Discussion of: Productivity and Capital Allocation in Europe by Gopinath, Kalemli-Ozcan, Karabarbounis, and Villegas-Sanchez Brent Neiman University of Chicago AEA Meetings 215
Static Misallocation (Quick Refresher) ( ) α ( ) 1 α ki li y i = z i α 1 α 1 p i = µ i mc i = µ i Ri α w 1 α i z i TFPR i = p i z i = µ i R α i w 1 α i (R i = R) + (w i = w) + (µ i = µ) = TFPR i = TFPR Otherwise (i.e. R i = R(1 + τ k i )) = TFPR i TFPR Paper is about Var(ln(TFPR i )) in South, but not in North
Plan for Discussion A really nice paper! My discussion will focus on: 1 Importance of joint distribution of productivity and wedges 2 Decomposition of Var(ln TFPR i ) into Var(ln MRPK i ) and Var(ln MRPL i ) 3 Compare TFPR i dynamics in model and data
Great Data Covering Full Firm Size Distribution Large emphasis placed on breadth of sample 99% of firms are private much broader than Compustat Match nicely with size distribution from census/eurostat Convinced me they did enormous amount of careful work Focus on Var(ln(TFPR i )) may overweight small firms Theory says TFPR i = p i z i essentially scale invariant (1 + τ k i ) impacts Var(ln(TFPR i )) same for big and small Why? Assumes joint multivariate log normality.
Great Data Covering Full Firm Size Distribution The exact expression for TFP is: [ N ( TFP exact = i z i TFPR ) ] 1/(σ 1) σ 1 TFPR i Under joint log normality, it is: TFP approx = 1 σ 1 ( ( )) ln N + ln E z σ 1 i σ 2 Var (ln (TFPR i)) α (1 α) ( ( Var ln 2 1 + τ k i )) Intuition? Case 1: Covar(ln z i, ln(1 + τi k )) = Case 2: Covar(ln z i, ln(1 + τi k ))
Potential to Overemphasize Small Firms 5 4 lnz i ln ( ) 1+τi k (Initial).25.2 Market Shares (Initial) 3 2 1.15.1.5 1 5 1 15 2 5 1 15 2 5 4 lnz i ln ( ) 1+τi k (Final).25.2 Market Shares (Final) 3 2 1.15.1.5 1 5 1 15 2 5 1 15 2 Initial distribution is joint lognormal, N = 2, only τ k, σ = 3 Initial 5.5943 = ln TFP exact ln TFP approx = 5.598
Potential to Overemphasize Small Firms 5 4 lnz i ln ( ) 1+τi k (Initial).25.2 Market Shares (Initial) 3 2 1.15.1.5 1 5 1 15 2 5 1 15 2 5 4 lnz i ln ( ) 1+τi k (Final).25.2 Market Shares (Final) 3 2 1.15.1.5 1 5 1 15 2 5 1 15 2 Initial distribution is joint lognormal, N = 2, only τ k, σ = 3 Initial 5.5943 = ln TFP exact ln TFP approx = 5.598 But.1 = ln TFP exact > ln TFP approx =.334
Potential to Underemphasize Big Firms 5 4 lnz i ln ( ) 1+τi k (Initial).25.2 Market Shares (Initial) 3 2 1.15.1.5 1 5 1 15 2 5 1 15 2 5 4 lnz i ln ( ) 1+τi k (Final).25.2 Market Shares (Final) 3 2 1.15.1.5 1 5 1 15 2 5 1 15 2 Initial distribution is joint lognormal, N = 2, only τ k, σ = 3 Initial 5.5943 = ln TFP exact ln TFP approx = 5.598 But.884 = ln TFP exact < ln TFP approx =.361
Great Data Covering Full Firm Size Distribution Why might this matter? Measurement error bigger on small/private firms? Policies and reporting incentives different? (Hsieh 22) Potentially explains sensitivity to treatment of entry/exit? Model has size-dependence of financial frictions and endogenously generates joint-distribution between ln z i and ln(1 + τi k ). What is it in model? What is it in data? Do I suspect this is big deal? No. Examples I showed were far from zero-mean noise. Still, easy and important to check.
