Transcript General issues

Loan Loss Provisioning and
Economic Slowdowns:
Too much, Too Late?
By
Luc Laeven and Giovanni Majnoni
Finance Forum 2002
June 19-21, 2002
A policy question…
Should banks make provisions when losses have
already materialized? affecting capital?
•
Traditional “capital crunch” effects may set
in and they may be stronger for emerging
economies;
•
Risk-weighted capital has already cyclical
effects that may be compounded by high
level of provisioning in bad times.
Probability
Probability
PDF of credit losses:
expected and unexpected losses, capital and LLP.
O
General
Credit risk capitalCredit risk capital
Provisions
99th percentile
General
Provisions
99th percentile
Expected
losses
Unexpected
losses
A
Loss
Expected
losses
O
B
Unexpected
losses
A
Loss
B
… and possible answers
•
•
•
LLP regulation in Spain: build up of LLR in
good times has income smoothing effects;
Minimum capital requirements should not lead
to minimum LLR requirements;
To consider general loan loss reserves simply
as a component of tier II capital may not be
appropriate.
Some positive aspects of LLP from a
regulatory perspective
•
•
If loan loss reserves are required to be held on
average over a maintenance period (increased during
upswings and depleted during downswings) they
may act as a buffer protecting capital from negative
shocks (Spain).
This behavior would better reflect the true
underlying profitability of a bank. Banks would set
aside every year (as provisions) the income from
credit risk premia, apparently smoothing reported
bank earnings over time but in reality properly
accounting for expected costs.
Some positive aspects of LLP from a
regulatory perspective
•
•
Volatility of reported earnings can be shown to derive
purely from accounting practices (i.e. from an
omitted registration of costs that inflates earnings)
and has no economic determinants.
Loan loss provisions are easier to measure than
capital (no correlation issues, smaller confidence
intervals);
Has income smoothing unequivocal
beneficial effects?
Due to the presence of information imperfections two
positions emerge:
• “Accounting profession” perspective:
–
•
Income smoothing is bad because it introduces judgmental
modifications of firm earnings that reduce the
comparability of results across firms
“Economic profession” perspective:
–
Income smoothing has evident benefits for risk averse
agents.
Do LLP follow an income smoothing pattern?
1. For the US, Greenwalt and Sinkey (1988) found an
average positive association between LLP and
operating profits coherent with an income smoothing
pattern;
2. More recently an empirical analysis of evidence for
European countries participating in the EMU (ECB,
2002) found a different patterns across countries but
on average a lack of income smoothing patterns.
The empirical questions
Income smoothing is not the only indicator of the
overall cyclical behavior of LLP we should look for :
1. Income smoothing: positive association of loan loss
provisions with pre-provision income (Do banks set aside
more in good times?);
2. Positive association of LLP with GDP growth (Do banks
condition LLP on the macroeconomic conditions?);
3. Positive association of LLP with bank loan growth (Do exante risk management consideration condition LLP
policies?);
The testing strategy
We adopt two empirical specifications to test for the
robustness of our findings to different model
specification:
• a static specification;
• a dynamic specification to test for the nature of
dynamic adjustment of LLP.
 LLP 
 EBP

  α  β1 
  β2ΔLi t  β3ΔGDPi t  β4 Tt  v i  εi t
 A i t
 A i t
The data
1. Banks’ balance sheets and income statements from
Bankscope;
2. Country-level variables such as GDP from World
Development Indicators, World Bank
3. 1,419 banks from 45 countries
4. Sample period: 1988-99
5. 8,176 bank-year observations
6. “Good” distribution across years and regions
Distribution of bank-year observations
Region name
Number of Bank-year
observations
Percentage of Total
951
11.64
Europe
2,477
30.29
Japan
1,016
12.43
570
6.96
USA
2,288
27.98
Total
8,176
100.0
Asia
Latin America
The testing strategy
Separate tests:
1. For the sample as a whole;
2. For regional groups of countries:
• Europe, US, Japan, Latin America, Asia;
