Do Financial Factors Drive Aggregate Productivity? Evidence from Indian Manufacturing Establishments N. Aaron Pancost University of Chicago January 29, 2016
Motivation Does financial development increase economic growth? Log aggregate labor productivity in India: average agg productivity 8.5 9 9.5 10 1990 1995 2000 2005 2010 Year Indian financial reforms Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 1 / 8
What I Do Derive a model in which financial development can increase aggregate productivity, by reallocating resources to more productive uses. Derive implications for cross-sectional distribution of size and productivity. Identify financial shocks that re-allocate resources vs. common shocks to productivity using establishment-level microdata from India. mechanism Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 2 / 8
What I Find Common shocks to productivity, not financial development, explain the time-series evolution of the cross-sectional distribution of size and productivity from 1995 2011. details If finance is to explain aggregate productivity growth, it must affect productivity within firms, not allocation of resources across them. Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 3 / 8
Single Agent s Problem Assumptions: Productivity shocks are exogenous, idiosyncratic, and persistent. Firm owner-operators have finite intertemporal elasticity of substitution. Firms can borrow but default occurs in equilibrium and is priced. Results: Un-productive firms do not borrow. Among productive firms, more-productive firms choose higher leverage ratios and grow faster on average. In equilibrium, positive (endogenous) correlation between size and productivity. Correlation depends on level of financial development. model detail Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 4 / 8
Log Aggregate Productivity Decomposition Y i value added i = e z i L i L i employment i w i L i j L j Log Agg Prod log i i Y i i L i 1 N N w i z i = z i + N i=1 i=1 }{{} average productivity: Z ( (z i Z) w i 1 ) N } {{ } OP covariance: C details across countries Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 5 / 8
Identification Increases in collateral rate: Lower cost of capital highly-productive firms lever up and grow faster on average C increases. Idiosyncratic productivity unaffected Z constant. Increases in productivity: All firms more productive Z increases. Lower cost of capital highly-productive firms lever up and grow faster on average C increases. (Z, θ) }{{} interpretable model (Z, C) }{{} observable Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 6 / 8
Aggregate Productivity Decomposition 12 10 Reallocation Productivity Joint 8 6 %p.a. 4 2 0 2 1990 1995 1995 2000 2000 2006 2006 2011 year range data Pancost (University of Chicago) Financial Factors in Aggregate Productivity January 2016 7 / 8
Conclusion Derived a model in which financial development affects aggregate productivity through the allocation of resources across firms. Calibrated model to Indian microdata: financial development may have increased aggregate productivity only in 1990 1995. Other factors that affect within-firm productivity may be more important than re-allocative financial frictions. Informational barriers (Bloom et al 2013). Entry and financial constraints (Buera Kaboski & Shin 2011).
Time Series for Z Z 1.4 1.2 1 avg productivity 0.8 0.6 0.4 0.2 0 1989 1994 1999 2004 2009 2014 year
Plugging in the guess for the value function yields a + b log x = const + log x + β [ (1 π) b + π ] E log x }{{} =E{log x+const} so that b satisfies b = 1 + βπ 1 β (1 π) Define β β [ (1 π) b + π ] Then β = β 1 + β
Why Is This Hard? Macro evidence: causality Previous financial development predicts future growth. But markets are forward-looking. Instruments (country of legal origin): static, don t satisfy exclusion restriction. Micro evidence: impact Suggests mechanism. But is mechanism important for aggregate productivity? Measurement error?
Log Aggregate Productivity Approximation log i Y i i L = log L i Y i i i j L = log w i e z i j L i i ( ) f z = log ( w i e z i 1z ) ( ) ( ) f + f 1z z 1z i f ( ) 1 z = z j i w ie z w je z j = w je z i 1 e z z i w = w i j 1 z 1 z w ( ) f 1z = w 1 = w ( ) f z z + w ( z 1z ) = z + w z ( ) w 1 z = i w i z i
Time Series for World Bank Doing Business Recovery Rate 0.25 0.2 0.15 zeta 0.1 0.05 0 2004 2005 2006 2007 2008 2009 2010 2011 year
More Motivation Rajan & Zingales (1998): One way to make progress on causality is to focus on the details of theoretical mechanisms through which financial development affects economic growth, and document their working. Levine (2005): If finance is to explain economic growth, we need theories that describe how financial development influences resource allocation decisions in ways that foster productivity growth.
