Efficiency in Asia-Pacific Airports and Corruption: a flexible estimation

Transcription

1 Efficiency in Asia-Pacific Airports and Corruption: a flexible estimation Randrianarisoa L., C. Ordas Criado and T. Oum 2015 ITEA, Oslo, Norway June 17th, / 28

2 Context Recent empirical evidence establishes a negative link between corruption and efficiency in the US and UE airport industry (Yan et Oum, 2014, JUE; Randrianarisoa et al, 2015, TRA). In Asia, empirical evidence suggests that corruption can be profit/growth enhancing, the so-called Asian paradox (Vial et Hanoteau, WD, 2010: Jiang et Nie, EL, 2014) From a theoretical standpoint, the impact of corruption on allocative efficiency and growth is debated. Sand or grease of the wheels of the economy? Corruption : misuse of public office for private gain (Svensson, 2005) What is the relationship between corruption and the efficiency of Asia-Pacific airports? Positive or negative? Linear or nonlinear? 2 / 28

3 Context Asian countries are among the most exposed to corruption according to the International Country Risk Guide corruption index (ICRG) Transparency International Corruption Perception Index (CPI) World Bank s Control of Corruption Index (CCI) Corruption scandals may have extreme political consequences. Two well-known cases in the Asian airport industry : Allegations of corruption during the development of Suvarnabhumi Airport used as justifications for military coup in Thailand in 2006 (The Nation, 2007). China executed the former chairman of a huge state-owned airport holding company... after he was convicted on bribery and embezzlement charges involving more than $14.6 million. (The New York Times, August 7, 2009) 3 / 28

4 Context Most empirical evidence on the efficiency-corruption link is based on the following panel model : E it = α i + α t + λc it + BC itβ + (C it BC it ) γ + X itδ + ε it linear individual/time effects model with selected additive interactions. In general, researchers focus on E(E it ). Main results for airports: the effect of corruption on efficiency is (-) and its strength depends on business characteristics (BC), such as the ownership form. Example: if private infrastructures are more sensitive to corruption efficiency gains from privatisation can be neutralized by efficiency losses due to corruption. 4 / 28

5 Objectives 1. We explore the efficiency-corruption link for Asia-Pacific airports with simple linear panel models for different samples of coutries (Asia-Pacific as a whole, Asia, Oceania). 2. We relax some important restrictions of the above models to avoid misspecification: more flexibility : allow the corruption effect to be nonlinear and non additive; unrestricted interactions: allow arbitrary interactions between the continuous covariates and airports characteristics; optimal pooling of the data in an heterogeneous context. 5 / 28

6 Objectives 2. To avoid misspecification, we rely on kernel estimators and we rather focus on : F (E i,t C i,t, BC i,t, X i,t, CTY i,t, TIME i,t ) We provide empirical evidence based on stochastic dominance relationships of the conditional distribution of efficiency; conditional quantiles Q τ (E i,t C i,t, BC i,t, X i,t, CTY i,t, TIME i,t ) 6 / 28

7 Literature Review Researchers have two conflicting views regarding the effects of corruption on economic/firms efficiency. Sands the wheels hypothesis : Theory: Murphy et al. (1991, 1993), Acemoglu et Verdier (2000, AER), Claessens et Laevens (2003, JF), Yan and Oum (2014, JUE, airports). Empirics: Mauro (1995, QJE),, Dal Bó and Rossi (2007, electricity), Fisman and Svensson (2007), Yan and Oum (2014, JUE), Randrianarisoa et al. (2015, TRA, airports) Greases the wheels hypothesis : Theory: Lui (1985, JPE), Beck and Maher (1986, EL), Acemoglu et Verdier (1998, EL; 2000, AER) Empirics: Leff (1964, ABS), Rock et Bonnett (2004, WD), Egger and Winner (2005, EJPE), Meon and Weill (2010, WD), Jiang and Nie (2014, EL). 7 / 28

8 Methodology Our parametric analysis relies on linear panel FE or RE-effects estimators. where E it = α i + α t + λc i,t + OF i,tβ + (C it OF it ) γ + X itδ + ε it C it is the country-level corruption index (ICRG retained) OF it = [PrivMaj it, GovMaj it, Public it ] is the ownership form of airport i at time t X it = [Hub it, Competition it, GDP it ] is a set of control variables we estimate the model for several airports (sub-)samples : all airports (Asia-Pacific) Asian airports Oceanian airports (Australia and New Zealand) 8 / 28

9 Methodology The kernel estimator of the conditional distribution is the one proposed by Li, Lin and Racine (2013, JBES): F (y z) = F (y, z) ˆf (z) N T i = (NT ) 1 G ((y y js )/h y ) K hz,js,z/ˆf (z) j=1 s=1 reliable in small samples (n=100 to 300) if # continuous relevant covariates is limited ( 2). The optimal values for h = (h y, h z ) are given by least-squares cross-validation (lscv) : 1 N T i CV (h) = {I (y it y js ) NT (NT 1) F (js) (y js z it )} 2 dy. Advantages of lscv: j=1 s=1 i j,t s 1. optimal pooling of the data across characteristics, 2. identification of the relevant determinants of the efficiency distribution. 9 / 28

