Corruption, Productivity and Transition *

Transcription

1 CENTRE FOR ECONOMIC REFORM AND TRANSFORMATION School of Management and Languages, Heriot-Watt University, Edinburgh, EH14 4AS Tel: /3485 Fax: World-Wide Web: Corruption, Productivity and Transition * Geoffrey Wyatt ** Heriot-Watt University May 2002 Discussion Paper 2002/05 Abstract The level of productivity is correlated across countries with measures of (lack of) corruption, but this appears to be due to a common association of these variables with measures of social infrastructure, here measured by a combination of governance indexes labelled rule of law and government effectiveness. Econometric use of instruments establishes that causation runs from social infrastructure to productivity. Socialism has had a positive direct effect on productivity but one that is dominated by a negative indirect effect via social infrastructure. Keywords: corruption; governance; productivity; social infrastructure; transition. JEL Classification: H10, O11, P20 * Paper prepared for PHARE-ACE project P R. Presented at the Project Workshop in Sofia, Bulgaria, May This research was undertaken with support from the European Community s Phare ACE Programme The content of the publication is solely the responsibility of the author and it in no way represents the view of the Commission or its services. ** Department of Economics, School of Management and Languages, Heriot-Watt University, Edinburgh EH14 4AS. G.J.Wyatt@hw.ac.uk. 1

2 1. Introduction Many empirical studies of economic growth have reported a connection between growth and diversion, meaning unproductive economic activity pursued for gain via legal or illegal transfers of resources. This is called corruption where it involves transfers to government sector employees arising from side transactions between the private and public sectors. Since much of this type of activity is illegal, the transactions are for the most part unrecorded, and data measuring the extent of corruption has to rely on perceptions of corruption as recorded in surveys of businessmen, country experts and the like. We make use of such data in the empirical analysis reported below, where we find a clear positive correlation across countries between measures of corruption and the level of output per worker (productivity), see Figures 1 and 2. The main question addressed in this paper concerns the interpretation of this correlation: is it cause or effect, or perhaps even spurious? We follow Hall and Jones (1999) in arguing that corruption is a reflection, or indeed an expression, of a wider concept called social infrastructure. We confirm that social infrastructure appears to be causally related to the level of productivity. The principal contribution of this paper is to consider the manner in which corruption features in this relationship and, further, how the former socialist countries, including the transition countries, fit into the picture. The plan of this paper is as follows. In Section 2 we consider the recent literature on the topic. Sections 3 discusses the available data. The empirical econometric analysis is reported in Section 4 and the results obtained are discussed in the concluding Section 5. The data used in the empirical analysis is listed in an Appendix. 1

3 3.5 Productivity and Control of Corruption USA LUX log(productivity) ZAR NER TKM TJK BEL JPN ITA ARE QAT LBY BHR KWT TWN SAU OMN GRC CZEMLT BHS KOR ARG VEN URY HUN GAB SVK MUS BWA RUS MEX ZAF PAN COL CRI MYS EST IRN POL TTO LTU BLR BRA HRV ROM LBN TURLVA NAM THA DZA KAZ ECU BGR GTM TUN UKRDOM MKD SUR GEO PER PRY FJI SWZ JOR SYR PHL JAM MAR BOL EGY IDN SLV AZE KGZ MDA ALB ZWE HND GUY UZBPNG CMR NIC ARM CHN AGO GIN COG LKA CIV SEN GHA PAK TGO SDN LSO HTI GMB BEN VNM IND KEN MNG UGA ZMB TCD BFA GNB MDG NGA BGD YEM MMR MLI MOZ MWI SVN CHL NOR CHE ISL SGPCANDNK FRA HKG AUT AUS DEU SWE GBR NLD FIN IRL ISR NZL ESP PRT CYP -0.5 TZA ETH SLE Control of corruption Figure 1. Productivity and Control of Corruption 2.5 Measures of "Governance" SVN ISR DNK SWE CAN FIN NZL CHE NLD SGP CYP ISL LUXNOR GBR DEU IRL AUS AUT USA FRA HKG PRT ESP CHL Control of corruption FJI GRC ITA JPN BEL EST KWT HUNMYS TWN CRI QAT BWA BHS TTOPOL MLT OMN URY NAM CZE ZAF MUS LSO MAR JOR KOR BRA SVKLTU SLE SUR SWZ TUN GUY GMB ARE CIV JAMLKA GNB MNG MWI THA PER TGO SEN BHR BGD CHN EGY PHL GHA MEX LVA IND ARG ZWE SLV VNM TUR MDA BFA BOL ETH LBN MDGCOL ROM PAN MLI UGA HRV HTI MOZ MKD COG TCD BGR SAU KEN BLRRUS ZMB VEN PAK GEO SYR IDN KGZ ARM BEN DOM GTM ECU AGOLBY DZA YEMKAZ NIC PNGGIN IRN UKR NGA PRY HND TZA UZB SDN GAB ALBAZE MMR CMR -1.5 ZAR TJK NER TKM Government Effectiveness + Rule of Law Figure 2. Control of Corruption and Government Effectiveness + Rule of Law 2

