SMEs, Growth, and Poverty Thorsten Beck, Asli Demirg

1
SMEs, Growth, and PovertyThorsten Beck, Asli
Demirgüç-Kunt, Ross Levine

• Discussion by Aart Kraay
• The World Bank
• October 14, 2004

2
What the Paper Does

• Measures of importance of SMEs, and quality of
  overall business environment (BE)
• Which matters more for growth SME or BE?
• Which matters more for changes in inequality
  SME or BE?

3
What I Liked About the Paper

• Paper asks a great question!
• Very useful questioning of conventional wisdom
• 10 bn is a lot of money to spend without benefit
  of evidence!
• Heroic efforts at cross-country data collection!

4
What We Need More of

• High-quality evidence from micro data on role of
  firm size
• but this is hard to doand hard to generalize
• Example 1 Firm Size and Productivity
• TFP is Production / Inputs Q/L
• We measure Revenue / Costs (PxQ)/(WxL)
• OK if all firms face same P, W
• But they dont! Market power gt Upward bias for
  large firms
• Example 2 Firm Size and Job Creation
• SME birth creates lots of jobs but SME death
  destroys lots of jobs too
• To capture this you need a census that credibly
  tracks entry and exit (including possibly into
  the informal sector)
• This kind of data hard to find in developing
  countries

5
What I Didnt Like Identification Strategy

• Paper rightly is concerned with direction of
  causation
• do SMEs cause growth or does growth create
  opportunities for SMEs?
• does BE spur growth or do perceptions of BE
  reflect growth?
• Lots of possible omitted variables
• BE, SME have substantial measurement error
• All of these problems justify recourse to IV

6
Identification Strategy, Contd

• Proposed instruments (ethnic diversity, dummies
  for LAC, Transition, Africa) are correlated with
  BE, SME
• But it is very hard to believe exclusion
  restriction required to validate the instruments!
  
• Ethnic diversity matters for growth only through
  worse BE?
• Only reason transition economies grew slowly in
  1990s is because they had fewer SMEs?
• Africas difficult geography and bad institutions
  dont matter for growth?
• What about all the other endogenous RHS variables
  (including initial income by construction)?
• selective instrumentation does not deliver
  consistent estimates of any of the coefficients
  of interest

7
Taking IV Seriously

• Table 4 Business Environment and Growth
• OLS ? 0.73
• IV ? 2.72
• Four possible explanations for IV gtgt OLS
• Perverse reverse causation
• Peculiar omitted variables
• Enormous measurement error in BE
• Invalid exclusion restrictions

8
Taking IV Seriously, Contd

• Perverse Reverse Causation
• Conventional wisdom (growth raises BE) implies IV
  lt OLS
• To justify IV gtgt OLS need high growth to cause
  much worse BE why might this be?
• Peculiar Omitted Variables
• Many likely candidates (e.g. good institutions)
  raise growth and improve BE, this implies IV lt
  OLS
• To justify IV gtgt OLS need omitted variables that
  raise growth and lower BE, or vice versa what
  could they be?

9
Taking IV Seriously, Contd

• Enormous Measurement Error in BE
• Attenuation bias
• ?(OLS) ? VAR(True BE)/VAR(Measured BE)
• ?(IV) 3 x ?(OLS) implies variance of
  measurement error in BE is twice variance of true
  BE
• This means BE measures are virtually
  uninformative! e.g. suppose observed BE is 1 SD
  above the mean, best forecast of true BE is that
  it is only 0.33 SD above mean

10
Taking IV Seriously, Contd

• Invalid Exclusion Restriction
• True model is
• Growth ? BE e
• BE ? ELF v
• Instrumental variables regression delivers
• ?(IV) ? (COV(e, ELF)/VAR(ELF))/ ?
• Bias is slope of regression of structural error
  (e) on the instrument (ELF), scaled by ?
• Does ELF matter for growth holding constant BE?
  Quite plausibly (higher inequality, higher social
  conflict, many other channels)

11
Summary

• Great question very relevant for policy!
• Firm-level data will help to get closer to
  definitive answers on role of SMEs in growth and
  poverty but doing this right will be hard
• Identification in cross-country (or
  cross-anything) regressions is hard. Useful to
• take it as seriously as possible
• recognize that there is only so much orthogonal
  and independent variation across countries in the
  things we care about (so many good things go
  together across countries!)