WPS5944 Policy Research Working Paper 5944 Services Liberalization and Productivity of Manufacturing Firms Evidence from Ukraine Oleksandr Shepotylo Volodymyr Vakhitov The World Bank World Bank Institute Growth and Competitiveness Unit January 2012 Policy Research Working Paper 5944 Abstract This paper brings new evidence on the impact of services The results indicate that a standard deviation increase liberalization on the performance of manufacturing in services liberalization is associated with a 9 percent firms. Using a unique database of Ukrainian firms increase in TFP. Allowing services liberalization to in 2001–2007, the authors utilize an external push dynamically influence TFP through the investment for liberalization in the services sector as a source of channel leads to an even larger effect. The effect is robust exogenous variation to identify the impact of services to different estimation methods and to different sub- liberalization on total factor productivity (TFP) of samples of the data. In particular, it is more pronounced manufacturing firms. for domestic and small firms. This paper is a product of the Growth and Competitiveness Unit, World Bank Institute. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at oshepotylo@eerc.kiev.ua or Vakhitov@kse.org.ua. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Services liberalization and productivity of manufacturing �rms: Evidence from Ukraine Oleksandr Shepotylo∗and Volodymyr Vakhitov† Abstract This paper brings new evidence on the impact of services liberalization on the per- formance of manufacturing �rms. Using a unique database of Ukrainian �rms in 2001- 2007, we utilize an external push for liberalization in the services sector as a source of exogenous variation to identify the impact of services liberalization on total factor productivity (TFP) of manufacturing �rms. The results indicate that a standard deviation increase in services liberalization is associated with a 9 percent increase in TFP. Allowing services liberalization to dynam- ically influence TFP through the investment channel leads to an even larger effect. The effect is robust to different estimation methods and to different sub-samples of the data. In particular, it is more pronounced for domestic and small �rms. JEL codes: F14, G28, L80 Keywords: services liberalization, productivity, trade ∗ Kyiv School of Economics and Kyiv Economic Institute, oshepotylo@kse.org.ua † Kyiv School of Economics and Kyiv Economic Institute, vakhitov@kse.org.ua 1 1 Introduction In the 2000’s, services sector in transition countries experienced rapid development due to major regulatory changes. Deregulation allowed new �rms to enter the market resulting in rapid expansion of services as a share of GDP. The focus of this paper is on analyzing the impact of those changes on productivity of manufacturing �rms. This question has recently received considerable attention due to the importance of services in the global economy and due to the ongoing debates on the Doha Agenda (Hoekman et al., 2010). The literature documents a positive effect of services deregulation on productivity of manufacturing �rms in the Czech Republic (Arnold et al., 2011) and in Chile (Fernandes and Paunov, 2011). Still, as pointed out by Francois and Hoekman (2010), works that try to establish a causal link from services to increase in productivity are plagued with the endogeneity problem and with the problem of disentangling the effect of services liberalization reform from the effect of other reforms. We look at the episode of services liberalization in Ukraine in 2001-2007, which was isolated from other major deregulatory changes and was driven by political pressure imposed by trading partners as a precondition for the Ukrainian WTO accession. We exploit rich data on Ukrainian manufacturing �rms, which allows us to construct a �rm-speci�c index of the services use intensity and interact it with sub-sector and time- varying indices of services liberalization provided by the European Bank for Reconstruction and Development (EBRD). We adopt the standard two-stage approach in the literature of estimating the effect of a policy change on productivity (Pavcnik, 2002; Javorcik, 2004; Amiti and Konings, 2007; Khandelwal and Topalova, 2011). At the �rst stage, we estimate the production function using the Olley-Pakes methodology (Olley and Pakes, 1996), controlling for demand shocks as suggested by De Loecker (2011), to extract total factor productivity (TFP) of manufacturing �rms. At the second stage, we regress TFP on the �rm-speci�c index of services liberalization, controlling for the �rm-speci�c heterogeneity and market structure of manufacturing industries. As a new contribution, we also implement a one- stage procedure of estimating the effect of the services liberalization on productivity, which 2 takes into account a dynamic effect of liberalization on investment decisions and, as a result, on exit and entry of �rms. Using the standard method, we �nd that a standard deviation increase in our measure of services liberalization is associated with a 9 percent increase in productivity. The size of the effect is stronger then in previous studies, probably reflecting the fact that the Ukrainian services sector before the reform was less developed than in the Czech Republic and Chile. The effect is stronger for domestic and small �rms, which makes services liberalization a very useful tool for local policymakers interested in promoting growth of domestic small and medium enterprises. Allowing for the dynamic effect of services liberalization on current investment decisions and on future productivity further reinforces the effect of services lib- eralization on productivity in manufacturing industries. We also document the uniformly positive but heterogeneous in size impact of the reform across manufacturing industries. We �nd that the effect of the reform is stronger for more aggregated data, reflecting the two sources of increase in productivity at the industry level. First, the reform increases within �rm productivity as described in the previous paragraph. Second, the reform leads to exit of low productivity �rms and induces entrance of new competitors due to the general equilibrium effect of liberalization (see Olley and Pakes, 1996; Melitz, 2003), which further increases industry productivity. The structure of the paper is as follows. Section 2 places this study within the existing literature. Section 3 describes progress of the services sector liberalization in Ukraine in 2001-2007 and its impact on the services sector. Section 4 discusses data, methodology and results. Section 5 concludes. 3 2 Services liberalization and productivity in manufactur- ing Competitiveness of manufacturing �rms in open economy hinges on availability of low-cost, high-quality producer services (Francois and Hoekman, 2010). Literature mentions several theoretical links from services liberalization to growth in productivity. Increased special- ization of producer services leads to gains from trade in services due to increased variety and expanded markets (Markusen, 1989). Lower price, better quality, and wider choice of services allow more complex organization of a manufacturing �rm through further frag- mentation of production activities. In turn, fragmentation of production requires support from internationally competitive transportation, communication, professional and �nancial services providers (Deardorff, 2001). Higher variety of services also generates knowledge, increase its diffusion and exchange (Burgess and Venables, 2004). Outsourcing of services by productive �rms in non-stagnant sectors results in more efficient factor allocation that expands output and production (Oulton, 2001). Since services are often a ’margin’ sector, characterized by network externalities, strin- gent regulations, and barriers to entry, the market power in services leads to loss in com- petitiveness of the economy as a whole and requires services deregulation. Such services as transportation, insurance, professional, or �nancial services play a very important role in de- termining the export competitiveness of manufacturing �rms. In turn, expansion of exports due to lower price margins in services could increase productivity through economies of scale. Importantly, trade liberalization without services liberalization lowers the competitiveness of domestic �rms and causes their exit, which leads to negative employment dynamics in the short run (Francois and Hoekman, 2010). Competition and further specialization in profes- sional services could reduce transaction and contracting costs, which are quite substantial. Lower transaction costs, in turn, encourage more outsourcing activities and arms-length trade (Williamson, 1973). 