institutions, (iv) has more than 2 years of data, (v) loans above 1% of
per capita income are reported, and (vi) borrowers have right to access data.
6
Estimations: Private Full-File Coverage
and Private Sector Borrowing
VARIABLE
Constant
Log of GDP per capita
(adjusted for PPP)
Avg. Change in GDP
(1995-2004)
Legal Rights of Creditors
(from 0 to 10)
Credit Information
(from 0 to 6)
Private Full-file Coverage
(0 to 100, as percentage of adults)
Private Negative-only Coverage
(0 to 100, as percentage of adults)
Public Full-file Coverage
(0 to 100, as percentage of adults)
Public Negative-only Coverage
(0 to 100, as percentage of adults)
R squared
F-stat
(p value)
Residual Standard Error
N
I
II
III
IV
-142.40***
(35.31)
20.31***
(4.65)
-1.20*
(0.70)
4.55**
(2.07)
-3.87
(2.88)
0.72***
(0.20)
-0.02
(0.86)
-0.11
(0.41)
0.16
(0.46)
-139.48***
(35.49)
18.37***
(4.45)
-0.82
(0.64)
4.99**
(2.06)
-133.97***
(35.41)
17.38***
(4.41)
-130.80***
(32.20)
16.85***
(3.87)
4.68**
(2.06)
4.80**
(1.97)
0.60**
(0.18)
-0.13
(0.46)
-0.26
(0.40)
-0.01
(0.86)
0.66***
(0.17)
-0.06
(0.46)
-0.17
(0.39)
-0.09
(0.86)
0.67***
(0.16)
0.7075
16.93
(1.88e-012)
29.45
65
0.698
18.82
(9.65e-013)
29.65
65
0.6895
21.46
(4.251e-013)
29.81
65
0.6883
44.9
(1.887e-015)
29.12
65
100% full-file, private bureau coverage can increase lending by more
than 60 percentage points of GDP over 0% coverage
* p < 0.1
** p < 0.05
***p < 0.01
7
Simulations: Change in
Acceptance Rates (Market Size)
ACCEPTANCE RATE*
Share of furnishers providing positive and negative information
Target Default rate
100%
75%
50%
25%
0%
3%
10.00%
6.64%
4.73%
4.80%
2.56%
5%
41.35%
28.96%
19.28%
9.69%
5.15%
7%
58.82%
45.59%
36.42%
25.71%
13.60%
10%
73.06%
68.09%
68.08%
68.09%
54.97%
12%
77.80%
77.21%
76.49%
75.06%
72.26%
At 5% target default rate--roughly non-performing loans rate in
Colombia-- the acceptance rate (market size) drops:
o by 53.4% when 50% of all data furnishers are providing only
negative information (from 41.35% to 19.28%)
o by 30% when even only 25% of data furnishers are providing
only negative information
For a healthy target default rate, market size drastically
shrinks with a loss of positive information.
8
*Full sample (5.1 million files)
Simulations: Change in
Default Rates (Profitability)
DEFAULT RATES
Share of furnishers providing positive and negative information
Target
Acceptance Rate
20%
30%
40%
50%
60%
100%
3.52%
4.12%
4.89%
5.86%
7.20%
75%
3.72%
4.62%
5.66%
6.70%
7.73%
50%
4.66%
5.74%
6.67%
7.49%
8.49%
25%
5.91%
6.78%
7.52%
8.22%
9.25%
0%
8.46%
9.06%
13.85%
14.40%
15.30%
At 40% target acceptance rate, the default rate (90+ days past due)
increases:
o from 4.89% to 6.67% (an increase of nearly 2 percentage points)
that is, by 36.4%, when only 50% of furnishers provide positive
information
o from 4.89% to 5.66% (an increase of nearly 1 percentage point)
that is, by 15.7%, when even 75% of furnishers provide positive
information
Loan performance worsens significantly
with less positive information
9
Simulations: Participation Rates &
Acceptance-Default Trade-Offs
15%
Default Rates
12%
9%
6%
3%
0%
0%
15%
30%
45%
60%
75%
90%
Acceptance Rates
100% Reporting Full File
75% Reporting Full File
25% Reporting Full File
0% Reporting Full File
50% Reporting Full File
10
Simulations: Change in Acceptance
Rates by Demographic Segment
Acceptance rates fall but
unevenly across sociodemographic groups
For a 7%
default rate
Scenario
Male
Female
Age categories
0-32
32-42
42-50
50-57
57+
100%
ACCPETANCE RATE
Share of furnishers providing positive and negative
informa tion
100%
75%
50%
25%
0%
64.92% 51.40% 44.31%
33.68% 10.99%
53.13% 42.24% 33.43%
22.30%
5.10%
16.48%
49.72%
58.31%
62.76%
77.13%
75%
15.47%
44.75%
45.20%
52.02%
72.98%
14.20%
28.42%
30.52%
39.61%
69.54%
50%
8.61%
13.71%
19.14%
19.13%
66.49%
25%
0.90%
7.67%
12.84%
13.00%
20.01%
0%
Women as a share of
borrowers declines as positive
information is lost
Those between 32-50 as a
share of borrowers declines as
positive information is lost
11
Worsening
Model Performance
SCALED K -S, Predictiveness
Share o f fur nishers providing positive and negat ive inf ormation
(with the remaind er providing sole ly negat ive informatio n)
Scenario
100%
100.00
75%
92.42
50%
90.27
25%
87.67
0%
86.78
CHANGES IN ERROR RATES
Change (+/-) in share of fur nishers providing positive and negative informatio n
over the full-fi le scenario (as percent of all credit-eligible adults)
75%
50%
25%
Type I (false positives)
Type II (fals e negatives)
+1.00%
+3.81%
+2.22%
+5.32%
+3.31%
+7.53%
Important to note: Ability to tell good risks from bad ones
worsens--why performance and market size deteriorate
12
Evaluating Payment History vs.
Socio-Demographic Information
Objective: to test and compare the Costa Rican model, which relies on extensive
socio-demographic information and only negatives, to the Colombian one of
extensive positive information.
For both countries, we created a hypothetical files made of common variables:
o a “Costa Rican restricted” purged of socio-demographic information not
present in the Columbian files, and
o a Colombian “negative only”
Research-grade scoring models were developed for these two sets. Another model
was developed to score the complete Costa Rican files. The results are then
compared:
o the “Costa Rican restricted” were compared to the Costa Rican complete
files;
o the Colombian negative-only compared to the Colombia full-file, ACIERTA
instance; and
o the differences in K-S score differences in the two sets
13
Positive Payment Info Provides More
Lift than Socio-Demographic Info
K-S Scores: Negative Only Simulations,
Costa Rica and Colombia
Costa Rica Restricted
40.5
Costa Rica Complete
49.3
Colombia Negative Only
Colombia Full-File (ACIERTA)
54.2
67.3
Measure the relative merit of approaches with K-S, the ability to discern
goods from bads (or true positives from false positives).
(CAUTION: Measure differences are simply suggestive, not an overall
indication of the magnitude of differences.)
K-S increases considerably in moving from the Colombian negative only
to the Colombian full-file scenario. (+13.1)
By contrast, socio-demographic information improves the ability to
distinguish goods from bads in Costa Rica files by much less of a
degree. (+8.8)
14
INFORMATION POLICY
INSTITUTE
306 Fifth Ave, Penthouse
New York, NY 10001
www.infopolicy.org
Phone: (212) 629 -4557