3. With and without a non-linear income term.
Purpose:
1. Verify differentiated patterns across regions;
2. Avoid the average results being affected by
countries with more bank/years observations;
3. Test for behavior of more profitable banks.
Static specification
EBP/Assets
Negative earnings dummy
*(EBP/Assets)
Loan growth
GDP growth
Whole sample Negative earnings dummy
***.158
***.225
(.008)
(.009)
***-.640
***-.008
(.001)
***-.075
(.004)
(.034)
***-.008
(.001)
***-.072
(.004)
Main results: Static Specification
1. On average we observe an income smoothing
behavior: + 1 st. dev. of earnings leads to + 0.22 of
the provision ratio to total assets…
2. … but the interaction term related to periods of
negative profitability captures a significant anticyclical pattern at low or negative profits.
3. … and real GDP growth is negatively associated with
provisions: + 1 st. dev. of GDP growth leads to - 0.20
of the provisions ratio to total assets;
4. … and real loan growth is negatively related to
provisions: +1 loan growth st. dev. leads to - 0.13 of
the provisions ratio to total assets;
Static specification with regional
regressions
EBP/Assets
Negative earnings dummy
*(EBP/Assets)
Loan growth
GDP growth
Europe
US
Japan
***.209
(.019)
***-.373
***.207
(.014)
***-.541
(.074)
***-.008
(.001)
***-.070
(.013)
(.069)
***-.006
(.001)
***-.160
(.012)
***.545
(.059)
***-1.329
Latin
America
***.280
(.029)
***-.444
*.053
(.032)
***-.590
(.131)
***-.016
(.002)
***-.148
(.009)
(.100)
***-.009
(.003)
.000
(.016)
(.079)
***-.010
(.002)
***-.091
(.010)
Asia
Main results: Static specification with
regional regressions
1. Only Asia has a clearly anti-cyclical average
provisioning pattern;
2. But banks in all the regions – and in Japan and in
Asia especially - have an anti-cyclical pattern for
low levels of income;
3. The negative association between LLP and credit
growth is confirmed for all the regions and is
strongest in Japan;
4. Similarly a negative association with GDP growth is
found for all the regions but for Latin America.
Dynamic Specification
Whole sample
First lag of Prov/Assets
Second lag of Prov/Assets
EBP/Assets
Negative earnings dummy
*(EBP/Assets)
Loan growth
GDP growth
***.346
(.028)
***.095
(.022)
***.154
(.021)
-
Negative earnings
dummy
***.334
(.028)
***.109
(.021)
***.252
(.025)
***-.390
***-.003
(.001)
***-.077
(.006)
(.059)
***-.004
(.001)
***-.076
(.006)
Main results: Dynamic specification
At the aggregate level previous finding appear confirmed;
At the regional level:
• The income smoothing hypothesis is rejected for Asia and
for Japan;
• Banks located in Japan and in Asia have an especially, anticyclical pattern for low levels of income;
• The negative association between LLP and credit growth is
confirmed for Europe, Japan and Latin America;
• Similarly a negative association with GDP growth is
confirmed for US, Japan, and Latin America.
• Time dummies show a differentiated patterns across
regions with a reduction in the US case that is consistent
with changes in regulation.
Conclusions
1. Generalized evidence of anti-cyclical pattern of provisions
has emerged on average and across geographical regions;
2. Most of the time anti-cyclical patterns are generated by an
inadequate level of provisioning in the good phases of the
cycle consistent with a delayed recognition of potential
losses;
3. Independently from income smoothing at the bank level, the
empirical evidence points to an inadequate consideration of
risk elements (as proxied by loan growth) and macro
conditioning factors (GDP growth).
4. Each region has appeared to be exposed to at least one of the
three indicators of anti-cyclical pattern of LLP.
Conclusions
•
Overall the empirical evidence supports the need of
additional regulatory effort to insure that a more timely
provisioning for loan losses may insulate more effectively
bank capital from negative shocks during downturns.