WPS4841 P olicy R eseaRch W oRking P aPeR 4841 What Drives Firm Productivity Growth? Paloma Anos-Casero Charles Udomsaph The World Bank Eastern Europe and Central Asia Department Economic Policy Sector February 2009 Policy ReseaRch WoRking PaPeR 4841 Abstract This paper presents new evidence on the causal links business environment are estimated from the World between changes in the business environment and firm Bank Enterprise Surveys conducted in 2002 and 2005. productivity growth. It contributes to the literature in Multicollinearity problems in the full model regression three important aspects. First, it constructs a unique are mitigated by constructing a set of six aggregate database merging information from two large firm-level indicators of the business environment (using principal databases. The samples of both databases are merged on component analysis). The paper finds that, over the four criteria--country, sub-national location, firm size, period 2001 to 2004, an increase of one standard and year--producing a panel of 22,004 firms in eight deviation in infrastructure quality, financial development, economies of Eastern Europe and the former Soviet governance, labor market flexibility, labor quality, and Union: Bulgaria, Croatia, Czech Republic, Estonia,, market competition raises the total factor productivity Poland, Romania, Serbia, and Ukraine. Second, the paper of the average firm by 9.8, 7.8, 3.2, 3.4, 5.8, and 3 addresses shortcomings of earlier studies, namely reverse percent, respectively. Lastly, the paper decomposes firm causation, multicollinearity, and unreliable productivity productivity growth and ranks the relative impact of estimates. Firm productivity growth is estimated drawing changes in these six aspects of the business environment on corporate financial data from manufacturing firms by country, by firm size, and by industry. included in the AMADEUS database. Changes in the This paper is a product of the Economic Policy Division, Eastern Europe and Central Asia Department. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at panoscasero@ worldbank.org. 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 What Drives Firm Productivity Growth? Paloma Anos-Casero* and Charles Udomsaph** World Bank Keywords: total factor productivity, business environment, transition economies JEL Codes: D24, O12, P27 *Paloma Anos-Casero is Senior Economist at the World Bank (email: panoscasero@worldbank.org). Charles Udomsaph is consultant (STC) at the World Bank (email: cudompsaph@worldbank.org) 1. Introduction This paper addresses a central question in the recent literature on the microeconomics of growth: What is the impact of changes in the business environment on firm productivity? Institutions and policies determine the business environment within which individuals accumulate skills and firms accumulate capital and produce output. Regulations and laws exist to protect against diversion, but are often instruments of predation in an economy. A good business environment reduces rent seeking activities, supports productive activities, and encourages skill acquisition, capital accumulation, and innovation. This paper builds upon the recent research by Dollar, Hallward-Driemeier, and Mengistae (2003), Bastos and Nasir (2004), and Escribano and Guasch (2005) in using data from recent World Bank Enterprise Surveys to link indicators of the business environment to firm-level productivity. These studies have done much to overcome the many shortcomings of the macroeconomic literature on this topic (Knack and Keefer, 1995; Hall and Jones, 1999; Acemoglu, Johnson, and Robinson, 2001). 1 However, these earlier papers that were the first to use data from the World Bank Enterprise Surveys suffer from two major estimation problems. First, the countries covered were surveyed just once and only a single year of business environment indicators is available. Given the cross-sectional nature of the data, regressions potentially suffer from a problem of reverse causality--some business environment indicators whose effect on firm productivity is estimated may themselves be affected by firm productivity. For example, financing from foreign banks may have a positive effect on firm productivity, but concurrently, financing from foreign banks may be influenced by firm productivity (i.e., foreign banks are more willing to lend to only the most productive domestic firms). Second, while multiple years of data are collected for production function variables in some of these countries, measurement error and non-response plague the recall data collected by these surveys. The paper contributes to the literature in two important respects. First, a unique dataset is constructed by merging information from two large databases of European firms--the Business 1 As Dollar, Hallward-Driemeier, and Mengistae (2003) note, the literature that examines the links between the business environment and productivity at the macroeconomic level suffers from three major shortcomings: (i) few countries have good data on the business environment that are necessary to derive robust statistical results (Levine and Renelt, 1992; Rodriguez and Rodrik, 2000); (ii) the proxies used as explanatory variables provide minimal guidance about what governments need to do to improve their business environment; and (iii) using national-level data assumes that the business environment is the same across locations within a country, but interesting variation may exist based on heterogeneous local governments and institutions. 1 Environment and Enterprise Performance Survey (BEEPS) and AMADEUS--in order to address the aforementioned shortcomings of the earlier studies. For the measurement of the business environment, the analysis draws on firm-level data from the BEEPS, which was conducted by the World Bank in conjunction with the European Bank for Reconstruction and Development (EBRD) and covers all countries of Central and Eastern Europe, the former Soviet Union, and Turkey. In an effort to track changes of evolving business environments and benchmark the effects of reforms, BEEPS was conducted in 2002 and again in 2005, asking an identical core set of questions (covering 367 variables) in both rounds to ensure comparability across countries and years. For the estimation of firm productivity, the analysis uses data from the May 2006 edition of the AMADEUS database, a comprehensive, pan-European commercial database compiled by Bureau van Dijk. For each firm, the database includes up to ten years of accounting data. The manufacturing sector of these two large databases are merged on four criteria--country, sub- national location, firm size, and year--producing a large 4-year panel of 22,004 manufacturing firms in 8 countries, Bulgaria, Croatia, the Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine. This unique dataset enables us to measure the effect of changes in the business environment on firm-level productivity growth over the period 2001 to 2004. Second, in order to mitigate the problems of multicollinearity in the full model regression, a new set of robust indicators is constructed, using principal component analysis on quantitative variables from the BEEPS manufacturing dataset, that summarizes the following five distinct aspects of the business environment: (a) infrastructure quality, (b) financial development, (c) governance, (d) labor market flexibility, and (e) labor quality. Variable selection for PCA is guided by the preference for quantitative over qualitative indicators for two reasons. First, quantitative responses link directly to objective, actionable policy actions, as opposed to firm perceptions. Second, there are numerous statistical and measurement problems associated with the use of perception-based data, such as Likert-scale survey responses. 2 The construction of each indicator meets the three variables per component minimum threshold recommended for exploratory factor analysis (Thurstone, 1935; Kim and Mueller, 1978b). Furthermore, all synthetic indicators are given by the first principal component of their 2 For example, based on data from Enterprise Surveys in 33 African and Latin American countries that used instruments similar to those in the BEEPS, González, López-Córdova, and Valladares (2007) show that perceptions adjust slowly to firms' experience with corrupt officials and hence are an imperfect proxy for the true incidence of graft. 2 respective set of underlying BEEPS variables, and three separate tests--the Guttman-Kaiser criterion, Cattell's scree test, and Humphrey-Ilgen parallel analysis--confirm the decision to retain only the first principal component. The estimation strategy follows a two-step approach and exploits cross-cell (defined by country, sub-national location, and firm size) variation in the changes across time of the five synthetic business environment indicators, as well as a sixth measuring the level of competition (based on the four-firm concentration ratio for each 4-digit NACE industry), to determine their effect on firm-level productivity. 3 First, a production function equation whose residuals measure total factor productivity (TFP) is estimated using the methodology of Levinsohn and Petrin (2003), which corrects for the crucial simultaneity bias arising from the fact that firms make input choices with knowledge of their productivity. Second, a first-differenced equation in firm characteristics, whose dependent variable is the two-year change in log TFP and whose main regressors of interest are the lagged two-year changes in six different business environment indicators, is estimated using ordinary least squares with White correction for heteroskedasticity. The availability of four years of production function data from the AMADEUS database allows the model specification to control for lagged productivity in this second step. This feature is particularly important for consistency given the assumption in Levinsohn and Petrin (2003) of a Markov process for productivity (Fernandes, 2007). The results of the regression analysis confirm that firm-level productivity growth is directly linked to important factors in the business environment and strongly support the presence of large TFP gains from successful efforts to improve these microeconomic foundations of economic development, even after controlling for unobserved firm, industry, sub-national location, and country heterogeneity. The main findings of the paper are as follows. Over the period 2001 to 2004, (i) a one standard deviation increase in the infrastructure indicator raises TFP of the average firm by 9.8 percent; (ii) a one standard deviation increase in the financial development indicator raises TFP of the average firm by 7.8 percent; (iii) a one standard deviation increase in the governance indicator raises TFP of the average firm by 3.2 percent; (iv) a one standard deviation increase in the labor market flexibility indicator raises TFP of the 3 NACE Rev.1 (Nomenclature générale des activités économiques dans les Communautés européennes), the standard industrial classification of economic activities within the European Communities, is identical to the United Nations Statistical Division's International Standard Industrial Classification of All Economic Activities (ISIC Rev. 3) at the one- and two-digit levels. 3 average firm by 3.4 percent; (v) a one standard deviation increase in the labor quality indicator raises TFP of the average firm by 5.8 percent; and (vi) a one standard deviation increase in the competition indicator raises TFP of the average firm by 3 percent. Lastly, to complement the productivity analysis that is based on regression analysis, productivity growth over the period 2002 to 2004 is decomposed following Olley and Pakes (1996) as a way to measure and rank the relative impact of these six aspects of the business environment on a country-by-country basis. In Bulgaria, relative to the total impact of changes in all six business environment indicators, improvements in infrastructure quality contributed 15 percent to log TFP growth over the period 2002 to 2004, whereas a decrease in the level of competition accounted for a 33 percent negative impact. In Croatia, the increases in the infrastructure quality and governance indicators led to relative contributions of 36 and 28 percent, respectively, while the decrease in the labor quality indicator accounted for a -12 percent impact. In the Czech Republic, the increases in the infrastructure quality and financial development indicators led to relative contributions of 25 and 17 percent, and conversely, the decline in the labor market flexibility indicator resulted in a -23 percent relative contribution. In Estonia, the increases in the labor quality and infrastructure quality indicators led to relative contributions of 30 and 27 percent, while the decrease in the labor market flexibility indicator resulted in a relative contribution of -27 percent. In Poland, the increase in the labor market flexibility indicator led to a relative contribution of 34 percent to log TFP growth, whereas the decline in the financial development indicator resulted in a relative contribution of -39 percent. In Romania, all aspects of the business environment improved over the period 2001 to 2003, with the change in the governance indicator accounting for 42 percent of the total positive impact on log TFP. In Serbia, the increase in the infrastructure quality indicator led to a relative contribution of 53 percent to log TFP growth, and conversely, the decline in the financial development indicator resulted in a relative contribution of -17 percent. In Ukraine, the increase in the infrastructure quality indicator led to a relative contribution of 85 percent to log TFP growth and dominated the relative contributions of the other five aspects of the business environment. The paper proceeds as follows. Section 2 describes the data. Section 3 presents the empirical methodology. Section 4 discusses results. Section 5 concludes. The annex presents descriptive statistics and main results. 4 2. Data The empirical analysis in the paper merges information from two large databases of European firms: the Business Environment and Enterprise Performance Survey (BEEPS) and AMADEUS databases. 2.1 Business Environment and Enterprise Productivity Survey (BEEPS) For the measurement of the business environment, the analysis draws on firm-level data from the BEEPS, which was conducted by the World Bank in conjunction with the European Bank for Reconstruction and Development (EBRD). BEEPS covers establishments of all sizes in many industries and provides a wide array of qualitative and quantitative information regarding the business environment in all countries of Central and Eastern Europe, the former Soviet Union, and Turkey. Topics covered in the BEEPS include the obstacles to doing business, infrastructure, finance, corruption and red tape, legal and judicial issues, labor market regulations, and the skills and education of available workers. Taken together, the qualitative and quantitative data capture all aspects of the business environment within countries that affect firm productivity and performance. In an effort to track changes of evolving business environments and benchmark the effects of reforms, the survey was conducted in 2002 and again in 2005. An identical core set of questions (covering 367 variables) was asked in all countries in both rounds to ensure comparability across countries and years, and all questionnaires in every country in both rounds of the BEEPS were implemented through face-to-face interviews with managers and owners. In each country, the sectoral composition of the sample in terms of industry (ISIC codes 10-14, 15- 37, 45) versus services (ISIC codes 50-52, 55, 60-64, 70-74) was determined by their relative contribution to GDP. Furthermore, the sampling design in both rounds included quotas for a set of firm characteristics to ensure sufficient numbers for statistical analysis, specifically, city/town (i.e., large, medium, small), firm size (i.e., small=2-49, medium=50-249, large=250-9,999), ownership (i.e., domestic, foreign, state), and exporters/non-exporters. The sampling approach was the same in both rounds of the BEEPS and was implemented nationwide. 