WPS6572 Policy Research Working Paper 6572 Drivers of Firm-Level Productivity in Russia’s Manufacturing Sector Željko Bogetić Olasupo Olusi The World Bank Europe and Central Asia Region Poverty Reduction and Economic Management Department August 2013 Policy Research Working Paper 6572 Abstract This note presents the results of an empirical analysis of structure of firm ownership are important determinants firm-level productivity growth in Russia’s manufacturing of productivity, as evidenced by positive effects of scale sector during the period 2003–08 using a rich Amadeus economies (large firm effect), agglomeration (Moscow- database as well as the recent EBRD/World Bank city effect), private ownership, and a firm’s industry Business Enterprise and Performance surveys (BEEPs). dominance. Supplemental analysis of the quality of The results show that productivity grew steadily between infrastructure—water, electricity, transport, and the 2003 and 2008, with an annual growth rate averaging 4 internet—using BEEPS data show that infrastructure percent over the period, showing no signs of a slowdown quality gaps reduce firm productivity with water supply from the previous period after the 1998 crisis. gaps having the largest impact. Firm characteristics such as size, location, age, and the This paper is a product of the Poverty Reduction and Economic Management Department, Europe and Central Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// econ.worldbank.org. The authors may be contacted at zbogetic@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 Drivers of Firm-Level Productivity in Russia’s Manufacturing Sector Željko Bogetić and Olasupo Olusi 1 Keywords: Productivity, Russia JEL classification: O4, O5, P2, P5 Sector: Economic Policy (EPOL) 1 The authors are lead economist and economist, respectively, in the Poverty Reduction and Economic Management Department (ECSPE) of the Europe and Central Asia Region, World Bank. Corresponding author: Željko Bogetić (e-mail: zbogetic@worldbank.org). 1. Introduction Russia experienced a well-documented productivity surge over the period 1999- 2005, following the 1998 crisis. This contributed to a dynamic growth and poverty reduction. Estimated annual total factor productivity (TFP) growth of 5.8 percent was the driving force behind the observed average annual real GDP growth of 6.5 percent over this period (Alam et al. 2008). Part of the productivity surge is explained by better utilization of excess capacity, especially in the manufacturing sector, a key driver of Russia’s growth, but also within-firm factors and inter-sectoral allocation of labor. Productivity in manufacturing itself––an important engine of Russia’s growth––grew at a healthy rate of over 4 percent in this period. At the same time, infrastructure constraints began to limit firm activities, as evidenced by several firm surveys. Extending the Alam et al. study, the questions arise whether Russia was able to sustain such productivity growth in a more recent period and how firm-level characteristics and infrastructure gaps affected productivity growth. This note, with a limited scope, aims to analyze the impact of firm characteristics and infrastructure deficiencies in Russia for the more recent period 2003-08 using the Amadeus database; next, we use the 2004 and 2008 BEEPs data for Russia to explore the firm-level links between infrastructure and productivity. More specifically, we seek to answer the following questions: Did Russia’s manufacturing productivity growth slow in the latter part of the past decade, as conjectured by some analysts who attributed Russia’s pre-global crisis growth to only high oil prices? What are the firm-specific characteristics that correlate with manufacturing productivity growth? Who are the productive firms in Russia? Which factors contributed most to the observed performance? And, finally, what role infrastructure gaps played in productivity dynamics. 