4 2. Brief review of literature Study of the causes and consequences of corruption has increased in recent years with greater availability of the output of international surveys of corruption, e.g. Kaufmann et al. (1999) and Hellman et al, (2000). The recent empirical literature relating corruption, among other determinants, to growth begins with Knack and Keefer (1995) and Mauro (1995). The Knack and Keefer study is based on Barro's (1991) cross country regression formulation. They demonstrate that an index of the quality of institutions, which included a (lack of) corruption component, was highly significant as an explanatory variable for economic growth, indeed as significant as the education variables which proxied for human capital. This study used two survey sources for the institutional data: ICRG and BERI. Mauro, using BI data showed that indexes for bureaucratic effectiveness and for corruption were significant explanatory variables in Barro regressions for the share of investment, but less significant for economic growth 1. This study instrumented the explanatory variables by an index of linguistic fractionalization in order to treat for their potential endogeneity. It concludes that bureaucratic efficacy and corruption affect growth through the investment ratio. In a later paper (Mauro, 1999) it is shown that a composite index of corruption is significantly associated with the share in gdp of government spending on education. This would seem to provide another channel from corruption to economic growth as human capital is usually positively associated with growth. In Barro and Sala-i-Martin (1995) growth regressions involving corruption and rule of law variables are reported; it is found that the corruption variable was insignificant when included together with a "Rule of Law" variable, which was highly significant. In an influential paper, Hall and Jones (1999) argue that "social infrastructure" explains a large part the cross-country variation in the level of per capita output, and does so through physical and human capital formation and productivity. The measure of social infrastructure they use is a combination of ICRG indexes and openness to trade. The potential endogeneity of social infrastructure is treated by the use of geographical and linguistic characteristics of the countries in the world-wide dataset. Kaufmann et al (2000) build on the Hall- 1 ICRG is International Country Risk Guide, BERI is Business Environment Risk Intelligence and BI is Business International, part of the Economist Intelligence Unit. 3

5 Jones analysis by applying their own compilation of measures of governance ("voice", "political violence", "government effectiveness", "regulatory burden", "rule of law" and "graft") separately, using a language instrument - the fraction of the population speaking a major European language. However they encountered a problem with the graft (i.e. corruption) variable, namely that as an explanatory variable in a regression of per capita output it fails a test of the overidentifying restrictions, implying either that it is insufficiently well accounted for by the chosen instruments or that it has a separate direct influence on output per worker. 3. Data 3.1 Governance data A World Bank study (Kaufmann et al, 1999) lists eleven sources of data on corruption. All of these sources conduct questionnaire based surveys, in some cases several surveys, which are very heterogeneous: they have different aims, scope, methodology and sampled population. This study makes use of three of the six indexes synthesised from these disparate sources by the Governance Research Unit at the World Bank, namely those for "Control of Corruption" (CC), "Rule of Law" (RL) and "Government Effectiveness" (GE). Table 1 shows the origin of the data used to compile the indexes further details can be found in the source cited. The World Bank Governance Research Unit created the indices for our three variables by using a statistical methodology based on an unidentified components model. Although there is a substantial range of uncertainty around the estimated indexes, they have the considerable advantages of being comparable across countries and world-wide in coverage. 4

