Corruption Measurement: the case of Russian Federation

Transcription

1 Corruption Measurement: te case of Russian Federation Zuravleva Tatyana January 16, 2013 Abstract Tis study develops te approac of corruption measurement based on te incomeexpenditure comparison. Using micro-level data on reported ouseold earnings, expenditures and assets provided by Russian Longitudinal Monitoring Survey for te period we find tat ouseolds wit workers in te public sector receive lower earnings tan teir private sector counterparts, but enjoy te same level of consumption expenditures, in oter words tere exists an expenditure-income gap in favor of te public sector. Controlling for te reported level of earnings, ouseolds wit workers in te private sector do not sow neiter a significantly iger probability of possessing country ouses, cars and computers, nor living in better ousing conditions, nor aving iger financial wealt. Te analysis of current and accumulated savings, risk aversion and volatility of wages does not sow any sign of distinction between two sectors. Tus, differences in assets and precautionary motives of workers cannot reconcile te sizeable expenditureincome gap. Unexplained differences are referred to unreported income, or bribes. Keywords: corruption measurement, expenditure-income gap, RLMS, Russian Federation. 1 Introduction Today corruption exists in almost all te countries of te world. Researcers investigate wic countries/regions are te most corrupt, ow corruption affects economic growt and development, sould te world figt against corruption and wic metods are te most efficient. To answer all tese questions first we sould be able to measure corruption. But corruption is a idden penomenon by its nature, tus its measurement is not a trivial task. 1

2 Te measures of corruption based on surveys may be understated as tey depend on voluntary disclosure, and places were corruption is most severe migt be less likely to disclose it. Te measures of corruption based on case studies may be overstated as researcers may concentrate on situations were tey expect to find corruption. Developing careful and rigorous metrics of corruption, tat are not subject to tis type of selection bias, is an important area of researc. In tis paper we develop te measure of corruption based on te income-expenditure comparison using micro-level data of Russian economy. Tis metod was proposed for te first time by Gorodnicenko and Peter [2007] and applied to Ukrainian data. In transition economies it is common to find tat workers in te public sector receive lower monetary compensations. Tis is usually explained by lower efforts, social status, prestige etc., but in tis case lower wages sould translate into lower consumption expenditures, unless public sector workers ave non-reported income. Tis logic may fail because of te unobserved eterogeneity, for instance private sector employees may save more, and tus consume less, since teir job is more uncertain. We study in details te consumption-savings beavior of workers, in particular teir assets and wealt, in order to control for potential suspects of te selection bias. In line wit previous researc of transition economies we find tat wage premium in te private sector in Russia is statistically large and economically significant. Along wit tis we document a significant negative expenditure-income gap, meaning tat ouseolds wit te same level of income and oter observable caracteristics consume less if teir members are employed in te private sector. In principle it is possible tat public ouseolds 1 save more because of teir job uncertainty, but analyzing current and accumulated savings we do not find support for tat. Controlling for te reported level of earnings, ouseolds wit workers in te private sector do not sow neiter a significantly iger probability of possessing country ouses, cars and computers, nor living in better ousing conditions, nor aving iger wealt. Tus, differences in assets can not reconcile te found expenditure-income gap. Unexplained differences are refereed to unreported income, or bribes. Te capter is organized as follows. Section 2 presents te literature review of te studies of corruption, wit a particular empasis on corruption measurement. Section 3 contains 1. Properly defined in te text. 2

3 te evidence of corruption in Russian Federation. Section 4 describes te data, presents te econometric specification of te model and estimation results. Section 5 concludes. Appendix C1 contains te detailed description of expenditure and income data. 2 Literature review We summarize existing researc on corruption along two directions. First, we summarize various determinants of corruption and teoretical models tat link corruption to te functioning of te society. Second, we summarize and analyze existing empirical approaces to corruption measurement. 2.1 Models of corruption By definition corruption is te use of resources by government officials and employees for personal gains by means of mismanagement, fraud, bribes. Corruption is a concern of many modern democracies. It weakens democratic institutions, restricts public services, lowers economic growt and development. All existing models may be divided into two blocks: te first block of models explains te causes of corruption; te second block explains te consequences of corruption for te society. Let us start wit te first block: causes of corruption. Rose-Ackerman [1978] was te first wo proposed te explanation of political corruption using an agency model. In er model government officials care for two tings: (1) teir income tat can be increased using power for private gain; (2) teir cances of being re-elected tat naturally decrease from suc a beavior. In te world of perfect information corruption is not possible since fully informed voters will lead politicians to implement teir promises or vote out for corrupt politicians. But in more realistic world were voters are not fully informed about actions of politicians tere is a trade-off for politicians between increasing gain from corruption and reducing cances for reelection and, tus, tere is a potential for corruption. Tere are autors wo developed tis reasoning recognizing tat some politicians may not only get economic benefits from teir positions but also design policies to increase tese benefits (see, for example, Lien [1990], Kurer [1993]). Anoter explanation of corruption is so-called rent-seeking beavior. Wile traditional 3