4 4 The BEEPS 2002 and 2005 datasets in Stata and CSV format as well as documentation on sampling and implementation are available for download from the following World Bank website: 5 2.2. AMADEUS Database For the estimation of firm productivity, the analysis uses data from the May 2006 edition of the AMADEUS database, a comprehensive, pan-European commercial database compiled by Bureau van Dijk. For each firm, AMADEUS provides accounting data in standardized financial format for 24 balance sheet items, 25 profit and loss account items, 26 financial ratios, and additional information including trade description and activity codes. The database includes up to ten years of information per firm through 2004, although coverage varies by country. AMADEUS is created by collecting standardized data received from 50 vendors across Europe, where the local source for these data is generally the office of the Registrar of Companies. 5 The accounts for each firm are transformed into a universal format to allow for comparison across countries. All accounting data is converted into U.S. dollars using period average exchange rates, based on monthly series from the International Monetary Fund, nearest to the end date of each respective financial account. Nominal values are deflated using country- level GDP deflators to express values in 2001 US dollars. In addition, all firms are categorized by industry according to NACE Rev.1., and for the analysis, industry dummy variables are coded based on the 4-digit activity code following NACE Rev.1 that AMADEUS assigns to each firm. 2.3. Sample Selection The econometric analysis of firm-level TFP growth and changes in the business environment uses a first-differenced equation in firm characteristics with two-year changes and requires a panel of manufacturing firms from the AMADEUS database with complete information on production function variables for the years 2001 through 2004, the period that correspond to the 2002 and 2005 rounds of the BEEPS. Specifically, output, labor, material inputs, and capital are given by the operating revenues, number of employees, material costs, and tangible fixed assets of firms in the AMADEUS database. Consequently, observations that are missing values in just one of these four production function variables must be dropped from the sample. Given these data requirements, sufficient information exists in the AMADEUS database http://web.worldbank.org/WBSITE/EXTERNAL/COUNTRIES/ECAEXT/EXTECAREGTOPANTCOR/0,,content MDK:21303980~pagePK:34004173~piPK:34003707~theSitePK:704666,00.html. 5 Further details about the AMADEUS database can be found on the product page of Bureau van Dijk's website: http://www.bvdep.com/en/AMADEUS.html. 6 to estimate TFP for manufacturing firms in eight countries: Bulgaria, Croatia, the Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine. A number of additional restrictions are imposed to reduce sample bias in the panel of AMADEUS firms with complete data on production function variables. First, observations that are "inactive", "dissolved", "in bankruptcy", or "in liquidation" are dropped from the panel. Bureau van Dijk removes firms from the AMADEUS database only when there is no reporting for at least five years; specifically a "not available/missing" is reported for four years following the last included filing. Second, observations with data sourced from consolidated statements are dropped from the panel in order to avoid the double-counting of firms and subsidiaries or operations abroad. For most firms in the AMADEUS database, unconsolidated statements are reported and consolidated statements are provided when available. Third, observations with a positive number of subsidiaries are also dropped from the panel to reduce double-counting. Fourth, observations with less than two employees are dropped from the sample. This criterion helps to exclude any dummy (phantom) firms established for tax or other purposes. Fifth, certain manufacturing industries are excluded when the activity is country-specific. Observations in the manufacture of tobacco products (NACE code 16) are dropped from panel because there are no such observations from Croatia, the Czech Republic, Estonia, Serbia, and Ukraine in the AMADEUS database. Similarly, observations in the manufacture of coke, refined petroleum products and nuclear fuel (NACE code 23) are dropped from panel because there are no such observations from Bulgaria, Croatia, the Czech Republic, Estonia, Poland, Serbia, and Ukraine in the AMADEUS database. Lastly, observations in recycling (NACE code 37) are dropped from the panel because there are no such observations from Bulgaria with complete information on production function variables in the AMADEUS database. A number of additional criteria are imposed on the set of four production function variables to reduce measurement error. First, observations with negative tangible fixed assets and material costs are dropped from the sample. Second, observations with material costs- operating revenues and cost of employees-operating revenues ratios greater than one are dropped from the sample. Lastly, observations with operating revenues-number of employees, tangible fixed assets-number of employees, material costs-number of employees, material costs- operating revenues, and cost of employees-operating revenues ratios that are greater (less) than 7 three times the standard deviation from the upper (lower) quartile in the corresponding two-digit NACE industry, country, and year are considered outliers and dropped from the sample. Given that respondents to the BEEPS were asked to answer questions with respect to business operations occurring in the previous year, BEEPS 2002 and 2005 data are assumed to capture the characteristics of the business environment in 2001 and 2003, respectively, in order to fit the first-difference model with two-year changes in firm productivity regressed on lagged two-year changes in the business environment. BEEPS 2002 and 2005, therefore, is match merged with Amadeus 2002 and 2004 observations, respectively, on country, sector, sub- national location, and firm size. Specifically, averages of variables from the BEEPS manufacturing dataset are first calculated for groups defined by country, sub-national location, and firm size in each respective year using only the responses of manufacturing firms (NACE codes 15-36). There are three sub-national location categories: capital city, large city (defined as a non-capital city having a population of 250,000 or greater), and small city (defined as a non- capital city having a population less than 250,000); and two firm size categories: small (defined as employing 2 to 49 full-time workers) and large (defined as employing 50 or more full-time workers). These country-location-size-year averages of BEEPS variables for the manufacturing sector are then match merged to each AMADEUS observation on this identical set of variables. To illustrate, the average number of days in 2001 that large-sized manufacturing firms located in small cities in Bulgaria experienced power outages or surges from the public grid is first calculated from the BEEPS 2002 database, and then this value is assigned to all observations in the 2002 AMADEUS sample that operate in the manufacturing sector, employ 50 or more full- time workers, and are located in cities with populations less than 250,000 in Bulgaria. The final sample that will be used for the econometric analysis of the effect of changes in the business environment on firm-level TFP growth over the period 2001 to 2004 consists of 22,004 manufacturing firms in 8 countries: Bulgaria, Croatia, the Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine. The distribution of the merged AMADEUS-BEEPS balanced panel dataset by countries is as follows: Bulgaria 221; Croatia 1,780; Czech Republic 964; Estonia 1,253; Poland 1,133; Romania 12,576; Serbia 2,237; and Ukraine 1,840. The above inclusion criteria create the most comparable sample of firms across countries. Note, however, that strong conclusions at the international level cannot be derived from direct cross-country 8 comparisons because data requirements for the estimation of log TFP result in varying sample attrition across countries, leading to non-representative country samples. 3. Estimation Methodology To estimate the impact of the business environment on firm performance, the two-year change in log TFP of manufacturing firms is regressed on lagged two-year changes in several aspects of the business environment as measured by a wide array of BEEPS variables. 3.1. Estimation of TFP in the Presence of Simultaneity Total factor productivity is measured as the residual from the estimation of a log-linear three factor Cobb-Douglas production function. For the analysis, the production function of firm i in NACE 2-digit manufacturing industry (15-36) j at time t is assumed to have the following form: Yijt Aijt L M ijt K ijt , ijt (1) where Y is a measure of output, and L, M, and K are the usage of labor, material inputs, and capital with output shares , , and , respectively. Drawing on the AMADEUS database, Y is measured by operating revenues (thousands of 2001 U.S. dollars), L is measured by the number of employees, M is measured by material cost (thousands of 2001 U.S. dollars), and K is measured by the value of tangible fixed assets (thousands of 2001 U.S. dollars). Aijt represents TFP and increases the marginal product of all factors simultaneously. Transforming equation (1) into logarithms allows for linear estimation of TFP with the equation for the general form written as: ln Aijt ln Yijt j ln Lijt j ln M ijt j ln K ijt , (2) where industry-specific coefficients--j, j, and j-- are given by the estimation of the production function. A simultaneity problem, however, arises when there is contemporaneous correlation between the factors of production and the errors, often thought as Hicks neutral productivity shocks. The firm, for example, may observe productivity shocks early enough to allow for a change in factor input decisions. In the context of the Cobb-Douglas production function, the error term is therefore assumed to be additively separable in two distinct components: 9 y ijt a j j lijt j mijt j k ijt ijt ijt , (3) where y is the logarithm of output; l and m are the logarithm of the freely variable inputs of labor and materials; k is the logarithm of the state variable capital; is the part of the error term that is observed by the firm when decisions on optimal factor input choices are being made, and thus, are correlated with the inputs, l, m, and k; and is a true error term uncorrelated with factor input choices that may contain both unobserved shocks (i.e., unpredictable zero-mean shocks realized after inputs are chosen) and measurement errors. As pointed out by Griliches and Mareisse (1998), profit-maximizing firms immediately adjust their inputs each time a productivity shock is observed, resulting in input levels correlated with in the regression. This simultaneity violates the OLS conditions for unbiased and consistent estimation. Olley and Pakes (1996) and Levinsohn and Petrin (2003) have developed two similar semi-parametric estimation procedures to overcome the simultaneity problem when estimating production functions. Olley and Pakes include the investment decision of the firm in the estimation equation to proxy for unobserved productivity shocks. Derived from a structural model of the optimizing firm, the proxy controls for the part of the error correlated with inputs, , by "annihilating" any variation that is possibly related to the productivity term. The method suggested by Olley and Pakes, however, generates consistent and unbiased estimates if and only if there is a strictly monotonous relationship between the proxy and output. Consequently, firms that make only intermittent investments will have their zero-investment observations truncated from the estimation routine because the monotonicity condition does not hold for these observations. For AMADEUS, this is a large portion of the data. 6 Given the considerable attrition in the AMADEUS sample when using the Olley and Pakes approach, the paper adopts the method developed by Levinsohn and Petrin (2003) to estimate production functions. Levinsohn and Petrin offer an estimation technique that is very close in spirit to the Olley and Pakes approach but uses intermediate inputs in lieu of investment as a proxy for unobserved productivity shocks. Nearly all firms in the AMADEUS database almost always report positive material costs. Therefore, the Levinsohn-Petrin intermediate input proxy estimator is the optimal choice for the AMADEUS sample. 6 Calculating investment as the year-to-year change in the real value of tangible fixed assets, only 1,947 (8.5 percent) of the 22,004 manufacturing firms in the final sample used for the econometric analysis in this paper have strictly positive investment in years 2001 through 2004. 10 Given differences in production technologies across industries, the analysis estimates heterogeneous, industry-specific (2-digit NACE) production functions using the Levinsohn and Petrin technique to obtain consistent and unbiased estimates of j, j, and j for the derivation of log TFP estimates according to equation (3), which takes two step. 7 In the first step, the coefficient on labor is obtained using semi-parametric techniques. Assuming that the firm's demand for material inputs increases monotonically with its productivity conditional on its capital, the inverse demand function for material inputs then depends only on observable materials usage and capital and its nonparametric estimate can be used to control for unobservable productivity, thus removing the simultaneity bias. 8 In the second step, the coefficients for material inputs and capital are obtained using generalized method of moments techniques. The identification assumption is that capital adjusts with a lag to productivity, specifically productivity is assumed to follow a Markov process, ijt E[ ijt | ijt 1 ] ijt , where ijt is the unexpected part of current productivity to which capital does not adjust. The estimates of firm log TFP are given by the residuals from equation (3), TFPijt ijt ijt , and capture the efficiency in transforming inputs into outputs and may include changes in factor utilization. 9 Table 1 presents descriptive statistics of log TFP estimates, calculated according to the technique of Levinsohn and Petrin (2003), for firms in the manufacturing sector (NACE 1500 to 3663) of Bulgaria, Croatia, the Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine. For years 2001 through 2004, means are provided for the whole sample, by country, and also by groups defined by country, sub-national location, and firm size upon which the BEEPS manufacturing dataset is merged. On average, firms experienced an overall increase in log TFP 7 Alternatively, a fixed effects model can be used to address the simultaneity problem if the part of the error that influences input factor decision, i, is assumed to be a firm-specific attribute and time invariant (e.g., managerial skills, organizational efficiency, etc.). In this case, unobserved firm heterogeneity that remains constant over time can be removed (for example, by subtracting the means from each variable for each observation) before estimating the production function so that l, m, and k are no longer correlated with the error term. However, evidence from BEEPS suggests that managerial skills and organizational structure have changed significantly over time, and therefore, preclude the adoption of fixed effect methods for the analysis in the paper. Among the 1,416 firms from Bulgaria, Croatia, Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine that comprise the BEEPS 2005 sample, 22 percent had "some reallocation of responsibility and resources between departments", 11 percent had "major reallocations of responsibility and resources between departments", and 5 percent had a "completely new organisational structure" over the last three years. 