Download presentation
Presentation is loading. Please wait.
Published byBernard Patrick Modified over 9 years ago
1
Estimating the Firm-Level Growth Effects of Small Loan Programs Using Universal Panel Data from Romania J. David Brown (US Census Bureau) John S. Earle (Upjohn Institute and CEU) June 2010
2
Motivation: small firms and the crisis What are the prospects for a “small business- fueled employment recovery”? What are the prospects for a “small business- fueled employment recovery”? Recent credit boom was smaller for small/young firms, and current credit crunch is worse Recent credit boom was smaller for small/young firms, and current credit crunch is worse New policy proposals around the world. In US: New policy proposals around the world. In US: SBA stimulus SBA stimulus Sen. Warner: Fed and TARP funds to small firms– including loss-sharing Sen. Warner: Fed and TARP funds to small firms– including loss-sharing FDIC: matching loans to small business FDIC: matching loans to small business
4
Do small business loan programs promote growth? Conceptually ambiguous: Conceptually ambiguous: Easier access to finance may enable expansion Easier access to finance may enable expansion But funds may be used for other purposes But funds may be used for other purposes Displacement and substitution effects Displacement and substitution effects Empirically difficult (absent an experiment): Empirically difficult (absent an experiment): Many factors influence firm growth (industry, region, size, age…) Many factors influence firm growth (industry, region, size, age…) Need long time series on factors and outcomes – pre and post Need long time series on factors and outcomes – pre and post Selection bias – loan could reflect growth potential Selection bias – loan could reflect growth potential Many studies of firm growth, but no rigorous evaluations Many studies of firm growth, but no rigorous evaluations N.B.: few such policy evaluations at firm-level more generally (except Jarmin, 1998) N.B.: few such policy evaluations at firm-level more generally (except Jarmin, 1998)
5
Broader question: growth and finance Do well-functioning financial markets enhance growth, or does economic growth improve financial markets? (Rajan & Zingales 1998; Beck et al. 2000; Fisman & Love 2007) Do well-functioning financial markets enhance growth, or does economic growth improve financial markets? (Rajan & Zingales 1998; Beck et al. 2000; Fisman & Love 2007) Macro debate: relationship of real and monetary economy Macro debate: relationship of real and monetary economy Aggregate studies of loans and growth find different results for US states and Euro-zone countries (Driscoll 2004; Cappiello et al. 2010) Aggregate studies of loans and growth find different results for US states and Euro-zone countries (Driscoll 2004; Cappiello et al. 2010) Relatively little micro evidence, especially rigorous estimates of causal effects Relatively little micro evidence, especially rigorous estimates of causal effects
6
Small business and finance in transition and development Transition Transition IFIs: size of new private business indicates progress IFIs: size of new private business indicates progress Policy debate: finance versus property rights, contracts, regulation Policy debate: finance versus property rights, contracts, regulation Development Development Microcredit is fashionable but few estimates of firm- level growth effects (many of repayment & poverty, Morduch 1999; Karlan&Morduch 2009) Microcredit is fashionable but few estimates of firm- level growth effects (many of repayment & poverty, Morduch 1999; Karlan&Morduch 2009) Alternatives: technical assistance, business environment Alternatives: technical assistance, business environment
7
Our case: small business loans in Romania USAID-supported programs through March 2001 USAID-supported programs through March 2001 Small size: 372 firms (=> rationing, few spillovers) Small size: 372 firms (=> rationing, few spillovers) Partial coverage: 18/41 counties (=> ineligibles) Partial coverage: 18/41 counties (=> ineligibles) Loan conditions: Loan conditions: “Commercial terms”; lenders “profit-oriented” “Commercial terms”; lenders “profit-oriented” Decisions based on past years’ accounting cash-flow Decisions based on past years’ accounting cash-flow State-owned and some sectors ineligible, startups not immediately eligible State-owned and some sectors ineligible, startups not immediately eligible Romanian context: credit markets poorly developed Romanian context: credit markets poorly developed For most recipients, first access to formal credit For most recipients, first access to formal credit