4 neo-classical economics consider te government as a benevolent exogenous agent, new neoclassical political economists argue tat te government as its own agenda. Rent-seeking activities of te firms consist in making illegal payments for te government for some services. Some researcers (Paul and A.Wilite [1994], Linster [1994]) use game teory approac to sow ow firms will beave wen tere is competition for rent-seeking and derive optimal level of bribes in te game between firms and politicians. Let us now move to te second block: consequences of corruption. Here we summarize researc tat analyze ow corruption affects economic performance. Firstly, corruption influences bureaucratic efficiency. In te corrupted country te most efficient producer does not necessary get a contract. Te most efficient producer as a large ability to offer a large bribe (since e as te igest surplus) but te corrupted producer is able to offer even larger bribe since e is compromising on te quality of is goods or services. Secondly, corruption affects resource allocation. Corruption may result in resource misallocation since corrupted politician considers bribes as one of decision criteria. And ranking of projects based on teir social value normally differs from ranking based on unofficial payments tat decision makers expect to receive. Anoter way in wic corruption may lead to inefficiency is its impact on government s ability to correct externalities. For example, if someone can bribe a police officer instead of paying an official fine, te marginal cost of breaking te law is reduced from te official fine to te amount of te bribe. And if te police officer extracts te same bribe regardless of weter te person as broken te law, te marginal cost of breaking te law tends to zero. Social losses of corruption also come from inefficient allocation of talent, tecnology and capital (Murpy, Sleifer, and Visny [1991], Murpy, Sleifer, and Visny [1993]). Wen entrepreneurs understand tat profits or potential profits will be taken away from firms troug corruption tey prefer not to start firms or not to expand tem. Moreover, entrepreneurs may illegally sift part or all of teir profit to teir personal accounts instead of developing teir business. If entrepreneurs expect to be forced to pay bribes in te future tey ave incentives to adopt inefficient fly-by-nigt wic allows tem to react more flexibly to demands from corrupt officials and more credibly treaten to sut down operations (Coi and Tum [1998]). Svensson [2003] provides an empirical support for inefficient fly-by-nigt ypotesis using firm-level capital stock data. He finds tat te amount of bribes a firm needs to pay is 4

5 negatively correlated wit te degree of reversibility of te capital stock. Te important social consequence of corruption is tat it affects te distribution of income in te society. It is common to tink tat corruption favors te ric. Gupta, Davoodi, and Alonso-Terme [1998, p.6] argue tat corruption affects income inequality troug biased tax system, poor targeting of social programs, uman capital formation, education inequalities, and uncertainty in factor accumulation. Tey confirm teir propositions empirically. Using instrumental variables, tey argue tat corruption causes poverty and increases income inequality. Some researcers mention positive effects from corruption. It is argued tat corruption elps to overcome bureaucratic rigidities and elps maintain allocation efficiency wen tere is competition between bribers (Bardan [1997]). It is also argued tat unofficial payments to politicians speed up bureaucratic processes and ence promote economic growt (Leff [1964]), tis is so-called grease-te-weels argument. But tese arguments fail empirically. Kaufmann and Wei [2000] using cross-country analysis argue tat tere is a positive relationsip between te incidence of bribery in a country and te time tat managers of firms spend wit bureaucrats. 2.2 Corruption measurement By its nature, corruption is unobserved and, ence, its measurement is te primary callenge in te empirical literature. A researcer wo develops quantitative measures of corruption faces te question of wat sould be included in suc a measurement and ten tries to measure someting tat everybody involved in te process is trying to ide. Te empirical measures of corruption are indispensable to test teoretical ypotesis and models, to quantify te magnitude of corruption, to assess anticorruption policies, terefore tat is an issue of current importance. In te literature tere are different approaces to measure corruption. Most measures of corruption are based on perceptions suc as surveys and expert opinions, in wic individuals and firms managers are asked about teir evaluation of corruption in teir environments. Jonston [2001] provides a detailed survey and discussion of suc measures. Te most wellknown index is te Transparency International Corruption Perception index. Transparency 5

6 International is te organization devoted to figting bribery around te world, it measures perception of corruption in different countries. Teir index is an aggregated indicator tat ranks countries in terms of te degree to wic corruption is perceived to exist among public officials and politicians. Tis is an aggregate index tat combines different sources of information, i.e. different surveys and assessments from independent institutions. 1 Tere are oters organizations tat publis some indexes of corruption. For example, Business International Corporation publises a number of ratings on countries, including an assessment of te level of corruption in various countries. Political Risk Services Inc. publises an annual report, International Country Risk Guide, tat includes a corruption index. Te main advantage of perception-based measures is teir breadt wic allows to run large cross-countries studies (for example, Mauro [1995]). Te main problem wit all measures based on perceptions is perception bias. In most cases surveys ask direct questions on weter individuals or firms pay bribes to government officials. In suc situations people tend to be afraid or, possibly, be asamed, and ide te information, tus results distort te dept and breadt of corruption. Second, te aggregate nature of te data gives us little information about te relationsip between corruption and individual agents, tus we are not able to explain te witin-country variation of corruption. And, finally, tis measure focuses on perceptions of corruption rater tan on te actual bribes paid or te actual teft of resources. Tis corruption index does not provide information about te type and level of corruption in a country. For example, in te Transparency International Corruption Perception Index for 2011 Russia, Azerbaijan, Belarus, Comoros, Mauritania, Nigeria, Togo and Uganda ave te same index value. However, tis value does not tell us wic is te form and te level of corruption in tese very different countries. Tere are metods tat provide pysical audits of governmental processes. For example, Bertrand, Djankov, Hanna, and Mullainatan [2007] followed individuals troug te process of obtaining a driver s license in India, and found extra-legal payments made and te rules tat were broken in excange for tese extra-legal payments. Similarly, Barron and Olken 1. In te year 2010 tese institutions were te African Development Bank, te Asian Development Bank, te Bertelsmann Foundation, te Economist Intelligence Unit, Freedom House, Global Insigt, International Institute for Management Development, Political and Economic Risk Consultancy, te World Economic Forum, and te World Bank. 6

