Insider Ownership and the Performance of Firms: Evidence from Russia. Boris Kuznetsov Interdepartmental Analytical Centre, Moscow

Transcription

1 Insider Ownership and the Performance of Firms: Evidence from Russia Boris Kuznetsov Interdepartmental Analytical Centre, Moscow Giovanni Mangiarotti CERT, Heriot-Watt University Mark Schaffer CERT, Heriot-Watt University, WDI and CEPR September 2002 Centre for Economic Reform and Transformation Heriot-Watt University, Edinburgh

2 Motivation Possible impact of insider ownership on firm performance: 1. Linking insiders payoffs to firm performance creates highpowered incentives, raises effort 2. But can also lead to excessively cautious behaviour by riskaverse insiders. 3. Insider ownership Low rate of managerial turnover Low rate of introduction of new managerial skills. 4. Weak investor protection provides incentive for insiders to expropriate minority/other shareholders. This paper: Focus on the expropriation channel (4).

3 Direction of Impact of Insider Ownership on Performance 1. High-powered incentive effects positively linked to performance 2. Risk-aversion effects negatively linked to performance (1) & (2) comprise the standard incentive-insurance tradeoff from principal-agent theory. 3. Managerial turnover effects are negatively linked to performance. 4. Expropriation effects are negatively linked to performance. or are they? We argue: Expropriation channel implies a U-shape relationship between insider ownership and performance, negatively-sloped at low-medium levels of insider ownership, positively-sloped at high levels of ownership. Intuition at very high levels of insider ownership, there s no one for insiders to expropriate.

4 Insider Stakes, Expropriation, and Firm Performance Expropriation generates a U-shape relationship between insider stake and performance: Very low Little expropriation possible Good Performance insider stake Controlling Expropriation possible Poor Performance insider stake and but not 100% Incentive to expropriate 100% No expropriation because... Good Performance insider stake No outsiders to expropriate Example: δ = Share of firm owned de jure by insiders -and- = Measure of insiders power to expropriate δ (1-δ) = Expropriation of outsiders = (Insider power) * (Outsiders de jure share) = 0 if insider power is zero = 0 if insiders own 100% of firm = max at δ=50%

5 Insider Stakes, Expropriation, and Firm Performance: Empirical Implications Empirical studies: An inverse U-shape relationship between insider/managerial stakes and performance. Peak at a 5-15% insider (managerial) ownership share. Risk aversion, managerial turnover, and expropriation effects rapidly dominate high-powered incentives as insider ownership increases. but these studies typically draw on Western experience, where the insider/managerial ownership share is usually relatively low. We argue: Look at the relationship at medium to high levels of insider ownership. A positive insider-ownership / performance relationship at high levels of insider ownership a large negative expropriation impact at medium levels of insider ownership. Large means: Marginal impact of removing expropriation channel by increasing the insider stake is not just > 0, it s > negative marginal impact of increased risk aversion and lower managerial turnover. NB: if net effect is positive, expect the major contribution was from removal of the expropriation channel, since the positive impact of additional high-powered incentives was already swamped at low levels of insider ownership. but where to find a set of firms with such an ownership structure? And how did they get that way?

6 The Endogeneity of Ownership Problem Demsetz and Lehn (1985): - Optimal ownership structure is firm-specific - Market forces drive ownership towards this optimum. Ownership is endogenous to expected performance. (Demsetz-Lehn: should not expect to find a relation between ownership structure and profitability, but the endogeneity problem is very general.) Here: If we find that firms with 100% insider ownership perform better than firms with 51% insider ownership, does this mean that Removal of the expropriation channel improves performance? or Insiders had better access to shares, and were able to buy 100% if they knew that the firm was a good one? NB: Law and Finance literature: Common law countries offer less minority shareholder protection than civil law countries Concentrated ownership more frequent in common law countries than in civil law countries

7 Expropriation and Corporate Governance in Russia - The Russian privatization programme generated firms with large very shareholdings by management and workers. - Protection of minority/outside shareholders is poor in Russia. - The Russian privatization programme gave insiders preferential access to shares in a well-understood process. Russia is an excellent natural experiment for our purposes: Many firms privatized with large insider shareholdings. Initial ownership structure is far from the equilibrium ownership structure. But still considerable cross-sectional variation in these insider stakes. This cross-sectional variation can be modelled and estimated.