8
Estimating effect of first international loan on growth (ATT): Our method Construct two control groups from universal panel data Construct two control groups from universal panel data Eligible non-recipients (same county) Eligible non-recipients (same county) Ineligibles (non-USAID counties) Ineligibles (non-USAID counties) Match on several years of pre-loan characteristics Match on several years of pre-loan characteristics Outcomes: growth (employment & sales); survival Outcomes: growth (employment & sales); survival Panel DiD regressions using matched samples, 1992- 2006 Panel DiD regressions using matched samples, 1992- 2006 Pre- and post-dynamics of the effect Pre- and post-dynamics of the effect Pre-loan: diagnose selection bias (“pseudo-outcomes”) Pre-loan: diagnose selection bias (“pseudo-outcomes”) Post-loan: long- versus short-term effects Post-loan: long- versus short-term effects
9
Data List of 372 firms receiving a USAID loan by March 2001 – most in 1999-2000 List of 372 firms receiving a USAID loan by March 2001 – most in 1999-2000 Annual balance sheet information for universe of registered firms from 1992-2006: about 200,000 firms per year Annual balance sheet information for universe of registered firms from 1992-2006: about 200,000 firms per year Exclusions Exclusions All “old” firms (ever have any state ownership) All “old” firms (ever have any state ownership) Ineligible industries (tobacco, weapons) Ineligible industries (tobacco, weapons) >49 employees (only small and micro start-ups left) >49 employees (only small and micro start-ups left)
10
Matching Heterogeneity: recipients versus nonrecipients Industry (manufacturing) Age (older) Size (smaller in early years, larger later) Exact matching Always: 2-digit industry, age, year Sometimes: county, +/- 10% t-1 outcome, 3-digit ind Propensity score matching Lagged outcomes(to t-4), other characteristics Nearest neighbor, radius, kernel methods
11
Control Groups Same county (eligible non-recipients) Exact match on county Selection problem (applicants and loan officers) Non-USAID county (ineligibles) No matching on county No self-selection problem Possible program selection & heterogeneity P-scores estimated from relationship in USAID counties
12
Romanian counties with USAID loans
13
Specification Checks Identifying assumption: unconfoundedness Balancing tests for covariates Rosenbaum-Rubin standardized differences (bias) t-tests Hotelling’s T 2 test by P-score quintiles Smith-Todd regression test “Pseudo-outcome” (Imbens-Wooldridge) tests Pre-treatment outcomes (Heckman-Hotz 1989) Estimation using two control groups (Rosenbaum 1987; Heckman et al. 1997)
14
Pseudo-outcome test: 2 control groups Definitions Y i (1), Y i (0) = outcomes for treatment, non-treatment G i ∊ {-1, 0, 1} (2 control groups, 1 treatment group) W i = 0 if G i = -1, 0; W i = 1 if G i = 1 Unconfoundedness requires Y i (0), Y i (1) ╨ W i |X i Stronger condition: Y i (0), Y i (1) ╨ G i |X i => Y i (0) ╨ G i | X i, G i ∊ {-1, 0} Test: E[E(Y i |G i =-1, X i ]- E[E(Y i |G i =0, X i ]=0
15
Characteristics: Employment Distribution (1999) Number of Employees USAIDNon-USAID 0-961.4%83.5% 10-4931.9%12.1% 50-2496.8%3.3% 250+0%1.0% Total Firms339218,759
16
Start-up Year (%) Full SamplesTruncated Samples USAIDNonUSAIDUSAIDNonUSAID 199225.7522.5724.6823.30 199314.6316.7614.9416.21 199422.2217.9923.0518.38 199514.9114.1215.2613.44 19967.326.737.146.94 19978.407.108.777.26 19984.348.054.228.53 19992.446.671.955.94
17
Exit Year (%) Full SamplesTruncated Samples USAIDNonUSAIDUSAIDNonUSAID 20000.815.090.00 20012.446.422.273.63 20022.986.182.923.28 20031.631.301.621.24 20044.8810.455.526.64 20055.155.415.194.58 20065.159.365.848.21 2007+76.9655.8076.6272.41
18
Distribution of Year of First International Loan Number of FirmsPercent of Firms 199310.3 199410.3 199541.1 199671.9 199782.2 19986216.9 199920054.4 20008222.3 200130.8 N368100.0
19