4 Mounting empirical evidence shows a positive impact of services deregulation on pro- ductivity in downstream manufacturing industries. Arnold et al. (2011) establish a positive link between TFP of manufacturing �rms and liberalization of the services sector by an- alyzing the impact of liberalization of services on the performance of approximately ten thousand manufacturing �rms in the Czech Republic in 1998-2003. The link is stronger for the �rms that use services inputs more intensively. A standard deviation increase in the foreign presence in services is associated with a 3.8 percent increase in TFP. Fernandes and Paunov (2011) �nd that forward linkages from foreign direct investment in services to down- stream manufacturing industries account for almost 5 percent of the observed increase in the Chilean manufacturing productivity growth. Deregulation and liberalization policies that increase competition among intermediate services providers are linked to increased export competitiveness for high-tech industries (Fink et al., 2005). Despite an unambiguously positive link between deregulation of services and manufactur- ing productivity, the endogeneity of services sector reforms makes it difficult to demonstrate that there is a direct causal effect of policy changes in services on productivity. For example, as pointed out by Francois and Hoekman (2010), the liberalization of the services sector in Eastern Europe coincided with a broad range of reforms carried out as the prerequisite for the EU accession. As a result, it is very difficult to disentangle the effect of a particular reform that was a part of the broader reform package. With this regard, investigation of liberalization of services in Ukraine brings some advantages because the reform package was very limited and the effect of the EU integration was not present. 3 Services liberalization in Ukraine 3.1 Services sector in Ukraine The services sector has been generally neglected under central planning (Ofer, 1973). Overem- phasis on the accelerated development of the producer goods industry, as the main driver 5 of economic growth, led to crowding out of investment in services connected to �nal con- sumption (retail trade, hotels and restaurants, personal services). Organization of central planning and abolishing of private ownership of productive assets resulted in underdevel- opment in wholesale trade, �nancial and business services. For instance, the state-owned banking system and central planning of investment decisions resulted in only 1 percent of employment allocated to banking and insurance (Bićanić and Škreb, 1991). Transition from the centrally-planned to market-based economic system required a larger and better developed services sector, which has been growing quite impressively. Figure 1 reports the dynamics of the services sector and �nancial and business sub-sector as a share of GDP of Ukraine in 1991-2009. By 2007, the share of services in the Ukrainian GDP had reached 42 percent. Still, the share was well below the average for the middle income countries, which was equal to 60 percent in 2007 (Francois and Hoekman, 2010) and much lower than the average for the EU countries, which had reached 65.5 percent in 20071 . The whole period could be split into two sub-periods. Ukraine had entered its inde- pendence in 1991 with a share of services to GDP of 20.5 percent, with only 5.5 percent attributed to �nancial and business services. In the �rst decade since independence, the services sector grew primarily due to expansion of telecommunication, retail and wholesale trade sub-sectors. Between 1991 and 2000, the share of �nancial and business services in- creased only marginally. Between 2001 and 2009, on the other hand, �nancial and business services expanded from 6.7 to 18.8 percent of GDP. 3.2 Liberalization of the Ukrainian services sector in 2001-2007 Liberalization of the services sector in Ukraine, �rst and foremost is linked to the WTO accession negotiations. Ukraine applied for accession on 30 November, 1993. The major obstacle on the way to the WTO accession was to bring the national legislation in compliance with the WTO rules and regulations. However, not much had been done till 2001, when 1 The WTO database on services reports the pro�le of the EU services sector in 2007. 6 Figure 1: Evolution of services in Ukraine in 1991-2009 Notes: Data from the the National Accounts Main Aggregates Database provided by the UN. President L. Kuchma “instructed his government to speed up all technical work related to accession negotiations� 2 . The favorable political situation – the coalition government had the majority in the Parliament – allowed it to pass more than 20 new laws related to harmonization of the national laws and regulations with the WTO requirements in 2001- 2003. Concerning services, the government developed new laws and amended existing ones that regulate activities of TV and broadcasting, information agencies, banks and banking activities, insurance, telecommunications, and business services. In telecommunication services, “Law on Telecommunications� of November 2003 provided the possibility for any legal person in Ukraine to operate, service or own telecommunications networks. A National Committee for Communication Regulation (NCCR), established ac- cording to the law, became the regulatory authority in telecommunications which made the 2 Report of the working party on the accession of Ukraine to the world trade organization, 25 January 2008, WT/ACC/UKR/152 7 sector more transparent and open to competition. The law declared principles of equal access and fair competition; introduced the policy towards standardization and harmonization with the world standards; speci�ed detailed procedures for frequency auctions and rules for licens- ing. The �nancial sub-sector has experienced a steady liberalization. In 2006, an amendment to the law “On Banks and Banking� permitted foreign banks to open branches in Ukraine, simpli�ed the procedure for opening of banks and subsidiaries, and clearly de�ned under which circumstances the National Bank of Ukraine can turn down the application by the foreign bank to operate in Ukraine. The law also de�ned limiting terms for accreditation of the foreign banks (up to 3 months). A sequence of amendments to the law on insurance substantially liberalized the insurance sub-sector. In professional services, the laws “On au- diting� and “On Bar� have been amended to remove the nationality requirements. The law on auditing allowed the competition from foreign services providers. The evidence on legislative improvements in the services sector regulations is supported by improvements in EBRD indices of services. According to Figure 2, reporting the progress of Ukraine in reforms of the services sub-sectors, Ukraine has substantially liberalized services in a number of the services sub-sectors. The market access has been improved and the barriers to entry considerably reduced in �nancial, telecommunication, and business services. The legislative effort leveled the playing �eld for local and foreign services providers, improved market access, and made laws and regulations more transparent. The progress to a large extent was exogenously imposed on the Ukrainian government by external economic agents as a prerequisite to the WTO accession. There was no similar progress in equally important infrastructure, utilities, and transport, hotels and restaurants sub-sectors, for which no demand for improved market access had been made. A noticeable exception that illustrates the rule was the process of harmonization of rail transportation tariffs that began in April 2005 after some WTO members asked Ukraine to apply railway tariffs in conformity with the WTO obligations. By June 2007, railway tariffs for most commodities had been equalized. In infrastructure sub-sectors, a state program for reforming and development of 8 Gas Telecom Electricity 4 Indices of infrastructure and services reforms 3 2 1 Trade Transportation Finance 4 3 2 1 2000 2002 2004 2006 2008 Other services Water 4 3 2 1 2000 2002 2004 2006 2008 2000 2002 2004 2006 2008 Notes: The progress of reforms in infrastructure and services is based on EBRD indeces of infrastructure and transition indicators. Mapping from EBRD indices to services sub-sectors is dicussed in Appendix. Progress of reforms in Gas sub-sector is based on the index for Gas sector developed by Institute of Economic Research which is compatible with EBRD index by methodology and scale. Other services category includes business services, hotels and restaurants, real estate, rent, information technologies, research and development Figure 2: Services sub-sectors liberalization in Ukraine in 2000-2009 9 utilities in 2005-2010 was accepted in 2004. The program foresaw dissolution of monopolies operating in the utilities sub-sectors in the long run, but the actual progress of the de- monopolization has been very limited. In parallel with the services liberalization, the WTO negotiations also led to further liberalization of trade in goods. As mentioned earlier, it could have created a problem of disentangling the effects of services liberalization on productivity from the effect of trade lib- eralization, which is positively linked to an increase in productivity in the literature (Pavcnik, 2002; Amiti and Konings, 2007; Khandelwal and Topalova, 2011). However, by 2001 Ukraine had already substantially liberalized its trade in goods. The average MFN tariff of Ukraine in 2002 was 7 percent and declined to 4.7 percent in 20083 . The view that the effect of trade liberalization for Ukraine had a limited impact can be further backed by the results of a computable general equilibrium analysis of the potential gains of WTO accession provided by Rutherford et al. (2005), who concluded that more than 70 percent of welfare gains of the WTO accession for Russia – a country similar to Ukraine in many respects – would come from liberalization of services. Taking into account the similarities of the two economies and the fact that Russia is less trade-liberalized relative to Ukraine4 , it is reasonable to assume that the effect of services liberalization for Ukraine was even more pronounced. Still, in the empirical analysis we control for the effect of exporting on productivity and interact it with the effect of services liberalization to control for potential complementarities. 