8 Making mild assumptions about the firm's production technology, Levinsohn and Petrin (2003) show that the demand function is monotonically increasing in ijt. 9 The estimated j , ^ j , ^ and j ^ show the importance of the simultaneity bias when compared to OLS. The production function parameters are available from the authors upon request. 11 of 0.062 log points from 2002 to 2004, but performance varied greatly among firms--the standard deviation of sample is 0.386. Log TFP of the average firm in Serbia grew the fastest, increasing by 0.219 log points, whereas log TFP of the average firm in Romania grew the slowest, increasing by only 0.019 log points. Figure 1 through 3 present kernel density estimations of log TFP for several different cuts of the sample using estimates in all years 2001 through 2004 for the panel of 22,004 firms, resulting in a total of 88,016 observations. An adaptive kernel density estimation method using a varying, rather than fixed, bandwidth is used to draw the distributions. The fixed bandwidth tends to oversmooth the middle of the log TFP distribution. On the contrary, the adaptive kernel estimate is smoother in the tails (especially in the higher tail). 10 All estimations use the Epanechnikov kernel function, start with an oversmoothed global bandwidth of 0.3, and specify 3,000 equally spaced grid points. Figure 1 presents the kernel density estimation for the sample as a whole. Figure 2a shows the kernel density estimation of log TFP by firm size. Not surprisingly, the distribution for large firms (250 or more employees) is higher than that for small firms (less than 250 employees). Figure 2b presents the kernel density estimation of log TFP by sub-national location. The order of the distributions from highest to lower are also as expected: firms located in capital cities, firms located in large cities (population greater or equal to 250,000), and lastly, firms located in small cities (populations less than 250,000). Figure 2c shows the kernel density estimation of log TFP by the average industry factor intensity. The distribution for firms in capital-intensity industries (i.e., 4-digit NACE industries with average tangible fixed assets per employee in the top two quintiles, specifically greater than or equal to $8,837.43) is higher than that for firms in labor-intensive industries (i.e., 4-digit NACE industries with average tangible fixed assets per employee in the bottom two quintiles, specifically less than or equal to $6,785.85). Figure 3 shows the kernel density estimation of log TFP by the country. 10 The advantages of varying or local bandwidths is widely acknowledged in the estimation of long-tailed density functions with kernel methods, when a fixed or global bandwidth approach may result in undersmoothing in areas with sparse observations, while oversmoothing in areas with abundant observations. Varying the bandwidth along the support of the sample data gives flexibility to reduce the variance of the estimates in areas with few observations and can reduce the bias of the estimates in areas with many observations. An adaptive kernel approach adapts to the sparseness of the data by varying the bandwidth inversely with the density using an iterative procedure. An initial (fixed bandwidth) density estimate is computed to get an approximation of the density at each of the specified grid points. Subsequently, this pilot estimate is used to adapt the size of the bandwidth over the data points when computing a new kernel density estimate. For a discussion, see Silverman (1986), Bowman and Azzalini (1997), and Van Kerm (2003). 12 The separations observed in the kernel density estimates presented in Figures 2 and 3 confirm the necessity to match merge the BEEPS data with the Amadeus observations on country, sector, sub-national location, and firm size. However, it is important to reiterate here that strong conclusions at the international level cannot be derived from direct cross-country comparisons because of data requirements and varying sample attrition across countries. For example, given the limited data sources available for Serbia, firms that have the prerequisite data on production function variables in all four years of the panel exhibit very high log TFP levels, resulting in a distribution much higher than those of the other countries. Nonetheless, even though sample biases may exist between countries, the basic test in this paper examines within- industry differences across countries and will not be affected unless there are systematic biases in sub-national location-size-year groups within industries in each country. 3.2. Identification Strategy The analysis exploits cross-cell (defined by country, sub-national location, and firm size) variation in the changes of the business environment variables across time to determine their effect on firm-level productivity. Estimated using ordinary least squares (OLS) with White correction for heteroskedasticity, the full regression model is a first-differenced equation in firm characteristics with two-year changes whose main regressors of interest are lagged two-year changes in business environment indicators and is formally specified as follows: ln TFPit 1INFRASTRUCTUREts,1,c 2 FINANCEts,1,c 3GOVERNANCEts,1,c l l l 4 LABOR _ MARKETt ,1l ,c 5 LABOR _ QUALITYt ,1l ,c 6 COMPETITIONtm1c s s , ln TFPi ,t 1 n Z in,t INDUSTRYm LOCATION l COUNTRYc i ,t (4) n m l c where ln TFPi ,t is the change in the logarithm of TFP of manufacturing establishment i from 2002 to 2004, estimated by the semiparametric estimation technique developed by Levinsohn and Petrin (2003); INFRASTRUC TURE ts,1,c , l FINANCE ts,1,c , l GOVERNANCE ts,1,c , l LABOR _ MARKETt ,1l ,c , and LABOR _ QUALITYt ,1l ,c are the changes from 2001 to 2003 in s s respective business environment indicators for groups of firm size s, location l, and country c; ln TFPi ,t 1 is the change in the logarithm of TFP from 2001 to 2003; Z in is a vector of logarithmic changes in firm characteristics from 2002 to 2004 that include the number of 13 employees, the value of tangible fixed assets (thousands of 2001 U.S. dollars), and cost of materials (thousands of 2001 U.S. dollars); INDUSTRYs is a vector of industry dummy variables defined at the 4-digit NACE level (1510 to 3663); LOCATIONs is a vector of location dummy variables including a capital city dummy variable (equal to 1 if the firm is located in a capital city--that is, Belgrade, Bucharest, Kyiv, Prague, Sofia, Tallinn, Warsaw, or Zagreb--and 0 otherwise) and a large city dummy variable (equal to 1 if the firm is located in a city with a population of 250,000 or greater, and 0 otherwise); and COUNTRY c is a vector of country dummy variables for Bulgaria, Croatia, the Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine. Lastly, COMPETITIO N tm1c is the change in the level of competition in each industry m , of country c from 2001 to 2003 and is equal to 1 minus the change in the four-firm concentration ratio for industries defined at the 4-digit NACE level (so that positive changes indicate higher levels of competition). The four-firm concentration of an industry is equal to the market share as measured by operating revenues of the four largest firms in each 4-digit NACE level industry and is country-specific. 11 Given the lower data requirements, a much larger AMADEUS sample is used to calculate the competition indicator: 69,116 firms in 1,935 4-digit NACE industries from the 2001 sample and 77,265 firms in 1,970 4-digit NACE industries from the 2003 sample. Table 2 presents descriptive statistics of the competition indicator. The above specification of the model addresses a number of econometric concerns. Given that the objective of the paper is to capture the effect of changes in the business environment on productivity growth of the average firm, the regression analysis opts for a balanced panel design, pooling observations across 296 NACE industries at the 4-digit level in eight countries with data in the years 2001 through 2004. Second, first-differencing firm characteristics and lagging business environment indicators by one year mitigates further endogeneity between unobservable firm heterogeneity and factor input choices. Third, the inclusion of lagged changes in log TFP addresses serial correlation that is not eliminated by first differencing. Given the necessary assumption made in the TFP estimation technique of Levinsohn and Petrin (2003) of a Markov process for productivity--that is, the conditional probability distribution of future states of the productivity, given the present state and all past 11 Market forms are often classified by their four-firm concentration ratio. Perfect competition is associated with a very low ratio, monopolistic competition with ratios below 0.4, oligopoly with ratios above 0.4, and monopoly with a near-1 four-firm measurement. 14 states, depends only upon the present state and not on any past states--lagged productivity must be included in the regression model for consistency (Fernandes, 2007). Fourth, the inclusion of industry, sub-national location, and country fixed effects controls for time trends and unobserved sub-national location-, industry-, and country-specific characteristics that might affect the correlation between productivity growth and changes in the business environment. Fifth, as described in the previous section, merging the AMADEUS and BEEPS manufacturing datasets on country, sub-national location, firm size, and year mitigates the endogeneity between firm productivity and business environment indicators. The econometric analysis in this paper treats BEEPS variables as exogenous determinants of firm productivity; however, firms can be proactive in reducing the constraints they face in the business environment, producing a simultaneity bias in the estimation exercise. For example, a well- managed firm with high productivity growth may have worked with authorities to secure a more reliable power supply or to relax hiring and firing restrictions. Statistically, a balance must be struck so that the set of variables, on which the AMADEUS and BEEPS manufacturing datasets are merged, is large enough so that resulting average values not only mitigates the endogeneity problem but also retain sufficient variation for regression analysis. To the extent that sub-sample groupings as defined are sufficiently aggregated so that individual firms are less likely to influence averages but varied enough so that heterogeneous "pockets" of business environments are reflected, using year-specific averages of BEEPS indicators taken across firms in the same country, sector, sub-national location, and size groups is a valid way to instrument out the simultaneity problem (Bastos and Nasir, 2004). Sixth, in order to mitigate the problems of multicollinearity in the full model regression, principal component analysis (PCA) is used to reduce the dimensionality of the BEEPS data and construct indicators that summarize various dimensions of the business environment. In the BEEPS database, there are typically several variables that address a particular issue that affect the productivity and growth of firms. Several questions, for example, collect information on the quality of infrastructure, namely the number of days of power outages or surges from the public grid, the number of days of insufficient water supply, and the number of days of unavailable mainline telephone service. Inclusion of two or more highly correlated explanatory variables in a regression model generally leads to difficulties in ascertaining the effects of individual factors 15 on the dependent variable. The follow section explains the construction of the five business environment indicators that are used in the paper. 3.3. Business Environment Indicators: Principal Component Analysis (PCA) Synthetic indicators are constructed using PCA on the BEEPS manufacturing dataset for the following five distinct aspects of the business environment: (a) infrastructure quality, (b) financial development, (c) governance, (d) labor market flexibility, and (e) labor quality. Intuitively, the method of principal components is used to describe a set of variables with a set of variables of lower dimensionality; for this paper, the objective of PCA is to construct one series that summarizes the behavior of a group of three or more underlying BEEPS variables that describe a particular aspect of the business environment. Statistically, PCA reduces the number of variables in the analysis by specifying linear combinations ("principal components") of the underlying BEEPS variables such that the resulting series contains most of the information, i.e. has maximum variance. 12 Specifically, BEEPS variables are first mapped into one of five distinct aspects of the business environment, and then the main variation commanded by each aspect is extracted through the use of their respective principal components. Before applying PCA, the underlying variables are rescaled so that higher values indicate improvements in the business environment and then standardized to having mean zero and standard deviation one in order to abstract from units of measurements. Variable selection for PCA is guided by the preference for quantitative over qualitative indicators. First, quantitative responses link directly to objective, actionable policy actions, as opposed to firm perceptions. Second, there are numerous statistical problems associated with the use of perception-based data, such as Likert-scale survey responses. The most fundamental is whether responses along a semantic continuum can be treated as if they were interval data. Additionally, there are several potential sources of measurement error with perception-based data. For example, individual respondents may differ in their use of the Likert scale owning to 12 Algebraically, this method locates n linear combinations of the n columns of the X'X matrix, all orthogonal to each other, with the following property: the first principal component pl minimizes tr ( X p1 a1 ) ( X p1 a1 ) , where a1 is the eigenvector of the XX matrix associated with the largest eigenvalue. Intuitively, pl summarizes the n variables in X by giving the best linear description of the columns of X in a least squares sense. The second principal component of p2 also describes what is not "captured" by the first component pl by minimizing the sum of squared residuals after subtracting pl, i.e. pl minimizes tr ( X p1 a1 p 2 a 2 ) ( X p1 a1 p 2 a 2 ) where a2 is now the eigenvector associated with the second largest eigenvalue, and so on. See Alesina and Perotti (1996). 