7 [2007] made a case study in wic surveyors accompanied truck drivers on 304 trips along teir regular routes in two Indonesian provinces. Tese surveyors accompanied truck drivers on teir regular routes and noted te illegal payments tat drivers made to te traffic police and military officers. Tey found tat, in total, te illegal payments represented 13% of te cost of te trip. Sequeira and Djankov [2010] used a similar metodology in Mozambique and Sout Africa, tey analyzed sadowing clearing agents wo processed customs for cargo as it passed troug te ports. Tey observed bribe payments to port and border post officials and found tat, on average, bribes represent 14% of sipping costs in Mozambique and 4% in Sout Africa. Recently te literature turned to evaluating corruption using policy experiments by comparing te amount of transfers disbursed from a federal grant to te amount tat actually reaces te intended recipient of te transfer. Te pioneering study by Reinikka and Svensson [2004] analyze total public expenditures spent on education in te mid-1990s in Uganda. Tey compare data from records on central government disbursements and a public expenditure tracking survey to measure te dissipation in a scool grants in Uganda. Tey found tat scools, on average, received only 13% of te grants, most scools received noting. Tey also found considerable variation in grants received across scools, meaning tat scools use teir bargaining power to secure greater funding. Tere is a similar paper by Mironov and Zuravskaya [2012] wic tries to provide a reliable measure of corruption by measuring te amount of cas tunneled illegally out of firms around te time of regional elections and relating it to te probability tat te firms obtained procurement contracts from te government. Tere are metods tat collect data from te bureaucrat documenting ow te government resources were spent in acieving a task and ten compare tem wit an objective measure of ow muc it costs to conduct te required task. Te difference ten is estimated as ow muc was stolen. For example, Olken [2007] compared te official expenditure reports wit an independent measure of wat it sould ave cost to build a road of a particular quality. To obtain independent measures, e sampled eac road to determine te materials and labor used, and ten used local prices to cost tese items. He found tat, on average, about 24% of expenditures across te roads were missing. Golden and Picci [2005] propose te measure of corruption tat consists of te difference between a measure of te pysical quantities of public infrastructure and te cumulative price government pays for public capital stocks. Were 7

8 te difference is larger between te money spent and te existing pysical infrastructure, more money is being siponed to mismanagement, fraud, bribes, tat is corruption is greater. In tis way Golden and Picci [2005] create te measure of corruption for Italy s 95 provinces and 20 regions. Te main problem of tis approac consists in te interweaving of corruption to te efficiency. Experiments tat allow corruption evaluation are rare and cover only a limited part of corrupt activities, tus we are not able to deeply analyze te determinants and consequences of corruption. And te main limitation of above-mentioned approaces is tat it is difficult to distinguis between corruption and mismeasurement or simple bureaucrat incompetence. In Olken [2007] it is possible tat road quality is mismeasured or tat te bureaucratic apparatus is not good in building roads. In Reinikka and Svensson [2004] it remains possible tat te resources tat sould ave gone to scools were actually spent on oter important programs and did not end up in te pockets of government officials. In te literature tere is an oter approac to measuring corruption tat uses equilibrium conditions in te labor market. People being in te public sector must be indifferent between teir public sector job and alternative job in te private sector. If controlling for teir job opportunities, salaries are lower in te public sector it sould simply reflect a compensating wage differentials (fringe benefits, job security etc). But if salaries in te public sector are lower and consumption level is te same, it may infer tat public sector workers receive some unobserved compensation. Tis approac proposed for te first time in Gorodnicenko and Peter [2007] requires important assumptions about equal consumption-savings beavior in two sectors. We borrow teir approac, develop it, and apply it to Russian data. 3 Evidence of corruption in Russia According to various estimates Russian Federation is one of te most corrupt countries in te world, and te issue of measuring corruption in tis country is particulary acute. Transparency International publised its 2011 Corruption Perceptions Index and ranked Russia 143 out of 180. Te country as te same rank (2.4 out of 10) as suc African countries as Nigeria, Togo, Uganda ave. For comparison US as 7.1, France - 7.0, Germany - 8.0, Spain