3.3 Performance of services in 2001-2007 Liberalization of the Ukrainian services sector in 2001-2007 was accompanied by increased share of output produced by private and foreign-owned �rms. The privatization process was limited by the fact that, by 2001, most services sub-sectors in Ukraine already had a high share of output produced by private �rms, including more than 90 percent of output produced by private �rms in the retail trade, �nancial, insurance, and business services sub- 3 Data on MFN tariffs are from UNCTAD - TRAINS (Trade Analysis and Information System) database. 4 The average Russian MFN tariff in 2002 was 9.6 percent. 10 Figure 3: FDI stock in Ukraine in 2001-2007 sectors. Utilities, land transport, and supporting transport activities sub-sectors, on the other hand, were largely state-owned in 2001 and remained state-owned in 2007. The period has been characterized by a surge of FDI in the services sector. Figure 3 shows that in 2001, 48 percent of inward FDI stock in Ukraine was in manufacturing. By 2007, the share of FDI stock in the services sector (excluding utilities) had reached 53 percent, while the share of FDI stock in manufacturing had declined to 30 percent. As a result, the output produced by foreign-owned services providers5 has been growing in almost all services sub-sectors, increasing from 5 percent of services sector output in 2001 to 11 percent in 2007. Labor productivity in the services sector more than doubled between 2001 and 2007. What factors correlate with productivity of services �rms? Table 1 reports labor productivity “premia� for services sector �rms depending on size, ownership, exporter status, sub-sector, 5 Foreign-owned services providers are de�ned as �rms with at least 10 percent of foreign ownership. 11 and year. The table reports the point estimates of the coefficients of the following regression ln(Yit /Lit ) = α + β · F DIshareit + γ · Exporterit + + ζ · ln(Lit ) + θDj + ϑDt + δDr + it (1) where Yit is value added of services sector �rm i in year t deflated by the sub-sector speci�c price deflator, Lit is the �rm’s employment, F DIshareit is the share of the equity owned by foreigners, Exporterit is the dummy variable taking one if the �rm exports and zero otherwise, and Dj , Dt , and Dr are sub-sector, time, and region �xed effects. The regression is estimated by ordinary least squares; signi�cance levels are reported based on robust standard errors. Large services sector �rms with foreign ownership and export activities have the highest labor productivity. Elasticity of labor productivity with respect to size is 0.11. Firms with 10 percent higher foreign ownership are also 9.2 percent more productive. Exporting services �rms have 206 percent higher value added per worker. There is substantial heterogeneity of services �rms across sub-sectors. For instance, post and telecommunications are three times more productive, while air transport is 77 percent more productive than the electricity, gas, and steam sub-sector. The evidence on opening up of services to foreign competition and on improvement in performance of services providers supports our claim about substantial regulatory changes in the Ukrainian services sector and give us a source of variation in the services sector to analyze the impact of the deregulation on performance of manufacturing �rms. The rest of the paper analyzes this question in more detail. 12 Variable Coefficient Variable Coefficient ln(Employment) 0.11** Insurance 1.06** FDI share [0,1] 0.92** Auxiliary �n. activities 0.40** Exporter, Yes=1 1.12** Real estate activities 0.19** Sub-Sector. Base = Electricity, gas, and steam Renting of equipment 0.43** Water -1.03** IT 0.50** Sale of motor vehicles 1.59** R&D 0.14** Retail trade 1.13** Other services 0.42** Hotels and restaurants -0.39** Year, Base = 2001 Land transport 0.10** 2002 0.12** Water transport 1.03** 2003 0.23** Air transport 0.57** 2004 0.38** Auxiliary transp. activities 0.78** 2005 0.52** Post & telecom 1.12** 2006 0.62** Financial intermediation 0.83** 2007 0.72** N=501797 * p < 0.05, ** p < 0.01 Notes: Table reports labor productivity premium conditional on size, foreign ownership, sub-sector, and year. Regional �xed effects are included but not reported. Results are based on OLS regression with robust standard errors. The dependent variable is log of value added per worker deflated by sub-sector speci�c price deflator. Table 1: Labor productivity premium depending on �rm ownership, export status, sub- sector, and year 4 Data, methodology, and results 4.1 Sample The data for the study come from several statistical statements annually submitted to the National Statistics Office (Derzhkomstat) by all commercial �rms in the country. The data are restricted and not available for public use. The sample covers seven years from 2001 to 2007. The total number of �rms in the data set exceeds 400,000 per year and covers all sectors except budgetary organizations (public schools, public hospitals, museums, etc.) and banks. We start with the sample of manufacturing �rms (NACE Section “D�) which never switched to another sector over the period of study. Since the Sectoral Expenditures Statement, required to construct the �rm-speci�c service liberalization index, is submitted by only relatively large �rms, our sample is restricted to the �rms with more than 150 employees on average. We further excluded observations with zero or negative output, capital stock or employment assuming that they indicated non-operational �rms in a year. Based 13 on the �les accompanying the Enterprise Performance Statement and the Balance Sheet Statement, we have created a comprehensive pro�le for every �rm which includes the industry (KVED/NACE) and territory codes, as well as exporting status in every year which were used as controls. The industry codes were used to assign manufacturing �rms into one of eleven sub-industries. In every sub-industry, we cut off the top 1 percentile of the sample (measured by employment, capital and output) to exclude outliers. As the measure of output, we used net sales after excise taxes from the Financial Results Statement. The Balance Sheet Statement is the source of the capital measure for which we used the end-of-year value of the tangible assets. For the production function estimation we used investments in tangible assets which come from the Enterprise Performance Statement. The same statement is also a source for our employment variable. It is measured as the “year- averaged number of enlisted employees�, which is a rough estimate of the full time equivalent of labor used. The material costs come from the same statement in 2001-2004, whereas since 2005 they have been available from a separate Sectoral Expenditures Statement. The statement provides detailed information about the �rm’s expenditures on purchases from 22 manufacturing sectors and 15 service sectors. Data from this statement were used to construct an individual �rm-speci�c index of services liberalization as we explain in the Appendix. All variables were deflated by the appropriate price deflators available from the National Statistical Office. The descriptive statistics for the sample are presented in Table 2. 