16 his or her subjective frame of reference. An issue perceived as a major obstacle to doing business in one country may actually impose a lower cost in actuality than it does in a country where the problem is rated as merely a minor problem. For example, based on data from Enterprise Surveys in 33 African and Latin American countries that used instruments similar to those used for the BEEPS, González, López-Córdova, and Valladares (2007) show that perceptions adjust slowly to firms' experience with corrupt officials and hence are an imperfect proxy for the true incidence of graft. Consequently, quantitative measures of an issue in the business environment are always selected over perception-based indicators whenever available. For example, the number of power outages or surges from the public grid is used rather than the perceptions of the manager on how problematic electricity is for the operation and growth of the business. Similarly, the level of bribes paid as a percentage of total annual sales number is used rather than the perceptions of the manager on how problematic corruption is for the operation and growth of the business. However, because of the inadequate number of quantitative measures available in the areas of governance (legal system), labor market flexibility, and labor quality, one perceptions- based question is used in the construction of these indicators in order to meet the three variables per component minimum threshold recommended for exploratory factor analysis (Thurstone, 1935; Kim and Mueller, 1978b). All synthetic indicators are given by the first principal component of their respective set of underlying BEEPS variables, and three separate tests confirm the decision to retain only the first principal component. The first test is the most frequently used Guttman-Kaiser criterion, which states all components with eigenvalues greater than 1 should be extracted as variables. The rationale behind this criterion is that the interpretation of proportions of variance smaller than the variance contribution of a single variable is of dubious value (Guttman, 1954; Kaiser, 1961). The second test is the Cattell's scree test, which plots the components along the X-axis and the corresponding eigenvalues along the Y-axis and is also a widely used criterion. Cattell (1996) suggests visual inspection to identify an inflection point of the resulting curve (scree), where components to the left are retained and those to the right are dropped. 13 13 "Scree" is the geological term referring to the debris that collects on the lower part of a rocky slope (Cattell, 1966). 17 The final test is Humphrey-Ilgen parallel analysis, which is now often recommended as the best method to assess the true number of factors (Velicer, Eaton, and Fava, 2000; Lance, Butts, and Michels, 2006). Parallel analysis compares obtained eigenvalues to those one would expect to obtain from random data. To use this procedure, a matrix of random numbers representing the same number of observations and variables is factor analyzed. If the first n eigenvalues given by the actual data are those which have values greater than those generated from random data, then n components are retained. Graphically, eigenvalues from the actual and random data are represented on the same scree plot; the intersection of the two lines determines the number of components to be retained. All three tests determined that for each set of BEEPS variables only one component should be retained. A detailed explanation for each of the underlying BEEPS variables used in the construction of the synthetic indicators for infrastructure quality, financial development, governance, labor market flexibility, and labor quality follows below. Given that principal components are used to summarize a group of variables that describe a particular aspect of the business environment, the resulting indices are expected to be correlated with their underlying BEEPS variables. Tables 3 through 7 show that all five indices are indeed strongly associated with their corresponding BEEPS variables. Figures 4 through 8 graphically show that the Guttman-Kaiser criterion, Cattell's scree test, and Humphrey-Ilgen parallel analysis all confirm the retention of only the first principal component for each set of BEEPS variables. Infrastructure Quality The infrastructure quality indicator measures the quality in the provision of infrastructure services. Underlying variables are rescaled as explained below so that higher values of the indicator signify higher levels of infrastructure quality. The indicator is based on a PCA of the following three BEEPS variables: Power outages. The number of days over the last 12 months that each establishment experienced power outages or surges from the public grid (multiplied by -1) (Question 23). Insufficient water supply. The number of days over the last 12 months that each establishment experienced insufficient water supply (multiplied by -1) (Question 23). 18 Unavailable mainline telephone service. The number of days over the last 12 months that each establishment experienced unavailable mainline telephone service (multiplied by -1) (Question 23). Financial Development The financial development indicator measures the reliance of firms on various sources of finance for new fixed investments (i.e., new machinery, equipment, buildings, and land). Underlying variables are rescaled as explained below so that higher values of the indicator signify higher levels of financial development. The indicator is based on a PCA of the following three BEEPS variables: Local private commercial banks. The percentage of new fixed investment financed by borrowing from "local private commercial banks" (Question 45a). Foreign banks. The percentage of new fixed investment financed by borrowing from "foreign banks" (Question 45a). Informal (family/friends/money lenders). The percentage of new fixed investment financed by borrowing from loans from family or friends, money lenders, or other informal sources (subtracted from 100 percent) (Question 45a). Governance The governance indicator measures the control of corruption, bureaucratic efficiency, and judicial effectiveness in resolving business disputes. Underlying variables are rescaled as explained below so that higher values of the indicator signify higher levels of good governance. The indicator is based on a PCA of the following three BEEPS variables: Bribe level. The estimated percentage of total annual sales firms typically pay in unofficial payments or gifts to public officials (subtracted from 100 percent) (Question 40). Tax compliance. The response of the firm to the question, "Recognizing the difficulties that many firms face in fully complying with taxes and regulations, what percentage of total annual sales would you estimate the typical firm in your area of business reports for tax purposes" (Question 43a). 19 Confidence in the legal system. The response of the firm on a six-point scale (1="strongly disagree" to 6="strongly agree") when asked the question, "To what degree do you agree with this statement. `I am confident that the legal system will uphold my contract and property rights in business disputes. (Question 27). Labor Market Flexibility The labor market flexibility indicator measures the efficiency of employment protection legislation and the degree to which labor markets can adapt to fluctuations and changes in the economy or the demands of production. Underlying variables are rescaled as explained below so that higher values of the indicator signify higher levels of labor market flexibility. The indicator is based on a PCA of the following three BEEPS variables: Underemployment and overemployment. The percentage of firms that either report underemployment because of labor restrictions regarding the hiring of workers (i.e., seeking and obtaining permission, etc.) or report overemployment because of labor restrictions regarding the firing of workers (i.e., making severance payments, etc.). Specifically, this dummy variable is equal to 1 if the optimal level of employment estimated by the firm is equal to or greater than 120 percent (underemployment) or equal to or less than 80 percent (overemployment) of their existing workforce, and is equal to 0 otherwise (subtracted from 100) (Question 73). Change in the use of temporary workers. The change in the number of part- time/temporary workers (as a percentage of permanent, full-time workers) over the last 36 months (Questions 66 and 67). Atkinson (1984) and Atkinson and Meager (1986) study the labor management strategies companies use and identify four types of labor market flexibility. One category is called "external numerical flexibility," which refers to the adjustment of labor intake, or the number of workers from the external market. External numerical flexibility can be achieved by employing workers on temporary or fixed-term contracts or through relaxed hiring and firing regulations, where employers can hire and fire permanent workers according to the needs of the firm. 20 Labor regulations as a constraint. The responses of firms on a four-point scale (1="major obstacle" to 4="no obstacle") to the question: How problematic are "labor regulations" to the operation and growth of your business? (Question 63). Labor Quality The labor quality indicator measures the skill level and educational attainment of workers. Underlying variables are rescaled as explained below so that higher values of the indicator signify higher levels of labor quality. The indicator is based on a PCA of the following three BEEPS variables: Skilled workers/Total employees. The percentage of the firm's current permanent, full-time workers that are managers, professionals, or skilled production workers (Question 68). Time to fill vacancy. The average number of weeks it took to fill the most recent vacancy for a manager, professional, or skilled production worker (multiplied by -1) (Question 70). Labor quality as a constraint. The responses of firms on a four-point scale (1="major obstacle" to 4="no obstacle") to the question: How problematic are the "skills and education of available workers'" to the operation and growth of your business? (Question 63). Table 8 presents descriptive statistics of the five synthetic indicators, constructed using PCA on the BEEPS manufacturing dataset for the following five distinct aspects of the business environment: (a) infrastructure quality, (b) financial development, (c) governance, (d) labor market flexibility, and (e) labor quality. For years 2001 and 2003, means are provided for the whole sample, by country, and also by groups defined by country, sub-national location, and firm size upon which the AMADEUS data is merged. On average, countries from 2001 to 2003 improved in the areas of infrastructure quality, governance, and labor quality, but faced worsening financial development and decreasing labor market flexibility. All countries improved infrastructure quality over this period, but results were mixed across countries in the other four areas. Labor market flexibility worsened in the largest number of countries, only improving in Poland and Romania. 21 3.4. The Olley and Pakes Decomposition: Relative Percentage Contribution of Changes in Business Environment Indicators to Log TFP Growth, 2001-2004 To complement the productivity analysis that is based on the OLS estimation of equation (4), the paper follows Escribano and Guasch (2005) and measures the partial direct effect of the change in each business environment indicator on average productivity for each country by calculating the average productivity term of the Olley and Pakes (1996) decomposition of productivity. The Olley and Pakes decomposition of productivity has two components: average Nc productivity and the efficiency or covariance term. Formally, let TFPt c sit TFPitc be the c Y i 1 productivity of country j at year t obtained as the weighted average productivity of firm i in c country c at year t, where N c is the number of firms in country c. The weights sit indicate the Y share of firm i in aggregate operating revenue of country c in year t, and is equal to the operating Yit revenue of firm i divided by the total operating revenue of country c at year t: sit c Y Nc . Let Y i 1 it Nc 1 TFP c TFP c t c it be the average productivity of the firms in country c at year t. Let N i 1 c c 1 sit sit stY and TFP t TFPitc TFP t be deviations to the mean. Since sit c c c c Y Y Y , the annual Nc aggregate productivity of country c can then be decomposed as: Nc TFPt c TFP t sit TFP it , c c c Y (5) i 1 The first term TFPt c is the average productivity of country c at year t and the second term Nc s TFP it N c cov sc ,it , TFPitc measures the allocative efficiency or covariance between the Yc c it Y i 1 share of operating revenue and productivity, cov sc ,it , TFPitc , multiplied by the number of firms, Y Nc, that operate in country c. A covariance that is negative indicates that there are allocation inefficiencies. That is, as the share of output for less productive firms increases, the covariance becomes more negative and the productivity of country c decreases. 22 For the calculation of the relative percentage contribution of changes in business environment indicators to log TFP growth over the period 2001 to 2004, the Olley and Pakes decomposition of productivity is also similarly computed for aggregate productivity in logs. Let Nc ln TFPt sit Y ln TFPitc be the log productivity of country j at year t obtained as the weighted c ln c i 1 c ln average log productivity of firm i in country c at year t. The weights sit Y indicate the share of firm i in aggregate log operating revenue of country c in year t, and is equal to the log operating revenue of firm i divided by the total log operating revenue of country c at year t: Nc ln Yit 1 ln TFP c sln Y c it Nc . Let ln TFP c t c it be the average log productivity of the firms in N ln Y i 1 it i 1 c c country c at year t. Let sit Y sit Y st ln Y and ln TFP t ln TFPitc ln TFP t be deviations to the c c c ln ln 1 Since sit Y c mean. ln , the annual aggregate log productivity of country c can then be Nc decomposed as: Nc ln TFPt ln TFP sit Y ln TFP it , c c c c ln t (6) i 1 The first term ln TFPt c is the average log productivity of country c at year t and the second term Nc s ln TFP it N c cov sclnit , ln TFPitc ln Y c c it , Y measures the allocative efficiency or covariance i 1 between the share of log operating revenue and log productivity. Equation (5) estimated by OLS with a constant term implies that the mean of the residuals is zero, and therefore, the estimation results of equation (5) can be evaluated at their sample mean values without including an error term (Escribano and Guasch, 2005). The corresponding expression for the first term of Olley and Pakes decomposition in changes then becomes: c c c c ln TFPt 1 INFRASTRUCTURE t 1 2 FINANCE t 1 3 GOVERNANCE t 1 ^ ^ s ,l , c s ,l , c m,c 4 LABOR _ MARKET t 1 5 LABOR _ QUALITY t 1 6 COMPETITION t 1 ^ ^ ^ ln TFP i ,t 1 n Z i ,t INDUSTRY m LOCATION l ^COUNTRY c (7) n ^ ^ ^ ^ n m l c 23 where the variables with bars on top indicate the country averages of each covariate. Following Escribano and Guasch (2005), the relative contribution of each business indicator is derived by dividing the change in each business environment indicators by the dependent variable c ln TFPt and multiplying by 100: c c s ,l , c 1 INFRASTRUCTURE t 1 ^ 2 FINANCE t 1 3 GOVERNANCE t 1 ^ 100 c 100 c 100 c 100 ln TFPt ln TFPt ln TFPt s ,l , c s ,l , c 4 LABOR _ MARKET t 1 ^ 5 LABOR _ QUALITY t 1 ^ c 100 c 100 ln TFPt ln TFPt m ,c 6 COMPETITION t 1 ^ c 100 (8) ln TFP t Equation (8) represents the sum of the percentage productivity gains and losses from the change in each business environment indicators relative to the average log TFP growth of country c over the period 2002 to 2004. In this way, the relative impact of the average change in each business environment indicator over the period 2001 to 2003 on average log TFP growth over the period 2002 to 2004 can be estimated. 