8 Modelling Performance Performance uses TFP approach on survey-based cross-section data: 170 privatized Russian manufacturing firms in Cobb-Douglas production function: y i = γxi + α ln li + βln ki + ui ln (1) Output: Log capital: Log labour: Log value added log fixed assets corrected by capacity utilization rate log annual average number of employees and γ is a vector of coefficients on the X variables that influence TFP: Degree of insider ownership (more on this shortly) Competitive position of the firm: Number of regional competitors in the firm s main market: - None (monopoly, the benchmark category) - 1 competitor competitors - > 5 competitors 3 dummy variables for regional competition

9 Modelling Performance (continued) Competitive environment of the firm (market concentration): Log (1/HHI) where HHI = 5-digit national Herfindahl-Hirschman index Interpretation: 1/HHI is the number of firms that would be observed in the market if firms were all the same size. Coefficient in the TFP regression is an elasticity: the % impact on TFP of a % change in the # of competitors 2-digit industry dummies (6 categories) Region dummies (12 categories) Time elapsed since privatization Note: measuring performance using TFP means we pick up some effects of expropriation, e.g., Extraction of surplus via transfer pricing Asset stripping, theft Diversion of managerial and labour effort from production to expropriation but not others, e.g., Diversion of profits/dividends into wages, bonuses

10 Size of insider shareholdings: what did Russian insiders choose? Amount of shares obtained by management and workers in the Russian privatization programme a choice variable... subject to constraints. Variant I, Variant II, etc. Purchase of additional shares through auctions Liquidity constraints facing managers, workers Size of firm, size of book value of capital Pre-1992 privatization, i.e., leaseholds Was the firm worth owning? Questions: How to formalize this? Is it reasonable to treat managers and workers as a single group of insiders? We address both questions with a little-known stylized fact.

11 Distribution of Insider Shareholdings following Privatization...was tri-modal: 1. Peak at 0% 2. Little between ~0% and ~50% 3. Peak at ~50% 4. Little between ~50% and ~100% 5. Peak at ~100% and shares of managers and workers considered separately do not show this pattern. Conditional on being privatized, insider decision was (approximately) binary: Choose to be 100% insider-owned or Choose to be majority insider-owned It is reasonable to consider manager and worker shares together. Note: We get similar results if manager and worker shares are handled separately. Note: 51% insider share at privatization shares were voting shares insider share was a controlling stake

12 Evidence: World Bank survey of 1994 Distribution of insider shareholdings at time of survey:.25 Fraction Insider share

13 This pattern does NOT emerge so clearly from managers or workers separately: Managerial shareholdings:.25 Fraction Managers' share Worker shareholdings:.25 Fraction Workers' share

14 Evidence: BEA/CERT/LBS Survey Data used in this paper. Privatized firms only. Insider shareholdings at time of privatization:.25 Fraction Insider share at privatization

15 Pattern again absent from managerial or worker stakes separately: Managerial shareholdings:.25 Fraction Managers' share at privatization Worker shareholdings:.25 Fraction Workers' share at privatization

16 Dispersion of Non-100% Insider Stakes since Privatization Insider share at privatization:.25 Fraction Insider share at privatization Insider share in 2000:.25 Fraction Insider share in 2000