6 Table 1. Sources of governance data Source coverage: number of countries GE RL CC Standard & Poor's DRI / McGraw-Hill n/a Economist Intelligence Unit Heritage Foundation / Wall Street Journal Political Risk Services n/a World Bank / University of Basel Business Environment Risk Intelligence Wall St. Journal Central Europ. Econ. Review Freedom House Gallup International World Economic Forum Political Economic Risk Consultancy Institute for Management Development Productivity The variable we seek to explain is the level of output per worker, not its growth over a number of years. An important reason for this is that the countries of particular interest in this study are the transition countries. Many of these are states that have emerged or re-emerged from the split-up of the Soviet Union. Obviously, for these countries growth data is at best available for a period of about 10 years, which is firstly rather short and secondly covers the exceptional period of transition, which is not yet over. Even the data for those formerly independent socialist states such as Romania and Bulgaria, which have data spanning back several decades, are affected by the distortions and corrections experienced through the transition years. Productivity is GDP at purchasing power parity divided by the labour force, defined as the population aged between 16 and 65. Globally comprehensive figures (excepting Taiwan) for real GDP up to the year 2000 are extracted from the IMF's website. These data extend the Penn World Tables, both in time and in the coverage of countries, to include importantly the transition economies. Labour 5

7 force data comes from the World Bank s online databank. In order to minimise the effects of any particular year which may be abnormal, the cross-country productivity data to be used in Section 4 are the averages over the 1991 to 2000 period. 3.3 Instruments Our aim is to investigate the connections between corruption, social infrastructure and productivity, and for this we need as instruments for social infrastructure variables that are expected to be independent of the level of output per worker in the 1990s. As Hall and Jones point out, it is not easy to find such instruments. Most economic variables are influenced by the level of productivity, which vitiates their use as instruments. Hence the search for valid instruments led Hall and Jones to consider two types of variable: (i) inherent physical characteristics of countries like their latitude (distance from the equator) or their proximity to other countries, and (ii) features which reflect the distant history of countries, such as their exposure to early European influence, possibly by colonization, as currently reflected in the everyday languages spoken. The languages spoken do of course change over time, partly in response to economic circumstances, implying some feedback. It is a matter of judgement whether there is enough feedback to invalidate these variables as instruments independent of the level of productivity. However, entry to the modern world of technology more or less involves the use of English, so the problem with the English language variable in particular may not be minor. The instruments we use are derived from two sources. One is the World Bank s file on Social Indicators and Fixed Factors. These include geographic variables (latitude, land area and landlocked dummy), oil-exporter and transition dummy variables. The other source is a geographic gazetteer listing the names, populations, administrative status and geographic positions of all cities with population greater than 50,000 plus others that have administrative functions as state or regional capitals (see There are over 10,000 such cities. These data were used to construct instruments as follows. 6

8 3.3(i) City sizes The size distribution of cities in most countries approximately follows a Pareto distribution (also called Zipf s law ). This is evident from visual plots of log(size) against log(rank). A roughly straight line is the indicator of a twoparameter Pareto distribution, with the crucial shape parameter being the slope. If it is equal to one then it may be said that "Zipf s law" applies. In fact the Pareto "alpha" (i.e. slope) parameters for empirical size distributions of cities vary across countries, mostly between about 0.5 and 2, see Figure 2. The histogram shows the international dispersion of alphas, calculated for each country by ordinary least squares regression of log(population) against log(size) with a dummy variable for the capital city, for all towns with over 50,000 inhabitants. 100 Histogram of Pareto alpha coefficients, all countries Figure 2. The international distribution of Pareto exponents Output per worker in a particular city and the city s population are surely causally connected, both ways. But what about the distribution of city sizes in a country? Recently there have been put forward theories (e.g. Gabaix, 2000) in terms of stochastic processes which lead to the typical Pareto shape. Empirically, 7