4.2 Methodology Following an insight from Rajan and Zingales (1998), who use variation in industries’ �nan- cial dependency and countries’ �nancial development to investigate the effect of �nancial liberalization on economic growth, several recent papers adopted this idea to investigate ef- fects of trade (Amiti and Konings, 2007; Pavcnik, 2002; Khandelwal and Topalova, 2011) and 14 Variable N Mean Std. Dev. Min Max R,thd.hryvnas 40440 8001.91 16948.02 0.08 548158.70 L, workers 40440 171.57 265.52 1 6779 K,thd.hryvnas 40440 3111.47 6964.35 0.07 183732.00 M,thd.hryvnas 40440 6971.82 17583.37 0.1 706991.3 I,thd.hryvnas 30357 693.39 2112.28 0 89370.69 Serv. Lib. 40440 0.36 0.57 0 4.85 Serv. Lib. (FDI) 40440 0.34 0.64 0 29.31 Exporter, Yes=1 40440 0.34 0.47 0 1 Table 2: Descriptive statistics services (Arnold et al., 2011; Fernandes and Paunov, 2011) liberalization on productivity. We follow a similar identi�cation strategy. The strategy relies on the assumption that manu- facturing �rms that use services more intensively gain more from services liberalization. For each period t, we construct a �rm-speci�c index of services liberalization by interacting a sub- sector-speci�c index of services liberalization with �rm- and sub-sector-speci�c intensity of services use. We further look at the within �rm variation in TFP and relate it to the changes in the �rm-speci�c index of services liberalization. To recover the TFP measure, we estimate the production function for each manufacturing industry (1-digit NACE classi�cation) by the Olley-Pakes procedure (Olley and Pakes, 1996), controlling for sub-industry-speci�c demand and price shocks as suggested by De Loecker (2011). We identify demand and price shocks by exploiting variation in sub-industry (4-digit NACE classi�cation) output at time t and by controlling for sub-industry and time �xed effects. Under the constant elasticity of sub- stitution (CES) demand system, unobserved prices are picked up by the variation in inputs and by aggregate demand and do not reflect differences in technology within an industry. If this assumption fails, we still are able to estimate the impact of services liberalization on productivity because our identi�cation strategy relies on within �rm variation in services intensity and time invariant differences in technology are not important. 15 Technology and market structure Consider a production technology of a single-product �rm i at time t described by a produc- tion function α Yit = Lαl Kitk Mit m exp(˜ it + uit ), it α ω ˜ (2) where Yit units of real output are produced using Lit units of labor, Kit units of capital, deflated by producer-price deflator, and Mit units of material and services inputs. Since we have a breakdown of inputs by sector, each component of Mit is deflated by the corre- ˜ sponding sector-speci�c price deflator. ωit is �rm-speci�c productivity, unobservable by an ˜ econometrician, but known to the �rm before it chooses variable inputs. uit is idiosyncratic shock to production that also captures measurement error. Yit is not observable, because we do not observe �rm-speci�c prices, pit , but sales, Rit = pit Yit , are known. Use of Rit as the dependent variable in estimation of production function parameters, without controlling for prices, determined among other things by market structure and demand shocks, would bias estimates of the production function if prices are correlated with inputs. Even more importantly, generating productivity estimates containing demand variation introduces a re- lationship between services liberalization and measured productivity through the impact of the liberalization on prices and demand. To separate the direct effect of services liberalization on productivity from the indirect effect on demand, we introduce a constant elasticity of substitution demand system σs pit ˜ Yit = Yst exp(ξit ), (3) Pst where Yst is total expenditures on goods produced by manufacturing industry s, in which �rm ˜ i operates. Pst is industry-wide price at time t. ξit is demand shock which is not observed by the �rm when it chooses variable inputs in production. Assuming monopolistic competition, this demand structure implies a constant mark-up price-setting rule, which depends on the industry-speci�c elasticity of substitution σs . It further implies the following expression for 16 the revenue function 1 σs +1 1 −σ (Yst )− σs Pst exp(ξit ) ˜ s Rit = (Yit ) σs . (4) Substituting (2) into (4) and taking logs yields rit = βl lit + βk kit + βm mit + βs yst + ωit + ξit + uit , (5) where rit = ln(Rit /Pst ) is log of revenue deflated by corresponding industry price deflator, σs +1 and other lower-case letters represent upper-case variables in the log form. βf = σs αf , where f = {l, k, m}. The elasticity of substitution in industry s can be retrieved from σs +1 1 ˜ σs +1 σs = −1/βs . Finally, ω it = σs ˜ ωit , ξit = − σs ξit , and uit = σs ˜ uit are error terms. Estimation of production function We estimate rit = βl lit + βk kit + βm mit + βs ygt + ωit + ξit + uit , (6) separately, for each industry s, keeping in mind our ultimate goal of measuring TFP net of price and demand shocks. In what follows we suppress index s for clarity of presentation. Instead of using overall output of industry s we use more disaggregated sub-industry g output, ygt , to add more variability to estimation of σs . It is valid since we assume that the elasticity of substitution is constant within the industry. We decompose the overall demand shock into the following components ˜ ξit = ξt + ξg + ξit , (7) where ξt is industry-speci�c shock common to all �rms at time t, ξg is demand factor affecting ˜ only �rms producing in sub-industry g, and ξit is an idiosyncratic shock. Plugging in (7) in 17 (6), we have rit = βl lit + βk kit + βm mit + βygt + δt Dt + δg Dg + ωit + εit (8) where Dt = ξt is a shock common to all �rms in the industry at time t and Dg is a dummy variable that takes the value of one if a �rm i operates in sub-industry g and zero otherwise. ˜ εit = ξit + uit is the error term which is not correlated with inputs and productivity. We estimate (8) by the Olley-Pakes methodology, which is described in the appendix. The point estimates of the coefficients of the production function are presented in Table 3. Total factor productivity net of price and demand effects is recovered as σs ln(T F Pit ) = (rit − βl lit − βk kit − βm mit − βs yst ) . (9) σs + 1 We do not factor out sub-industry and time effects because we control for those effects in the second stage of the estimation described in the next subsection. 4.3 Main results Impact of services liberalization on TFP of manufacturing �rms The estimated T F P is further regressed on the index of services liberalization that is �rm- speci�c, reflecting the variation in �rm-level intensity of usage of various services inputs. Following Arnold et al. (2011), the index is computed according to the following formula serv libit = aijt · indexjt (10) j where serv libit is the �rm-speci�c index of services liberalization, aijt is the share of input sourced from the services sub-sector j in the total input for a �rm i at time t, and indexjt is the EBRD measure of liberalization in the service sub-sector j at time t. Mapping from the EBRD indices to services sub-sectors is described in the appendix. The constructed index of services liberalization takes into account liberalization of services 18 ln(K) ln(L) ln(M ) ln(Y ) Firms N βK αK βL αL βM αM βs Food and Tobacco 0.043* 0.043 0.204*** 0.204 0.751*** 0.752 0.001 2567 11253 (0.018) (0.014) (0.012) (0.024) Textile and Leather 0.107*** 0.104 0.445*** 0.434 0.469*** 0.458 -0.025 816 3104 (0.029) (0.025) (0.019) (0.045) Wood and Paper 0.052 0.064 0.154*** 0.189 0.712*** 0.876 0.187* 513 2025 (0.029) (0.026) (0.035) (0.090) Printing 0.084*** 0.079 0.402*** 0.377 0.497*** 0.467 -0.065 848 3363 (0.023) (0.033) (0.021) (0.053) Coke, chemistry 0.107** 0.118 0.156*** 0.172 0.697*** 0.767 0.091** 798 3662 plastics (0.036) (0.019) (0.027) (0.030) Non-metallic minerals 0.041 0.040 0.170*** 0.166 0.784*** 0.764 -0.026 758 3269 (0.027) (0.021) (0.016) (0.042) Metallurgy 0.074** 0.079 0.179*** 0.192 0.670*** 0.717 0.066 747 2999 (0.028) (0.030) (0.043) (0.035) Machinery and 0.025 0.024 0.365*** 0.355 0.570*** 0.554 -0.029 1033 4291 equipment (0.021) (0.021) (0.022) (0.030) High-tech machinery 0.060 0.061 0.207*** 0.209 0.594*** 0.600 0.010 705 3111 (0.037) (0.032) (0.020) (0.055) Vehicles and transport -0.015 -0.017 0.279*** 0.314 0.551*** 0.620 0.111* 305 1355 (0.050) (0.057) (0.051) (0.053) Furniture and others 0.095* 0.090 0.312*** 0.294 0.536*** 0.505 -0.061 555 2159 (0.038) (0.042) (0.041) (0.075) Notes: * p<0.05, ** p<0.01, *** p<0.001. Bootstrap standard errors are presented in parentheses. Table reports point σs estimates of revenue function parameters, β and production function paramters α = σs +1 β, where σs = −1/βs for Ukrainian manufactruing �rms for 2001-2007. Each row in the table represents Olley-Pakes estimation of production function for eleven manufacturing industries, de�ned according to NACE Revision 1 classi�cation. Each estimation is performed with year and sub-industry dummies, which are not reported for brevity. Table 3: Estimation of production function by Olley-Pakes procedure 19 sub-sectors and weights more heavily services sub-sectors used by the �rm i more intensively. The assumption here is that the liberalization of the sub-sector used by the �rm more intensively is more important and has a larger impact on the �rm’s performance. We control for export status of the �rm by including an indicator variable that takes the value of 1 if the �rm i has exported at time t. We also control for industry-time speci�c �xed effect to take into account unobservable industry characteristics such as market structure, cost and demand shocks, and technological changes. Finally, we