Split Var(ln TFPR) into Var(ln MRPK) and Var(ln MRPL) ln TFPR i = γ + α ln MRPK i + (1 α) ln MRPL i Upward trend in Var(ln TFPR i ) is clearly driven by upward trend in Var(ln MRPK i ) and not in Var(ln MRPL i ) Surprising and interesting, suggestive of key shocks, one of coolest results in paper! Clear and compelling split between South and North Nicely motivates focus on capital in model
Split Var(ln TFPR) into Var(ln MRPK) and Var(ln MRPL) Implies heterogeneous markups (or changes in them) aren t the story. MRPL i = 1 µ i α p iy i l i MRPK i = 1 µ i (1 α) p iy i k i Authors should emphasize this more. Peters (213) generates what would look like misallocation in CES models with variable markups Fernald and Neiman (211) model impact of dynamic misallocation from variable markups on TFP in Singapore Surprising? External validity? Again, elevates importance of approximation...
Dynamic Model with Risk, Financial Frictions, Adj. Costs Model (but with b i,t+1 θk i,t+1 ) yields user cost expression (when constraints binding): u i,t = E [MRPK i,t+1 ] = (r t+1 + δ) + (1 θ) (1 E [m i,t+1] (1 + r t+1 )) [m i,t+1 ] + AC i,t 1 k i,t+1 E [m i,t+1 ] + AC i,t+1 k i,t+1 Risk, financial frictions, and adjustment costs do the work
Dynamic Model with Risk, Financial Frictions, Adj. Costs Model (but with b i,t+1 < θk i,t+1 ) yields user cost expression (when constraints binding): u i,t = E [MRPK i,t+1 ] = (r t+1 + δ) + (1 θ) (1 E [m i,t+1] (1 + r t+1 )) [m i,t+1 ] + (Without Adjustment Costs) Risk, financial frictions, and adjustment costs do the work Kill Adjustment Costs
Dynamic Model with Risk, Financial Frictions, Adj. Costs Model (but with b i,t+1 < θk i,t+1 ) yields user cost expression (when constraints binding): u i,t = E [MRPK i,t+1 ] = (r t+1 + δ) + (θ = 1 Implies Borrowing Unconstrained) + (Without Adjustment Costs) Risk, financial frictions, and adjustment costs do the work Kill Adjustment Costs Kill Borrowing Constraints
Dynamic Model with Risk, Financial Frictions, Adj. Costs Model (but with b i,t+1 < θk i,t+1 ) yields user cost expression (when constraints binding): u i,t = MRPK i,t+1 = (r t+1 + δ) (If No Risk) + (θ = 1 Implies Borrowing Unconstrained) + (Without Adjustment Costs) Risk, financial frictions, and adjustment costs do the work Kill Adjustment Costs Kill Borrowing Constraints Kill Risk
Dynamic Model with Risk, Financial Frictions, Adj. Costs Nice dynamics that I think are missing from much of misallocation literature Opportunity to use panel structure of data and relate to dynamics in model How persistent is a firm s TFPR in the model? In the data?
Spain s entry to Euro Zone Authors represent inflows to Spain with decline in interest rate. Very cool/important application. Even from perspective of model, didn t other important things occur in tandem? Structural change? Authors capture within sector dispersion and explain nearly all for Spain. Very different from between-sector stories about tradable/non-tradable. FX-driven relative prices? Trade-induced changes in market shares? VAT changes? How think about capital inflows to U.S. over same period? Different only due to initial conditions, or average productivity growth? Comparative statics on the model would help.
Conclusion Great paper! Helpful next step in misallocation literature Adds quantative rigor to familiar stories about entry into euro zone Can be strengthened by: 1 Thinking more about size-wedge joint distribution, 2 Highlighting that markups aren t doing anything, 3 Using model to test dynamic behavior of wedges