10 Methodology We compare conditional distributions of efficiency across airports characteristics with a simple stochastic dominance test proposed by Li, Lin and Racine (2013, JBES): 1. FOSD is measured by the KS statistic D F = min {sup[f B (E it x) F A (E it x)], sup[f B (E it x) F A (E it x)]} F A FOSD F B if D F < 0: at least as high proportion of efficient airports in A as in B and E A (rvfp) > E B (rvfp). 2. SOSD measured by the difference D S = max(eit ) min(e it ) [F B (E it x) F A (E it x)]de it F A SOSD F B if D S > 0: E A (rvfp) > E B (rvfp). 3. The distribution of both statistics under the null F A = F B relies on a double-bootstrap procedure (no asymptotics). 10 / 28

11 Methodology red distribution FOSDs and SOSDs black distribution f(e_it x) f(e_it x) Efficiency_it Efficiency_it red distribution SOSDs black distribution F(E_it x) F(E_it x) Efficiency_it Efficiency_it 11 / 28

12 Data The data on Asian-Pacific airports are compiled by the ATRS. We use an unbalanced panel of the 37 major airports located in 12 countries and observed between 2003 and 2011 (260 observations). Efficiency = residual (or net) variable factor productivity (rvfp): rvfp is used by the ATRS (airports benchmarking annual reports) and in Randrianarisoa et al (2015) for EU airports; (a) agg. outputs (# aircraft mvts, PAX, non-aero. revenues) / agg. inputs (labor (eft), purchased G-M-S but NO capital) (b) clean-out effects of factors that managers can t control in the short run: vfp cargo shr + capacity constraints + avg airport and aircraft sizes + macro shocks. We explored three corruption indices (ICRG, CPI and CCI): we retained the ICRG index. 12 / 28

13 Parametric models: Asia-Pacific Airport-effects Country-effects Explanatory variables Pooled OLS Random Effects Fixed Effects Fixed Effects Intercept (0.137) (0.196) - (0.273) Corruption (icrg index) (0.045) (0.042) (0.183) (0.089) Ownership form Reference category : public ownership Gov. majority (0.325) (0.348) - (0.297) Private majority (0.228) (0.301) - (0.229) Interaction between ownership and corruption Reference category: public ownership Gov. majority corruption (0.084) (0.080) (0.209) (0.092) Priv. majority corruption (0.101) (0.122) (0.189) (0.101) Hub status Reference category: not hub Hub (0.061) (0.074) Competition level Reference category: weak Competition (0.054) (0.061) - (0.036) GDP per capita -1.66e e e e-06 (2.62e-06) (1.67e-06) (2.16e-06) (2.29e-06) Obs Adj. R Hausman test / 28

14 Parametric models: Asia Airport-effects Country-effects Explanatory variables Pooled OLS Random Effects Fixed Effects Fixed Effects Intercept (0.280) (0.340) (0.372) (0.339) Corruption (ICRG index) (0.062) (0.063) (0.161) (0.087) Ownership form Reference category: public ownership Gov majority (0.369) (0.456) - (0.358) Interaction between ownership and corruption Hub status Reference category: public ownership Gov majority (0.095) (0.104) (0.188) (0.104) Competition level Reference category: not hub (0.075) (0.068) - (0.061) Reference category: weak (0.071) (0.076) - (0.066) GDP per capita (0.000) (0.000) (0.000) (0.000) Obs Adj.R Hausman test / 28

15 Parametric models: Oceania Airport-effects Country-effects Explanatory variables Pooled OLS Random Effects Fixed Effects Fixed Effects Intercept (0.468) (0.180) (0.254) (0.664) Corruption (ICRG index) (0.292) (0.130) (0.199) (0.347) Ownership form Ref category: public and gov majority Private majority (0.474) (0.338) - (0.487) Interaction between ownership and corruption Ref category: public and gov majority Private majority (0.297) (0.172) (0.159) (0.304) Hub status Competition level Reference category: not hub (0.084) (0.072) - (0.084) Reference category: weak (0.047) (0.073) - (0.047) GDP per capita (0.000) (0.000) (0.000) (0.000) Obs Adj. R Hausman test / 28

16 Main regression results de(efficiency C, X)/dC is mainly (-) for public airports (+) for airports with government majority not significant for private airports Airports with government majority ownership are less efficient than private and public airports. Hub status and competition level do not seem to be relevant determinant of airports efficiency in Asia-Pacific. We need to rely on several estimators and sub-samples to better capture cross-country heterogeneity. Effects of some determinants are not always identified (lack of time variability in airports characteristics). Let s consider the conditional distributions. 16 / 28

17 Distribution of efficiency conditioning upon one covariate UB f (y x) F (y x) x variables y x y x y x corruption (x c ) ownership (x d ) hub (x d ) competition (x d ) gdp per capita (x c ) countries (x d ) time (x o ) We rely on F (y x). It is optimal to pool the data over time It is optimal to semi-pool across hub status. We observe heterogeneity across ownership form, competition levels and countries. Continuous covariates are not smoothed out, they do have an impact on the distribution of efficiency. 17 / 28