17 Modelling Privatization What factors ( instrumental variables, exogenous to the insider stake decision) influenced how many shares insiders obtained at privatization? Capital intensity: 1992 book value of capital determined (low but not zero) price at which insiders could buy shares. Variable: Log K/L ratio in 1992 Expected sign of coefficient: negative Size/coordination costs: Managers & workers had to coordinate to buy 100% of firm. More insiders harder to coordinate Variable: Log L in 1992 Expected sign of coefficient: negative Profitability: More profitable firms more attractive for insiders to acquire. NB: Profitability derives from various factors, e.g., market power. Variable: position of the firm (percentile) in the distribution of average profit per worker at time of privatization. Distribution based on industrial registry. Expected sign of coefficient: positive Wages: Higher-wage firms more attractive for insiders to acquire. Variable: position of the firm (percentile) in the distribution of average wage at time of privatization. Distribution based on industrial registry. Expected sign of coefficient: positive

18 Modelling Privatization (continued) Percentile position of wage and profit per worker used because: - Not sensitive to outliers - Not necessary to convert to constant rubles ( a period of high inflation) - Coefficients on wage and profit/l directly comparable but in fact the results are similar if other measures are used. Competitive environment Variable: Log (1/HHI) Expected sign of coefficient:? 2-digit industry dummies (6 categories) Region dummies (12 categories) Note that we do not include variables that are themselves endogenous and did not independently influence the privatization decision, e.g., the choice of Variant 1, Variant 2, etc.

19 Basic Sample Statistics: BEA/CERT/LBS Survey Survey conducted in Interview-based. Firms randomly sampled from industrial registry, subject to constraints/stratification. Privatized firms only Firms employing 100-5,000 people only Stratified on employment ( , , ) 4 macro-regions only (European Russia, Volga, Urals, Siberia) Manufacturing firms (excluding metals) Stratified on 2-digit manufacturing industries Historical data back to 1985 taken from the industrial registry Total sample is 437 firms, but missing historical data and removal of minority insider-owned firms reduces sample to 170 firms: 76 are 100% insider-owned 94 are majority (51-99%) insider-owned

20 1997 Employment Mean Median Log 1997 VA Mean Median Years since priv. Mean Median /HHI Mean Median 1/HHI at priv. Mean Median Prof/L at priv (%-tile) Mean Median Wage at priv (%-tile) Mean Median 1992 Log K/L Mean Median Number of regional competitors (%) None >5 Total Basic Stats (continued) Majority insider-owned % 12% 28% 28% 100% 100% insider-owned % 11% 30% 33% 100%

21 Estimation Strategy Two related estimation methods: Method 1: Single equation instrumental variables estimation Instrumental variables, GMM, or LIML estimation of equation (1), instrumenting for insider ownership. Insider ownership measured as a continuous variable or as a binary variable (100% vs. majority-insider). Results very similar in both cases. Single-equation estimates consistent under a wide range of assumptions, but not efficient. Method 2: System estimation ( treatment effect model) Full maximum-likelihood estimation of equation (1) and explicit equation for insider-ownership decision: Zi = 1 if δwi + ei > 0 and corr( ei, ui ) = ρ (2) where Z=1 if the firm is 100% insider-owned at time of privatization Z=0 if it is majority insider-owned at time of privatization and W is the set of variables that determine the decision of insiders to choose 100% vs. majority ownership. Efficient form of estimation, but less robust.

22 Single-Equation IV/GMM/LIML Regression Diagnostics and Tests 1. Endogeneity of labour and capacity-adjusted capital Instrument ownership, and labour and capacity-adjusted capital using labour and unadjusted 1997 capital as additional instruments. Report this regression as benchmark result. Next: endogeneity tests using (Durbin-)Wu-Hausman test (an F-test). (In GMM-speak : Can we add additional orthogonality conditions?) (a) Test endogeneity of L and K Result: K exogenous, L endogenous but cannot reject constant returns to scale (b) Reestimate CRS production function (L-intensive form). Result: K/L exogenous Conclusion: cannot reject specification of CRS production function with one endogenous binary regressor. Can use standard treatment effect system estimation. 2. Heteroskedasticity IV, LIML, treatment effect system estimation consistent in the presence of heteroskedasticity, but inefficient and should use robust standard errors for inference. GMM estimation efficient in the presence of heteroskedasticity, but over-rejects the null in finite samples. Pagan-Hall (1983) test for heteroskedasticity in single-equation IV estimation: no heteroskedasticity IV, LIML, treatreg efficient.