9 while the Pareto curve does move over time (e.g. Parr, 1985), and with it the slope coefficient alpha, that movement, even over many decades, is slight in comparison with the cross-country variations in alpha. For the UK, for example, while it is true that the second city, now Birmingham, used to be Liverpool and a century before that, Glasgow, these swaps can take place without alpha changing by much. Alpha seems to be characteristic of a country s urban make-up over very long periods of time, even centuries. A lower absolute value for alpha implies a greater size equivalence among cities, suggesting perhaps greater autonomy and independence, whereas a higher alpha suggests something of a hierarchical structure among cities, and perhaps less social cohesion. A key question for us is therefore, whether the distribution of city sizes influences social infrastructure and hence national productivity, or whether it has a separate effect of its own on productivity? Accordingly, we shall consider the validity of alpha as an instrument for social infrastructure in a Hall and Jones type regression. 3.3(ii) City density The size distribution of the cities within a country says nothing about their geographical spread. But the spatial distribution of economic activity could also affect social infrastructure and hence national productivity. External economies are often invoked to suggest that agglomerations of economic activity have lower costs than dispersed activity. That would represent a direct effect on productivity. But the spatial distribution of urban centres could have an indirect effect via social infrastructure. The clustering of economic activity may both lower the costs of diversionary activity and increase the opportunities for it, especially where this requires networking among the diverters. On the other hand, concentrated economic activity offers greater opportunities for the observation and control of miscreants. But, again, we shall need to consider whether the spatial dispersion of urban economic activity is exogenous for the level of productivity when there are clear economic incentives to drive clustering. We attempt to construct a simple measure from our city size database which reflects the degree to which cities are clumped together or spread out across 8

10 the country. It should of course allow meaningful comparisons to be made across countries. Consider two cities, of sizes C 1 and C 2 respectively, and let them be separated by a distance d 12. Let us suppose that these two cities make up the whole country. We wish to formulate a measure that tells how clumped or dispersed it is. This will depend on the three variables we have defined. The measure must be increasing in C 1 and C 2, and decreasing in d 12, though always non-negative. Also C 1 and C 2 should be interchangeable, and if d 12 = 0 (so that the cities are contiguous), it would be desirable for the conurbation to reflect simply the combined size, C 1 + C 2. A simple formula that meets these desiderata is: (C 1 + C 2 ).exp( d 12 ). If there are, say, n cities then the extent to which the others are clustered with respect to the first would then be represented: n (C 1 + C i )e d 12 i =1 and summing this over all n cities we arrive at: n 1 n i =1 j= i +1 (C i + C j )e d ij which is basis of the index we use. Now, bearing in mind the Pareto law for city sizes: C i AR α where R is the city rank within a country, we use ranks instead of population numbers in the formula. And to make them comparable across countries, we use not the within-country ranks, but the ranks of the cities in the world distribution, W. For the choice of α, a benchmark of 0.5 seems acceptable based on the data, so our index becomes: n 1 n D = ( W i + W j )e d ij. i =1 j = i+1 Figure 4 shows how the logarithm of city density is related to the Pareto alpha coefficient across countries. The log-transform was taken to avoid excessive clustering on the vertical axis, and indeed the transformed variable seems better 9

11 behaved overall, and is used in the regression analysis of Section 4. Although alpha and log(d) are positively correlated, the dispersion of the scatter in Figure 4 confirms that they are basically distinct variables. 0 Alpha and urban density KOR -0.5 EGY TWN JPN -1 BGD PHL GBR DEU NLD NGA IDN MEX BRA IND PAK VEN ISR log(city Density) DJI BHS SUR QAT GRCZAF POL COL CHN DZA THA MYS FRA ITA JOR TUR ESP DOM MAR UKR CAN CHL BEL SLV IRNLKA SYR ECU VNM ROM ARG LBN HTI CIV MMR PRY PRT USA ARE TUN KEN ARMGTM NIC SAU GHA RUS CZE HUN BGR YEM AUS SEN BLR CMR UZB ZWE PER HND LBY KWT OMN BOL ZMB JAM GEO SDN AZE BHR MUS PAN MDA BEN ETHNPL COG ZAR MKD NZL KHM LTU BFA AUT SWE ALBTJK TZASVK CHE SLE DNK GIN MWI GAB GMB TGO RWA LVAUGA IRL MOZ FIN AGO BDI KGZ URY GNB HRV NOR KAZ CRI MLI MDG MNG GUY LAO SVN NER PNG CAF EST TCD NAM TKM FJI LSO BWA MRT -3.5 NEA SLU CPV BTN STP TTO GNQ ISL SLB SWZ BLZ CYP LUX MDV Alpha Figure 4. Pareto "alpha" and urban density, D 4. Empirical analysis 4.1 Government Effectiveness, Rule of Law and Control of Corruption These three indicators probably capture different aspects of social infrastructure. While Government Effectiveness (GE) and the Rule of Law (RL) relate fairly obviously to institutions of economic policy, Control of Corruption (CC) is an outcome and reflects the behavioural response of government officials and those with whom they do business to the opportunities for diversion. 10

