Motivation

1. Trade Liberalization and Embedded Institutional Reform: Evidence from Chinese ExportersAmit K. Khandelwal, Columbia Business SchoolPeter K. Schott, Yale School of ManagementShang-Jin Wei, Columbia Business School

2. Motivation • Institutions that distort the efficient allocation of resources can have sizeable effects on aggregate outcomes • Hsieh and Klenow (2009): aggregate Chinese productivity nearly doubles if capital and labor are properly allocated among existing firms • Trade barriers distort resource allocation along both “intensive” and “extensive” margins • Institutions that manage trade barriers can cause additional distortions • Key idea: productivity gains from trade liberalization may be larger than expected if institutions that manage the trade barriers are inefficient: • Gain from removal of the “embedded institution” • Gain from removal of the trade barrieritself

3. China and The Multifiber Arrangement (MFA) • This paper examines the distortions associated with institutions that manage quota licensing • The global MFA restricted Chinese exports of textile and clothing to the US, EU and Canada until 2005 • Quotas were assigned by the Chinese government • Our question: were quotas assigned to the most productive firms? • Comparison of quota-bound vs quota-free goods before/after 2005 suggests entrants are more productive than incumbents, i.e., the most productive firms were not allocated licenses • Use key feature of empirical analysis to simulate “political allocation” and compute contribution of eliminating licensing to overall gain • Eliminating actual institution accounts for ~70% of overall gain • Replacing actual institution with auction raises productivity ~13%

4. Related Literature • Growing literature on misallocation • Hsieh and Klenow (2009), Brandt et al. (2010), Dollar and Wei (2007), Restuccia and Rogerson (2010), Alfaro et al. (2008) • Extensive-margin misallocation • Banerjee and Duflo (2005), Banerjee and Moll (2010), Buera et al. (2010), Chari (2010) • Inefficient implementation of quotas; studies of MFA/ATC • Krishna and Tan (1998), Anderson (1985) • Harrigan & Barrows (2009), Brambilla et al (2010), Bernhofen et al. (2011)

5. Outline • Auction-allocation model of quota licenses • Data and Identification Strategy • Evidence of misallocation • “Political allocation” and counterfactual exercise • Conclusion

6. Overview of Auction-Allocation Model • Same basic structure as in Melitz/Chaney • Two countries, one industry • Monopolistic competition, CES utility • Firms are heterogeneous in productivity (j) • Exporting requires fixed and iceberg trade costs (t) • Firms optimize under quantity restriction • Quota license fee is like a per-unit trade cost (aod) to export from origin country o to destination country d (Irrazabal et al. 2010) • Price of variety with productivity j: aod > 0 imposes a disproportionate penalty on high productivity (i.e., high j) firms Analytical solutions to model not possible when aod > 0

7. Three Empirical Implications of Quota Removal • Export growth following quota removal is driven by the intensive margin • High productivity firms are most constrained under quotas • Their exports jump disproportionately as quotas are removed • Low-productivity enter because license fee goes to zero when quotas are removed • (Depends on TFP distribution: if density of very high TFP firms is high enough, there will be no entry and the lowest TFP firms will exit) • Incumbentsand entrants make opposing contributions to export prices • Incumbents’ prices fall as the license fee goes to zero • But removal of license fee allows high price (i.e., low-productivity) firms to enter • (Will come back to quality variant of model later)

8. Outline • Auction-allocation of quota licenses • Data and Identification Strategy • Evidence of misallocation • “Political allocation” and counterfactual exercise • Conclusion

9. Quotas Under the MFA/ATC • During the Uruguay Round (early 1990s), the US, EU and Canada committed to a schedule for withdrawing textile and clothing quotas in four phases • At the start of 1995, 1998, 2002 and 2005 • China’s quotas on goods in first three phases were relaxed in early 2002 following its entry into the WTO in late 2001 • We focus on the final phase • Chinese quotas were allocated by the government • Details are scarce but predominantly on the basis of “past performance” • Black-market sales of licenses complicates our analysis; appears to be a bigger issue during the 1980s than our sample period (Moore, 2002) • (More about this later)

10. Aggregate Chinese Textile & Clothing Exports Quota-free exports rise 29% in 2005 Quota-bound exports rise 119% in 2005 Quotas Relaxed Quota-Bound Quota-Free Notes: Quota-bound = any export constrained by a quota; quota-free = other textile and clothing goods not bound by quotas