include �rm-speci�c �xed effects to control for unobservable managerial abilities and other �rm-speci�c characteristics that can be correlated with intensity of services use and productivity. The estimated regression takes the following parametric form ln(T F Pit ) = α + serv libit β + exporterit γ (11) + exporterit · serv libit δ + Di µ + Dst λ + it where T F Pit is �rm i s level of total factor productivity at time t, exporterit is the dummy variable that takes the value of one if �rm i exported in year t and zero otherwise, Di are �rm-speci�c �xed effects, and Dst are industry-time speci�c �xed effects capturing the market structure of industry s, and industry-speci�c macroeconomic shocks at time t. Results are presented in panel A of Table 4 in columns (1) - (3). In column (1), the ln(T F P ) estimated by the Olley-Pakes method is regressed on the index of services liberal- ization, controlling for �rm and industry-time �xed effects. Hence, we estimate within �rm effect of liberalization on productivity, removing any impact of current market structure and demand shocks within the industry. The coefficient of the services liberalization is positive and signi�cant. Increase in the index of services liberalization by a standard deviation is associated with 9 percent increase in productivity. In column (2), we add the export status of the �rm to capture the fact that exporters are both more productive (see, for example, Bernard et al., 2003) and use services more intensively in order to coordinate their overseas 20 (1) (2) (3) (4) (5) (6) A. Main results Serv. lib. 0.135∗∗∗ 0.137∗∗∗ 0.127∗∗∗ (0.014) (0.014) (0.016) Serv. lib. (FDI) 0.082∗∗∗ 0.083∗∗∗ 0.065∗∗ (0.019) (0.019) (0.020) Exporter 0.100∗∗∗ 0.091 ∗∗∗ 0.098∗∗∗ 0.083∗∗∗ (0.011) (0.012) (0.011) (0.013) Serv. Lib. × Exporter 0.028 (0.025) Serv. Lib. (FDI) × Exporter 0.049 (0.031) Firms 11057 11057 11057 11057 11057 11057 N 40440 40440 40440 40440 40440 40440 R2 0.65 0.66 0.66 0.64 0.65 0.65 B. Main results. IV Serv. lib. 0.153∗∗∗ 0.155 ∗∗∗ 0.144∗∗∗ (0.015) (0.015) (0.017) Serv. lib. (FDI) 0.106∗∗∗ 0.108∗∗∗ 0.099∗∗∗ (0.013) (0.013) (0.015) ∗∗∗ ∗∗∗ Exporter 0.100 0.089 0.100∗∗∗ 0.091∗∗∗ (0.011) (0.012) (0.011) (0.012) Serv. Lib. × Exporter 0.033 (0.028) Serv. Lib. (FDI) × Exporter 0.027 (0.024) Firms 11057 11057 11057 11057 11057 11057 N 40440 40440 40440 40440 40440 40440 R2 0.65 0.66 0.66 0.65 0.65 0.65 ∗ ∗∗ ∗∗∗ p < 0.05, p < 0.01, p < 0.001 Notes: Dependent variable is ln(T F P ) estimated by Olley-Pakes procedure. Robust standard errors are reported in parentheses. Each estimation is performed with industry-time cross-effects and �rms’ �xed effects, which are not reported for brevity. Table 4: Services liberalization and total factor productivity in Ukraine in 2001-2007. 21 activities. Inclusion of the export status only marginally changes the point estimate of the coefficient of the services liberalization. At the same time, we �nd that �rms that change their status from non-exporters to exporters are about 10 percent more productive relative to �rms that operate only domestically. In column (3), our preferred speci�cation, we add an interaction between the exporter status and the services liberalization to see whether the exporters respond differently to services liberalization as discussed in the literature (Dear- dorff, 2001; Francois and Woerz, 2008; Francois and Hoekman, 2010). We �nd that exporters additionally gain in TFP due to services liberalization, but the effect is not signi�cant. Alternative measure of services liberalization The EBRD measure of services liberalization can be criticized for being subjective, because the indices are based on experts’ judgment. To check whether the subjectivity drives the result, we introduce an alternative measure of services sub-sector liberalization based on the share of employment of services providers with foreign ownership in total employment in the sub-sector6 . This measure is outcome-based and it proxies the degree of openness of services sub-sectors to foreign competition. The FDI based index of services liberalization is computed as serv lib(F DI)it = aijt · F DIsharejt (12) j where serv lib(F DI)it is the �rm-speci�c index of services liberalization, aijt is the share of input sourced from service sub-sector j in the total input of �rm i at time t, and F DIsharejt is the share of labor of majority foreign-owned companies in sub-sector j at time t. The results with the alternative measure of services liberalization are presented in columns (4) - (6) of Table 4. In terms of the direction and signi�cance of the effect of services liberalization, the results are similar to the results with EBRD indices. A standard deviation increase in the services liberalization measured by foreign presence is associated with an increase in 6 A measure based on output produced by the foreign services providers gives very similar results. 22 productivity by 5.5 percent. Arnold et al. (2011) �nd that a standard deviation increase in foreign presence in the services sectors in the Czech republic is associated with a 3.8 percent increase in the productivity of manufacturing �rms, which is in line with our �ndings. Endogeneity issues Industries may lobby the government to liberalize services. More productive �rms that are larger and better politically connected have a stronger inluence on the government’s decision which services sub-sectors to liberalize. Hence, the positive link between services liberalization and productivity may be due to reverse causality. To adress the concern, we instrument indexjt and F DIsharejt by the log of the sub-sector speci�c outward services FDI of the EU to the rest of the world, ln(F DIjt )7 . The argument goes as follows. The EU has been a major bilateral negotiator over the WTO accession of Ukraine. We expect that the EU put more pressure on liberalization of those services sub-sectors in which there are large FDI outflows from the EU. Results of the �rst stage IV regression, presented in the appendix, indicate the EU FDI outflows are good predictors for both the EBRD index of liberalization and for the FDI share of employment. In the second stage we replace our indices of services liberalization with the indices of services liberalization instrumented with outward FDI in services sub-sectors and report the results in panel B of Table 4. 4.4 Sub-sample results We further test the robustness of our results by looking at sub-samples of data along the time, ownership and size dimensions. The results are presented in Table 5. 7 Data is available from Eurostat 23 (1) (2) (3) (4) (5) (6) 2001-2004 2005-2007 Domestic Foreign Small Large Serv. lib. 0.146∗∗∗ 0.115∗∗∗ 0.131∗∗∗ 0.151 0.167∗∗∗ 0.102∗∗∗ (0.021) (0.028) (0.016) (0.082) (0.024) (0.019) Exporter 0.113∗∗∗ 0.031 0.084 ∗∗∗ 0.125∗ 0.086 0.075∗∗∗ (0.016) (0.021) (0.012) (0.055) (0.044) (0.011) Serv. Lib. × Exporter 0.018 0.043 0.043 -0.056 0.145 0.019 (0.035) (0.049) (0.029) (0.082) (0.094) (0.024) Firms 8813 8174 10444 948 5752 7290 N 25710 14730 37418 3022 12041 28399 R2 0.60 0.74 0.64 0.76 0.60 0.69 ∗ ∗∗ ∗∗∗ p < 0.05, p < 0.01, p < 0.001 Notes: Dependent variable is ln(T F P ) estimated by Olley-Pakes procedure. Robust standard errors are reported in parenthesis. Each estimation is performed with year dummies, industry-time cross-effects, and �rm �xed effects, which are not reported for brevity. Column (1) is estimated for 2001-2004. Column (2) is estimated for 2005-2007. Column (3) is estimated for domestic �rms with share of FDI less than 10 percent. Column (4) is estimated for foreign-owned �rms, with share of FDI above 10 percent. Column (5) is estimated for �rms with employment below 50 workers. Column (6) is estimated for �rms with employment 50 workers and above Table 5: Results for different sub-samples Services liberalization before and after 2005 We split the sample into two sub-periods – 2001-2004 and 2005-2007 – to control for possible effect of the political regime switch, because Ukrainian governments before and after the Orange revolution of 2005 represented interests of different �nancial and industrial groups. Also, there was a constitutional reform that shifted political power from the president to the Parliament. The �rst period was characterized by a more coordinated legislative effort be- tween the president and the Parliament, but the privatization process was non-transparent, resulting in poor investment climate8 . In addition, the government has been indecisive on the integration strategy for the country. There was a discussion on bene�ts of EU vs. Com- monwealth of Independent States (CIS) integration. The second period was characterized by the surge of FDI due to improvements in investment climate and clearly-stated strategy 8 Privatization of the Kryvorozhstal, Ukraine’s largest and most modern steelworks, illustrates the irreg- ularities in pre-2005 privatization procedures. In 2004 it has been privatized for 800 million US dollars by Ukrainian oligarchs Akhmetov and Pinchuk in an auction that left out international bidders due to highly protectionist conditions of the tender. In 2005, the steelworks has been re-privatized by Arcelor Mittal for 4.8 billion US dollars. 