12 Corruption is an opportunistic reflection of weak social norms and constraints, which are themselves an aspect of social infrastructure. It is a form of behaviour in which the parties to the corrupt exchange lack respect for the social rules. But it can also be seen as a perversion of the profit seeking which is the basis of efficient resource allocation in well functioning markets. With appropriate constraints in place, this form of socially harmful profit seeking could presumably be channelled into more socially useful directions. Whereas CC directly reflects the extent and prevalence of corruption, GE and RL probably reflect restraints to diversionary activity, and thereby the cost of engaging in a corrupt transaction. They are also, presumably, variables which can be influenced by government policies and actions. Since a loosening of the restraints reduces the costs of corruption, the three variables should be related, and in particular it seems likely that GE and RL together cause CC, at least in part. Furthermore, one might suppose that the impact on CC of an increment in GE would be larger the higher the level of RL, and similarly that the impact of an increment in RL on CC would probably be greater the higher the level of GE. This argues for the plausibility of an interaction effect, which suggests that an appropriate regression model might be: CC = a + b.ge + c.rl + d.(ge*rl) + eps (1) Table 2. Regressions of equation (1), dependent variable: CC Full dataset Hall & Jones subset constant (0.032) (0.033) GE (0.061) (0.063) RL (0.057) (0.059) GE*RL (0.029) (0.028) R s.e.e nobs DWH p-value Instruments Note: Standard errors in brackets a, A, a*a, ll, lat, log(d), a*log(d), -- see Section 4.2 EngFrac EurFrac log(frankrom) Latitude This regression model was estimated on the World Bank dataset. The possibility of feedback, i.e. that CC also causes GE and RL, which would render ordinary least squares estimates inconsistent, was examined by a Durbin-Wu- 11

13 Hausmann (DHW) test. The regression was estimated both for the full dataset, and also for a subset of the data for which the social infrastructure instruments of Hall and Jones were available. The results reported in Table 2 imply that GE and RL are very good predictors of CC, and that the interaction term is highly significant. There is no evidence of inconsistency of ordinary least squares from the DHW tests 2. The lack of evidence of feedback in this regression is taken to be consistent with our interpretation that Government Effectiveness and the Rule of Law act as constraints on corrupt behaviour. Following Hall and Jones, we take these two variables to be indicators of social infrastructure, and use a simple average of the two to represent social infrastructure in the analysis that follows. 4.2 Social infrastructure, transition and the level of productivity Between them, social infrastructure and productivity probably affect almost every economic variable we can define. Nevertheless, as Hall and Jones demonstrate, it is possible to analyse the effect of social infrastructure on productivity by careful use, as instruments, of the very limited set of exogenous variables that determine social infrastructure. In their analysis these variables included geographic position, distance between countries (in the Frankel-Romer gravity model predictor of trade) and linguistic heritage. We follow their analysis but with certain variations which stem partly from our wish to examine how the long experience of socialism has influenced outcomes for the current and former socialist states. In particular, we do not have the Frankel-Romer predictions for states which did not exist prior to the break up of the Soviet Union, and also we have doubts about exogeneity with respect to productivity of the two language variables (proportions of the population speaking English and another Western European language). However, we retain distance from the equator as an instrument. Empirically, it was the most important instrument for Hall and Jones, and their conceptual justification for it is persuasive. The other instruments we use include the slope and intercept of the city log(size):log(rank) regressions and the city density variables describes in Section 3. In addition, inspection of the 12

