Transcript Partial Credit Guarantee Schemes Conference

Partial Credit Guarantee Schemes
Conference
Discussion Session V
Erik Feyen, World Bank
March 14, 2008, Washington, D.C.
The Big Picture:
Why do governments use CGs?
•
One of main goals of CGs: Spur development by targeting SMEs
– Information issues, externalities, etc. (Honohan)
•
Papers study CG schemes in the US and Japan over time
•
Both papers find that CGs increase the supply of loans, more so than
increases in bank capital…
•
...And both (implicitly) find positive association with economic outcomes
– More directly tested in the US paper and postulated in the Japan paper
•
However, these papers describe schemes in two completely different
settings...
•
…so besides Spur, they seem to imply two other (related) motives:
– Stabilize access to finance since CGs less sensitive to the credit cycle
– Salvage insolvent companies (“zombies”)
Emphasis differences
Emphasis
Success factors
SPUR
Show additionality
Address moral hazard
(like most papers in this conference)
STABILIZE
Reduce impact of monetary tightening and lower growth of
lending scheme
(US paper)
SALVAGE
Bail out insolvent companies
(Japan paper)
Stabilize:
The US experience
• Context of Small Business Administration:
– Guaranteed up to around $13 billion in 2005 (could be
described a bit better)
– Collateral needed
– Screening: Coverage up to 80%; prior other sources needed
• Methodology:
– Aggregated data up to the state-level for period 1990-2000
– Simple OLS, but taking first differences amounts to fixed effects
• Main results:
– SBA lending less cyclical
• Less responsive to bank capital and income
• Rises if delinquencies rise
– SBA negatively associated with failures and bankrupties (not
significantly though)
– Channel to affect SMEs mainly through small banks
– SBA positively associated with employment, # of firms, payroll
US: Questions
•
Methodology:
– Endogeneity: why not an IV approach like in Japan paper?
– SBA loan disbursements: not just a proxy for total (SME) loans?
– Why not use county-level data to exploit more sources of variation?
Data available to pinpoint channel better
•
Statistical issues:
– Interpretation of results: puzzling signs cannot be easily addressed by
just counting the “right” number of signs
– Should use formal tests to compare results in different regimes?
– Why no interaction effects to test effect for different climate?
– Isn’t an insignificant GSP variable for loans troubling?
– Clustering SE on the state level?
– Where are the descriptive statistics? Correlations?
•
Main results:
– SME intensity decreased and the scheme is not that big: Why should
there be a positive effect on aggregate welfare? (E.g. Italy is pretty
small: Ventura, Zecchini)
– Explain the link with lower default rates? No moral hazard?
– Does it really spur innovation? Interesting to link it to patents?
•
Context:
–
–
Economic turmoil; many insolvent companies
Japan scheme is huge
•
•
–
–
–
–
No seniority
No risk-based pricing
Japan low entry requirements (only not to be blacklisted)
Calculate LGD with and without guarantee
Comparative statics: ∂ l/ ∂ g>0, given dLGD=0 if r>e/l
•
Makes sense: it already bad, so there is only upside for the bank
Empirical methodology:
–
–
•
Arping et al. showed the negative consequence of that
High compared to world standards (Klapper, Beck, Mendoza)
Theory model:
–
–
•
(40% of SME loans covered; 10% of total bank loans; $250 billion, GDP is around $4
trillion)
Developed countries typically 0.3% of GDP; Asia 5% (Klapper, Beck, Mendoza)
Japan 100% coverage, lower capital requirements
•
•
•
Salvage:
The Japan Experience
IV estimates-> instrument initial share of guarantees
Bank-specific effects
Main results:
–
–
Massive increase of non-guaranteed loans b/c of guarantees
Capital injections had smaller effect than effect of guarantees
Japan: Questions and comparison to US
•
Theory model:
– Low net worth induces gambling; should be modeled too?
– Validity assumption of keeping LGD constant?
– Only works in practice when net worth is negative. Can you show the
numbers?
•
Empirical methodology:
– Why not also show simple regressions and the first stage to build
intuition?
– Why use stock of guarantees variables instead of the flow?
•
Statistical issues:
– Where are the normal IV tests (overidentifying restrictions, relevance,
etc.)
– Where are the descriptive statistics? Correlations?
•
Results:
– Low effort, bad pricing, risk shifting, bad screening, refinancing poor
loans: how can this be good for the economy?
• Interesting to see how this worked out!
– This was not a guarantee scheme, this looked like a bail out!
• “True” additionality seems impossible
Partial Credit Guarantee Schemes
Conference
Discussion Session V
Erik Feyen, World Bank
March 14, 2008, Washington, D.C.