9 It is widely accepted tat corruption is one of te main obstacles to te country s economic growt and development. It is a major barrier to business in Russia. BNP Paribas estimates tat perceptions of corruption costs Russia te equivalent of 4 per cent of GDP in lost foreign investment eac year. Te general public identified it as te biggest block to economic growt in a recent survey by independent pollsters Levada Centre. Te Association of European Business agrees: extras can account for 20% of te cost of doing business in Russia 1. President Dmitry Medvedev as made figting corruption one of te top goals of is presidency, and as launced an anti-corruption campaign. Building on Putin s Anti-Corruption Commission, establised in 2004, e introduced a National Plan to Counteract Corruption only a mont after taking office. New legislation came into force on January 10, 2009, requiring civil servants to disclose all income and assets, putting in place a framework for determining conflicts of interest, and simplifying te Criminal Code to old corrupt judges to account. But most Russians remain sceptical about te efficiency of tese reforms. Analysts argue tat te problem is not wit te legislation but wit its implementation. Some officials apparently agree. Russian Prosecutor-General, Yuri Caika, claimed tat Russian law-enforcers were only rooting out minor cases of bribery, ignoring large-scale corruption, wile many believe te corruption cases brougt by te autorities are little more tan a pretext to dismiss officials wo ave fallen out of favour. Let us now explore te mecanisms of corruption in different industries in Russia. According to Transparency International s Global Corruption Barometer 2010, te Russian police is considered to be te most corrupt public institution in te country. Demands for bribes are especially uge among te traffic police, despite te recent tigtening of legislation, drivers prefer pay bribes since it is time-consuming, nobody wants wait in bank lines to pay te fines. Bribes are paid to get unjustified driving licenses, to falsify te information on road accidents for te benefit of interested parties. Bribes affect te decision of te Interior Ministry employees to open or close te case; to take unjust decision; to contradict decisions of different judges on te same case. Corruption is ig on customs: firms pay bribes for letting forbidden goods across border, return of confiscated goods and currencies, reduction of customs fees, to avoid ungrounded delays at te border, unjustified delays of customs payments etc. Corruption in te tax 1. ttp:// 9

10 autorities is manifested in te not collection of taxes, te return of VAT, te suspension of production caused by competitor firms. Corruption in ealtcare and education is also uge. Healtcare is supposed to be free in Russia, but Russians know tat every ospital as its under-te-table price list, and for te care of quality you ave to pay. It is common to pay bribes to get false medical papers and certificates or to get served faster at te expense of oter customers. Corruption in education may take many different forms. Te state registration and accreditation of iger education institutions is carried out troug bribes. A common practice for teacers, bot in scools and especially iger educational institutions, to suggest to students tat in order to improve grades or guarantee certain exam scores, tey must receive extra tutoring, and te income of te professor in prestigious University for one private lesson may exceed is montly salary. Hidden costs wait any student trying to get into a good university or a prestigious faculty, and quite often people wit unsatisfactory level of knowledge are admitted to prestigious Universities (mainly ones in law and economics), tat reduces significantly te value of te diploma. Moreover, tere is a market for diploma: it is possible to buy and sell it. We described te main areas of corruption in Russia but tere are various oter forms: paying bribes to get necessary papers and permits, to release from army service, to get certification and accreditation of some institutions, to construct buildings, notarize transactions etc. Despite all tese facts te evidence of corruption in Russia is scarce in te academic literature. Tere are some papers tat examine te impact of corruption on te Russian economy. Tere are papers tat igligt te negative effect of corruption on te development of small and medium entreprises in Russia. For example, Safavian, D.Graam, and C.Gonzalez-Vega [2001] construct a teoretical model to explorer te mecanism troug wic corruption affects micro and small firms in Russia. Tey conclude tat corruption in Russia prevents te development of small entreprises in Russia and tus as a negative effect on investment activities and economic growt. Later Yakovlev and Zuravskaya [2009] igligt one more time tat corruption is an obstacle to te development of small and medium entreprises in Russia. Using te data on Russian state capture 1 constructed by Slinko, Yakovlev, and Zu- 1. To measure capture and firms political power tey construct a database of regional laws and regulations tat treat selected large firms in tese regions preferentially. Tey deem an enterprise to be treated preferentially if it receives any of te following benefits: tax breaks, investment credits, subsidies, subsidized loans, 10

11 ravskaya [2005], te autors argue tat large firms wic receive privileges from te regional government tend to be profitable and faster developing. A ig state capture index reflects te political influence of large firms and implies worse performance for small business. Tere are some papers tat analyze differences in corruption between Russian regions. For example, Popova [2010] investigates te effects of regional differences in corruption on electoral results in Russia and on te voting beavior of people wit different employment status. Te autor uses two measures of corruption: Corruption Perception Index of Transparency International measured for Russian regions and te state capture index constructed by Slinko, Yakovlev, and Zuravskaya [2005]. Te paper finds tat corruption positively influences participation in elections, moreover, it sows tat even after controlling for corruption, people wit different employment statuses vote differently. P.Dininio and R.Orttung [2005] analyze regional differences in corruption using corruption perception data. Te autors find tat corruption in Russia is a structural problem, and not te one related to its institutions. Tey find tat witin eac region te amount of corruption increases as te size of regional economy grows, te GDP per capita decreases, and te population decreases. Te recent Mironov and Zuravskaya [2012] paper proposes an innovative measure of corruption tat consists in measuring te amount of cas tunneled illegally out of firms around te time of regional elections and relating it to te probability tat tese firms obtain procurement contracts from te government. Using tis measure tey reject te efficient greasing ypotesis by sowing tat in more corrupt regions, public procurement contracts are allocated to less efficient firms, and terefore ave negative implications to economic growt. We see tat several effects of corruption in Russia ave been analyzed in te literature but te issue of te measurement of corruption is not well developed. To te best of our knowledge tere are only tree approaces: Te Corruption Perception Index for Russian regions measured by Transparency International. It involves te perception bias problem related to te difficulty of inducing people to talk about corruption. State capture index developed by Slinko, Yakovlev, and Zuravskaya [2005]. official delays in tax payments, free grants of state property, or a special open economic zone status for teir territory. 11

