The Productivity Effects of Privatization Longitudinal Estimates using Comprehensive Manufacturing Firm Data from Hungary, Romania, Russia, and Ukraine J. David Brown, John S. Earle, and Almos Telegdy Presentation to the World Bank Workshop on Productivity, 31 October 2005
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal)
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic Dynamics of privatization effect
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic Dynamics of privatization effect 5 years pre to 5 post
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic Dynamics of privatization effect 5 years pre to 7 post Program evaluation methods and specification tests
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic Dynamics of privatization effect 5 years pre to 5 post Program evaluation methods and specification tests Firm specific trends
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic Dynamics of privatization effect 5 years pre to 5 post Program evaluation methods and specification tests Firm specific trends Extensive robustness checks
Does Privatization Raise Productivity? Many previous studies, but none has all the following: This paper: Large, representative sample of firms 28,000 (nearly universal) Long time series before and after privatization 11 years average Comparable variables/samples for several countries 4 transition economies Different types of private owners Foreign v. domestic Dynamics of privatization effect 5 years pre to 5 post Program evaluation methods and specification tests Firm specific trends Extensive robustness checks 40 different methods
Policy and Business Environment Four Economies Reform Progress Policy and Business Environment Privatization World Bank grouping “Government Effectiveness” “Regulatory Quality” “Rule of Law” Speed Type Hungary 1 (advanced) 0.78 1.21 0.90 Fast Sales Romania 2 (second) -0.33 0.04 -0.12 Slow Mixed Russia 3 (lagging) -0.40 -0.30 -0.78 Insider/mass Ukraine 4 (way behind) -0.74 -0.62 -0.79 Insider
Cross-Country Hypotheses 1. (alphabetical) The magnitude of the privatization effect is increasing in alphabetic ranking of a country (i.e., strongest for Hungary, second in Romania, third in Russia, and lowest in Ukraine). 2. (reverse alphabetical) The privatization effect is decreasing in alphabetic ranking.
Data Comparable samples and variables. Sources: statistical offices, tax authorities, privatization agencies, private data providers. Cover nearly all “old” manufacturing firms (existing prior to 1990 or had any state ownership): the firms “at risk” of privatization. Coverage by year: Hungary: 1986-2002 Romania: 1992-2002 Russia: 1985-2002 Ukraine: 1989, 1992-2002 Number of annual observations per firm as high as 18, with an average of about 11 across countries. Total number of firm-year observations approximately 300,000.
Data (2): Coverage in 1992 Number of firms Percent of all old firms Total employment Percent of old firm Hungary 1,800 82.2 513,596 92.1 Romania 1,943 95.1 2,399,904 98.8 Russia 17,505 92.2 13,511,852 93.7 Ukraine 5,587 99.3 3,894,269 99.9
Percentage of Firms Privatized Majority Private and Majority Foreign
Pre-Privatization Relative Employment and Output
Pre-Privatization Relative Productivity
Estimation Strategy: Previous Research In general literature on effects of private and foreign ownership, two approaches: 1. Cross-sectional comparisons. Regression-adjusted version: yi = δ0 + δ1Privatei + Xi δ2 + ui Problem of endogenous ownership. (e.g., ui = μi + Є i, where μi is the “quality” (inherent productivity) of firm i) => selection bias. 2. Time-series comparison of pre- and post-ownership change: yit = δ0 + δ1Postt + Xi δ2 + uit , estimated using a sample of changers. But other factors may be changing in the same period (e.g., ui = γt + Є i, where γt is a time effect that cannot be distinguished from Postt).
Estimation Strategy: This Paper (1) Use panel data, large in both the cross-section (i) and time (t) dimensions. Start with a basic 2-factor Cobb-Douglas specification, treating any private effect as Hicks-neutral technical change: yit = αkit + βlit + δ1Privateit-1 + uit. Permit shape of production function to vary across industries (j, 10 in sample) and include unrestricted industry-year effects (up to 170 dummies): yit = jαjkit*Dj + jβjlit*Dj + jtjtDjDt + δ1Privateit-1 + uit, δ1 = mean within-country-industry-year difference in multifactor productivity between firms majority private and majority state-owned.