4. Results 4.1. OLS Regression Estimation As presented in Table 9, the results obtained from the estimation of equation (4) by OLS with robust standard errors (White correction for heteroskedasticity) show a positive and statistically significant impact of improvements in each of the six aspects of the business environment on firm TFP over the period 2001 to 2004. Entering changes in the PCA indicators into the model one by one, the effects are statistically significant at the 1 percent level for infrastructure quality (column 1), financial development (column 2), governance (column 3), labor market flexibility (column 4), and labor quality (column 5), and at the 5 percent level for competition (column 6). Entering changes in all six aspects of the business environment into the model jointly (column 7), the effects remain strong. Changes in all BEEPS-based indicators are again statistically significant at the 1 percent level, with changes in the competition indicator significant at the 5 percent level. 24 From the point estimates of the full regression model presented in column 7 of Table 9, and given the joint significance of the coefficients on the changes in all six business environment indicators, the following causal relationships can be inferred 14 : A one standard deviation increase in the infrastructure indicator over the period 2001 to 2004 (1.532) raises TFP of the average firm by 9.8 percent. A one standard deviation increase in the financial development indicator over period 2001 to 2004 (1.177) raises TFP of the average firm by 7.8 percent. A one standard deviation increase in the governance indicator over period 2001 to 2004 (1.392) raises TFP of the average firm by 3.2 percent. A one standard deviation increase in the labor market flexibility indicator over period 2001 to 2004 (1.198) raises TFP of the average firm by 3.4 percent. A one standard deviation increase in the labor quality indicator over period 2001 to 2004 (1.175) raises TFP of the average firm by 5.8 percent. A one standard deviation increase in the competition indicator over period 2001 to 2004 (0.234) raises TFP of the average firm by 3 percent. The results of the regression analysis confirm that firm-level productivity growth is directly linked to each of these factors in the business environment and strongly support the presence of large TFP gains from successful efforts to improve the business environment. On the whole, while evidence shows that each of the six dimensions of the business environment is important and significant, one caveat is that the results do not provide clear implications for reform priorities in specific countries. 4.2. Olley and Pakes Decomposition Figure 9 presents the results of the Olley and Pakes decomposition in levels by country for 2001, 2002, 2003, and 2004. There are no significant differences across years. Poland has the largest aggregate productivity followed by Serbia and the Czech Republic. 15 The efficiency terms are likewise high for these three countries, whereas their role in the other five countries is 14 With the dependent variable in logarithmic form, the exact percentage change in the predicted TFP associated with a change in the regressor is calculated as [exp( i xi ) 1] 100 where i is the estimated coefficient. ^ ^ 15 Again, given the limited data sources available for Serbia, firms that have the prerequisite data on production function variables in all four years of the panel exhibit very high log TFP levels, resulting in a distribution much higher than those of the other countries at similar levels of economic development. 25 marginal. Nonetheless, the efficiency term is positive in all countries, indicating no allocative inefficiencies in any of the eight countries over the period 2001 to 2004. Figure 10 graphically presents the relative percentage contribution of changes in each business environment indicator to log TFP growth over the period 2001 to 2004 calculated according to equation (8) by country. That is, each bar in Figure 10 shows the relative weight of the average change in each business environment indicator with respect to the total impact of the changes in all six business environment indicators for the respective country sample. Because all coefficient estimates from the OLS regression of equation (4) are positive (see column 7 of Table 9), a positive (negative) relative percentage indicates an improvement (worsening), on average, in the respective business environment indicator. For example, infrastructure quality improved, on average, in all countries over the period 2001 to 2003, while labor quality, on average, increased in the Czech Republic, Estonia, Poland, Romania, and Ukraine, but decreased in Bulgaria, Croatia, and Serbia. For the sample as a whole (first column in Figure 10), all aspects of the business environment, on average, improve over the period 2001 to 2003. Improvements in the infrastructure quality and governance indicators have relative contributions of 27.8 and 22.7 percent. Changes in the labor quality (7.5 percent), financial development (5.6 percent), labor market flexibility (2.6 percent), and competition (2.2 percent) indicators account for the remaining positive business environment impacts on log TFP growth. These results from the Olley Pakes decomposition of log TFP growth by country are consistent with the OLS regression results for the full sample presented in Table 9. In Bulgaria, only two aspects of the business environment improve over the period 2001 to 2003. Relative to the total change in all six business environment indicators, increases in the infrastructure quality and financial development indicators contribute 14.7 and 6.0 percent, respectively, to log TFP growth over the period 2002 to 2004. Conversely, negative changes in the governance, labor market flexibility, labor quality, and competition indicators dominate the positive contributions of increases in infrastructure quality and financial development. A worsening in the competition indicator accounts for a third (33.3 percent) of the total impact of business environment changes on log TFP growth over the period. Negative changes in labor quality (-24.8 percent), labor market flexibility (-14.7 percent), and governance (-6.5 percent) account for the remaining impacts on log TFP growth. 26 In Croatia, several aspects of the business environment improve over the period 2001 to 2003 and have large relative contributions, while indicators with negative changes have relatively little impact, in sharp contrast to Bulgaria. Improvements in the infrastructure quality and governance indicators have relative contributions of 35.8 and 27.9 percent. Changes in the financial development (16.2 percent) and labor market flexibility (7.5 percent) indicators account for the remaining positive business environment impacts on log TFP growth. A worsening in labor quality (-11.7 percent) and competition (-0.9 percent) have limited negative impact on log TFP growth relative to the positive changes in other aspects of the business environment. In the Czech Republic, several aspects of the business environment also improve over the period 2001 to 2003, but have more moderate relative contributions, in comparison to Croatia, while indicators with negative changes have larger relative impacts on log TFP growth. Improvements in the infrastructure quality and financial development indicators have relative contributions of 25.1 and 16.5 percent. Changes in the competition (13.3 percent) and labor market quality (10 percent) indicators account for the remaining positive business environment impacts on log TFP growth. Worsening labor market flexibility (-22.9 percent) and governance (-12.2 percent) over the period have significant negative impacts on log TFP growth relative to the positive changes in other aspects of the business environment. In Estonia, the positive impacts in several aspects of the business environment are also somewhat diminished by the large negative relative contribution of worsening labor market flexibility, similar to the Czech Republic. Improvements in the labor quality and infrastructure quality indicators have relative contributions of 29.5 and 26.7 percent. Changes in the financial development (13.8 percent) and competition (1.4 percent) indicators account for the remaining positive business environment impacts on log TFP growth. A worsening in labor market flexibility has a significant relative contribution of -27.3 percent on log TFP growth relative to the positive changes in other aspects of the business environment. The relative contribution of the change in the governance indicator is -1.3 percent. In Poland, the positive impacts in several aspects of the business environment are diminished by the large negative relative contribution of worsening financial development. Improvements in the labor market flexibility and labor quality indicators have relative contributions of 34.3 and 14.3 percent. Changes in the infrastructure quality (8.3 percent) and competition (0.5 percent) indicators account for the remaining positive business environment 27 impacts on log TFP growth. A worsening in financial development had a significant contribution of -38.6 percent on log TFP growth relative to the positive changes in other aspects of the business environment. The relative contribution of the change in the governance indicator is -4 percent. In Romania, all aspects of the business environment improve over the period 2001 to 2003. The relative contribution of improvements in the governance indicator lead the way with 42 percent of the total positive impact on log TFP growth over the period 2002 to 2004. Labor quality (17.7 percent), infrastructure quality (13 percent), and financial development (12.8 percent) have double digit relative contributions. Changes in the labor market flexibility (9.4 percent) and competition (5.1 percent) indicators account for the remaining positive business environment impacts on log TFP growth. In Serbia, several aspects of the business environment improve over the period 2001 to 2003 and have large relative contributions, while indicators with negative changes have relatively little impact, similar to Croatia. Improvements in the infrastructure quality and governance indicators have relative contributions of 52.5 and 20.2 percent. Changes in the competition (2.5 percent) and labor market flexibility (1.4 percent) indicators account for the remaining positive business environment impacts on log TFP growth. A worsening in financial development (-17.4 percent) and labor quality (-6 percent) have limited negative impact on log TFP growth relative to the positive changes in other aspects of the business environment. In Ukraine, improvements in infrastructure dominate the relative contributions of the changes in all other aspects of the business environment. The increase in the infrastructure quality indicator has a relative contribution 85.4 percent. Changes in the financial development (1.1 percent), labor quality (1.1 percent), and labor market flexibility (0.6 percent) indicators account for the remaining positive business environment impacts on log TFP growth. A worsening in competition (-9.4 percent) and governance (-2.4 percent) have limited negative impact on log TFP growth relative to the positive change in infrastructure quality. Figure 11 graphically presents the relative percentage contribution of changes in each business environment indicator to log TFP growth by firm size. That is, each bar in Figure 11 shows the relative weight of the average change in each business environment indicator with respect to the total impact of the changes in all six business environment indicators for the respective samples of small-sized firms (less that 50 employees) and large-sized firms (50 or 28 more employees). Small-size firms in the sample experienced on average an increase in all the business environment indicators over this period, with improvements in governance (40.1 percent) and infrastructure quality (30.3 percent) accounting for the largest relative contributions to log TFP growth. Conversely, large-sized firms experienced a worsening in labor market flexibility that accounted for a -9.3 percent relative contribution. For large-sized firms, the largest relative contributions to log TFP growth are attributed to increases in the indicators for infrastructure quality (39.3 percent), labor quality (25.4 percent), and financial development (15.7 percent). Figure 12 graphically presents the relative percentage contribution of changes in each business environment indicator to log TFP growth by NACE industries defined at the 2-digit level. That is, each bar in Figure 12 shows the relative weight of the average change in each business environment indicator with respect to the total impact of the changes in all six business environment indicators for the respective sample of firms operating in the indicated industry. Improvements in infrastructure and governance account for the two largest relative contributions, to log TFP growth in all industries (41.6 and 29.9 percent on average, respectively), with the exception of garments (NACE 18) and leather (NACE 19). While the increase in the governance indicator has the largest relative contribution for firms in both the garments (26.6 percent) and leather (23.5 percent) industries, improvements in labor quality (23.4 and 15.7 percent, respectively, for garments and leather) and financial development (19.8 and 16.1 percent) have higher relative contributions than that in infrastructure quality (18.3 and 14.5 percent) for garment- and leather-producing firms. Improvement in labor quality also contributes significantly to productivity growth in many of the industries (11.1 percent, on average) and is ranked third in terms of relative contribution in food processing (NACE 15), textiles (NACE 17), other non-metallic products (NACE 26), basic metals (NACE 27), motor vehicles (NACE 34), and other transport equipment (NACE 35), and furniture (NACE 36). Increases in competition relative to other changes in the business environment contributes at least 5 percent to productivity growth in textiles (NACE 17), leather (NACE 19), chemicals (NACE 24), rubber and plastic (NACE 25), other non-metallic products (NACE 26), basic metals (NACE 27), machinery and equipment (NACE 29), electric machinery (NACE 31), radio, television, and communication equipment (NACE 32), medical, precision, and optical instruments (NACE 33), other transport equipment (NACE 35), and furniture (NACE 36). 29 Worsening of certain aspects of the business environment over the period 2001 to 2004 for firms in leather (NACE 19), paper (NACE 21), office, accounting, and computing machinery (NACE 30), radio, television, and communication equipment (NACE 32), and, motor vehicles (NACE 34) led to significant negative impacts on productivity growth. Decreases in the indicator for labor quality negatively impacted log TFP growth in office, accounting, and computing machinery (-13.8 percent), radio, television, and communication equipment (-7.3 percent), and paper (-2.0 percent). Decreases in the indicator for labor market flexibility negatively impacted log TFP growth in office, accounting, and computing machinery (-7.1 percent), motor vehicles (-4.0 percent), and radio, television, and communication equipment (-3.5 percent). Decreases in the indicator for competition negatively impacted log TFP growth in leather (-6.1 percent), radio, television, and communication equipment (-5.8 percent), and paper (-3.1 percent). Lastly, the decrease in the indicator for financial development negatively impacted log TFP growth in motor vehicles (-8.3 percent). 