NN matching) Results: Estimates of the Loan Impact on Employment ( NN matching) Same county controls OLS OLS with covariates FE Post-Loan Dummy 0.2320.2020.168 (0.056)(0.048)(0.045) Number of Treated Firms 203 Number of Observations 3,956
20
NN matching) Estimates of the Loan Impact on Employment ( NN matching) Non-eligible controls OLS OLS, covariates FE Post-Loan Dummy 0.2950.2420.234 (0.044)(0.041)0.046 Number of Treated Firms 267 Number of Observations 5,203
21
Estimates of the Loan Impact on Employment (Kernel matching) Same County MatchesNon-eligible Matches OLSFEOLSFE Post Loan0.242***0.221***0.323***0.176*** (0.034) (0.041) Age0.157***0.139*** (0.017)(0.023) Age 2 -0.011***-0.010*** (0.001)(0.002) Firms6,026 52,373 Obs57,40757,592503,658504,794
22
Estimates of the Loan Impact on Sales (Kernel matching) Same County MatchesNon-eligible Matches OLSFEOLSFE Post Loan0.326***0.295***0.601***0.376*** (0.044)(0.051)(0.052)(0.058) Age0.180***0.165*** (0.031)(0.034) Age 2 -0.013***-0.015*** (0.002) Firms4,209 48,182 Obs46,980 521,993
23
Dynamics of Employment Effect (Same county controls)
24
Dynamics of Employment Effect (Non-eligible controls)
25
Dynamics of Sales Effect (Same county controls)
26
Dynamics of Sales Effect (Non-eligible controls)
27
Estimates of Loan Effects on Exit (Cox proportional hazards) Log-odds ratios Same county matchesNon-eligible matches no E t-1 restriction E t-1 within 10% no E t-1 restriction E t-1 within 10% Without covariates 0.9110.9540.9451.120 (0.150)(0.166)(0.190)(0.242) With covariates 0.9190.9360.9771.137 (0.154)(0.165)(0.204)(0.252)
28
Conclusion Results suggest loans have long-lasting effects on job and sales growth, but little on survival Results suggest loans have long-lasting effects on job and sales growth, but little on survival Evidence of causal link: finance -> growth Evidence of causal link: finance -> growth Mechanism unclear Mechanism unclear lower cost of capital lower cost of capital alleviate credit rationing alleviate credit rationing open access to formal credit markets open access to formal credit markets Approach (data, methods) widely applicable Approach (data, methods) widely applicable Loan programs in other countries (SBA) Loan programs in other countries (SBA) Other policies with differential effects on firms Other policies with differential effects on firms
29
Implications for the US? SBA Project Previous Research: Studies of local employment growth and per capita income as a function of the amount of SBA loans, but not effect on loan recipients Studies of local employment growth and per capita income as a function of the amount of SBA loans, but not effect on loan recipients Urban Institute (2008) study for SBA Urban Institute (2008) study for SBA Dun & Bradstreet data Dun & Bradstreet data Incomplete coverage of small firms Incomplete coverage of small firms Biased toward larger, more successful recipients Biased toward larger, more successful recipients No control group of non-recipients No control group of non-recipients
30
SBA Project (with Census Bureau) SBA has detailed data on loan recipients: SBA has detailed data on loan recipients: Name and address Name and address Loan date Loan date Loan amount Loan amount Interest rate Interest rate Credit score at time of application Credit score at time of application Demographic information about borrower Demographic information about borrower Loan performance Loan performance Also some data on rejected applicants Also some data on rejected applicants
31
SBA Project (with Census Bureau) Link SBA to Census data for all establishments in 1976-2008 on age, employment, payroll, industry code, etc. Link SBA to Census data for all establishments in 1976-2008 on age, employment, payroll, industry code, etc. Subset of the Census Bureau establishments have sales, capital stock, and profit Subset of the Census Bureau establishments have sales, capital stock, and profit Match to select controls most similar to loan recipients prior to loan Match to select controls most similar to loan recipients prior to loan Compare performance of recipients and non- recipients before and after the loan Compare performance of recipients and non- recipients before and after the loan 2 nd control group: rejected applicants 2 nd control group: rejected applicants
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.