24 of integration into the EU structures. However, the legislative effort has been stalled due to less effective coordination between the branches of government. The results of the baseline regression for the two sub-periods, presented in columns (1) and (2) of the table, indicate that the effect of services liberalization on productivity was positive and signi�cant in both sub-periods, ruling out the possibility that our main result was driven by the shift in the political environment. We also can not reject the test that the point estimates for the services liberalization were different in the two sub-periods. Ownership type We further split the sample into domestic- and foreign-owned �rms, de�ning the foreign ownership threshold at 10 percent . Since the foreign-owned �rms often have better access to services from the international services providers, we expect that the services liberalization should have a smaller impact on them. The results are presented in columns (3) and (4) of the table. Indeed, only the coefficient of services liberalization for the domestic sub-sample is signi�cant. However, it should be noted that the coefficient for the foreign sub-sample is still positive and large in size. The loss in signi�cance might be driven by a considerably smaller sample size of the foreign-owned �rms. Firm size Finally, we split the sample into small and large �rms, de�ning a small �rm as a �rm that employs less than 50 workers. We expect the small �rms gain more from the services liberal- ization because larger �rms can produce some services internally (i.e. having a transportation or auditing departments), while small �rms rely on external services providers more heavily. The results, presented in columns (5) and (6) of the table, indicate that the effect of the ser- vices liberalization on the small �rms is about 50 percent larger. The result has an important policy implication that improved services encourage development of small and medium enter- prises. A caveat to this conclusion is that small manufacturing �rms are under-represented 25 in our sample, because a considerable number of small manufacturing �rms do not report their use of services and are excluded from the sample. 4.5 Robustness checks One-stage method with exit depending on TFP The literature, while giving advantage to the two-stage procedure of estimating the impact of policy change on TFP (Pavcnik, 2002; Amiti and Konings, 2007; Arnold et al., 2011), also utilizes a one-step approach (Fernandes and Paunov, 2008; Javorcik, 2004). As pointed out by De Loecker (2011), the standard two-stage procedure of estimating the impact of liberal- ization on productivity implicitly assumes that services liberalization does not impact prices and variable inputs in production, is not related to returns to scale. Even more importantly, this approach does not allow liberalization to dynamically impact the evolution of produc- tivity, which is crucial for the exit decision by the �rm. However, the �ndings presented in the previous subsections directly contradict these assumptions. In particular, increase in contemporaneous TFP due to services liberalization induces higher capital accumulation due to expectation of even higher TFP in the future. It also has an effect on the exit decision. To investigate how our results change if we allow services liberalization to interact with variable inputs, investment, and exit, we implement a one-stage procedure that simultane- ously estimates parameters of the production function and the effect of services liberalization on productivity. We introduce two possible channels of influence of liberalization on TFP and exit. One of the channels comes from overall trade liberalization either due to selection process (Melitz, 2003) or due to learning by exporting (De Loecker, 2007; Amiti and Konings, 2007). We control for this effect by including the export status as one of the variables that influences TFP either directly in the production function or indirectly through the selection process. The second channel, the one we are focused on, is from services liberalization to productivity. We modify the model by changing the productivity process to depend on export status 26 and services liberalization ωit = ht (kit , iit , exportit , servlibit ). This creates two effects: a contemporaneous effect on current level of productivity and the dynamic effect on future productivity due to current investment decisions. While we capture the �rst effect in the two-stage procedure, the dynamic effect is ignored. The results of the one-stage estimation by manufacturing industries is presented in panel B of Table 6 and compared with the results of the two-stage procedure by industries, presented in panel A of the table. In general, the one-stage procedure estimations of the effect of services liberalization are higher relative to the two-stage, which is an indication of the dynamic interaction of services and export liberalization with productivity and choice of variable inputs. At the same time, the one-stage procedure suffers from the fact that the services liberalization indicator is an endogenous variable because it depends on the choice of services inputs. Therefore, these results should be taken with care. Industry level results and �rm dynamics with exit depending on productivity We expect that the effect of services liberalization on industry level productivity should be greater then the within effect on �rm level productivity. Additional channels of increase in industry productivity work through exit of low productive �rms and reallocation of capital, labor, and materials towards more productive �rms, which expand their output and boost industry level productivity. The services sector liberalization, which according to our results increases productivity of �rms that use services more intensively, shifts the TFP distribution within an industry to the right. The size of the shift varies across �rms – heavy services users gain more – and is exogenous to the �rm. Analyzing the effect of the shift of productivity on exit and entry of �rms and distribution of �rms within the industry in the framework of the Melitz (2003) model brings the following conclusions: high productive �rms that use services more intensively expand their output, revenues, and pro�ts, while low productive �rms exit the market. Therefore, ceteris paribus, services liberalization should lead to a higher aggregate 27 A. Two-stage Olley-Pakes results by manufacturing industries (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Serv. Lib. 0.279*** 0.085 0.239* 0.235*** 0.195* 0.069 0.223*** 0.231*** 0.099 0.309*** 0.366*** (0.043) (0.052) (0.103) (0.025) (0.083) (0.054) (0.051) (0.049) (0.066) (0.079) (0.066) Exporter 0.156*** 0.288*** 0.167*** 0.213* 0.077 0.06 0.121* 0.108*** 0.249*** 0.465*** 0.200** (0.018) (0.038) (0.048) (0.086) (0.068) (0.035) (0.057) (0.031) (0.049) (0.072) (0.064) Servlib × 0.118 0.104 0.034 -0.111 0.397 0.182* 0.235 -0.005 0.146 -0.11 0.155 Exporter (0.071) (0.056) (0.130) (0.077) (0.206) (0.087) (0.195) (0.073) (0.085) (0.105) (0.159) Firms 3011 1074 600 960 874 918 866 1258 816 343 662 N 11616 3728 1759 2826 3087 3516 3066 4592 2854 1323 2073 ∗ ∗∗ ∗∗∗ p < 0.05, p < 0.01, p < 0.001 Robust standard errors are presented in parentheses. B. One-stage Olley-Pakes results by manufacturing industries (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Serv. Lib. 0.355*** 0.160** 0.324*** 0.287*** 0.256** 0.138** 0.205*** 0.308*** 0.173** 0.329*** 0.456*** (0.049) (0.055) (0.086) (0.035) (0.087) (0.043) (0.055) (0.049) (0.053) (0.050) (0.086) Exporter 0.081*** 0.262*** 0.081 0.106 0.101* 0.043 0.093* 0.069* 0.124** 0.098 0.193** (0.017) (0.039) (0.043) (0.084) (0.042) (0.032) (0.046) (0.032) (0.039) (0.063) (0.065) Servlib × 0.088 0.077 -0.090 -0.124* 0.229 0.100 0.263 -0.103 0.107 -0.088 -0.051 Exporter (0.077) (0.054) (0.095) (0.061) (0.175) (0.056) (0.175) (0.067) (0.063) (0.075) (0.113) 28 Revenue function parameters ln(K) 0.046 0.109** 0.023 0.068 0.127* 0.046 0.061* 0.062*** 0.028 -0.000 0.111* (0.024) (0.039) (0.035) (0.047) (0.064) (0.032) (0.026) (0.018) (0.039) (0.043) (0.046) ln(L) 0.185*** 0.326*** 0.117*** 0.346*** 0.082** 0.157*** 0.124*** 0.329*** 0.197*** 0.280*** 0.304*** (0.015) (0.027) (0.025) (0.041) (0.025) (0.028) (0.032) (0.027) (0.028) (0.054) (0.060) ln(M ) 0.772*** 0.532*** 0.794*** 0.547*** 0.764*** 0.823*** 0.678*** 0.639*** 0.666*** 0.638*** 0.558*** (0.013) (0.024) (0.026) (0.034) (0.025) (0.019) (0.040) (0.021) (0.024) (0.047) (0.056) ln(Y ) 0.019 -0.012 0.238** -0.056 0.056 0.044 0.091* -0.023 -0.012 0.079 -0.092 (0.028) (0.048) (0.086) (0.072) (0.033) (0.052) (0.038) (0.033) (0.070) (0.058) (0.088) Log like. 6443*** 1481*** 768*** 1017*** 1792*** 1793*** 1563*** 2255*** 1461*** 790*** 916*** Firms 2567 816 513 848 798 758 747 1033 705 305 555 N 11253 3104 2025 3363 3662 3269 2999 4291 3111 1355 2159 ∗ ∗∗ ∗∗∗ p < 0.05, p < 0.01, p < 0.001 