Robustness to η η Terminal θ Percent of Growth from θ % NSS Share 1990 1995 1995 2011 Model Data 0.3 0.29 70% -8.4 10.0% -14.0% -11.2% graph 0.4 0.33 71% -3.5 8.6% -12.3% -11.2% graph 0.5 0.38 71% 2.4 7.5% -11.1% -11.2% graph 0.6 0.45 71% 6.6 9.9% -10.0% -11.2% graph
Aggregate Productivity Decomposition, η = 0.3 14 12 Reallocation Productivity Joint 10 8 %p.a. 6 4 2 0 2 1990 1995 1995 2000 2000 2006 2006 2011 year range
Aggregate Productivity Decomposition, η = 0.4 12 10 Reallocation Productivity Joint 8 6 %p.a. 4 2 0 2 1990 1995 1995 2000 2000 2006 2006 2011 year range
Aggregate Productivity Decomposition, η = 0.5 12 10 Reallocation Productivity Joint 8 6 %p.a. 4 2 0 2 1990 1995 1995 2000 2000 2006 2006 2011 year range
Aggregate Productivity Decomposition, η = 0.6 12 10 Reallocation Productivity Joint 8 6 %p.a. 4 2 0 2 1990 1995 1995 2000 2000 2006 2006 2011 year range
Labor Productivity At the beginning of the period, after idiosyncratic shocks are realized, agent has total resources k. Working capital constraint: labor must be paid before production occurs. Agent s problem at start of period: max k,l Ae z k α L 1 α wl + k s.t. k + wl k (model setup) (model OP) timing
Labor Productivity Solution: m (z) L K solves and Labor productivity is Ae z [ 1 α w ] αm = 2m α k (z) = k/ [ 1 + wm (z) ]. Y L = Aez m 1 α k mk [ ] 1 α 1 = 2 w αm
Mechanism Growth & productivity correlated with financial intermediation. Growth of externally-dependent industries, constrained firms more correlated with financial intermediation. Suggests the following mechanism: financial development lower cost of capital improved resource allocation higher aggregate productivity more motivation
Across Countries: OP Covariance Growth Country change (1993 1996 1997 2001) United States 0.09 United Kingdom 0.06 Germany 0.14 Netherlands 0.11 Hungary 0.18 Romania 0.25 Slovenia 0.16 source: Bartelsman Haltiwanger & Scarpetta (2013) India 1990 1995 0.18 India 1995 2000 0.09 India 2000 2006 0.06 India 2006 2011 0
Indian Financial Reforms 1991 Improved accounting rules, bank transparency Recapitalized failing public-sector banks Allow public-sector banks to raise equity Loosened restrictions on FDI After 1991 Debt Recovery Tribunals Act Sarfaesi Act Continual reduction in government ownership of banking sector
Average Productivity Z avg log labor productivity 8 8.5 9 9.5 1990 1995 2000 2005 2010 Year 1990 weights 2011 weights dynamic weights
OP Covariance C OP covariance.2.3.4.5 1990 1995 2000 2005 2010 Year 1990 weights 2011 weights dynamic weights
Default Decision Net wealth tomorrow: pay b : x = default on b : x = (1 ζ) k ( ) PAe z + 1 k + b Value function will be increasing in net wealth, so indifference point is { 1 [ ] ε = σ log (l ζ) log A log P ρz Z l > ζ otherwise l b k In default, the lender recovers χ min Default iff ε < ε, so bond price is { } { } 1, ζ k b = min 1, ζ l. q ( k, b, z; Z, ζ, P ) = q (l, ρz + Z + log P; ζ) = 1 [ ] 1 (1 χ) Φ {ε} 1 + r
Individual Agent s Problem Individual agent s problem: { V (x, z; Z, ζ, P, F ) = max u(c) + βπe u ( x )} k 0,b 0 + β (1 π) E {V ( x, z ; Z, ζ, P, F )} s.t. c x qb k ( ) b q q k, z; ζ z ρz + Z + log P { [ ] } x = max PAe ρz+z+σε + 1 k + b, (1 ζ) k Exogenous aggregate demand curve log P = η log Y + D.
Special Case: Log Utility Implies that the optimal decisions satisfy k + q (l, ρz + Z + log P; ζ) b = β x for a calculable β < β. details c = ( 1 β ) x Optimal ( k, b ) as a proportion of x depend only on current value of (ρz + Z + log P, ζ). Wealth distribution F only affects equilibrium through current P.
Decision Rules Define z ρz + Z + log P. First-order condition for l is q + q l l 1 ql = ε(l, z) φ (ε) dε Ae z+σε + 1 l Let z (r) solve equation (1) for l = 0. Then (1) z z b = 0 k = β x z(r) b < 0 k > β x l > 0
Model: Approach Two aggregate shocks: Common productivity Z Financial friction θ: recovery rate for risky loans / fraction borrowers can steal. Z or θ more-productive firms grow faster increase OP covariance. Only Z shocks increase unweighted average productivity. (Z, θ) }{{} interpretable model (Z, C) }{{} observable