14 productivity data reveals the importance of natural resources, especially oil, for certain countries. It is improbable that the impact of oil on output works through social infrastructure, so this is treated as a specifically included exogenous variable, having a direct effect on productivity. Finally, two dummy variables for socialist and former socialist regimes are used to represent those countries that were already socialist before WWII and those that became socialist after WWII respectively. These variables could affect productivity directly, and also indirectly via social infrastructure. Reduced form regressions The exogenous variables in the system are: oil exporting dummy (Oil); early socialist regime (T1); later socialist regime (T2); Pareto exponent from distribution of city sizes (a); Pareto intercept (A); interaction between A and a (A*a); landlocked dummy (ll); latitude (lat); city density (log(d)); and interaction between log(d) and a (log(d)*a). The first three of these variables are exogenous variables assumed to have a direct effect on productivity, and the remainder are assumed to be uncorrelated with the disturbance term in the productivity equation, but correlated with social infrastructure. The reduced form regressions are shown in Table 3. The reduced form regression of GE+RL can be regarded as producing the instrument for the structural equation for productivity to be discussed later, and for that the F ratio in the table relates to the joint significance of the excluded instruments. Recent papers (e.g. Staiger and Stock, 1997) suggest that an F ratio of this size may be adequate to avoid a serious problem of finite-sample bias associated with weak instruments in IV regression. Note that the dummy variables for both the pre-wwii and the post-wwii socialist regimes, T1 and T2 respectively, are negative and highly significant in both equations. Moreover, in the GE+RL regression, the absolute size of the coefficient on T1 exceeds that on T2 by a margin that is statistically significant at the 5% critical level on a one-way test. But in the log(y) regression, the null that the coefficients on these dummies are equal cannot be rejected. 2 Not reported are further regressions with dummy variables for the 12 pre-wwii socialist regimes and for the 13 post-wwii socialist regimes which turn out to be insignificant. 13

15 Table 3. Reduced form regressions Dep Variable GE+RL log(y) const (7.741) (4.657) Oil (0.285) (0.171) T (0.413) (0.248) T (0.369) (0.222) a (7.386) (4.443) A (0.460) (0.277) A*a (0.4297) (0.258) ll (0.271) (0.163) lat (0.007) (0.004) log(d) (0.857) (0.516) log(d)*a (0.848) (0.510) F ratio R-squared s.e.e nobs OLS regressions, standard errors in brackets 1 F ratio for last seven variables (excluded instruments) Structural regression for productivity The reduced form regressions raise the question how the socialist regime variables affect productivity. Is the effect direct or indirect, working through the social infrastructure? The structural equation for productivity is: ln y = α + βs(x ) + γ 1 T1 + γ 2 T2 + δoil + v (2) in which the coefficients can be consistently estimated using the set of variables in X as instruments for S. It is assumed that the Oil variable only affects productivity, not social infrastructure. This regression is reported in Table 4. 14

16 Table 4. IV estimation of productivity equation Dep Variable log(y) log(y) const (0.059) (0.054) (0.040) (0.038) S=GE+RL T (0.191) T (0.174) Oil (0.149) (0.150) R-squared s.e.e Nobs OverID test, p-value Instruments a, A, A*a, ll, lat, log(d), log(d)*a Standard errors in brackets a, A, A*a, ll, Lat, log(d), log(d)*a, T1, T2 The panel on the left of Table 4 shows that the coefficients of the productivity equation are all significant and the instruments appear to be valid according to the test for overidentifying restrictions 3. The regression reported in the panel on the right assumes that the socialist dummy variables do not affect productivity directly, but only indirectly via social infrastructure, so these variables are in the excluded instrument set for this regression. However, the test for overidentifying restrictions now rejects the expanded set of instruments, so our preferred formulation is the regression reported in the left panel. The socialist dummy variables are significantly positive, and a test for equality of their coefficients does not reject. A significant positive effect from socialism was also found by Hall and Jones, for whom it was a puzzle. As they note, it implies that (former) socialist countries produce substantially more output per worker than otherwise similar capitalist countries 4. But this may not be such a puzzle. The reduced form equations give the total effects of socialism on productivity, including feedback effects between productivity and social infrastructure. These total effects were found from the reduced form regression to be negative, and were 3 Note that the regression reported in the left panel of Table 5 was also estimated with control of corruption CC as the proxy for social infrastructure S instead of GE+RL, with the result that the overidentifying restrictions were decisively rejected (at the 1% critical level). This confirms the earlier findings of Mauro and of Kaufmann that this formulation appears to be a misspecification. 4 They conjectured that the unexpected finding was due to their definition of capitalism, which included a number of sub-saharan African economies. 15