18 Distribution of efficiency conditioning upon one covariate... by hub status f( Efficiency Hub ) All airports Hub Not hub F( Efficiency Hub ) Hub Not hub Efficiency (rvfp) Efficiency (rvfp) The grey and solid line shows the unconditional density of efficiency. rvfp distributions are very similar across hub status (semi-pooling). 18 / 28

19 Distribution of efficiency conditioning upon one covariate... by ownership form f( Efficiency Ownership ) All airports Private majority Government majority Public F( Efficiency Ownership form, X* ) Private majority Government majority Public Efficiency (rvfp) Efficiency (rvfp) Strong heterogeneity across ownership forms. Peaks of the global distribution of rvfp are linked to ownership. PrivMaj and Public categories almost FOS-dominates the GovMaj one. 19 / 28

20 Distribution of efficiency conditioning upon one covariate... for low and high corruption levels f( Efficiency Corruption ) All airports Corruption low (q25) Corruption high (q75) F( Efficiency Corruption ) Low corruption (q25) High corruption (q75) Efficiency (rvfp) Efficiency (rvfp) The efficiency distribution of airports located in cleaner environments seems to FOS dominate the one of airports exposed to higher corruption. 20 / 28

21 Distribution of efficiency conditioning upon one covariate Inference based on the stochastic dominance analysis FOSD SOSD (dominance relationship) D S -stat p-value D F -stat p-value (corlow corhigh) (privmaj gov) (privmaj govmaj) (gov govmaj) (hub=yes hub=no) (comp=weak comp=strong) (ctyf=tha ctyf=aus) (ctyf=tha ctyf=jap ) (ctyf=tha ctyf=chi) Corruption hurts efficiency. Public airports are not less efficient than privmaj ones, govmaj structures are the least efficient. More competition (proximity) across airports hurts efficiency. Country effects are relevant. 21 / 28

22 Conditioning efficiency upon all covariates f (y x) F (y x) variables UB icrg cpi icrg cpi efficiency (y c ) corruption (x c ) 3e ownership (x d ) hub (x d ) competition (x d ) gdp per capita (x c ) 3e+10 8e+11 4e+09 8e+11 countries (x d ) time (x o ) We rely on F (y x). Most covariates are relevant, except GDP (smoothed out). Semi-pooling is identified over time. No pooling across business characteristics or countries. 22 / 28

23 Conditioning efficiency upon all covariates Stochastic dominance with other covariates at median/modal values (modal country is Australia) FOSD SOSD (dominance relationship) covariates D F -stat p-value D S -stat p-value (corlow corhigh) privmaj, X (corlow corhigh) govmaj, X (privmaj govmaj) X (hub=yes hub=no) X (comp=weak comp=strong) X (yearo=2003 yearo=2011) X Effect of corruption on efficiency depends on the ownership form, all else equal. Private airports are more efficient than govmaj ones. Hubs are way more efficient and more competition hurts efficiency. Time effects are relevant and negative for Australia. 23 / 28

24 Conditional bivariate distributions for selected covariates Conditional quantiles also capture strong nonlinearities and interactions. 1.0 f( Efficiency Corruption, GovMaj, X* ) Efficiency (rvfp) Corruption (icrg) F( Efficiency Corruption (icrg), GovMaj, X* ) Low corruption High corruption Efficiency (rvfp) Government-majority airports are more sensitive to corruption in Australia. 24 / 28

25 Conditioning efficiency upon all covariates Browsing across the levels of the country variable unveils strong heterogeneity FOSD SOSD (dominance relationship) covariates D F -stat p-value D S-stat p-value China (corlow corhigh) govmaj, X (hub=yes hub=no) X (comp=weak comp=strong) X (year=2003 year=2011) X Thailand (corlow corhigh) govmaj, X (hub=yes hub=no) X (comp=weak comp=strong) X (yearo=2003 yearo=2011) X Corruption is efficiency-enhancing in China but the distributional shift is small. Competition hurts strongly efficiency and we find a positive trend in efficiency between 2003 and Corruption is efficiency-enhancing in Taiwan but the main driver of efficiency is the time factor (strong negative shock between 2003 and 2011). 25 / 28

26 Conclusion We focus on the residuals variable factor productivity - a widely-used measure in the airport industry - to compare airports efficiency in Asia-Pacific. Standard linear panel models had difficulties to identify some determinants of airports efficiency. The distributional analysis optimally pools the data and accounts for non linearities and interactions between continuous and discrete covariates. Corruption, ownership form, hub status, competition level, country/years effects are relevant determinants of airports efficiency while GDP appear to be irrelevant. 26 / 28

27 Conclusion We find strong heterogeneity in the relationships depending on the country. Corruption has no effect on the efficiency of private airports (mainly located in Australia and New Zealand); it hurst the efficiency of airports with public ownership and slightly improves the efficiency of airports with gov-majority ownership (short run Asian Paradox?). Business characteristics have in general a larger impact on efficiency than corruption: More competition hurts efficiency Hubs are in general more efficient In many countries, time-shocks do have a significant impact on efficiency. 27 / 28

28 Thank You For Your Attention 28 / 28