5. Conclusions This paper provides new evidence on the impact that changes in the business environment have on firm productivity, and contributes to the literature in two important respects. First, a unique dataset is constructed by merging information from two large databases in order to address shortcomings of earlier studies, namely reverse causation and unreliable TFP estimates. Second, in order to mitigate the problems of multicollinearity in the full model regression, a new set of robust indicators is constructed using principal component analysis on quantitative variables from the BEEPS manufacturing dataset to summarize the following five distinct aspects of the business environment: (a) infrastructure quality, (b) financial development, (c) governance, (d) labor market flexibility, and (e) labor quality. Regression analysis is based on production function estimates that correct the bias arising from the simultaneity between inputs and productivity. Furthermore, the paper exploits cross-cell (defined by country, sub- national location, and firm size) variation in the changes of the business environment indicators across years to determine their effect on firm-level productivity, thereby circumventing the shortfalls of previous studies that focus only on a single year of business environment variables. Results indicate that successful efforts to improve the business environment has a strong positive impact on firm productivity, even after controlling for unobserved firm, industry, sub- 30 national location, and country heterogeneity. Evidence from the BEEPS-AMADEUS dataset confirms that a good business environment encourages firms to operate efficiently and promote productivity growth by lowering risks, costs, and barriers to entry. In a global economy where technology diffuses rapidly, the persistence of productivity differences across countries may be largely explained by differences in the business environment in which firms operate. These microeconomic foundations of economic development--infrastructure quality, financial development, governance, labor market flexibility, labor quality, and competition--are critically linked to the success and growth of firms. 31 References Acemoglu, Daron, Simon Johnson, and James A. Robinson (2001). "The Colonial Origins of Comparative Development: An Empirical Investigation." American Economic Review 91(5): 1369-1401. Alesina, Alberto and Roberto Perotti (1996). "Income Distribution, Political Instability, and Investment." European Economic Review 40: 1202-1229. Atkinson, John (1984). "Flexibility, Uncertainty and Manpower Management." IMS Report No.89. Brighton, UK: University of Sussex, Institute for Manpower Studies. Atkinson, John and Nigel Meager (1986). "Changing Working Patterns: How companies achieve flexibility to meet new needs." London: Institute of Manpower Studies, National Economic Development Office. Bastos, Fabiano and John Nasir (2004). "Productivity and the Investment Climate: What Matters Most?" World Bank Policy Research Working Paper 3335. Bosworth, Barry and Susan Collins (2003). "The Empirics of Growth: An Update". The Brookings Institution. Washington, D.C. Bowman, Adrian W. and Adelchi Azzalini (1997). Applied Smoothing Techniques for Data Analysis: The Kernel Approach with S-Plus Illustrations, vol. 18. Oxford, UK: Oxford University Press. Cattell, R. B. (1966). The Scree Test for the Number of Factors. Multivariate Behavioral Research, 1(2): 245-276. Dollar, David, Mary Hallward-Driemeier, and Taye Mengistae (2004). "Investment Climate and Firm Performance in Developing Economies." World Bank Policy Research Working Paper 3323. Escribano, Alvaro and J. Luis Guasch (2005). "Assessing the impact of the investment climate on productivity using firm-level data: methodology and the cases of Guatemala, Honduras, and Nicaragua." World Bank Policy Research Working Paper 3621. Fernandes, Ana M. (2007). "Trade policy, trade volumes and plant-level productivity in Colombian manufacturing industries." Journal of International Economics 71(1): 52-71. González, Alvaro, J., J. Ernesto López-Córdova, and Elio E. Valladares (2007). "The Incidence of Graft on Developing-Country Firms." World Bank Policy Research Working Paper 4394. Griliches, Zvi, and Jacques Mareisse (1998). "Production Function: The Search for Identification." In Econometrics and Economic Theory in the Twentieth Century: The Ragnar Prisch Centennial Symposium. Cambridge University Press: 169-203. Guttman, Louis (1954). "Some Necessary Conditions for Common Factor Analysis." Psychometrika XM: 149-61. Hall, Robert E., and Charles Jones (1999). "Why Do Some Countries Produce So Much More Output per Worker Than Others?" Quarterly Journal of Economics 114(1): 83­116. 32 Kaiser, Henry F. (1961). "A Note on Guttman's Lower Bound for the Number of Common Factors." British Journal of Psychology 14: 1-2. Kim, Jae-On and Charles W. Mueller (1978a). Introduction to factor analysis: What it is and how to do it. Thousand Oaks, CA: Sage Publications, Quantitative Applications in the Social Sciences Series, No. 13. Kim, Jae-On and Charles W. Mueller (1978b). Factor Analysis: Statistical methods and practical issues. Thousand Oaks, CA: Sage Publications, Quantitative Applications in the Social Sciences Series, No. 14. Knack, Steven, and Philip Keefer (1995). "Institutions and Economic Performance: Cross- Country Tests Using Alternative Measures." Economics and Politics 7(3): 207-227. Lance, Charles E, Marcus M. Butts, and Lawrence C. Michels (2006). "The sources of four commonly reported cutoff criteria: What did they really say?" Organizational Research Methods 9(2): 202-220. Levine, Ross, and David Renelt (1992). "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4): 942-963. Levinsohn, James and Amil Petrin (2003). "Estimating Production Functions Using Inputs to Control for Unobservables." Review of Economic Studies 70(2): 317-342. Levinsohn, James, Amil Petrin, and Brian P. Poi (2003). "Production Function Estimation in Stata using Inputs to Control for Unobservables." Stata Journal 4(2): 113-123. Olley, G. Steven and Ariel Pakes (1996). "The Dynamics of Productivity in the Telecommunications Equipment Industry." Econometrica 64(6):1263­1297. Rodriguez, Francisco, and Dani Rodrik (2000). "Trade Policy and Economic Growth: A Skeptic's Guide to the Cross-National Evidence." Macroeconomics Annual 2000, Ben Bernanke and Kenneth Rogoff, eds., MIT Press for NBER. Silverman, B. W. (1986). Density Estimation for Statistics and Data Analysis. Monographs on Statistics and Applied Probability, London: Chapman & Hall. Thurstone, Louis Leon (1935). The Vectors of Mind - Multiple-Factor Analysis for the Isolation of Primary Traits. Chicago: University of Chicago Press. Van Kerm, Philippe (2003). "Adaptive kernel density estimation." Stata Journal 3(2):148-156. Velicer, W. F., Eaton, C. A., and Fava, J. L. (2000). "Construct explication through factor or component analysis: A review and evaluation of alternative procedures for determining the number of factors or components." In R. D. Goffin and E. Helmes, eds., Problems and solutions in human assessment. Boston: Kluwer. 33 Figure 1. Kernel Density Estimation (a) for the Sample of all Firms, 2001-2004. 0.6 0.5 0.4 Density 0.3 0.2 0.1 0 -1 0 1 2 3 4 5 6 Log TFP All Firms Figure 2. Kernel Density Estimation by (a) Firm Size, (b) Sub-national Location, and (c) Average Industry Factor Intensity, 2001-2004. (a) (b) 0.6 0.6 0.5 0.5 0.4 0.4 Density Density 0.3 0.3 0.2 0.2 0.1 0.1 0 0 -1 0 1 2 3 4 5 6 -1 0 1 2 3 4 5 6 Log TFP Log TFP small: 1-249 employees large: 250 or more employees capital large (pop>=250,000) small (pop<250,000) (c) 0.6 0.50.4 Density 0.3 0.2 0.1 0 -1 0 1 2 3 4 5 6 Log TFP Capital-Intensive Labor-Intensive 34 Figure 3. Kernel Density Estimation by Country, 2001-2004. 0.8 0.6 Density 0.4 0.2 -2 0 2 4 6 8 Log TFP Bulgaria Croatia Czech Republic Estonia Poland Romania Serbia Ukraine 35 Figure 4. Infrastructure Quality Indicator: Guttman-Kaiser Criterion, Cattell's Scree Test, and Humphrey- Ilgen Parallel Analysis. Scree plot of eigenvalues after pca Parallel Analysis 1.2 1.2 1.1 1.1 Eigenvalues Eigenvalues 1 1 .9 .9 1 1.5 2 2.5 3 Component 1 1.5 2 2.5 3 PCA Parallel Analysis Number Figure 5. Financial Development Indicator: Guttman-Kaiser Criterion, Cattell's Scree Test, and Humphrey- Ilgen Parallel Analysis. Scree plot of eigenvalues after pca Parallel Analysis 1.04 1.04 1.02 1.02 Eigenvalues Eigenvalues 1 1 .98 .98 .96 1 1.5 2 2.5 3 .96 Component 1 1.5 2 2.5 3 PCA Parallel Analysis Number Figure 6. Governance Indicator: Guttman-Kaiser Criterion, Cattell's Scree Test, and Humphrey-Ilgen Parallel Analysis. Scree plot of eigenvalues after pca Parallel Analysis 1.2 1.2 1.1 1.1 Eigenvalues Eigenvalues 1 1 .9 .9 .8 1 1.5 2 2.5 3 .8 Component 1 1.5 2 2.5 3 PCA Parallel Analysis Number 36 Figure 7. Labor Market Flexibility Indicator Guttman-Kaiser Criterion, Cattell's Scree Test, and Humphrey- Ilgen Parallel Analysis. Scree plot of eigenvalues after pca Parallel Analysis 1.06 1.06 1.04 1.04 1.02 Eigenvalues 1.02 Eigenvalues 1 1 .98 .98 .96 1 1.5 2 2.5 3 .96 Component 1 1.5 2 2.5 3 PCA Parallel Analysis Number Figure 8. Labor Quality Indicator: Guttman-Kaiser Criterion, Cattell's Scree Test, and Humphrey-Ilgen Parallel Analysis. Scree plot of eigenvalues after pca Parallel Analysis 1.15 1.15 1.1 1.1 1.05 Eigenvalues 1.05 Eigenvalues 1 1 .95 .95 .9 .9 1 1.5 2 2.5 3 Component 1 1.5 2 2.5 3 PCA Parallel Analysis Number 37 Figure 9. Olley and Pakes Decomposition of Aggregate Productivity (TFP à la Levinsohn and Petrin) in Levels by Country, 2001-2004 2001 2002 Aggregate Productivity Average Productivity Efficiency Term Aggregate Productivity Average Productivity Efficiency Term 450 450 400 400 350 350 300 300 250 250 200 200 150 150 100 100 50 50 0 0 Bulgaria Croatia Czech Estonia Poland Romania Serbia Ukraine Bulgaria Croatia Czech Estonia Poland Romania Serbia Ukraine Republic Republic 2003 2004 Aggregate Productivity Average Productivity Efficiency Term Aggregate Productivity Average Productivity Efficiency Term 450 450 400 400 350 350 300 300 250 250 200 200 150 150 100 100 50 50 0 0 Bulgaria Croatia Czech Estonia Poland Romania Serbia Ukraine Bulgaria Croatia Czech Estonia Poland Romania Serbia Ukraine Republic Republic 38 Figure 10. Relative Percentage Contribution of Changes in Business Environment Indicators to Log TFP Growth over the Period 2001-2004 by Country. 100% Infrastructure 80% 60% Financial Development 40% 20% Governance 0% Labor Market -20% Flexibility -40% Labor Quality -60% -80% Competition All Bulgaria Croatia Czech Estonia Poland Romania Serbia Ukraine Countries Republic Figure 11. Relative Percentage Contribution of Changes in Business Environment Indicators to Log TFP Growth over the Period 2001-2004 by Size. 100% Infrastructure 80% Financial Development 60% Governance 40% Labor Market 20% Flexibility Labor Quality 0% -20% Competition Small (less than 50 employees) Large (50 or more employees) 39 Figure 12. Relative Percentage Contribution of Changes in Business Environment Indicators to Log TFP Growth over the Period 2001-2004 by Industry. 100% Infrastructure 80% Financial 60% Development Governance 40% 20% Labor Market Flexibility 0% Labor Quality -20% Competition -40% NACE15 NACE17 NACE18 NACE19 NACE20 NACE21 NACE22 NACE24 NACE25 NACE26 100% Infrastructure 80% Financial 60% Development Governance 40% 20% Labor Market Flexibility 0% Labor Quality -20% Competition -40% NACE27 NACE28 NACE29 NACE30 NACE31 NACE32 NACE33 NACE34 NACE35 NACE36 NACE Industry Descriptions (15) food products and beverages; (17) textiles; (18) garments; (19) tanning and dressing of leather; luggage, handbags, saddlery, harness, and footwear; (20) wood and products of wood and cork, except furniture; articles of straw and plaiting materials; (21) paper and paper products; (22) Publishing, printing and reproduction of recorded media; (24) chemicals and chemical products; (25) rubber and plastics products; (26) other non-metallic mineral products; (27) basic metals; (28) fabricated metal products, except machinery and equipment; (29)machinery and equipment not elsewhere classified t elsewhere classified (n.e.c.); (30) office, accounting, and computing machinery; (31) electrical machinery and apparatus n.e.c.; (32) radio, television, and communication equipment and apparatus; (33) medical, precision, and optical instruments, watches, and clocks; (34) motor vehicles, trailers, and semi-trailers; (35) other transport equipment; and (36) furniture; manufacturing n.e.c. 40 Table 1. Descriptive Statistics of Total Factor Productivity (TFP) Estimates [calculated according to the method described in Levinsohn and Petrin (2003)]. lnTFP Country Location Firm Size Obs lnTFP t lnTFP t-1 2001 2002 2003 2004 All Countries1 Total 22,004 0.062 0.030 1.770 1.782 1.800 1.844 (0.386) (0.384) (1.095) (1.103) (1.105) (1.157) Sofia large 102 0.100 0.128 1.931 1.968 2.059 2.068 small city small 32 0.325 0.256 1.921 1.979 2.177 2.305 Bulgaria small city large 77 0.183 0.184 1.672 1.729 1.856 1.912 large city large 10 0.077 0.172 1.152 1.183 1.325 1.260 Total 221 0.160 0.168 1.804 1.851 1.972 2.012 Zagreb small 433 0.062 0.121 2.100 2.152 2.221 2.214 Zagreb large 65 0.043 0.096 1.667 1.696 1.763 1.739 Croatia small city small 987 0.045 0.086 1.657 1.699 1.743 1.744 small city large 295 0.058 0.089 1.941 1.988 2.030 2.046 Total 1,780 0.051 0.095 1.812 1.857 1.908 1.908 small city small 134 0.031 0.030 2.215 2.226 2.245 2.257 small city large 769 0.070 0.079 2.187 2.234 2.266 2.304 Czech Republic large city small 11 0.216 0.097 2.535 2.609 2.632 2.825 large city large 50 0.036 0.059 2.222 2.301 2.281 2.337 Total 964 0.064 0.072 2.197 2.241 2.268 2.306 Tallinn small 369 0.113 0.117 1.637 1.694 1.754 1.806 Tallinn large 40 0.113 0.103 1.769 1.813 1.873 1.926 Estonia small city small 728 0.074 0.101 1.413 1.460 1.514 1.534 small city large 116 0.067 0.119 1.469 1.531 1.588 1.598 Total 1,253 0.086 0.108 1.495 1.547 1.603 1.633 Warsaw small 14 0.108 -0.010 3.625 3.604 3.615 3.712 Warsaw large 48 0.159 0.092 4.084 4.120 4.176 4.279 small city small 117 0.088 0.037 2.515 2.554 2.552 2.642 Poland small city large 754 0.129 -0.031 2.655 2.577 2.624 2.706 large city small 46 0.108 -0.014 3.028 3.004 3.014 3.112 large city large 154 0.166 0.019 2.954 2.918 2.973 3.084 Total 1,133 0.130 -0.011 2.769 2.716 2.758 2.846 Romania Bucharest small 1,157 0.003 -0.025 1.691 1.681 1.666 1.684 Bucharest large 273 0.055 -0.033 1.956 1.917 1.923 1.972 small city small 7,067 0.015 0.005 1.262 1.265 1.267 1.280 small city large 2,173 0.030 0.005 1.658 1.650 1.662 1.680 large city small 1,505 0.022 -0.024 1.541 1.517 1.517 1.539 large city large 401 0.036 0.010 1.852 1.839 1.862 1.876 Total 12,576 0.019 -0.002 1.437 1.433 1.435 1.451 Belgrade small 430 0.387 0.139 2.971 3.104 3.110 3.490 Belgrade large 103 0.214 0.153 3.978 4.128 4.131 4.342 Serbia small city small 1,120 0.249 0.108 3.047 3.151 3.155 3.400 small city large 584 0.039 -0.033 3.859 3.871 3.826 3.909 Total 2,237 0.219 0.079 3.287 3.375 3.367 3.594 Ukraine Kiev small 10 -0.093 0.194 2.129 2.244 2.323 2.151 Kiev large 98 0.151 0.095 1.591 1.607 1.687 1.758 small city small 124 0.066 -0.119 1.111 1.024 0.991 1.090 small city large 1,022 0.122 0.071 1.486 1.490 1.558 1.612 large city small 60 0.064 0.093 1.501 1.415 1.594 1.479 large city large 526 0.091 0.056 1.623 1.610 1.679 1.700 Total 1,840 0.108 0.057 1.510 1.501 1.567 1.608 Source : Authors' calculations based on AMADEUS database (May 2006 edition). 1 Notes : Standard deviation in parentheses. Appendix Table A1 presents descriptive statistics of labor, materials, and capital in levels for the years 2001 through 2004. A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 41 Table 2. Descriptive Statistics of the Competition Indicator. Number of Industries 2001 Number of Industries 2003 Change All Countries 1,935 0.190 1,970 0.185 -0.005 (0.234) (0.231) Bulgaria 280 0.215 254 0.106 -0.108 (0.233) (0.165) Croatia 230 0.135 233 0.131 -0.005 (0.198) (0.195) Czech Republic 265 0.163 284 0.220 0.057 (0.206) (0.230) Estonia 204 0.132 214 0.129 -0.003 (0.196) (0.193) Poland 276 0.217 288 0.226 0.009 (0.246) (0.239) Romania 247 0.313 248 0.340 0.027 (0.267) (0.268) Serbia 208 0.167 220 0.175 0.008 (0.238) (0.242) Ukraine 225 0.152 229 0.131 -0.021 (0.218) (0.201) Note : The competition indicator is equal to 1 minus the four-firm concentration ratio for industries defined at the 4-digit NACE level (1500-3663) and calculated using the full AMADEUS sample. 