17 highly significant. With a positive and highly significant coefficient of social infrastructure in the IV regression reported in Table 4, the explanation for the positive signs on T1 and T2 could only be that socialism has a strongly negative effect on social infrastructure, and that the implied indirect negative effects on productivity outweigh the direct positive effects reported in Table 4. But it should not be surprising that, allowing for social infrastructure, socialism enhances productivity. The socialist regimes generally placed a great weight, even if misguidedly, on output and productivity. The failure of socialism for productivity may well be due to the overriding importance of the indirect effects working through social infrastructure, which was not nurtured in those regimes. 5. Concluding discussion The cross-country correlation between "control of corruption" and productivity evident in Figure 1 stems from the fact that both variables are correlated with social infrastructure. The causal relationships between these variables were explored in this paper, which finds that the underlying factor connecting governance indicators and productivity is social infrastructure, as summarized in Figure 5. There, X and ξ represent respectively exogenous variables (the instruments in the regression analysis) and disturbances affecting social infrastructure, while Z and ζ represent respectively an exogenous "oil exporter" dummy variable and other unspecified factore (disturbances) affecting productivity. The (control of) corruption variable reflects social infrastructure but does not directly influence productivity. The transition countries of Eastern Europe and the former Soviet Union are represented by the exogenous "socialism" node in Figure 5, which indicates that their socialist history has depressed social infrastructure and thereby productivity, to such a degree that it offsets the directly positive effect on productivity. Consistent with this, we also find that these effects of socialism are larger the longer the country experienced a socialist regime. 16

18 ε X, ξ Social Infrastructure + Corruption Socialism + + (+) Z, ζ Productivity Figure 5. Causal structure This interpretation of the empirical results suggests a policy prescription for the former socialist countries currently undergoing transition to a market economy. Namely that, in order to improve the economy i.e. productivity the most effective route could be to repair the social infrastructure, in particular the institutions of governance such as the legal framework (e.g. the law of contract and property rights) and the quality and integrity of the civil service. In addition, perhaps, the State should desist from direct interference in production where possible, thus reducing opportunities for diversion and corruption. 17

19 References Barro, R. J. (1991) "Economic growth in a cross-section of countries", Quarterly Journal of Economics, 106, Barro, R. J. and X. Sala-i-Martin (1995) Economic Growth, McGraw-Hill, New York. Gabaix, X. (1999) "Zipf's Law for cities: an explanation", Quarterly Journal of Economics 114(1) Hall, R. E. and C. I. Jones (1999) "Why do some countries produce so much more output per worker than others?" Quarterly Journal of Economics Hellman, J. S., G. Jones, D. Kaufmann and M. Schankerman (2000) "Measuring governance, corruption and state capture", Policy Research Working Paper 2312, The World Bank, Washington USA. Kaufmann, D., A. Kraay and P. Zoido-Lobaton (1999) "Aggregating governance indicators ", Policy Research Working Paper 2195, The World Bank, Washington USA. Kaufmann, D., A. Kraay and P. Zoido-Lobaton (2000) "Governance matters", Policy Research Working Paper 2196, The World Bank, Washington USA. Knack, S. and P. Keefer (1995) "Institutions and economic performance: crosscountry tests using alternative institutional measures" Economics and Politics, 7(3) Lambsdorff, J. G. (1999) "The Transparency International corruption perceptions index - Framework Document" mimeo. Mauro, P. (1995) "Corruption and growth" Quarterly Journal of Economics Mauro, P. (1998) "Corruption and the composition of government expenditure" Journal of Public Economics 69, Parr, J. B. (1985) "A note on the size distribution of cities over time", Journal of Urban Economics, 18,