12 Te measure of corruption recently developed by Mironov and Zuravskaya [2012]. In te literature anoter approac, so called income-expenditure approac, exists. For te first time it was proposed by Gorodnicenko and Peter [2007] for te analysis of Ukrainian economy. In te next section we apply it to Russian economy. 4 Measuring bribery: expenditures versus income 4.1 Data description Te data used in tis study stem from te Russian Longitudinal Monitoring Survey. Te information comes from bot individual and ouseold questionnaires. Te data from individual questionnaire are aggregated to ouseold level. We consider te time period Labor compensation data are taken from individual questionnaire and are aggregated to ouseold level as te sum of te reported income of all ouseold s earners. To make our analysis more accurate we use tree different definitions of labor compensation. Actual wage denotes all wages received by all ouseold members at teir primary jobs during last 30 days. If some ouseold member does not ave a job is contribution is equal to zero. Contractual wage denotes te sum of contractual wages of all ouseold members at teir primary jobs. Contractual wages of individuals wo are on leave, including maternity leave, are also included. Contractual wage is an important indicator since it is less affected by transitory income socks and may serve as a proxy for permanent income, tus te problem of different wage volatility in two sectors may be partially solved. All wages are net of taxes. Total earnings denote all actual wages received by all ouseold members at all teir jobs during te last 30 days plus te income from selling livestock, crops, agricultural products or collected in te forest berries and musrooms. To account for non-labor compensation we use total income wic is defined as te sum of total earnings and social payments tat include fuel subsidies, pensions, stipends, unemployment benefits, payments for cildren, rents on property, interests on deposits and dividends Financial income, defined as te sum of rents on property, interests on deposits and dividends, is not a big part of te income of workers in our sample, tere are less tan 2% wo report to ave non-zero financial income. People may also underreport tis type of income in te survey because tey ide it from te tax. 12

13 Houseold composition data are constructed using individual questionnaire. Earner is a member of ouseold wo works or as paid (or unpaid) leave, including maternity leave. A member of ouseold under 18 is considered as a cild. Individual is a pensioner if e is older tan 60 years (55 for women). Te definition of te private/public ouseold is tricky. It is clear tat a ouseold may consist of two people, one of wic is employed in te public sector, anoter in te private one. In tis situation it is difficult to assign it to be public or private. We define te ouseold to be private (public) if more tan 75% of earners are employed in te private (public) sector. Wit tis specification we loose te 50/50 ouseolds, it is 19% of observations in our sample. We also employ te ead of ouseold specification: te ouseold follows te sector of its ead, were te ead of ouseold is te person wit te igest observed wage. Te Mincerian variables used in our regressions include age, ours of work, number of males, experience and tree types of education (secondary complete, university diploma and PD degree). We also add dummies for urban areas and Moscow/Saint-Petersburg. All tese data come from te individual dataset. For te aggregation we compute te average age of working ouseold members, te sum of ours of work of all earners, te sum of work experience computed separately for males and females, te number of employed ouseold members wit complete secondary education, university diploma, PD degree. Houseold expenditures data come from te family questionnaire. In te survey te responsible person in te family answers te question about te total expenditures of te ouseold during te last 30 days. In our analysis we use different subsets of permanent 1 expenditures: (1) food expenditures bot for eating at ome and away from ome; (2) clots expenditures bot for adults and cildren; (3) services including transport, repair, laundry, airdressing, mobile pone, Internet; (4) leisure expenditures including travels, cinema, teater; (5) all medical expenditures including drug purcasing. Oter expenditures include insurance and loan payments, alimony and taxes. Total expenditures on non-durables are computed as te sum of all of te above. We ave ouseold assets data tat also come from te family questionnaire. We ave te information about ousing conditions, including value of ousing (evaluated by te ouseold) 1. We do not ave data on expenditures for durable goods, but we ave information about te presence of tese goods in te ouseold wic is used later in te robustness analysis. 13