23 Single-Equation Regression Diagnostics (continued) 3. Weak instruments problem F-test of excluded instruments in 1 st -stage regression for insiderownership is significant at 1-5% level but F-stat not huge (about 2.5-3) a weak instrument problem, IV biased towards OLS. Hahn-Hausman-Kuersteiner (2001) recommend Fuller s modified LIML when instruments are weak. We use two variants of Fuller s LIML (α=1 and α=4; former is mean-unbiased, latter has smaller MSE). Results: IV, GMM, LIML all yield similar estimates. Weak instrument problem disappears with system (treatment effect) estimation, a result of the efficiency gain from system estimation. 4. Tests of overidentifying restrictions Sargan (IV) / Hansen (GMM) / Anderson-Rubin (LIML) test of overidentifying restrictions, a joint test of: Are the excluded instruments W correlated with the error term? Do the instruments belong explicitly in the main regression? (In GMM-speak : Are the orthogonality conditions valid?) Pass test in all specifications (instrumenting for L, K and ownership, or instrumenting for just ownership) Structure of the model is sound, endogeneity of ownership successfully addressed. Can use standard treatment effect system estimation.

24 Single-Equation Results (1): 100% vs. Majority Insider-Owned IV, GMM, Fuller LIML Estimates (1) IV (2) GMM (3) F-LIML α=1 Dependent variable: Log VA Log VA Log VA Endogenous variables: ownership, capacity-adj K, L Regressors Log K (capacity-adj) 0.473** (0.146) 0.508** (0.143) 0.486** (0.156) Log L 0.645** (0.237) 0.617** (0.225) 0.652** (0.253) 100% insider-owned (1/0 dummy) 1.93** (0.76) 2.03** (0.70) 2.14** (0.81) Log (1/HHI) (0.084) (0.083) (0.090) Regional competitors insig insig insig (3 dummies) Yrs since priv (0.124) (0.136) (0.132) Industry dummies Yes Yes Yes Region dummies Yes Yes Yes Instruments: Log K, Log L 92-93, Log K/L 92, Wage at priv. (p tile), Profit/L at priv. (p tile), Log (1/HHI) at priv. 1 st stage F-tests of own: F(7,140)=2.29, p-value=0.031 excluded IVs K: F(7,140)=75.5, p-value=0.000 L: F(7,140)=102, p-value=0.000 Test of overidentifying restrictictions χ 2 (4)=2.58 p-value=0.631 χ 2 (4)=2.28 p-value=0.685 χ 2 (4)=2.23 p-value=0.694 Pagan-Hall test of heteroskedasticity (w/predicted y & y 2 ) χ 2 (2)=0.000 p-value=0.999 χ 2 (2)=0.020 p-value=0.990 χ 2 (2)=0.003 p-value=0.998 Significance: *=5% **=1%

25 Single-Equation Results (2): % of Shares Owned by Insiders IV, GMM, Fuller LIML Estimates (1) IV (2) GMM (3) F-LIML α=1 Dependent variable: Log VA Log VA Log VA Endogenous variables: ownership, capacity-adj K, L Regressors Log K (capacity-adj) 0.431** (0.130) 0.439** (0.138) 0.460** (0.124) Log L 0.653** (0.218) 0.660** (0.231) 0.606** (0.216) % owned by insiders 0.044** (0.016) 0.048** (0.016) 0.043** (0.014) Log (1/HHI) (0.076) (0.081) (0.073) Regional competitors insig insig insig (3 dummies) Yrs since priv (0.117) (0.124) (0.134) Industry dummies Yes Yes Yes Region dummies Yes Yes Yes Instruments: Log K, Log L 92-93, Log K/L 92, Wage at priv. (p tile), Profit/L at priv. (p tile) 1 st stage F-tests of own: F(6,141)=2.61, p-value=0.020 excluded IVs K: F(6,141)=88.6, p-value=0.000 L: F(6,141)=120, p-value=0.000 Test of overidentifying restrictictions χ 2 (3)=2.63 p-value=0.453 χ 2 (3)=2.34 p-value=0.506 χ 2 (3)=2.52 p-value=0.472 Pagan-Hall test of heteroskedasticity (w/predicted y & y 2 ) χ 2 (2)=0.427 p-value=0.808 χ 2 (2)=0.453 p-value=0.797 χ 2 (2)=0.594 p-value=0.743 Significance: *=5% **=1%