11. Firm-Level Chinese Customs Data • Value and quantity exported • By firm, HS8 product, destination country and year • Focus on 2003-2005 exports to US, EU and Canada • Observe exporter’s ownership type • “SOE”: state-owned enterprise • “Domestic”: privately-owned domestic firm • “Foreign”: privately-owned foreign firm • Create two sets of HS8-country (hd) groups : • Quota-bound: subject to quota until 2004 by subset of US/EU/Canada • “Men’s cotton pajamas” to US/Canada • Quota-free: not subject to quota by subset of US/EU/Canada • “Men’s cotton pajamas ” to EU

12. Identification Strategy • Sample • Start with 547 HS8 products are subject to quotas by US/EU/Canada • Drop the 188 of these that are subject to quotas by all three countries • The remaining 359 HS8 products are our sample • Difference-in-differences comparison • Quota-bound (“treatment”) vs quota-free (“control”) for 2004-5 versus the same difference for 2003-4 • Changes in control group account for trends in textile-and-clothing supply (e.g., privatization) or demand (e.g., preferences) • Attribute any differential response to the removal of quotas

13. Quota-Bound vs Quota-Free • Compare treatment and control groups pre- and post-reform • SOE share differs substantially ex ante, but not ex post

14. Regression Specification • Where DYhdt is • Change in market share of incumbent SOEs • Change in market share of privately owned entrants • Etc. • Just report α3: quota-bound vs quota free in 2004-5 versus 2003-4 • Full regression results in appendix • Also do “placebo” diff-in-diffs for prior year, i.e., 2003-4 versus 2002-3 • Also add country-product FEs to control for underlying trends

15. Outline • Auction-allocation of quota licenses • Identification Strategy • Evidence of misallocation • “Political allocation” and counterfactual exercise • Conclusion

16. Decompositions (DY) • Quantity market share changes • By margins of adjustment, ownership • Examine quantity growth to avoid price effects • Can’t aggregate quantity across HS8, so compute changes for each HS8-country pair and then average across pairs, by group • The tables I show you will be these averages • Price changes • By margins of adjustment, ownership

17. Margins of Adjustment • Intensive: • Incumbent: firm exports same HS8 to same country in both t-1 and t(note: EU considered single country) • Extensive • Exiter: firm exports HS8-country in t-1 but not t • Entrant: firm exports HS8-country in t but no exports in t-1 • Adder: firm exports HS8-country in tbut not t-1AND was an exporter of some other HS8-country in t-1

18. Decompose Change in Market Shares(Summary of Diff-in-Diff Terms from Regression) Notes: Bold indicates statistical significance at conventional levels.

19. Decompose Change in Market Shares(Summary of Diff-in-Diff Terms from Regression) Notes: Bold indicates statistical significance at conventional levels.

20. Decompose Change in Market Shares(Summary of Diff-in-Diff Terms from Regression) Notes: Bold indicates statistical significance at conventional levels.

21. Decompose Change in Market Shares(Summary of Diff-in-Diff Terms from Regression) Notes: Bold indicates statistical significance at conventional levels.

22. Decompose Change in Market Shares(Summary of Diff-in-Diff Terms from Regression) Notes: Bold indicates statistical significance at conventional levels.

23. Pre-Reform “Placebo” Market-Share Decomposition Notes: Bold indicates statistical significance at conventional levels.

24. Each line is the lowess-smoothed relationship between initial market share and subsequent change

25. Quota relationships are steeper, especially for SOEs

26. Price Changes Before/After Quota Removal Quota-Bound Exports

27. Export Price Decomposition • Decompose change in the overall MFA price between 2004-5 by margin and compare with OTC • where {f,h,d,t} index {firm,product,country,year} • Quantity-weighted avg log export price • Product-country price change ΔOverall = ΔIncumbents + ΔNet Entrants

28. Distribution of Prices, by Margin

29. Distribution of Prices, by Margin(Comparison Groups)

30. Distribution of Prices, by Margin(Comparison Groups)

31. Decompose Price Response

32. Decompose Price Response

33. Pre-Reform “Placebo” Diff-in-Diff (Prices)

34. Quality Downgrading? • Might expect prices to decline due to quality downgrading in response to quotas (Aw and Roberts 1986; Boorstein and Feenstra 1991; Harrigan and Barrows 2009) • We see prices fall in the data, but declines are concentrated among privately owned entrants (assumed to be more productive) • Nevertheless, we can compute quality-adjusted prices to check • Approach is similar to Hummels and Klenow (2005), Khandelwal (2010), Hallak and Schott (2011) • Find similar results….