Estimation Strategy: This Paper (2) Identification of δ1 involves assumptions about the error term uit. Suppose δ0 = difference in productivity between firms that become privatized at some point or other and those that do not, so that uit = δ0 + νit for firms that are privatized. The difference-in-differences (DD) estimator adds a group fixed effect, EverPrivatei = 1 if Privateit = 1 for any t in the data: (DD) yit = jαjkit*Dj + jβjlit*Dj + jtjtDjDt + δ0EverPrivatei + δ1Privateit-1 + εit. Key identifying assumption is strict exogeneity conditional on all independent variables and the group fixed effects: cov(Privateit-1, εis) = 0 for s, t = 1,…,T, so that the error term in any year s is uncorrelated with the ownership of the firm in any year t.
Estimation Strategy: This Paper (3) Or suppose uit = μi + εit, with μi a fixed firm effect (FE): yit = jαjkit*Dj + jβjlit*Dj + jtjtDjDt + δ1Privateit-1 + μi + εit. Identifying assumption is strict exogeneity conditional on all independent variables and the fixed effects μi : cov(Privateit-1, εis) = 0 for s, t = 1,…,T. Or suppose uit = μi + ωit + εit, with ωi a firm-specific time trend (FT): yit = jαjkit*Dj + jβjlit*Dj + jtjtDjDt + δ1Privateit-1 + μi + ωit + εit. Identifying assumption is again strict exogeneity conditional on all independent variables, the fixed effects μi and the time trends ωi. Similar equations can be specified for Foreign and Domestic private ownership.
Estimation Strategy: This Paper (4) Specification tests: F-tests on joint significance of fixed effects μi and the time trends ωi. “Pre-program test”: is E(εisPrivateτ|s<τ)=0? τ is the privatization date. If privatization appears to affect firm productivity well before the privatization date, then selection bias may be affecting estimates. Similar tests can be specified for Foreign and Domestic private ownership. Many other robustness checks: constant-returns to scale, translog, labor productivity, assumed technologies, controlling for investment, controlling for survival selection bias, instrumenting factors of production, etc.
Estimation Strategy: This Paper (5) Dynamics of the Privatization Effect: yit = jαjkit*Dj + jβjlit*Dj + jtjtDjDt + π-4Pre4it + π-3Pre3it+ π-2Pre2it + π-1Pre1it + π0PrivYearit + π1Post1it + π2Post2it + π3Post3it + π4Post4it + π5Post5it + νit Preprogram test: π-4 = π-3 = π-2 = 0. Similarly for FE (fixed effect) and FT (firm specific trend) specifications. Similar for foreign and domestic.
Estimated Productivity Effects of Privatization
Foreign and Domestic Effects
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Romania
Effect of Foreign Ownership on Productivity: Hungary
Effect of Foreign Ownership on Productivity: Hungary
Effect of Foreign Ownership on Productivity: Hungary
Effect of Foreign Ownership on Productivity: Russia
Effect of Foreign Ownership on Productivity: Russia
Effect of Foreign Ownership on Productivity: Russia
Effect of Foreign Ownership on Productivity: Ukraine
Effect of Foreign Ownership on Productivity: Ukraine
Effect of Foreign Ownership on Productivity: Ukraine
Effect of Domestic Ownership on Productivity: Hungary
Effect of Domestic Ownership on Productivity: Hungary
Effect of Domestic Ownership on Productivity: Hungary
Effect of Domestic Ownership on Productivity: Romania
Effect of Domestic Ownership on Productivity: Romania
Effect of Domestic Ownership on Productivity: Romania
Effect of Domestic Ownership on Productivity: Russia
Effect of Domestic Ownership on Productivity: Russia
Effect of Domestic Ownership on Productivity: Russia
Effect of Domestic Ownership on Productivity: Ukraine
Effect of Domestic Ownership on Productivity: Ukraine
Effect of Domestic Ownership on Productivity: Ukraine
Pre-Program Tests: F-Statistics (P-Values)
Estimates of the Productivity Effect of Privatization Assuming Alternative Production Functions Hungary
Estimates of the Productivity Effect of Privatization Assuming Alternative Production Functions Romania
Estimates of the Productivity Effect of Privatization Assuming Alternative Production Functions Russia
Estimates of the Productivity Effect of Privatization Assuming Alternative Production Functions Ukraine
Conclusions Estimates of the effect of privatization on productivity: 0.15 in Romania 0.08 in Hungary 0.02 in Ukraine -0.03 in Russia Foreign ownership effects are much stronger than domestic: 0.21 to 0.35 in all countries. Controlling for selection bias is a much bigger concern than the method of productivity measurement. Results are inconsistent with simple stories about the differences in policy and business environment and in the methods of privatization programs in these countries.