14 and ousing space; te presence of a computer, a car and a country ouse. 4.2 Houseold income estimates First, we estimate te following income equation. ln w = β 0 + β 1 N P R + β 2 N T OT + β 3 N EAR + β 4 N CH + β 5 N P N + γ X + ɛ (4.2.1) were w denotes te sum of total labor compensation of all ouseold earners; N P R te number of ouseold members employed in te private sector; N T OT total ouseold members. N EAR, N CH cildren and pensioners respectively. aggregated to te ouseold level. and N P N denotes is te number of denote te number of ouseold earners, X is a vector of Mincer s covariates tat affect income To make estimates in ouseold income and expenditures comparable we need to add te non-labor portion of ouseold income. ln Y = ˆβ 0 + ˆβ 1 N P R + ˆβ 2 N T OT + ˆβ 3 N EAR + ˆβ 4 N CH + ˆβ 5 N P N + ˆγ X + ˆɛ (4.2.2) were Y is total ouseold income including labor income of all te ouseold earners and non-labor compensation (stipends, pensions, oter social payments, alimonies, rental income etc). For te elimination of te inflation effect all income measures are considered in 2000-year prices. Table 1 reports results. We observe tat if one member of ouseold working in te public sector was in te private sector tis would increase te total ouseold income by 12%, tis olds for all considered income measures. Te income premium associated wit te ouseold member working in te private sector is large and statistically significant, as it was at te individual level. Te coefficients on oter covariates are consistent wit economic teory, for instance we observe a positive and statistically significant return on all levels of education and experience of bot males and females. Estimates state tat one working male brings 25% of te additional income to te ouseold. Te revenues of families in urban areas are significantly iger, especially in Moscow and Saint-Petersburg. Te presence of pensioners and cildren in te ouseold reduce teir total earnings but positively affect total income because of social payments (pensions, cild support). 14

15 Table 1: Houseold income estimates, Actual wage Contractual Wage Total earnings Total income Earners (0.00) (0.01) (0.01) (0.00) Pensioners (0.00) (0.00) (0.01) (0.00) Cildren (0.01) (0.00) (0.01) (0.00) Age work (0.00) (0.00) (0.00) (0.00) Hours (0.00) (0.00) (0.00) Male work (0.01) (0.01) (0.02) (0.02) Urban (0.01) (0.01) (0.01) (0.01) Moscow/St.Petersburg (0.01) (0.01) (0.01) (0.01) Complete secondary (0.00) (0.00) (0.01) (0.00) University work (0.01) (0.01) (0.01) (0.00) PD work (0.03) (0.03) (0.03) (0.03) Experience male work (0.000) (0.000) (0.000) (0.000) Experience female work (0.000) (0.000) (0.000) (0.000) Year (0.00) (0.00) (0.00) (0.00) Private sector (0.00) (0.00) (0.00) (0.00) N R F-statistics Te table reports te ˆβ 1 estimate of te eq.(4.2.2) for different income measures (defined in te text) tat all are used in logaritmic form. All covariates are also defined in te text. Robust standard errors are given in parentesis. We exclude ouseolds if at least one member does not report is income or sector (10% of observations). 15

16 In te Table 2 te first line OLS(1) provides te ˆβ 1 estimate of te eq.(4.2.2) discussed above. Te second line OLS(2) provides te estimates of te equation wit te dummy for te private ouseold instead of te number of members employed in te private sector: ln Y = ˆβ 0 + ˆβ 1 Z + ˆβ 2 N T OT + ˆβ 3 N EAR + ˆβ 4 N CH + ˆβ 5 N P N + ˆγ X + ˆɛ (4.2.3) were Z = 1 if more tan 75% of earners are employed in te private sector, and Z = 0 if more tan 75% of earners are employed in te public sector. All te oter variables are defined as above. Table 2: Houseold income estimates, alternative specifications. Actual wage Contractual Wage Total earnings Total income OLS(1) (0.00) (0.00) (0.00) (0.00) OLS(2) (0.01) (0.00) (0.01) (0.01) OLS(3) (0.01) (0.01) (0.00) (0.01) FE (0.01) (0.00) (0.01) (0.02) N of obs Robust standard errors are given in parentesis. ** significant at 5%; *** significant at 1%. Te specifications OLS(1), OLS(2), OLS(3) and (FE) are defined in te text. OLS(3) provides te ˆβ 1 estimates of te eq.(4.2.3) were Z = 1 if te ead of ouseold is employed in te private sector, Z = 0 oterwise (later tis is called te ead of ouseold specification). To control for te unobserved variables tat can affect ouseold income (preferences, friends influence etc) we use fixed effect estimates tat account for all timeinvariant omitted variables. ln Y t = β 0 + β 1 Z t + β 2 N t + γx t + α + u t (4.2.4) were N t combines te variables of te ouseold composition, and α denotes individual fixed effect. Te estimates β 1 of te eq.(4.2.4) are reported in te last line of te Table 2. OLS(2) and OLS(3) estimates indicate tat if we switc from te public ouseold to te private one aving te same composition (te number of earners, cildren, pensioners) and te 16

17 same Mincerian caracteristics (age, education, experience) we will observed a 17% increase in te ouseold income. FE estimates confirm tese results, but wit smaller premium. We conclude tat private sector premium is economically large and statistically significant. 4.3 Measuring bribery: metodology Te metod presented ere was first proposed by Gorodnicenko and Peter [2007, p ]. We ave found tat ouseolds wit workers in te private sector are associated wit large and significant premium in te observed income, it is common result for economies in transition. Tis is usually explained by lower efforts in te public sector, bonuses, social status, prestige etc. However in tis case lower wage must translate into lower consumption expenditures 1. In oter worlds, te sectoral gap in expenditures sould be equal to te gap in earnings, unless public sector workers ave unobservable, non-reported income (bribes). To verify tis we estimate te following equation: ln E ln Y = β 0 +β 1 N P R +β 2 N T OT were N P R +β 3 N EAR +β 4 N CH +β 5 N P N +γ X +θy +ɛ (4.3.1) follows te ead of ouseold specification. E are te ouseold total expenditures. We control for te level of earnings Y. As te measure of income Y we use contractual wages rater tan actual wages because tey are less affected by transitory income socks and tus may serve as a proxy for permanent income. All oter variables are defined as above. Te strong property of tis specification is tat it controls for any omitted variable tat affects bot ouseold income and consumption. Te weak property is tat is involves identical specifications of ouseold income and ouseold consumption. β 1 denotes te expendituresincome gap. For example, ˆβ1 = 0.1 means tat if we consider two identical private and public ouseolds (identical in terms of ouseold composition and Mincerian covariates), private ouseolds consume 10% smaller sare of teir income. Te main critics of tis metod is te unobserved eterogeneity, in particular te fact tat public and private sector employees may ave different consumption-savings beavior tat may be due to different risk in two sectors. Private sector employees may save more, and tus consume less, because teir job is more uncertain, and tis is also te reason of te expenditure-income gap between two sectors discussed above. In te following section 1. Unless private sector employees save more because of risk and wage uncertainty, tis is discussed later. 17