35. Quality-Adjusted Prices • Put quality in CES preferences • Quantity demanded for each variety • Impose σ = 4, use dt fixed effects to capture price index/income, h fixed effect compares quantities and prices within products • Log quality is • Quality-adjusted prices:

36. Decompose Quality-Adjusted Price Response

37. Pre-Reform “Placebo” Diff-in-Diff (QA Prices)

38. Coarse, Back-of-Envelope Productivity Calculation • Identify textile and clothing exporters in the Annual Survey of Industrial Production • (Match with trade data is imperfect) • Calculate TFP of each firm assuming Cobb-Douglas, constant returns to scale • Labor coefficient is the share of wages in value added • Capital coefficient = 1 - labor coefficient • Among textile/clothing exporters • Average SOEs is 1/4 to 1/3 as productive as the average domestic and foreign firm, respectively • Consistent with literature

39. Coarse, Back-of-Envelope Productivity Calculation

40. Coarse, Back-of-Envelope Productivity Calculation These numbers are from the market share table Multiply changes in market share by each ownership type’s mean TFP to gauge TFP gain from reallocation of 21.3% (Calculation assumes homogenous firms within ownership type)

41. Outline • Allocation of quota licenses via an auction • MFA Background, Identification Strategy • Evidence of misallocation • “Political allocation” and counterfactual exercise • Conclusion

42. Decomposing Productivity Gains • We want to decompose the overall productivity gain from quota removal into two parts No Quota AuctionAllocation PoliticalAllocation

43. Decomposing Productivity Gains • We want to decompose the overall productivity gain from quota removal into two parts • Part due to removal of licensing regime No Quota AuctionAllocation PoliticalAllocation

44. Decomposing Productivity Gains • We want to decompose the overall productivity gain from quota removal into two parts • Part due to removal of licensing regime • Part due to removal of quota No Quota AuctionAllocation PoliticalAllocation

45. Decomposing Productivity Gains • We want to decompose the overall productivity gain from quota removal into two parts • Part due to removal of licensing regime • Part due to removal of quota • In order to do this, we use numerical solutions of the model to compute weighted-average firm productivity under three scenarios • No quota • Auction allocation • Political allocation: a perturbation of the auction-allocation model that matches our empirical evidence of misallocation No Quota AuctionAllocation PoliticalAllocation

46. Numerical Solutions for No-Quota Scenario • Choose parameters of the no-quota scenario • Elasticity of substitution σ=4 (from Brodaet al. 2006) • Country sizes • Fixed and variable trade costs • Log Normal productivity distribution, LN(μ,q) • Choose (μ,q), iceberg trade costs and ratio of export to domestic fixed cost to match: • Export size distribution • Share of Chinese and U.S. textile and clothing firms that export • U.S. and Chinese import penetration in each others’ markets • Simulate productivity draws, compute cutoffs, total exports, market shares and prices

47. TFP vs Market Share Under No Quotas

48. Numerical Solutions for Auction-Allocation Scenario • Use the no-quota scenario but impose the quota restrictiveness observed in data • Export quantities jump 161% in quota-bound versus quota-free goods when quotas are removed • Solve for endogenous license fee that clears the market • This license price is ~10% of the average price of an exporter • Re-compute aggregate export TFP

49. TFP vs Market Share, No Quota vs Auction Allocation Disproportionate penalty on high-TFP firms

50. Numerical Solutions for Political-Allocation Scenario • Firms have second, political draw • Correlation ρwith TFP • Re-assign market shares from auction-allocation based on this draw • Assign highest market share to the most politically connected firm, second most connected firm gets second highest share, etc. • Firm prices continue to based on true underlying productivity • Low TFP firms with high political draw get high market share • Decompose aggregate price decline between political allocation and “no quota” allocation as we did in empirical tables • Calculate contribution of price decline attributed to net extensive margin • Choose ρ to match observed 67.5% extensive-margin contribution to quality-adjusted price decline (at ρ=0.15)