42 Table 3. Correlation between Infrastructure Quality Indicator and Underlying BEEPS Variables Overall 2001 2003 Corr. Mean S.D. Corr. Mean S.D. Corr. Mean S.D. Power outages 0.720 3.7 11.5 0.739 6.2 15.3 0.686 2.7 9.4 Insufficient water supply 0.362 4.6 32.9 0.466 11.9 56.3 0.184 1.7 14.9 Unavailable mainline telephone service 0.701 1.5 10.1 0.665 3.2 17.7 0.746 0.8 3.9 Table 4. Correlation between Financial Development Indicator and Underlying BEEPS Variables Overall 2001 2003 Corr. Mean S.D. Corr. Mean S.D. Corr. Mean S.D. Local private commercial banks 0.574 10.5 25.1 0.462 6.3 19.3 0.635 12.1 26.7 Foreign banks 0.468 2.8 14.2 0.617 4.8 18.4 0.344 2.0 12.1 Informal (family/friends/money lenders) 0.701 3.6 15.4 0.654 4.1 17.2 0.730 3.5 14.7 Table 5. Correlation between Governance Indicator and Underlying BEEPS Variables Overall 2001 2003 Corr. Mean S.D. Corr. Mean S.D. Corr. Mean S.D. Bribe level (officials) 0.735 0.9 2.3 0.825 1.5 3.1 0.656 0.7 1.9 Tax compliance 0.669 90.4 17.4 0.630 89.4 19.7 0.697 90.8 16.5 Confidence in legal system 0.461 3.5 1.4 0.355 3.5 1.4 0.532 3.5 1.4 Table 6. Correlation between Labor Market Flexibility Indicator and Underlying BEEPS Variables Overall 2001 2003 Corr. Mean S.D. Corr. Mean S.D. Corr. Mean S.D. Underemployment and overemployment 0.680 30.2 45.9 0.616 27.0 44.4 0.718 31.4 46.4 Temporary workers to permanent 0.509 0.3 41.5 0.690 2.1 58.7 0.356 -0.5 32.4 workers Labor regulations as a constraint 0.576 2.8 1.1 0.503 2.9 1.0 0.613 2.7 1.1 Table 7. Correlation between Labor Quality Indicator and Underlying BEEPS Variables Overall 2001 2003 Corr. Mean S.D. Corr. Mean S.D. Corr. Mean S.D. Skilled workers to total employees 0.640 83.0 21.5 0.568 82.5 20.4 0.674 83.3 21.9 Time to fill vacancy for skilled worker 0.689 4.0 5.4 0.765 5.1 7.6 0.646 3.6 4.2 Labor quality as a constraint 0.517 2.8 1.1 0.482 2.9 1.0 0.536 2.8 1.1 43 Table 8. Descriptive Statistics of Business Environment PCA Indicators, 2001 and 2003. Infrastructure Quality Financial Development Governance Labor Market Flexibility Labor Quality Country Location Firm Size Obs 2001 Obs 2003 Change Obs 2001 Obs 2003 Change Obs 2001 Obs 2003 Change Obs 2001 Obs 2003 Change Obs 2001 Obs 2003 Change All Countries Total 376 -0.273 1,082 0.106 0.379 370 0.027 927 -0.012 -0.039 340 -0.202 1,029 0.082 0.284 356 0.109 914 -0.043 -0.152 348 -0.110 889 0.034 0.144 (1.532) (0.823) (1.177) (0.951) (1.392) (0.951) (1.198) (0.947) (1.175) (1.029) Sofia large 5 0.042 7 0.135 0.093 8 0.284 4 -0.018 -0.302 8 -0.618 6 -1.189 -0.570 7 -0.306 4 -0.782 -0.476 6 -0.088 6 -0.288 -0.200 small city small 7 -0.306 14 -0.026 0.281 8 0.161 12 0.439 0.278 10 -0.886 15 -0.330 0.556 9 0.560 16 -0.120 -0.680 7 0.553 12 0.322 -0.231 Bulgaria small city large 15 -0.114 16 0.047 0.161 16 0.464 12 0.934 0.470 12 0.018 15 0.016 -0.003 15 0.404 15 0.491 0.087 15 0.239 16 -0.191 -0.430 large city large 2 -0.323 6 -0.336 -0.014 2 -0.018 4 -0.055 -0.036 3 0.684 5 0.432 -0.252 2 1.019 4 0.492 -0.527 3 0.162 6 -0.196 -0.359 Total 29 -0.148 43 -0.016 0.132 34 0.322 32 0.506 0.184 33 -0.350 41 -0.236 0.113 33 0.333 39 0.110 -0.223 31 0.239 40 -0.053 -0.291 Zagreb small 7 -0.208 11 0.303 0.511 4 -0.018 6 0.470 0.488 7 0.225 13 0.633 0.408 5 1.727 14 -0.030 -1.758 6 -0.153 6 0.207 0.360 Zagreb large 4 0.264 12 0.209 -0.055 3 -0.018 7 0.301 0.319 4 0.331 10 0.303 -0.028 2 0.139 11 0.141 0.002 2 -0.823 9 -0.139 0.684 Croatia small city small 11 -0.406 18 0.323 0.729 9 0.252 17 0.409 0.157 5 -1.690 16 0.288 1.978 5 -0.719 15 0.479 1.197 6 0.452 13 -0.300 -0.752 small city large 9 0.033 16 0.290 0.257 8 1.491 11 1.668 0.176 6 0.423 16 0.445 0.021 9 0.241 14 0.388 0.147 11 -0.940 16 -0.523 0.416 Total 31 -0.148 57 0.286 0.433 24 0.586 41 0.737 0.151 22 -0.137 55 0.418 0.555 21 0.357 54 0.254 -0.102 25 -0.408 44 -0.279 0.128 small city small 23 -0.050 37 0.170 0.220 23 -0.329 30 0.144 0.473 12 0.284 34 -0.394 -0.679 18 0.469 28 -0.191 -0.659 14 0.374 15 0.289 -0.085 small city large 14 -0.046 22 0.308 0.355 12 0.130 20 0.261 0.131 11 0.424 14 0.039 -0.385 10 0.643 16 -0.018 -0.661 11 -0.674 15 -0.398 0.276 Czech Republic large city small 3 0.007 5 -0.824 -0.831 4 -1.111 6 -0.643 0.468 2 0.842 6 0.039 -0.802 3 0.601 6 0.155 -0.446 3 -0.894 5 0.656 1.550 large city large 3 0.336 3 0.336 0.000 4 -0.018 2 -0.018 0.000 2 0.166 3 0.367 0.200 3 0.274 3 0.508 0.234 4 0.088 3 -0.931 -1.020 Total 43 -0.018 67 0.149 0.167 43 -0.245 58 0.097 0.342 27 0.374 57 -0.202 -0.576 34 0.515 53 -0.060 -0.574 32 -0.141 38 -0.030 0.111 Tallinn small 10 -0.107 13 0.161 0.268 6 -0.018 7 0.168 0.186 4 0.442 10 0.317 -0.125 7 0.972 12 0.043 -0.929 10 -0.187 5 0.511 0.698 Tallinn large 4 -0.470 7 -0.174 0.296 4 0.075 5 0.800 0.725 2 -0.022 5 0.520 0.542 4 0.471 7 0.103 -0.368 4 -0.672 6 -0.920 -0.248 Estonia small city small 6 -0.286 10 0.131 0.418 5 -0.018 8 0.051 0.070 5 0.541 5 0.569 0.029 3 0.843 8 0.269 -0.574 4 -1.081 7 -0.818 0.263 small city large 5 -0.056 8 -0.059 -0.003 6 0.695 6 1.141 0.447 4 0.654 2 0.212 -0.441 3 0.961 6 -0.296 -1.257 6 -1.211 7 0.017 1.228 Total 25 -0.198 38 0.045 0.243 21 0.203 26 0.478 0.275 15 0.469 22 0.411 -0.059 17 0.830 33 0.049 -0.781 24 -0.673 25 -0.343 0.330 Warsaw small 6 0.274 39 0.296 0.022 9 -0.018 36 -0.238 -0.220 3 -0.709 36 0.179 0.888 5 -0.439 32 -0.430 0.009 2 -0.872 31 0.521 1.393 Warsaw large 6 0.336 8 0.321 -0.016 4 -0.018 7 0.061 0.080 2 0.016 9 0.205 0.188 7 -0.045 8 -0.095 -0.050 5 0.416 10 0.022 -0.394 small city small 19 0.145 129 0.244 0.100 16 0.571 113 -0.166 -0.737 20 -0.103 130 0.066 0.170 18 -0.111 115 -0.188 -0.077 13 0.139 75 0.275 0.136 Poland small city large 23 0.207 44 0.243 0.036 20 0.163 36 -0.002 -0.165 18 0.511 46 0.418 -0.093 24 -0.315 34 0.051 0.366 22 -0.092 36 -0.070 0.022 large city small 10 0.229 153 0.293 0.064 14 -0.613 135 -0.248 0.365 9 0.667 155 -0.010 -0.677 11 -0.778 116 -0.182 0.596 9 0.167 113 0.161 -0.006 large city large 11 0.303 29 0.306 0.004 8 -0.018 22 -0.018 0.000 5 0.606 29 0.565 -0.041 10 -0.695 26 -0.035 0.660 7 -0.496 29 -0.073 0.423 Total 75 0.224 402 0.274 0.050 71 0.048 349 -0.174 -0.223 57 0.247 405 0.126 -0.121 75 -0.368 331 -0.170 0.197 58 -0.032 294 0.172 0.204 Romania Bucharest small 5 -0.011 20 0.103 0.114 7 -0.379 22 -0.247 0.132 4 0.024 19 -0.052 -0.075 4 -0.164 17 0.277 0.441 6 1.132 23 0.078 -1.054 Bucharest large 4 0.181 16 0.195 0.014 3 -0.018 18 0.083 0.101 3 -0.239 16 0.405 0.644 2 1.019 15 0.187 -0.832 4 0.042 20 -0.046 -0.088 small city small 15 -0.336 74 -0.115 0.221 26 -0.031 67 -0.037 -0.006 23 -0.826 72 0.176 1.002 22 -0.258 65 -0.015 0.242 20 -0.095 61 0.208 0.302 small city large 12 -0.153 45 -0.192 -0.038 16 -0.317 38 0.162 0.479 14 -0.097 43 0.413 0.511 12 0.289 36 0.136 -0.154 10 -0.602 39 0.158 0.760 large city small 8 0.336 111 0.072 -0.265 7 -0.050 91 -0.096 -0.046 8 -1.067 97 0.168 1.235 5 -0.276 83 -0.123 0.153 5 0.552 93 0.347 -0.205 large city large 4 0.243 44 0.128 -0.116 5 0.056 34 0.109 0.053 7 -0.003 44 0.346 0.349 5 0.140 39 -0.062 -0.201 6 0.501 43 0.120 -0.381 Total 48 -0.053 310 0.005 0.058 64 -0.135 270 -0.020 0.116 59 -0.500 291 0.232 0.732 50 -0.030 255 -0.005 0.025 51 0.094 279 0.205 0.110 Belgrade small 8 -2.148 12 -0.765 1.383 12 -0.018 11 -0.018 0.000 3 -0.222 10 -0.315 -0.093 10 0.409 10 0.418 0.009 2 -0.809 5 0.183 0.991 Belgrade large 10 0.187 13 0.240 0.053 10 -0.018 10 -0.018 0.000 6 -0.591 12 -0.132 0.460 5 0.618 10 -0.001 -0.619 7 0.495 9 -0.600 -1.095 Serbia small city small 7 -1.059 8 -0.083 0.976 10 -0.040 3 -0.018 0.022 5 -1.783 9 0.177 1.959 7 -1.045 5 -0.632 0.413 6 0.173 6 -0.238 -0.411 small city large 10 -1.386 20 -1.165 0.221 8 1.346 11 0.311 -1.035 8 0.344 13 -0.144 -0.488 11 0.016 18 -0.485 -0.500 5 -0.271 13 -0.494 -0.223 Total 35 -1.045 53 -0.567 0.479 40 0.249 35 0.085 -0.164 22 -0.472 44 -0.114 0.358 33 0.001 43 -0.179 -0.180 20 0.077 33 -0.374 -0.450 Ukraine Kiev small 7 -1.773 8 0.122 1.895 3 -0.018 10 -0.293 -0.275 6 -0.841 10 -0.608 0.233 4 0.923 10 -0.159 -1.081 6 0.098 11 -0.541 -0.639 Kiev large 8 0.088 10 0.187 0.099 4 -0.018 12 -0.365 -0.347 6 -0.491 7 -0.215 0.277 4 0.820 10 0.491 -0.329 8 -0.702 14 -0.247 0.455 small city small 20 -1.644 30 0.118 1.762 16 -0.448 28 -0.020 0.428 23 -0.378 33 -0.185 0.194 25 0.028 28 -0.204 -0.233 21 -0.324 28 -0.112 0.212 small city large 10 -0.916 12 0.112 1.028 9 -0.018 15 0.131 0.149 13 0.069 14 -0.039 -0.108 13 0.399 11 0.524 0.125 14 -0.423 15 -0.235 0.188 large city small 23 -0.355 32 -0.164 0.192 25 -0.235 32 -0.069 0.166 32 -0.527 34 -0.515 0.012 24 0.014 27 0.072 0.058 33 0.077 41 -0.192 -0.269 large city large 22 -0.181 20 0.146 0.326 16 -0.018 19 -0.238 -0.220 25 -0.514 16 -0.533 -0.020 23 0.195 20 0.197 0.002 25 -0.167 27 -0.617 -0.450 Total 90 -0.732 112 0.048 0.780 73 -0.187 116 -0.109 0.078 105 -0.433 114 -0.353 0.080 93 0.190 106 0.087 -0.103 107 -0.181 136 -0.299 -0.118 Source : Authors' calculations based on BEEPS 2002 and 2005 databases. 1 Note : Standard deviation in parentheses. Appendix Tables A2 through A6 present descriptive statistics of underlying BEEPS variables in levels for years 2001 and 2003. A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 44 Table 9. Firm-Level Productivity Growth and Changes in the Business Environment, 2001-2004. [dependent variable: ln(TFP) t (Levinsohn-Petrin)] (1) (2) (3) (4) (5) (6) (7) nfrastructure Quality t-1 0.112*** - - - - - 0.061*** (0.018) (0.020) Financial Development t-1 - 0.100*** - - - - 0.064*** (0.011) (0.012) Governance t-1 - - 0.037*** - - - 0.023*** (0.006) (0.0080) Labor Market Flexibility t-1 - - - 0.019*** - - 0.028*** (0.005) (0.010) Labor Quality t-1 - - - - 0.033*** - 0.048*** (0.007) (0.010) Competition t-1 - - - - - 0.117** 0.127** (0.057) (0.057) ln(TFP) t-1 0.253*** 0.252*** 0.253*** 0.255*** 0.254*** 0.255*** 0.249*** (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012) Firm characteristics Yes Yes Yes Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Yes Yes Yes Location dummies Yes Yes Yes Yes Yes Yes Yes Country dummies Yes Yes Yes Yes Yes Yes Yes Observations 22,004 22,004 22,004 22,004 22,004 22,004 22,004 2 Adjusted R 0.187 0.187 0.185 0.184 0.185 0.184 0.191 ** *** Note : significant at 5%; significant at 1%. Robust standard errors are denoted in parentheses. Firm characteristics include logarithmic changes in the number of employees, the value of tangible fixed assets (thousands of 2001 U.S. dollars), and cost of materials (thousands of 2001 U.S. dollars). Industry dummies are defined at the 4-digit NACE level (1500 to 3663). Location dummies include capital city dummy variable ((equal to 1 if the firm is located in a capital city--that is, Belgrade, Bucharest, Kyiv, Prague, Sofia, Tallinn, Warsaw, or Zagreb--and 0 otherwise) and large city dummy variable (equal to 1 if the firm is located in a city with a population of 250,000 and over, and 0 otherwise). Countries (8 total) include: Bulgaria, Croatia, Czech Republic, Estonia, Poland, Romania, Serbia, and Ukraine. 45 Table A1. Descriptive Statistics of Labor, Materials, and Capital, 2001-2004. Labor (number of employees) Materials (material costs 2001 US$000) Capital (tangible fixed assets 2001 US$000) Country Location Firm Size Obs 2001 2002 2003 2004 2001 2002 2003 2004 2001 2002 2003 2004 All Countries Total 22,004 98 97 95 94 1,063 1,285 1,588 2,136 699 853 1,032 1,278 (258) (241) (231) (227) (5,562) (6,937) (8,896) (12,962) (3,401) (4,165) (5,045) (6,072) Sofia large 102 171 173 171 187 738 983 1,383 1,747 865 1,117 1,578 1,819 small city small 32 41 38 30 24 121 149 171 217 195 223 272 272 Bulgaria small city large 77 245 256 256 268 1,270 1,766 2,493 3,129 1,183 1,693 2,424 2,854 large city large 10 203 205 201 206 1,011 1,158 1,734 2,083 745 996 1,696 2,429 Total 221 179 184 182 192 846 1,143 1,610 2,022 873 1,183 1,689 1,983 Zagreb small 433 10 11 11 12 388 496 689 750 97 134 178 209 Zagreb large 65 224 232 236 235 4,905 6,306 8,462 9,936 4,311 5,536 6,506 7,189 Croatia small city small 987 12 13 14 13 395 499 652 747 187 237 312 360 small city large 295 173 183 185 187 3,686 4,579 5,658 6,573 2,655 3,273 3,986 4,597 Total 1,780 46 49 50 50 1,103 1,387 1,776 2,048 725 909 1,114 1,275 small city small 134 31 30 30 29 844 981 1,239 1,573 394 499 642 815 small city large 769 279 281 285 304 7,018 8,887 11,288 15,021 4,264 5,318 6,454 7,527 Czech Republic large city small 11 27 27 27 27 701 649 818 963 381 507 576 647 large city large 50 298 288 285 267 6,582 7,383 7,792 10,655 5,450 6,798 7,824 9,080 Total 964 242 243 246 261 6,065 7,616 9,590 12,765 3,743 4,670 5,650 6,596 Tallinn small 369 12 13 13 13 149 189 233 291 40 55 83 97 Tallinn large 40 94 97 100 96 1,720 2,177 2,936 3,321 629 829 910 1,003 Estonia small city small 728 15 15 16 15 169 223 286 348 64 91 125 159 small city large 116 105 108 113 118 1,557 1,955 2,723 3,490 672 863 1,131 1,404 Total 1,253 25 26 26 27 341 435 581 717 131 176 231 283 Warsaw small 14 30 27 26 26 1,773 1,946 2,055 3,410 138 134 151 329 Warsaw large 48 293 281 278 273 12,948 13,726 15,551 24,139 8,057 9,943 10,021 12,554 small city small 117 28 27 27 27 1,164 1,498 1,669 2,369 418 473 493 615 Poland small city large 754 243 247 255 264 7,162 8,144 9,726 15,260 3,253 3,720 4,100 5,681 large city small 46 26 25 25 26 949 1,065 1,243 1,690 262 329 394 531 large city large 154 198 216 217 219 4,132 5,180 6,463 9,917 2,550 2,899 2,987 4,225 Total 1,133 205 210 215 221 6,057 6,927 8,258 12,881 2,908 3,355 3,628 4,976 Romania Bucharest small 1,157 14 14 14 14 111 134 172 226 34 44 61 86 Bucharest large 273 227 226 224 216 1,351 1,580 2,200 3,103 1,050 1,237 1,655 2,168 small city small 7,067 17 16 16 15 100 122 160 210 36 46 63 91 small city large 2,173 225 229 230 229 1,050 1,299 1,700 2,349 648 841 1,191 1,602 large city small 1,505 16 16 16 15 102 120 162 213 34 43 59 84 large city large 401 233 226 219 214 1,029 1,163 1,577 2,240 850 1,085 1,415 1,854 Total 12,576 64 64 64 63 322 391 517 709 189 242 335 452 Belgrade small 430 11 11 12 11 171 272 336 214 58 89 116 141 Belgrade large 103 326 309 296 291 3,601 5,260 5,941 5,803 3,416 4,434 5,973 7,200 Serbia small city small 1,120 12 13 13 12 198 280 336 228 101 138 172 209 small city large 584 443 409 370 338 3,285 4,249 4,543 5,000 4,013 5,228 6,441 7,818 Total 2,237 139 130 119 110 1,155 1,544 1,692 1,728 1,267 1,655 2,065 2,504 Ukraine Kiev small 10 67 59 37 29 131 161 200 262 57 55 75 86 Kiev large 98 230 226 220 221 1,137 1,006 1,296 1,929 1,002 1,060 1,167 1,346 small city small 124 139 94 56 29 249 172 143 133 514 480 427 382 small city large 1,022 238 231 220 221 873 856 1,103 1,408 863 885 924 986 large city small 60 137 92 50 28 162 141 113 108 431 356 262 259 large city large 526 249 235 225 226 815 823 987 1,268 1,016 1,010 1,046 1,086 Total 1,840 230 218 204 202 801 781 978 1,261 872 881 912 965 Source : Authors' calculations based on AMADEUS database (May 2006 edition). Note : 1Standard deviation in parentheses. A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 46 Table A2. Descriptive Statistics of Underlying BEEPS Variables for Infrastructure Quality PCA Indicator, 2001 and 2003. Power outages (days) Insufficient water supply (days) Unavailable mainline telephone service (days) Country Location Firm Size Obs 2001 Obs 2003 Obs 2001 Obs 2003 Obs 2001 Obs 2003 Sofia large 9 2.4 8 2.4 9 4.3 8 1.3 9 4.8 8 0.6 small city small 13 20.1 18 2.7 12 10.8 17 11.8 13 7.3 17 0.9 Bulgaria small city large 18 3.3 19 1.6 18 1.3 19 4.9 17 0.4 19 3.7 large city large 3 3.3 6 3.0 3 13.3 6 5.8 3 1.7 6 0.3 Total 43 8.2 51 2.3 42 5.5 50 6.8 42 3.6 50 1.9 Zagreb small 8 1.9 14 1.9 7 