18 we explorer furter te consumption-savings beavior, compare assets and wealt between sectors. Table 3 reports results. We observe tat expenditure-income gap is negative and significant. It means tat workers wit comparable levels of earnings consume a iger part of teir income if tey are employed in te public sector. Consider two identical families consisting of two adults, usband and wife, and two cildren. In bot families men ave a university degree and full-time job wit 10-years experience, women do not work. Bot earn RuR per mont. Te only difference tat te first one is employed in te public sector, wile te second one works in te private sector. Our estimates predict tat if te second family periodically spends RuR of teir income for current expenditures, te public one will spend 10% more, i.e RuR. 1 We tend to explain it by te fact tat workers in te public sector systematically ave additional, non-reported compensation, and tus are able to increase teir expenditures. All te oter variables are consistent wit te economic intuition. For instance, we observe tat one pensioner decreases te expenditures for clots by 9% but increases medical consumption by 23%. Te expected result is tat te net effect of ours of work and experience on consumption is null. We can extend furter tis analysis by exploring te consumption-income gap in te eq.(4.3.1) across different occupational subgroups (managers, professionals etc) and subsectors (education, ealt care). If te bribery ypotesis olds, ten te groups tat perceive more bribes sould detect larger discrepancies between expenditures and income. In Section 3 we ave discussed tat ealt care and education are among te most corrupt industries in Russia. Tese industries provide services tat ave igly inelastic demand and, tus, create te greatest opportunities to extract bribes. Te bribery ypotesis of our analysis is consistent wit tis evidence. We find tat consumption-income gap is considerably iger in education and ealt care: 0.13 and 0.15 log points respectively, versus 0.09 found previously. Table 4 also provides te estimates of te gap across occupations. We do not find te evidence of te gap among managers (it may be due to insufficient number of observations), but we find tat expenditure-income differentials between private and public sectors are large and significant for professionals, service and clerks. 1. Tis metod allows us to capture monetary bribes, and ignores bribes in kind. 18

19 Table 3: Expenditure-income gap estimates, Food Clots Leisure Medecine Total current expenditures Earners (0.01) (0.02) (0.07) (0.03) (0.02) Pensioners (0.01) (0.01) (0.01) (0.01) (0.00) Cildren (0.01) (0.01) (0.03) (0.01) (0.01) Age work (0.00) (0.00) (0.00) (0.00) (0.00) Hours (0.00) (0.00) (0.00) (0.00) Male work (0.01) (0.02) (0.06) (0.02) (0.01) Urban (0.01) (0.01) (0.06) (0.02) (0.01) Moscow/St.Petersburg (0.01) (0.02) (0.05) (0.02) (0.01) Complete secondary (0.01) (0.00) (0.03) (0.01) (0.00) University work (0.01) (0.01) (0.02) (0.01) (0.00) PD work (0.03) (0.03) (0.1) (0.03) (0.02) Experience male work (0.00) (0.00) (0.00) (0.00) (0.00) Experience female work (0.00) (0.00) (0.00) (0.00) (0.00) Year (0.00) (0.00) (0.00) (0.00) Te gap (0.00) (0.00) (0.00) (0.00) N R F-statistics Te line te gap reports te β 1 OLS estimate in te eq.(4.3.1). Te log of contractual wage is an income measure. Standard errors are given in parentesis. Expenditure measures are defined in te text. 19

20 Table 4: Te expenditure-income gap by subsector and occupational group, ead of ouseold specification, All sectors, all occupations Te gap by subsector 0.09 (0.00) [24372] Healt care Education Te gap by occupation 0.15 (0.05) [2030] 0.13 (0.04) [1744] Managers 0.02 (0.04) [1136] Professionals 0.14 (0.01) [6226] Service Clerks 0.08 (0.03) [1402] 0.12 (0.04) [885] Te table reports te β 1 OLS estimate in te eq.(4.3.1). Standard errors are given in parentesis, number of observations is given in brackets. * significant at 1%; ** significant at 5%; *** significant at 1%. Our findings clearly suggest tat workers wit te same level of income and oter caracteristics consume less if tey are in te private sector. But in principle, private sector employees aving te same level of true earnings as teir public sector counterparts could save more (and, tus, consume less) because of teir job uncertainty. We provided a partial treatment to tis effect using contractual wage as a measure of income tat is less affected by income socks. To make one more step in tis direction let us now analyze te volatility of wages in bot sectors and compare savings and wealt (accumulated savings) of ouseolds employed in te public and private sectors. 20