26 Single-Equation Results (3): 100% vs. Majority Insider-Owned Endogeneity of K and L in IV Estimation (1) (2) CRS (3) CRS Dependent variable: Log VA Log VA/L Log VA/L Endogenous variables: ownership, K, ownership, ownership L K/L Independent variables: Log K (capacity-adj) 0.473** (0.146) Log L 0.645** (0.237) Log (K/L) 0.497** (0.131) 0.454** (0.094) 100% insider-owned 1.93** (0.76) 1.66** (0.611) 1.63** (0.611) Log (1/HHI) (0.084) (0.076) (0.074) Regional competitors insig insig insig (3 dummies) Yrs since priv (0.123) (0.113) (0.111) Industry dummies Yes Yes Yes Region dummies Yes Yes Yes Instruments (shared): Log K/L 92, Log L 92, Log (1/HHI) at priv., Wage at priv. (p tile), Profit/L at priv. (p tile) Instruments (other): Log K, Log L 93 n.a. DWH test of L: pval=0.05 K/L: n.a. exogeneity of L, K/L K: pval=0.56 p-value=0.63 Test of overidentifying restrictictions χ 2 (4)=2.58 p-value=0.631 χ 2 (4)=3.69 p-value=0.595 χ 2 (4)=3.53 p-value=0.473 Wald test of constant χ 2 (1)=0.56 n.a. n.a. returns to scale p-value=0.453 Pagan-Hall test of heteroskedasticity χ 2 (2)=0.000 p-value=0.999 χ 2 (2)=0.453 p-value=0.797 χ 2 (2)=1.25 p-value=0.536 Significance: *=5% **=1%

27 System Estimation: Full Maximum Likelihood (1) Productivity (2) Ownership Dependent variable: Log VA/L 100% insider-owned Independent variables: Log (K/L) 0.360** (0.066) 100% insider-owned 0.894** (0.270) Log (1/HHI) (0.052) Regional competition insig (3 dummies) Yrs. since priv (0.068) Log K/L * (0.161) Log L ** (0.118) Profit/L at priv. (p tile) ** (0.0048) Wage at priv. (p tile) (0.0044) Log (1/HHI) at priv (0.086) Industry dummies Yes Yes Region dummies Yes Yes ρ=covariance of error terms in (1) and (2) ** (0.109) Joint test of significance of variables in (2) but excluded in (1): χ 2 (5)=36.9, p-value=0.000 Weak instruments problem largely gone. Significance: *=5% **=1% Robust standard errors in parentheses.

28 Summary of Results 100% insider ownership has a positive impact on productivity vs. alternative of majority insider ownership. Interpretation: 100% insider-owned no one to expropriate no (real/apparent) resources lost in expropriation activity General implication: expropriation channel is large and significant. Endogeneity of ownership problem successfully addressed using the Russian privatization program as a natural experiment. Plausible production function estimates. ρ negative and significant clear evidence of endogeneity of ownership, but opposite of anticipated direction. Explains why, in OLS estimation, no significant impact of 100% insider ownership. Insider choice of 100% insider ownership: - More likely if firm had large profits per worker... but - High wage firms no more likely to be taken 100% by insiders - Less likely if K/L ratio was high ( more expensive) - Less likely if L large ( coordination more difficult) Findings hold under a wide variety of specifications: IV, GMM, LIML, log sales as dependent variable, different measures of wage and profit/l, correcting for labour utilization. In a Russian-type environment with poor investor protection, 100% insider ownership has short-term advantages vs. majority insider ownership... implications for ownership structure in the medium term?