Does Improved Market Access Raise Plant-Level Productivity? Alla Lileeva. Daniel Trefler

Transcription

1 Does Improved Market Access Raise Plant-Level Productivity? Alla Lileeva Department of Economics, York Univ. Statistics Canada Daniel Trefler Rotman School of Management, Univ. of Toronto CIAR

2 Motivation Exporters are more productive than nonexporters. Does exporting raise productivity? 1. Raising productivity is not a passive activity. e.g., learning-by-exporting, down the AC curve. 2. What is the policy-relevant question being answered or policy-relevant parameter being estimated? 3. Is Trefler (AER, 2004) inconsistent with Clerides, Lach and Tybout (QJE, 1998) and Bernard and Jensen (JIE, 1999; OREP, 2004)? Results for developing countries are more mixed, possibly because of differences in why firms export.

3 Basic Insight Estimate effects of a policy improved market access. Improved market access leads to reorganization: Heterogeneity: Firms with low levels of productivity who reorganize and export must be getting a big kick from reorganization.

4 Why reorganization rather than conventional explanations? Moving down and average cost curve Not a big enough effect (Lileeva, Ph.D., 2005) Bilateral tariff reductions under monopolistic competition leads to import competition which raises productivity No (Trefler, AER, 2004) Changes in product mix Yes? (Baldwin, Caves, and Beckstead, Mimeo, 2004). The strategic management literature often claims that there are benefits Mork and Yueng (JB, 1991), Baily and Gersback (Brookings, 1995), Berry and Sakakibara (JEOB, 2005), Cassiman and Martínez-Ros (Mimeo, 2005), Salomon and Shaver (JEMS, forthcoming)

5 Figure 1. Number of Products per Multi-Product Plant.43 Product Dispersion (A measure of the number of products produced per plant) Pre-FTA FTA Source : Baldwin, Beckstead, and Caves (2001).

6 A Simple Model of Optimizing Behaviour Framework: Clarides-Lach-Tybout (QJE, 1997) and Melitz (Econometrica, 2003). Optimization problem only: Ignore markets and focus on selection. Static, full information. No general equilibrium feedbacks.

7 Demand Two countries, home (Canada) and foreign (, United States) CES monopolistic competition. A home firm producing variety i faces demands q(i) = p(i) σ A and q (i) = p (i) σ A. The foreign country imposes a tariff τ(i) 1 on firm i s products.

8 Cost A standardized bundle of inputs costs c and produces ϕ(i) units of output. Hence marginal costs are: c ϕ(i) Profit maximizing prices are Profit Maximization p(i) = σ σ 1 c ϕ(i) and p (i) = τ(i)p(i).

9 Entry into Export Markets A fixed cost F E must be incurred to enter the foreign market. Let E(i) = 1 if the firm enters and E(i) = 0 if it does not enter. Home firm s profits are (dropping the i index) for E = 0,1 where Enter foreign markets when π(e) = ϕ [ A + Eτ σ A ] EF E ϕ = (σ 1)σ 1 σ σ c σ 1 ϕ σ 1. ϕ > ϕ FE τ σ A. where τ σ A measures the size of the foreign market.

10 Reorganizing to Raise Productivity The firm can choose to pay a fixed reorganization cost F R that raises productivity from ϕ 0 to ϕ 1. ϕ 0 and ϕ 1 are known. Let R = 1 if the firm reorganizes, R = 0 if not. Profits are [ π(r = 1,E) = ϕ 1 A + Eτ σ A ] EF E F R [ π(r = 0,E) = ϕ 0 A + Eτ σ A ] EF E

11 ~ 1 ~ ϕ ϕ 0 F A R E + F A +τ σ * Export Reorganize Exporting Reorganize F R σ * A A +τ No Exporting No Reorganization Exporting No Reorganization ϕ = τ F E σ A * R F + F σ * τ A E ~ ϕ 0

12 ~ 1 ~ ϕ ϕ 0 F A R E + F A +τ σ * Improved market access induces reorganization and exporting F R σ * A A +τ F E τ σ Induces exporting without reorganization~ A * ϕ 0

13 F A R ~ 1 ~ ϕ ϕ E + F A +τ σ * 0 Causal Effects of Improved Market Access on Productivity Growth For a firm that is initially indifferent between exporting and not exporting. Assume E F < σ τ A* F A If low F R, reorganize immediately lack of dynamics. R If A* is higher for low ϕ 0 firms then the line is lower (and steeper). F R σ * A A +τ Expect opposite results if F R is higher for low ϕ 0 firms (but functional form kills this). F E τ σ A * ~ ϕ 0

14 Econometric Model ϕ 0 = β 0 (X) + U 0 (Core Eqn. 1) ϕ 1 = β 1 (X) + U 1 (Core Eqn. 2) 1 P(X, τ) U E E = (Core Eqn. 3) 0 P(X, τ) < U E where where X includes ϕ 0 and P(X, τ) Pr{E = 1 X, τ} Causal Effects of Improved Market Acces (suppress X): C = ϕ 1 ϕ 0 = (β 1 β 0 ) + (U 1 U 0 ) U 1 = U 0 means heterogeneous causal effects.