21 4.4 Robustness of te previous results Savings and wealt comparison If te found discrepancies are explained by different consumption-savings beavior, tan savings and wealt sould differ between two sectors. questions about ouseolds current monetary savings. sare of savings S Y Te family questionnaire contains Using tese data we compute te using two different measures of te ouseold income: total income and total earnings. Table 5 reports results. Table 5: Houseolds current savings, pooling S cur Earn tot Private sector Public sector N=7983 N=12653 S cur Inc tot S cur Earn tot S cur Inc tot % of ouseolds wit zero savings 84.6% 85.3% % of ouseolds witout answer 1.3% 1.2% min max mean standard deviation Note: Descriptive statistics (mean and standard deviation) are computed for te subsample of ouseolds wo report non zero savings. Analyzing te descriptive statistics we do not find support for te claim tat ouseolds wit workers in te private sector ave iger current savings. Te paired T-test predicts tat tere is no significant difference between two subsamples. Moreover, we observe tat 85% of ouseolds report to ave zero savings. Among tose wit non-zero savings families in bot sectors on average save 30% of teir income. We run OLS and FE regressions wit te same controls as in te eq.(4.3.1), we do not find tat sector significantly affects te sare of savings in earnings. Te survey also contains information about accumulated monetary savings, i.e. wealt, tat includes deposits (bot in ome and in a bank) in rubles or currency, securities and bonds. Tis question as been started to ask in te year 2006, tus we possess te 4-years information. Table 6 makes an overview of te descriptive statistics. We compare mean savings in two sectors using te paired T-test. Te test predicts tat tere is no significant difference between two subsamples. As well as wit current savings we observe tat te majority of ouseolds 21

22 Table 6: Houseolds accumulated savings, Private Public Private Public N = 1269 N = 1643 N = 1260 N = 1586 % of ouseolds wit zero savings 70% 70.9% 68% 69.7% % of ouseolds witout answer 10% 10% 12.1% 11.6% min max mean standard deviation Private Public Private Public N = 1331 N = 1468 N = 1385 N = 1446 % of ouseolds wit zero savings 69.8% 67.8% 64.8% 62.7% % of ouseolds witout answer 12.4% 11.5% 15.2% 15.2% min max mean standard deviation Note: All numbers are given in Russian rubles. For information on average 1$ 30RuR in (70%) report zero monetary savings. We estimate te eq.(4.3.1) restricting te sample only to ouseolds wit zero wealt. Previous results old, we find te significant expenditureincome gap of 9% between private and public sectors. It means tat te found discrepancies can not be attributed to te differences in savings. It is possible tat private sector workers convert quickly teir savings into assets. In our data set we ave te information about te value of ousing and ousing space. Also we know weter te family as a country ouse, a car (domestic or foreign) and a computer. To evaluate te differences in possessing tese assets we estimate OLS and probit regressions were asset measure is a dependent variable. Controls are te same as in te basic equation (4.3.1). Table 7 reports results. Houseolds wit workers in te private sector do not sow neiter a significantly iger probability of possessing country ouses, cars and computers nor living in better ousing conditions. At te same time te coefficients on oter covariates are consistent wit standard teoretical predictions: education and earnings increase te probability of aving eac of tose goods. Tese results are verified wen we apply te ead of ouseold 22

23 Table 7: Houseold assets estimates, pooling Private sector University degree Earnings N Total wealt /OLS/ (0.09) (0.04) (0.04) Value of ousing /OLS/ (0.00) (0.01) (0.00) Housing space /OLS/ (0.006) (0.01) (0.00) Country ouse /probit/ (0.006) (0.01) (0.00) Car /probit/ (0.00) (0.00) (0.00) Car, foreign brand /probit/ (0.01) (0.00) (0.00) Computer /probit/ (0.00) (0.00) (0.00) Houseold is te unit of observation. Value of ousing, total earnings and total wealt are used in te logaritmic form. Private ouseold: more tan 75% of earners are employed in te private sector. Standard errors are given in parenteses. * significant at 10%; ** significant at 5%; *** significant at 1%. All specifications include covariates from te eq.(4.3.1). Marginal effects are reported for te probit estimates. specification. Tus, differences in assets can not reconcile te found expenditure-income gap. Volatility of income We first try to verify tat te attitude to risk does not differ between individuals employed in te private and public sectors. In te year 2006 RLMS starts to ask questions about insurance programs. Tey ask about life and accident insurance, ome insurance, summerouse insurance, veicle insurance (not including compulsory liability coverage) and agricultural property insurance. As te measure of risk aversion we compute te indicator equals to te number of insurance programs in wic individual participates divided by te total number of insurance to wic e is applicable. We find tat te majority of individuals do not participate in any insurance program. Among tose wo participate we do not observe te evidence for te difference in risk aversion between two sectors. Te paired T-test predicts tat tere is no significant difference between two subsamples. We run control regressions (wit te same controls as in te eq.4.3.1) and do not find tat sector affects te risk. Subject to te correctness of te risk aversion measure 23