Bootstrap standard errors are presented in parentheses. Notes: Panel A reports point estimates of regression of ln(T F P ), estimated by Olley-Pakes procedure, on services liberalization and export status. Robust standard errors are reported in parenthesis. Each estimation is performed with year dummies and industry �xed effects, which are not reported for brevity. Panel B reports point estimates of regression of ln(Sales) on services liberalization and export status, as well as revenue function parameters for Ukrainian manufacturing �rms for 2001-2007. Each column in the table represents Olley-Pakes estimation of production function for eleven manufacturing industriess, de�ned according to NACE Revision 1 classi�cation. Each estimation is performed with year and industry dummies, which are not reported for brevity. Table 6: One- and two-stage results by manufacturing industries (1) (2) (3) Serv. Lib. 1.348* 0.446*** 0.133* (0.483) (0.086) (0.061) Exporter 3.114*** 0.294 0.094 (0.330) (0.223) (0.080) Serv. Lib. × -1.241 -0.007 0.078 Exporter (0.558) (0.126) (0.078) Industries 11 102 238 N 77 498 1574 * p<0.05, ** p<0.01, *** p<0.001 Notes: Dependent variable is ln(T F P ) estimated by Olley-Pakes procedure. Robust standard errors are reported in parentheses. Each estimation is performed with industry and time �xed effects which are not reported for brevity. Table 7: Industry level results productivity within the industry, which exceeds the effect for a particular �rm in the industry, and to higher dispersion of output, revenues, and pro�ts.9 The industry productivity is de�ned as output-weighted average TFP of �rms operating in the industry. The measure of TFP is taken from the baseline estimation of the production function presented in Table 3. Figure 4 shows kernel densities of TFP for each manufacturing industry in 2001 and 200610 . The distribution has shifted to the right for Textile and Leather, Printing, High-tech machinery, Vehicles and transport. At the same time, productivity has shifted to the left in Food and tobacco, Wood and Paper, Coke, chemistry and plastics, Non-metallic minerals, and Metallurgy. Industry level regressions, presented in Table 7, con�rm our conjecture of large and pos- itive impact of services liberalization on within industry productivity. Results are presented for industry-level aggregation in column (1), NACE 2 digit aggregation in column (2), and NACE 3 digit aggregation in column (3). Finally, we estimate the effect of productivity, capital, and liberalization on probability of exit. We expect that services liberalization should encourage exit of low productive �rms. We 9 The analysis is almost identical to the analysis of trade liberalization presented by Melitz. The important difference is that Melitz assumes that the distribution of productivity is constant over time, while we consider the case when the distribution shifts exogenously. 10 We have chosen TFP in 2006 rather than in 2007, because the sample of �rms in 2007 is much smaller and comparison of distributions in 2001 and 2007 might be misleading. 29 Figure 4: TFP in 2001 and 2006 Notes: Figure presents kernel density of TFP in 2001 (dashed line) and in 2006 (solid line) by each industry 30 High productive �rms Low productive �rms ln(T F Pi,t−1 ) -0.149* -0.103* (0.069) (0.042) ln(Ki,t−1 ) -0.067*** -0.084*** (0.016) (0.014) Serv.Lib.i,t−1 0.082 0.215*** (0.054) (0.052) Exporteri,t−1 -0.194* -0.271** (0.087) (0.084) Serv.Lib.i,t−1 × Exporteri,t−1 -0.020 0.063 (0.090) (0.134) N 6300 8103 Log Likelihood -773 -1608 * p<0.05, ** p<0.01, *** p<0.001 Table 8: Exit split the sample of �rms according to their productivity into quartiles and report estimates of probit for �rms in the �rst productivity quartile (low productive �rms) and the fourth productivity quartile (high productive �rms) in Table 8. Indeed, low productive �rms are more likely to exit, when the services sector liberalizes. All other variables influence the probability of exit in the expected direction. More productive �rms and �rms with more capital are less likely to exit. Exporters are less likely to exit. Alternative methods There is inherent difficulty of and methodological debates on estimating the production function (Ackerberg et al., 2005; Van Biesebroeck, 2007), which is the crucial element of our empirical procedure. We try several alternative methods of estimation of the production function to reassure the robustness of our results, presented in Table 9. The estimation results of the two-stage Olley-Pakes procedure estimated on the pooled manufacturing sample (columns (1) and (2) of the table) are compared with the estimation results by several other methods. The one-stage Olley-Pakes procedure, analogous to the method presented in panel B of Table 6, but estimated on the pooled manufacturing sample, is presented in column (3). Similarly to the results by manufacturing industries, the one-stage method estimate 31 of services liberalization is considerably larger relative to the two-stage method. A one standard deviation increase in services liberalization is associated with 16 percent increase in productivity. The one-stage Levinsohn-Petrin (Levinsohn and Petrin, 2003) procedure, which treats the selection of materials to infer unobserved productivity, is presented in column (4). It also estimates the effect of services liberalization to be higher relative to the two- stage method. As results of OLS with �rm �xed effects in column (5) and the estimate in �rst differences in column (6) demonstrate, the much higher estimates of the effect by the one-stage OP and LP methods are not due to the inherent differences between the one- and two-stage methods, but due to the dynamic effects of services liberalization on exit and entrance of �rms through the effect on future productivity. Finally, column (7) presents the Blundell-Bond estimate of the services liberalization effect (Blundell and Bond, 2000), which is smaller but still positive and signi�cant. 32 (1) (2) (3) (4) (5) (6) (7) OP 2 stage OP LP FE Dif BB 1st stage 2nd stage Serv. Lib. 0.130*** 0.314*** 0.284*** 0.100*** 0.075*** 0.049*** (0.020) (0.018) (0.017) (0.016) (0.014) (0.008) Exporter 0.075*** 0.117*** 0.097*** 0.097*** 0.095*** 0.085*** (0.012) (0.013) (0.011) (0.011) (0.010) (0.010) Serv. Lib × 0.011 0.032 0.148*** 0.040 0.029 0.034** Exporter (0.025) (0.031) (0.029) (0.024) (0.020) (0.012) Revenue function parameters ln(K) 0.054*** 0.050*** 0.134*** 0.027*** 0.035*** 0.029*** (0.008) (0.014) (0.020) (0.007) (0.008) (0.005) ln(L) 0.272*** 0.229*** 0.242*** 0.398*** 0.414*** 0.378*** (0.009) (0.010) (0.011) (0.017) (0.017) (0.010) ln(M ) 0.635*** 0.688*** 0.569*** 0.560*** 0.505*** 0.502*** (0.009) (0.009) (0.047) (0.013) (0.011) (0.004) ln(Y ) 0.024* 0.022 0.023 0.048*** 0.033*** 0.065*** (0.011) (0.015) (0.013) (0.012) (0.009) (0.012) ln(ri,t−1 ) 0.064*** (0.014) Time Yes Yes Yes Yes Yes Yes Industry Yes Yes Yes Yes Yes Time × Industry Yes Firm Yes Yes Firms 9411 11057 9411 11057 Obs. 41127 40440 41127 40440 40440 29041 32306 Notes: * p<0.05, ** p<0.01, *** p<0.001. Table reports point estimates of regression of ln(T F P ) (column (2)) or ln(Sales) (columns (3)-(7)) on services liberalization and export status, as well as revenue function parameters for Ukrainian manufacturing �rms for 2001-2007. Production function in all models is estimated for all manufacturing industries pooled in one regression. Columns (1) and (2) are estimated by Olley-Pakes two-stage procedure. Column (3) is estimated by Olley-Pakes one-stage procedure. Column (4) is estimated by Levinsohn-Petrin procedure. Column (5) is estimated by OLS with �rm �xed effects. Column (6) is estimated by OLS in �rst differences. Column (7) is estimated by Blundell-Bond. Standard errors, reported in parentheses, are either bootstrapped for OP and LP methods, or robust for all other methods. Table 9: Alternative methods 5 Conclusions This paper �nds that liberalization of services has a positive effect on productivity of manu- facturing �rms. We consider an episode of a limited liberalization in Ukraine, which primarily targeted the services sector as a prerequisite to the WTO accession. These particular fea- tures of the episode allow to separate the effect of services liberalization from the effects of 33 other reforms and to reduce concerns on the endogeneity of the reform. We pay particu- lar attention to unbiased estimation of TFP by employing the Olley-Pakes methodology of estimation of the production function with important innovations laid out in De Loecker (2011). In addition, we compare one- and two-stage methods of estimation of the effect of liberalization and �nd that the two-stage procedure biases the effect of policy downwards by failing to account for the effect of the policy on exit and on the variable inputs mix. According to more conservative results from the two-stage estimation procedure, a stan- dard deviation increase in the services liberalization boosts the productivity by 9 percent. An alternative measure of the services liberalization, based on FDI penetration in the ser- vices sector, indicates that a standard deviation increase in the FDI based index is associated with a 5.5 percent increase in productivity. The size of the effect is higher than in other studies (Arnold et al., 2011; Fernandes and Paunov, 2011), probably reflecting the fact that the initial conditions in the services sector in Ukraine were much worse than in the Czech Republic or Chile. The effect is stronger for domestic �rms and for small �rms, which gives policymakers a nice tool to support and develop small and medium size domestic enterprises. Another important �nding is much stronger estimates of the effect by the one-stage method. It shows that the services liberalization has an important effect on exit decision and on dynamics of TFP through its impact on investment decisions. This conclusion is supported by the fact that the effect of services liberalization is stronger at the industry level and the exit decision of low productive �rms strongly depends on our measure of services liberalization. However, the one-stage method introduces the endogeneity problem which is not resolved in this paper. We leave this task for future research. References Ackerberg, D., Caves, K., and Frazer, G. (2005). Structural identi�cation of production functions. Preliminary, March. 