15 Importance of Conditioning on U E Consider a firm on our line. Set X = ϕ 0 : P(ϕ 0, τ = 0) = U E. P(ϕ 0, τ = 0) is increasing in ϕ 0. On the line, firms with a small ϕ 0 have: A small U E = P(ϕ 0, τ = 0). A large U 1 U 0. U E and U 1 U 0 are negatively correlated Conditioning on U E is informative. Our object of interest is E [β 1 β 0 + U 1 U 0 ϕ 0,U E = P(ϕ 0, τ)] (1) or E [C P]. (2)

16 Importance of Conditioning on U E CES example: Go back to the old notation so that ϕ i is replaced by ϕ i = β i + U i. For simplicity set U 0 = 0. Then on our line ] σ 1 [ 1/(σ 1) Aβ 0 + F E + F R β 1 A + τ σ A }{{} P = U 1 }{{} U E where P/ τ < 0 so that a higher tariff makes entry less likely.

17 Estimation of the Line E(C P) Observables With Heterogeneity (U 1 = U 0 ) ϕ = ϕ 0 + ( ϕ 1 ϕ 0 )E = β 0 + U 0 + [(β 1 β 0 ) + (U 1 U 0 )]E C = (β 1 β 0 ) + (U 1 U 0 ) Observables Without Heterogeneity (U 1 = U 0 ) ϕ = β 0 + (β 1 β 0 )E + U 0 C = β 1 β 0

18 Estimation with No Heterogeneity E[ ϕ E] = β 0 + (β 1 β 0 )E + E[U 0 E] C OLS = ϕ E=1 ϕ E=0 C IV = ϕ τ=1 ϕ τ=0 E τ=1 E τ=0 E[ ϕ P] = β 0 + (β 1 β 0 )P + E[U 0 ] Local IV using plants with the same P τ=0 P(X, τ = 0) and P τ=1 P(X, τ = 1) E[C P] = E[ ϕ P]/ P Ê[C P] = ϕ τ=1 ϕ τ=0 P τ=1 P τ=0

19 Estimation with Heterogeneity ϕ = β 0 + (β 1 β 0 )E + U 0 + (U 1 U 0 )E C = β 1 β 0 + U 1 U 0 E[ ϕ E] = β 0 + (β 1 β 0 )E + E[U 0 E] + E[(U 1 U 0 )E E] IV fails because Cov[E, τ] < 0 so that the residual is correlated with the instrument. E[ ϕ P] = β 0 + (β 1 β 0 )P + E[(U 1 U 0 )E P] Theorem 1 (Heckman and Vytlacil, PNAS, 1999, MTE Identification) E[C P] = E[ ϕ P] P.

20 Proof (1) We do not observe whether the firm has reorganized. However, in the region we are interested in R = E = 1. Thus, ϕ ϕ 0 + E( ϕ 1 ϕ 0 ). Consider a selection rule based on. g(z) > V E Let F V be the cdf of V. Then the rule is equivalent to F V (Z) > F V (V E ). Define U E = F V (V E ). Then U E is uniformily distributed on (0,1), P(Z) = Pr(E = 1 Z) = F V (g(z)), and the distribution of U E conditional on U E < P(z) is 1/P(z).

21 Proof (2) E[ ϕ P(Z) = P(z)] = = P(z) 0 P(z) 0 E[ ϕ 1 U E p] E[ ϕ 1 U E p]dp + 1 P(z) dp P(z) P(z) P(z) E[ ϕ 0 U E > p]dp E[ ϕ 0 U E > p] 1 dp (1 1 P(z) Theorem 2 (Heckman and Vylacil, 1999) E[C P] = E[ ϕ P(Z) = P(z)] P(z)

22 Relationship to LATE LATE: E [C U E is in the Switcher set] Switcher Set: P(X, τ = 0) < U E < P(X, τ = 1) MTE: E [C U E is on the Line] Firms on the Line: P(X, τ) = U E MTE is the limit of LATE as the Switcher interval shrinks.

23 Ê[C P] = E[ ϕ P] P = K( P) P Assume β 1 (x) = β 0 (x) = βx. ϕ = βx + U 0 + (U 1 U 0 )E. Estimation E[ ϕ P] = βx + E[(U 1 U 0 )E P] = βx + E[(U 1 U 0 ) P,E = 1] P }{{} = βx + K(P) Four-Stage Procedure: 1. Estimate a probit explaining entry E on Z = (ϕ 0, τ) P 2. Local linear regression of observed productivity growth ϕ on x and P K( P). 3. Differentiate K( P):

24 Hypothesis Testing Bootstrap standard errors. There is heterogeneous treatment effects (U 1 = U 0 ) if the derivative is not a constant. The theory is true if the derivative looks like the profiles from the theory.