0.0 14 0.2 7 1.1 14 0.3 Zagreb large 5 1.6 13 1.0 5 3.0 13 0.0 5 0.2 13 0.3 Croatia small city small 13 3.0 24 0.6 13 2.7 22 0.0 13 1.6 24 1.6 small city large 11 2.5 17 0.5 11 0.5 17 0.1 11 0.9 17 0.1 Total 37 2.4 68 0.9 36 1.5 66 0.1 36 1.1 68 0.7 small city small 28 3.7 40 1.3 27 0.4 39 0.3 27 0.5 39 0.2 small city large 16 3.1 22 0.2 16 1.6 22 0.0 16 0.7 22 0.0 Czech Republic large city small 4 1.3 6 0.0 4 0.5 6 0.0 4 1.5 6 5.0 large city large 4 2.5 3 0.0 4 0.5 3 0.0 4 0.5 3 0.0 Total 52 3.2 71 0.8 51 0.8 70 0.2 51 0.6 70 0.5 Tallinn small 11 2.7 14 0.7 11 0.5 14 0.7 11 1.5 14 0.3 Tallinn large 5 3.2 7 1.6 5 1.4 7 0.7 5 2.4 7 1.0 Estonia small city small 7 2.0 11 2.4 6 0.8 11 0.3 6 1.3 11 0.2 small city large 7 3.3 8 2.8 7 1.3 8 0.5 7 1.7 8 0.1 Total 30 2.8 40 1.7 29 0.9 40 0.6 29 1.7 40 0.4 Warsaw small 10 0.5 46 0.4 10 0.4 46 0.1 10 1.9 46 0.9 Warsaw large 7 0.4 12 2.3 7 0.4 12 0.4 7 0.1 12 0.4 small city small 28 4.0 173 1.5 28 0.1 173 0.3 28 1.5 173 0.6 Poland small city large 35 3.6 55 0.9 35 0.5 55 0.1 35 1.1 55 0.6 large city small 18 4.4 205 1.1 18 0.9 205 0.2 18 0.6 205 1.0 large city large 15 3.5 36 0.8 15 0.8 36 0.1 15 1.0 36 0.3 Total 113 3.3 527 1.1 113 0.5 527 0.2 113 1.1 527 0.7 Romania Bucharest small 9 3.8 30 10.9 9 1.8 30 2.7 9 1.2 30 1.1 Bucharest large 4 1.3 22 2.9 4 0.0 22 1.5 4 0.0 22 2.2 small city small 30 16.0 93 5.0 30 26.7 93 1.6 30 1.8 93 0.9 small city large 20 9.9 53 4.6 20 23.6 53 1.2 20 2.7 52 0.0 large city small 9 0.0 131 4.3 9 1.1 131 1.7 9 0.2 130 0.4 large city large 8 4.5 55 4.4 8 2.1 55 6.9 8 0.9 55 0.2 Total 80 9.4 384 4.9 80 16.5 384 2.4 80 1.6 382 0.6 Belgrade small 12 14.6 14 3.0 12 2.1 14 0.6 12 17.1 14 2.2 Belgrade large 10 1.2 14 0.9 10 0.0 14 0.9 10 0.0 14 0.0 Serbia small city small 13 10.8 9 2.2 13 0.9 9 2.2 13 10.2 9 0.3 small city large 12 13.7 22 5.3 12 0.0 22 0.0 12 33.5 22 4.4 Total 47 10.4 59 3.2 47 0.8 59 0.7 47 15.7 59 2.2 Ukraine Kiev small 8 6.4 13 9.5 8 0.4 13 2.5 8 4.0 13 1.9 Kiev large 9 2.0 13 1.2 9 0.6 12 0.0 9 0.3 12 0.3 small city small 31 12.8 42 3.5 30 42.8 37 1.7 31 3.5 36 0.6 small city large 16 10.3 15 1.5 16 30.3 13 0.4 16 3.8 13 0.4 large city small 44 6.4 44 5.5 43 54.1 43 14.4 43 4.4 43 1.1 large city large 31 4.0 26 2.5 31 16.7 26 2.7 31 1.7 27 0.6 Total 139 7.4 153 4.0 137 33.7 144 5.5 138 3.2 144 0.8 All Countries Total 541 6.2 1,353 2.7 535 11.9 1,340 1.7 536 3.2 1,340 0.8 Source : Authors' calculations based on BEEPS 2002 and 2005 databases. Note : A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 47 Table A3. Descriptive Statistics of Underlying BEEPS Variables for Financial Development PCA Indicator, 2001 and 2003. Sources of Finance (percentage of new fixed investment) Country Location Firm Size Local private commercial banks Foreign banks Informal (family/friends/ money lenders) Obs 2001 Obs 2003 Obs 2001 Obs 2003 Obs 2001 Obs 2003 Sofia large 8 8.1 8 15.0 8 0.0 8 7.5 8 0.0 8 1.9 small city small 9 11.1 12 14.2 9 22.2 12 0.0 9 22.2 12 1.7 Bulgaria small city large 17 6.5 15 14.7 17 7.1 15 9.3 17 0.0 15 0.0 large city large 2 0.0 5 36.0 2 0.0 5 0.0 2 0.0 5 15.0 Total 36 7.6 40 17.3 36 8.9 40 5.0 36 5.6 40 2.8 Zagreb small 7 0.0 7 5.7 7 25.7 7 4.3 7 0.0 7 1.4 Zagreb large 5 18.0 10 16.0 5 5.0 10 20.0 5 0.0 10 0.0 Croatia small city small 9 10.0 20 24.5 9 4.4 20 0.0 9 8.9 20 2.0 small city large 8 0.0 12 22.1 8 25.0 12 12.5 8 0.0 12 0.0 Total 29 6.2 49 19.5 29 15.3 49 7.8 29 2.8 49 1.0 small city small 25 7.6 35 14.3 25 0.4 35 0.9 25 10.8 35 3.7 small city large 14 7.8 22 12.3 14 0.0 22 0.0 14 0.7 22 0.0 Czech Republic large city small 4 0.0 6 0.0 4 0.0 6 0.0 4 26.3 6 15.0 large city large 4 0.0 3 20.0 4 0.0 3 6.7 4 0.0 3 0.0 Total 47 6.4 66 12.6 47 0.2 66 0.8 47 8.2 66 3.3 Tallinn small 6 0.0 8 16.9 6 0.0 8 0.0 6 0.0 8 0.0 Tallinn large 4 2.5 5 22.0 4 0.0 5 0.0 4 0.0 5 0.0 Estonia small city small 5 0.0 8 1.9 5 0.0 8 0.0 5 0.0 8 0.0 small city large 7 17.0 6 35.8 7 0.0 6 0.0 7 0.0 6 4.2 Total 22 5.9 27 17.6 22 0.0 27 0.0 22 0.0 27 0.9 Warsaw small 9 0.0 42 11.2 9 0.0 42 0.0 9 0.0 42 4.5 Warsaw large 5 0.0 10 8.0 5 10.0 10 5.0 5 0.0 10 0.0 small city small 21 12.9 143 9.7 21 9.5 143 1.6 21 0.0 143 3.7 Poland small city large 27 16.3 51 12.2 27 2.4 51 0.2 27 0.7 51 0.0 large city small 15 0.0 163 7.8 15 4.0 163 1.6 15 13.3 163 5.2 large city large 12 14.2 32 15.3 12 4.2 32 0.9 12 0.0 32 0.0 Total 89 9.9 441 9.8 89 4.8 441 1.3 89 2.5 441 3.6 Romania Bucharest small 8 12.5 24 7.3 8 12.5 24 1.3 8 18.8 24 8.3 Bucharest large 4 0.0 20 5.5 4 10.0 20 1.3 4 0.0 20 0.5 small city small 29 8.6 77 12.2 29 3.4 77 2.6 29 4.0 77 4.7 small city large 18 1.7 47 14.6 18 5.6 47 4.3 18 6.4 47 0.2 large city small 9 12.2 113 13.5 9 0.0 113 3.7 9 6.7 113 3.8 large city large 6 11.7 49 19.5 6 0.0 49 1.6 6 0.0 49 0.0 Total 74 7.6 330 13.3 74 4.6 330 2.9 74 5.9 330 3.0 Belgrade small 12 0.0 11 0.0 12 0.0 11 0.0 12 0.0 11 0.0 Belgrade large 10 0.0 10 0.0 10 0.0 10 0.0 10 0.0 10 0.0 Serbia small city small 10 5.0 7 18.6 10 0.0 7 0.0 10 5.0 7 0.0 small city large 10 13.0 12 7.9 10 18.0 12 1.7 10 0.0 12 0.0 Total 42 4.3 40 5.6 42 4.3 40 0.5 42 1.2 40 0.0 Ukraine Kiev small 3 0.0 11 13.6 3 0.0 11 4.5 3 0.0 11 18.2 Kiev large 4 0.0 13 3.8 4 0.0 13 0.0 4 0.0 13 7.7 small city small 17 0.0 34 14.7 17 5.9 34 0.0 17 9.7 34 4.7 small city large 10 1.0 16 4.0 10 0.0 16 0.0 10 0.0 16 0.0 large city small 27 1.9 38 15.4 27 4.4 38 1.3 27 6.5 38 6.6 large city large 20 5.0 26 17.5 20 4.5 26 0.0 20 1.0 26 8.8 Total 81 2.0 138 13.1 81 3.8 138 0.7 81 4.4 138 6.8 All Countries Total 420 6.3 1131 12.1 420 4.8 1131 2.0 420 4.1 1131 3.5 Source : Authors' calculations based on BEEPS 2002 and 2005 databases. Note : A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 48 Table A4. Descriptive Statistics of Underlying BEEPS Variables for Governance PCA Indicator, 2001 and 2003. Tax compliance (percentage of total annual sales Confidence in legal system (1="strongly disagree" to Bribe level (percentage of total annual sales ) Country Location Firm Size reported) 6="strongly agree") Obs 2001 Obs 2003 Obs 2001 Obs 2003 Obs 2001 Obs 2003 Sofia large 9 1.3 6 1.2 9 86.7 8 75.6 9 3.1 8 2.9 small city small 12 2.5 16 1.1 12 90.4 18 89.4 13 3.2 18 2.9 Bulgaria small city large 15 0.5 18 1.2 15 96.0 18 94.2 18 3.2 19 3.0 large city large 3 0.0 6 1.7 3 100.0 6 100.0 3 4.0 6 3.0 Total 39 1.3 46 1.2 39 92.4 50 90.2 43 3.2 51 3.0 Zagreb small 8 0.1 14 0.1 8 93.1 14 97.9 8 3.6 14 4.1 Zagreb large 5 0.0 13 0.2 4 95.0 11 88.6 5 3.0 13 4.2 Croatia small city small 13 2.7 25 0.6 10 77.5 22 93.2 12 3.1 24 3.9 small city large 12 0.1 18 0.1 8 93.8 17 93.2 12 3.9 18 4.1 Total 38 1.0 70 0.3 30 88.3 64 93.4 37 3.5 69 4.1 small city small 26 0.9 38 0.4 23 93.0 39 83.6 19 3.4 39 3.0 small city large 16 0.1 21 0.2 14 90.6 22 87.7 13 4.1 20 3.3 Czech Republic large city small 4 1.0 6 0.2 4 77.5 6 90.8 4 4.0 6 3.3 large city large 3 3.4 3 0.1 3 83.3 3 92.7 4 3.5 3 4.0 Total 49 0.8 68 0.3 44 90.2 70 85.9 40 3.7 68 3.1 Tallinn small 11 1.1 14 0.1 10 89.4 12 95.8 9 3.2 11 3.7 Tallinn large 3 0.3 7 0.0 4 92.5 5 95.8 5 3.6 7 4.0 Estonia small city small 7 0.0 9 0.1 5 96.8 8 96.8 7 3.9 9 4.0 small city large 5 0.0 3 0.0 4 97.5 6 100.0 6 4.2 6 3.8 Total 26 0.5 33 0.1 23 93.0 31 96.9 27 3.7 33 3.9 Warsaw small 7 1.1 46 0.7 9 92.2 46 89.3 9 3.7 41 3.5 Warsaw large 3 0.0 11 0.4 6 98.3 12 89.2 5 3.2 11 3.2 small city small 27 1.6 172 0.8 26 91.9 173 90.5 27 3.4 164 3.4 Poland small city large 33 0.7 55 0.4 30 95.7 55 94.9 33 4.1 53 3.8 large city small 16 0.7 203 0.5 17 85.9 204 88.6 16 3.8 191 3.1 large city large 13 1.7 36 0.3 12 94.6 36 96.8 14 3.7 36 3.8 Total 99 1.1 523 0.6 100 92.8 526 90.5 104 3.7 496 3.3 Romania Bucharest small 9 4.7 25 1.5 9 88.3 30 89.5 9 3.7 29 3.7 Bucharest large 4 1.8 16 0.6 3 98.3 22 98.2 4 4.0 22 4.1 small city small 30 2.7 83 0.6 29 88.6 87 93.9 28 3.8 85 3.2 small city large 20 2.0 51 0.4 16 94.4 51 94.1 20 3.9 51 3.8 large city small 9 3.4 116 0.8 9 94.4 126 91.9 9 2.9 129 3.6 large city large 8 1.8 48 0.5 8 90.0 55 93.7 8 3.6 54 3.9 Total 80 2.7 339 0.7 74 91.1 371 93.1 78 3.7 370 3.6 Belgrade small 8 2.5 10 0.4 10 80.0 14 90.0 10 3.2 14 3.1 Belgrade large 7 0.5 13 0.3 8 81.3 14 83.2 8 4.0 14 3.8 Serbia small city small 8 1.3 9 0.1 10 63.5 9 88.3 11 4.1 9 4.0 small city large 11 0.2 14 0.5 9 90.0 21 86.6 12 4.6 22 3.8 Total 34 1.0 46 0.3 37 78.2 58 86.9 41 4.0 59 3.7 Ukraine Kiev small 8 4.9 14 1.8 7 92.9 14 74.3 8 2.5 14 3.0 Kiev large 9 3.0 11 1.5 8 81.3 12 92.5 9 4.0 14 3.5 small city small 30 0.9 40 1.2 27 86.6 48 88.0 28 3.2 48 3.6 small city large 16 0.8 14 0.9 14 96.4 16 89.4 16 3.5 17 3.9 large city small 41 2.5 44 1.5 41 85.6 50 86.0 44 3.2 50 3.4 large city large 30 1.4 26 2.2 30 86.6 28 93.7 30 3.1 28 3.2 Total 134 1.9 149 1.5 127 87.4 168 87.7 135 3.2 171 3.5 All Countries Total 499 1.5 1,274 0.7 474 89.4 1,338 90.8 505 3.5 1,317 3.5 Source : Authors' calculations based on BEEPS 2002 and 2005 databases. Note : A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 49 Table A5. Descriptive Statistics of Underlying BEEPS Variables for Labor Market Flexibility PCA Indicator, 2001 and 2003. Underemployment/overemployment (percentage of Change in temporary workers/permanent workers over Labor regulations as a constraint (1="major obstacle" Country Location Firm Size firms) last 3 years to 4="no obstacle") Obs 2001 Obs 2003 Obs 2001 Obs 2003 Obs 2001 Obs 2003 Sofia large 9 33.3 8 50.0 9 -1.9 8 -1.5 9 2.4 8 2.3 small city small 13 30.8 18 33.3 13 15.6 18 -1.9 13 3.1 18 2.9 Bulgaria small city large 18 16.7 19 5.3 18 -0.5 17 -1.5 18 3.4 19 3.1 large city large 3 0.0 6 0.0 3 -4.3 6 0.5 3 4.0 6 3.3 Total 43 23.3 51 21.6 43 3.8 49 -1.4 43 3.1 51 2.9 Zagreb small 8 50.0 14 57.1 8 12.3 14 0.8 8 3.6 14 3.5 Zagreb large 4 25.0 13 30.8 3 5.1 13 -1.4 5 2.8 13 3.5 Croatia small city small 10 50.0 22 27.3 11 84.5 22 -1.2 13 3.6 25 3.8 small city large 11 36.4 17 29.4 10 2.2 18 -5.4 12 3.2 18 3.2 Total 33 42.4 66 34.8 32 33.3 67 -2.0 38 3.4 70 3.5 small city small 25 12.0 39 41.0 27 0.9 37 0.4 24 3.1 40 2.5 small city large 13 0.0 22 13.6 15 -0.8 22 2.6 16 2.9 22 2.0 Czech Republic large city small 4 0.0 6 16.7 4 5.3 6 0.0 4 3.0 6 2.8 large city large 4 25.0 3 0.0 4 2.5 3 0.2 3 3.0 3 3.0 Total 46 8.7 70 28.6 50 0.9 68 1.1 47 3.0 71 2.4 Tallinn small 10 30.0 14 14.3 10 6.2 14 -0.7 11 3.2 14 2.4 Tallinn large 4 0.0 7 0.0 5 -1.3 7 0.7 5 3.0 7 2.1 Estonia small city small 6 16.7 10 20.0 5 2.5 11 0.4 7 3.4 10 2.3 small city large 5 20.0 8 12.5 3 5.5 8 0.6 7 3.1 8 1.9 Total 25 20.0 39 12.8 23 3.7 40 0.1 30 3.2 39 2.2 Warsaw small 8 25.0 46 52.2 10 19.5 46 -11.5 10 2.5 44 2.8 Warsaw large 7 14.3 12 8.3 7 -0.3 10 -2.8 7 2.4 12 2.6 small city small 27 18.5 173 39.3 28 -0.2 173 -0.4 28 2.4 172 2.6 Poland small city large 31 19.4 55 14.5 35 1.2 54 0.6 35 2.3 55 2.6 large city small 16 25.0 205 35.6 18 22.0 204 3.4 18 2.3 202 2.5 large city large 15 40.0 36 11.1 14 -3.6 35 -0.3 15 2.0 36 2.2 Total 104 23.1 527 33.8 112 5.1 522 0.2 113 2.3 521 2.5 Romania Bucharest small 9 55.6 30 33.3 9 -2.0 30 1.0 9 3.1 30 2.9 Bucharest large 3 33.3 22 13.6 4 -8.2 22 -1.9 4 3.5 22 2.7 small city small 30 43.3 93 29.0 30 -2.0 90 2.4 29 2.9 90 2.9 small city large 19 31.6 53 24.5 19 -2.8 53 -1.2 19 3.6 52 2.7 large city small 9 22.2 131 38.2 9 -11.6 128 -1.4 9 3.0 128 2.5 large city large 8 25.0 55 21.8 8 -0.9 52 -3.1 8 3.0 54 2.5 Total 78 37.2 384 29.9 79 -3.5 375 -0.5 78 3.1 376 2.7 Belgrade small 12 16.7 13 38.5 12 11.3 14 -14.4 12 3.1 13 2.8 Belgrade large 8 12.5 14 28.6 9 -1.0 12 -0.2 9 2.7 14 3.0 Serbia small city small 13 61.5 8 37.5 10 -86.6 9 0.4 11 3.1 8 2.5 small city large 12 16.7 23 34.8 11 -1.3 20 -1.0 12 2.8 23 2.3 Total 45 28.9 58 34.5 42 -17.9 55 -4.0 44 2.9 58 2.6 Ukraine Kiev small 8 12.5 14 71.4 8 9.7 13 -0.5 7 3.4 13 3.6 Kiev large 9 11.1 15 33.3 9 2.1 14 0.8 9 3.6 15 3.5 small city small 31 32.3 48 41.7 31 -13.6 41 10.7 29 3.1 48 3.1 small city large 16 12.5 17 23.5 15 4.5 16 -31.6 16 3.1 17 3.2 large city small 44 50.0 52 28.8 44 12.7 46 -2.9 43 3.0 51 3.5 large city large 30 10.0 31 16.1 31 -3.0 27 0.5 31 3.1 31 2.9 Total 138 28.3 177 33.3 138 1.5 157 -1.2 135 3.1 175 3.2 All Countries Total 512 27.0 1,372 31.4 519 2.1 1,333 -0.5 528 2.9 1,361 2.7 Source : Authors' calculations based on BEEPS 2002 and 2005 databases. Note : A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 50 Table A6. Descriptive Statistics of Underlying BEEPS Variables for Labor Quality PCA Indicator, 2001 and 2003. Table A6. Descriptive Statistics of Underlying BEEPS Variables for Labor Quality PCA Indicator, 2001 and 2003. Labor quality\ as a constraint (1="major obstacle" to Skilled workers/total employees (percentage) Time to fill vacancy for skilled worker (weeks) Country Location Firm Size 4="no obstacle") Obs 2001 Obs 2003 Obs 2001 Obs 2003 Obs 2001 Obs 2003 Sofia large 9 83.3 8 81.0 6 4.2 7 5.8 9 3.0 8 3.0 small city small 13 76.3 18 83.2 8 2.0 12 2.1 13 3.1 18 2.8 Bulgaria small city large 18 73.6 19 71.6 15 2.7 16 2.6 18 3.4 19 2.7 large city large 3 62.7 6 75.0 3 2.3 6 3.4 3 4.0 6 2.8 Total 43 75.7 51 77.5 32 2.8 41 3.1 43 3.3 51 2.8 Zagreb small 8 81.9 14 78.5 6 4.9 6 4.4 8 3.0 14 3.5 Zagreb large 5 74.4 13 63.6 2 4.2 9 4.2 5 2.8 13 3.2 Croatia small city small 13 93.5 25 88.4 6 3.1 14 9.9 12 3.3 25 3.1 small city large 12 74.8 18 71.0 11 6.9 16 4.9 12 2.8 18 2.9 Total 38 82.6 70 77.3 25 5.3 45 6.3 37 3.0 70 3.2 small city small 28 86.1 26 81.5 15 5.1 15 3.6 26 3.1 38 2.7 small city large 14 74.9 17 74.8 11 5.3 15 4.5 16 2.4 21 2.6 Czech Republic large city small 4 80.0 6 75.8 3 9.3 5 1.8 4 3.3 6 3.5 large city large 4 81.5 3 57.7 4 5.8 3 3.4 4 3.0 3 2.3 Total 50 82.1 52 77.3 33 5.7 38 3.7 50 2.9 68 2.7 Tallinn small 11 90.3 14 81.2 10 5.7 5 2.8 11 2.5 14 3.4 Tallinn large 5 72.0 7 64.1 5 6.1 6 5.9 4 2.5 7 3.3 Estonia small city small 7 72.4 11 75.9 4 5.0 8 7.8 7 2.1 11 3.0 small city large 7 68.3 7 81.1 6 4.4 7 3.8 7 1.6 8 2.9 Total 30 77.9 39 76.6 25 5.4 26 5.3 29 2.2 40 3.2 Warsaw small 10 84.9 46 92.2 2 3.5 31 3.2 10 3.0 46 2.9 Warsaw large 7 86.4 12 78.6 5 6.1 10 2.8 7 3.3 12 2.9 small city small 27 86.7 173 90.2 14 2.4 75 3.2 28 2.5 172 2.8 Poland small city large 35 85.2 54 77.6 23 5.6 36 3.2 35 2.7 55 2.9 large city small 18 92.1 205 90.9 9 3.1 118 3.5 18 2.6 204 2.6 large city large 15 81.0 36 85.6 7 3.0 29 3.6 15 2.7 36 2.6 Total 112 86.2 526 88.8 60 4.1 299 3.4 113 2.7 525 2.7 Romania Bucharest small 9 86.0 30 81.6 7 3.7 24 3.1 9 3.6 30 2.6 Bucharest large 4 84.5 22 78.7 4 4.8 20 3.2 4 3.3 22 2.6 small city small 29 74.1 92 76.4 21 3.0 62 2.1 30 3.0 93 3.0 small city large 20 79.9 53 77.4 12 9.8 41 2.5 19 3.1 51 2.9 large city small 9 91.2 131 85.9 6 5.2 97 2.4 9 3.0 128 2.6 large city large 8 87.9 55 81.8 6 3.0 44 2.6 8 3.1 54 2.5 Total 79 80.8 383 81.1 56 4.9 288 2.5 79 3.1 378 2.7 Belgrade small 12 85.3 14 90.1 2 4.0 6 5.0 12 3.3 13 3.6 Belgrade large 10 80.9 13 73.6 7 2.5 9 5.3 10 3.6 14 3.2 Serbia small city small 13 82.5 9 63.9 6 2.3 6 4.3 11 2.7 9 3.0 small city large 12 59.2 22 75.3 5 2.4 13 4.7 11 2.9 23 2.8 Total 47 76.9 58 76.7 20 2.6 34 4.8 44 3.1 59 3.1 Ukraine Kiev small 8 88.0 14 71.3 6 6.6 12 3.6 8 3.4 13 2.2 Kiev large 9 71.6 15 82.1 8 5.3 15 8.6 9 2.8 15 3.1 small city small 31 88.5 48 81.5 21 5.6 28 3.9 31 2.8 48 2.8 small city large 16 75.0 17 82.8 15 9.1 15 4.4 16 2.8 17 2.5 large city small 42 90.4 53 83.6 34 4.8 42 5.9 44 2.7 53 2.7 large city large 30 85.7 31 76.6 26 8.3 27 5.3 31 2.5 31 2.5 Total 136 85.7 178 80.7 110 6.5 139 5.3 139 2.7 177 2.7 All Countries Total 535 82.5 1,357 83.3 361 5.1 910 3.6 534 2.9 1,368 2.8 Source : Authors' calculations based on BEEPS 2002 and 2005 databases. Note : A small city is defined as having a population less than 250,000, and a large city is defined as having a population equal to or greater than 250,000. A small-sized firm is defined as employing 2 to 49 full-time workers, and a large-sized firm is defined as employing 50 or more full-time workers. 51