34 Amiti, M. and Konings, J. (2007). Trade liberalization, intermediate inputs, and productiv- ity: Evidence from indonesia. American Economic Review, 97(5):1611 – 1638. Arnold, J., Javorcik, B., and Mattoo, A. (2011). Does services liberalization bene�t manu- facturing �rms?: Evidence from the czech republic. Journal of International Economics, 85(1):136–146. Bernard, A., Eaton, J., Jensen, J., and Kortum, S. (2003). Plants and productivity in international trade. The American Economic Review, 93(4):1268–1290. Bićanić, I. and Škreb, M. (1991). The service sector in east European economies: What role can it play in future development? Post-Communist Economies, 3(2):221–233. Blundell, R. and Bond, S. (2000). GMM estimation with persistent panel data: an application to production functions. Econometric Reviews, 19(3):321–340. Burgess, R. and Venables, A. (2004). Toward a Microeconomics of Growth. The World Bank, Policy Research Working Paper Series, 3257. De Loecker, J. (2007). Do exports generate higher productivity? evidence from slovenia. Journal of International Economics, 73(1):69–98. De Loecker, J. (2011). Product differentiation, multiproduct �rms, and estimating the impact of trade liberalization on productivity. Econometrica, 79(5):1407–1451. Deardorff, A. (2001). International provision of trade services, trade, and fragmentation. Review of International Economics, 9(2):233–248. Fernandes, A. M. and Paunov, C. (2008). Foreign direct investment in services and manu- facturing productivity growth: evidence for chile. Fernandes, A. M. and Paunov, C. (2011). Foreign direct investment in services and manu- facturing productivity: Evidence for chile. Journal of Development Economics, In Press, Corrected Proof:–. 35 Fink, C., Mattoo, A., and Neagu, I. C. (2005). Assessing the impact of communication costs on international trade. Journal of International Economics, 67(2):428 – 445. Francois, J. and Hoekman, B. (2010). Services trade and policy. Journal of Economic Literature, 48(3):642. Francois, J. and Woerz, J. (2008). Producer services, manufacturing linkages, and trade. Journal of Industry, Competition and Trade, 8(3-4):199 – 229. Hoekman, B., Martin, W., and Mattoo, A. (2010). Conclude doha: it matters! World Trade Review, 9(03):505–530. Javorcik, B. S. (2004). Does foreign direct investment increase the productivity of domestic �rms? in search of spillovers through backward linkages. American Economic Review, 94(3):605 – 627. Khandelwal, A. and Topalova, P. (2011). Trade liberalization and �rm productivity: The case of india. Review of Economics and Statistics, forthcoming. Levinsohn, J. and Petrin, A. (2003). Estimating production functions using inputs to control for unobservables. The Review of Economic Studies, 70(2):317–341. Markusen, J. R. (1989). Trade in producer services and in other specialized intermediate inputs. The American Economic Review, 79(1):pp. 85–95. Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industry productivity. Econometrica, 71(6):1695 – 1725. Ofer, G. (1973). The service sector in Soviet economic growth: a comparative study, volume 141. Harvard University Press. Olley, G. S. and Pakes, A. (1996). The dynamics of productivity in the telecommunications equipment industry. Econometrica, 64(6):1263 – 1297. 36 Oulton, N. (2001). Must the growth rate decline? baumol’s unbalanced growth revisited. Oxford Economic Papers, 53(4):605–627. Pavcnik, N. (2002). Trade liberalization, exit, and productivity improvement: Evidence from chilean plants. Review of Economic Studies, 69(1):245 – 276. Rajan, R. G. and Zingales, L. (1998). Financial dependence and growth. American Economic Review, 88(3):559 – 586. Rutherford, T., Tarr, D., and Shepotylo, O. (2005). The impact on russia of wto accession and the doha agenda: The importance of liberalization of barriers against foreign direct investment in services for growth and poverty reduction. World Bank Policy Research Working Paper No. 3725. Van Biesebroeck, J. (2007). Robustness of productivity estimates. The Journal of Industrial Economics, 55(3):529–569. Williamson, O. (1973). Markets and hierarchies: some elementary considerations. The American Economic Review, 63(2):316–325. 6 Appendix 6.1 Olley-Pakes procedure Unobserved productivity follows an exogenous �rst order Markov process p(ωit |Iit ) = p(ωit |ωit−1 ), where Iit is the �rms information set at time t. Capital accumulated by �rms is determined as kit = kit−1 (1 − δ) + iit−1 , where i is log of investment. Solving a dynamic problem of pro�t maximization yields the following investment function iit = it (kt , ωit ). (13) 37 Assuming (13) is strictly increasing in ωit , we invert it to generate ωit = ht (kit , iit ). (14) Substituting (14) into (8) yields ˜ rit = βl lit + βm mit + g(kit , iit ) + βygt + δt Dt + δg Dg + uit , (15) where g(kit , iit ) = βk kit + ht (kit , iit ). Clearly, βk can not be identi�ed from (15), but βl and βm are identi�ed, using a third- order polynomial approximation of g(kit , iit ). The capital coefficient is further identi�ed from E[rit |It , χit = 1] = βl lit + βk kit + βm mit + βygt + δt Dt + δg Dg + ψ(Πit , ωit−1 ) or rit = E[rit |It , χit = 1] + eit . (16) where we preliminary estimate the survival probability, χit = 1, given by P r{χit = 1|ωit (kit ), Iit−1 } = ϕ(kit−1 , iit−1 ) = Πit and approximate ψ(Πit , ωit−1 ), using predicted probability of survival Πit and a third degree linear approximation of ωit−1 = ht−1 (kit−1 , iit−1 ). 38 (1) (2) indexjt F DIsharejt ln(F DIjt ) 0.156*** 0.013** (0.041) (0.005) Constant 0.466 -0.133* (0.534) (0.063) N 32 32 2 R .32 .23 * p<0.05, ** p<0.01, *** p<0.001 Table 10: IV. First stage 6.2 IV �rst stage We regress our services sub-sector speci�c indicators of liberalization indexjt and F DIsharejt on the log of services sub-sector speci�c ourward FDI from the EU towards the rest of the world, ln(F DIjt ). The results are presented in Table 10. We further construct serv libit = j aijt ·indexjt and serv lib(F DI)it = j aijt ·F DIsharejt , where indexjt and F DIsharejt are linear predictions taken from the �rst stage regressions. 6.3 Mapping EBRD indices to services sub-sectors We have constructed two indices of services liberalization: one that includes utilities and retail trade and another that does not. All results in the paper are reported for the index that includes transportation, telecom, �nancial services, and business services. The results for the index that includes utilities are very similar. Index with utilities and retail trade For eight services sub-sectors – Electricity, Gas, Water and water waste, Retail trade and repair, Transport, Telecom, Finance, and Other business-related services (hotels and restau- rants, real estate, rent, informatization, R&D, agencies) – we map the sub-sector with EBRD indices of reforms as follows: E: Electricity - (electric) 39 E1: Gas (IER index: 2 all the time) E2: Water and water waste (water) G: Retail trade and repair I: Transportation 1/2(rail + roads) I1: Telecom (telecom) J: Finance 1/2(banking + �nancial ) H+K: Other business-related services (hotels and restaurants, real estate, rent, informa- tization, R&D, agencies) 1/5( ssp + price_lib + trade_lib+ competition+ �nancial) Index without utilities and retail trade For four services sub-sectors – Transport, Telecom, Finance, and Other business-related services (hotels and restaurants, real estate, rent, informatization, R&D, agencies) – we map the sub-sector with EBRD indices of reforms as follows: I: Transportation 1/2(rail + roads) I1: Telecom (telecom) J: Finance 1/2(banking + �nancial ) H+K: Other business-related services (hotels and restaurants, real estate, rent, informa- tization, R&D, agencies) 1/5( ssp + price_lib + trade_lib+ competition+ �nancial) 40