25 Canada-U.S. Free Trade Agreement Timeline September 1985 November 1988: implementation uncertainty. January 1, 1989: Tariff concessions begin. 1 in 4 industries faced tariffs in excess of 10%. December 31, 1996: Average tariff under 1%, nothing above 5%.

26 Canadian Exports to the United States: Little Entry Between 1984 and 1988? 120 Exports (1988=100) Pre-FTA Period FTA Period

27 The Sample Data on export status for 1984 and If the FTA was anticipated, we want pre-1985 data 1984 baseline. If not anticipated, we want pre-1989 data 1988 baseline. Data on the tariff concessions against each of the HS6 outputs produced by a plant. Aggregate using 1996 weights to obtain plant-specific tariff concessions τ. Binary tariffs. Prohibitive tariffs, fat tails, Swiss formulas, probits. τ = 1 if τ > median (industry or all). τ = 0 otherwise.

28 The Sample (continued) Three types of plants: Nonexporters : Did not export in 1984 or 1996 (2,133 plants) New Exporters : Did not export in 1984, exported in 1996 (3,114) Exported in 1984 and 1996 (4,000 plants)

29 Table 1. Average Plant Characteristics After Deviating from Industry Medians New Exporters Nonexporters Difference Variable E = 1 E = 0 µ 1 µ 0 t % with Tariff Concessions 64% 34% - - Log Employment, Log Productivity, Annual Log Productivity Growth % -0.5% 1.8% 7.32

30 Table 4. Probit Estimates of the Probability of Entry Between 1984 and 1996 Coeff. Prob. τ FTA Log Productivity, Log Employment, Log Productivity Growth, SIC4 Fixed Effects Yes Observations 5,247

31 Marginal Effect of Tariffs on Entry.18 Impact of Tariff Concessions on Entry: P( x, τ FTA =1 ) P( x, τ FTA = The Probability of Entry Without a Tariff Concession: P(x, τ FTA =0) = P( E=1 X=x, τ FTA =0

32 18 Ratio of Nonexporters to New Exporters By Probability of Entry 16 Ratio of Nonexporters to New Exporters Probability of Entry: P( x, τ FTA )

33 Number of Plants by Probability of Entry Nonexporters New Exporters 500 Number of Plants Probability of Entry: P( x, τ FTA )

34 ~ 1 ~ ϕ ϕ 0 Causal Effects of Improved Market Access on Productivity Growth F A R E + F A +τ σ * F R σ * A A +τ F E τ σ A * ~ ϕ 0

35 Marginal Treatment Effects: % Average Annual Change in Productivity 10% 8% 6% 4% 2% 0% % -4% Probability of Entry: P (z ) = Pr ( E =1 x, τ FTA )

36 12% Marginal Treatment Effects: Industries with Large Tariff Cuts 10% All Above Median Avg. Annual Change in Productivity 8% 6% 4% 2% 0% Above 1/ % Probability of Entry: P (z ) = Pr ( E =1 x, τ FTA )

37 8% Marginal Treatment Effects: By Ownership Avg. Annual Change in Productivity 6% 4% 2% 0% Foreign-Owned Domestic All % Probability of Entry: P (z ) = Pr ( E =1 x, τ FTA )

38 Table. New Product Innovation and Productivity Gains Commodity turnover index based on valuesof commodities New Exporters Non-Exporters Difference ,140 High Productivity Plants: P(z) > 0.5 Low Productivity Plants: P(z) < 0.5 Double Difference Mean t-stat Mean t-stat (Low - High) N Product Innovation New Exporters Non-exporters Difference Design & Engineering Adoption New Exporters Non-exporters Difference Design & Engineering, Plans 93 New Exporters Non-exporters Difference Design & Engineering, Plans 98 New Exporters Non-exporters Difference Number of commodities added, all plants New Exporters Non-Exporters Difference ,121

39 Table. Process Innovation and Productivity Gains High Productivity Plants: P(z) > 0.5 Low Productivity Plants: P(z) < 0.5 Double Difference Mean t-stat Mean t-stat (Low - High) N Fabrication and Assembly New Exporters Non-exporters Difference Infromation and Communication New Exporters Non-exporters Difference Table. Scale / Scope and Productivity Gains High Productivity Plants: P(z) > 0.5 Low Productivity Plants: P(z) < 0.5 Double Difference Mean t-stat Mean t-stat (Low - High) N Change in output per commodity New Exporters Non-exporters Difference ,179

40 Conclusions Improved market access matters. Articulating a policy-relevant question. With heterogeneity, traditional approaches fail: RT, BJ condition on pre-entry growth IV larger than OLS A theory of reorganization helps to explain the heterogeneity of productivity responses to increased market access.