20 Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Albania ALB Algeria DZA Angola AGO Antigua & Barbuda ANB NaN NaN NaN NaN NaN Argentina ARG Armenia ARM Australia AUS Austria AUT Azerbaijan AZE Bahamas BHS Bahrain BHR Bangladesh BGD Barbados BRB NaN NaN NaN Belarus BLR Belgium BEL Belize BLZ NaN NaN Benin BEN Bhutan BTN NaN NaN NaN Bolivia BOL Botswana BWA Brazil BRA Bulgaria BGR Burkina Faso BFA Burundi BDI NaN NaN Cambodia KHM NaN NaN NaN Cameroon CMR Canada CAN Cape Verde CPV NaN NaN Central Africa CAF NaN NaN Chad TCD Chile CHL China CHN China, Hong Kong HKG NaN NaN NaN Colombia COL i

21 Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Comoros COM NaN NaN NaN NaN Congo (Dem. Rep.) ZAR Congo COG Costa Rica CRI Cote d'ivoire CIV Croatia HRV Cyprus CYP Czech Republic CZE Denmark DNK Djibouti DJI NaN NaN Dominica DCA NaN NaN NaN NaN Dominican Republic DOM Ecuador ECU Egypt EGY El Salvador SLV Equatorial Guinea GNQ NaN NaN Estonia EST Ethiopia ETH Fiji FJI Finland FIN France FRA Gabon GAB Gambia GMB Georgia GEO Germany DEU Ghana GHA Greece GRC Grenada GRE NaN NaN NaN NaN Guatemala GTM Guinea GIN Guinea Bissau GNB Guyana GUY Haiti HTI Honduras HND Hungary HUN Iceland ISL ii

22 Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil India IND Indonesia IDN Iran IRN Ireland IRL Israel ISR Italy ITA Jamaica JAM Japan JPN Jordan JOR Kazakhstan KAZ Kenya KEN Kiribati KIR NaN NaN NaN NaN NaN Korea (South) KOR Kuwait KWT Kyrgyzstan KGZ Laos LAO NaN NaN Latvia LVA Lebanon LBN Lesotho LSO Libya LBY Lithuania LTU Luxembourg LUX Macedonia MKD Madagascar MDG Malawi MWI Malaysia MYS Maldives MDV NaN NaN NaN Mali MLI Malta MLT NaN Mauritania MRT NaN NaN Mauritius MUS Mexico MEX Moldova MDA Mongolia MNG Morocco MAR Mozambique MOZ iii

23 Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Myanmar MMR Namibia NAM Nepal NPL NaN NaN Netherlands NLD Netherlands Antilles NEA NaN NaN NaN NaN New Zealand NZL Nicaragua NIC Niger NER Nigeria NGA Norway NOR Oman OMN Pakistan PAK Panama PAN Papua New Guinea PNG Paraguay PRY Peru PER Philippines PHL Poland POL Portugal PRT Qatar QAT Romania ROM Russia RUS Rwanda RWA NaN NaN Samoa SAM NaN NaN NaN NaN Sao Tome STP NaN NaN NaN NaN Saudi Arabia SAU Senegal SEN Seychelles SEY NaN NaN NaN NaN Sierra Leone SLE Singapore SGP NaN Slovakia SVK Slovenia SVN Solomon Islands SLB NaN NaN NaN South Africa ZAF Spain ESP Sri Lanka LKA iv

24 Data Appendix ppp output per worker Transition countries Average Government Rule of Control of older younger oil Effectiveness Law Corruption Pareto coefficients City density latitude socialist socialist landlock exporter country y GE RL CC A alpha D lat T1 T2 ll Oil Saint Kitts SKN NaN NaN NaN NaN Saint Lucia SLU NaN NaN NaN Saint Vincent SVG NaN NaN NaN NaN Sudan SDN Suriname SUR Swaziland SWZ Sweden SWE Switzerland CHE Syria SYR Taiwan TWN Tajikistan TJK Tanzania TZA Thailand THA Togo TGO Tonga TGA NaN NaN NaN NaN Trinidad & Tobago TTO Tunisia TUN Turkey TUR Turkmenistan TKM Uganda UGA Ukraine UKR United Arab Emira. ARE United Kingdom GBR United States USA Uruguay URY Uzbekistan UZB Vanuatu VAN NaN NaN NaN NaN Venezuela VEN Vietnam VNM Yemen YEM Zambia ZMB Zimbabwe ZWE v