2. The Empirical Approach (i) Estimating TFP We estimate total factor productivity (TFP) for Russian manufacturing firms using the Generalized Method of Moments (GMM) model developed in Levinsohn and Petrin (2003) that accounts for endogeneity in productivity functions. This approach recognizes that profit-maximizing firms respond to positive productivity shocks by expanding output via mobilization of additional inputs. By contrast, firms experiencing negative productivity shocks reduce output by reducing the use of inputs. The GMM model is superior in this regard to the alternative OLS functions which yield biased estimates of productivity under these circumstances. Levinsohn and Petrin (2003) assume a Cobb-Douglas production function of the form: 2 𝑦𝑡 = 𝛽0 + 𝛽𝑙 𝑙𝑡 + 𝛽𝑘 𝑘𝑡 + 𝛽𝑚 𝑚𝑡 + 𝜔𝑡 + 𝜂𝑡 (1) where 𝑦𝑡 is the logarithm of the firm’s output, most often measured as gross revenue or value added (measured as gross revenue in this paper); 𝑙𝑡 and 𝑚𝑡 are the logarithm of the freely variable inputs labor and the intermediate input (materials); and 𝑘𝑡 is the logarithm of the state variable capital (fixed assets). The demand for the intermediate input 𝑚𝑡 is assumed to depend on the firm’s state variables 𝑘𝑡 and 𝜔𝑡 ; 𝑚𝑡 = 𝑚𝑡 (𝑘𝑡 , 𝜔𝑡 ) (2) The demand function is assumed to be monotonically increasing in 𝜔𝑡 , allowing the inversion of the intermediate demand function, so 𝜔𝑡 can be written as a function of 𝑘𝑡 and 𝑚𝑡 : 𝜔𝑡 = 𝜔𝑡 (𝑘𝑡 , 𝑚𝑡 ) (3) Substituting into (1) yields: 𝜔 � � � � 𝑡 = 𝑒𝑥��𝑦𝑡 − 𝛽𝑙 𝑙𝑡 − 𝛽𝑘 𝑘𝑡 − 𝛽𝑚 𝑚𝑡 � (4) which is solved using GMM. (ii) Growth and Productivity Decomposition Using standard growth accounting methodology, we assess the impact of each of the factor inputs (TFP, labor, capital) on the growth of firms’ output. Taking logarithms and differentiating a standard Cobb-Douglas, three-factor production function, we obtain the following growth accounting equation, expressed in value-added form (i.e. output less materials): 𝑑𝑌 𝑑𝐾 𝑑𝐿 𝑑𝐴 = 𝛼𝑖 + 𝛼𝑖𝑖 + , and, (5) 𝑌 𝐾 𝐿 𝐴 𝛼𝑖 + 𝛼𝑖𝑖 = 1 (6) This shows that firms’ output growth (first derivative of Y) can be decomposed into three sources of growth: (a) the contribution from growth in the physical capital stock (the derivative of K), (b) the contribution from growth in the labor force (the derivative of L), 3 and (c) the contribution from growth in TFP. Furthermore, 𝛼𝑖 , 𝑎𝑛𝑑 𝛼𝑖𝑖 are the respective coefficients of the factor inputs, which are assumed to be subject to constant returns to scale. We investigate the allocative efficiency of each industry within the manufacturing sector by decomposing total productivity along the lines of Olley and Pakes (1996): 𝑃𝑟𝑜𝑑𝑡 = � 𝜔𝑖𝑡 𝑃𝑟𝑜𝑑𝑖𝑡 = �������� 𝑃𝑟𝑜𝑑𝑡 + �(𝜔𝑡 − 𝜔 �������𝑡 � �𝑡 )�𝑃𝑟𝑜𝑑𝑖𝑡 − 𝑃𝑟𝑜𝑑 𝑖 𝑖 where 𝑃𝑟𝑜𝑑𝑡 is the total productivity of the industry, 𝑃𝑟𝑜𝑑𝑖𝑡 is firm-level productivity, 𝜔𝑖𝑡 is the share of activity for the firm, and a bar over a variable represents the unweighted industry average of the firm-level measure. The simple interpretation of this decomposition is that aggregate productivity is composed of two terms: un-weighted average firm-level productivity and a cross-term (i.e. the sample covariance between productivity and output share) that reflects the extent to which firms with higher than average productivity have a greater market share. The higher the covariance, the higher the share of output that goes to more productive firms, and the higher is industry productivity. In other words, when efficiency is positive, it means that most productive firms have a larger market share. To better understand the dynamics of productivity growth, we breakdown the sources of aggregate productivity growth across manufacturing industries into various components using the methodology in Foster et al (2001) as follows: ∆𝑃𝑡 = � 𝜃𝑖𝑡−𝑘 ∆�𝑖𝑡 + � ∆ 𝜃𝑖𝑡 (�𝑖𝑡−𝑘 − 𝑃𝑡−𝑘 ) + � ∆ 𝜃𝑖𝑡 ∆�𝑖𝑡 𝐶𝑜𝑛𝑡𝑖𝑛𝑢𝑒𝑟𝑠 𝐶𝑜𝑛𝑡𝑖𝑛𝑢𝑒𝑟𝑠 𝐶𝑜𝑛𝑡𝑖𝑛𝑢𝑒𝑟𝑠 + � ∆ 𝜃𝑖𝑡 (�𝑖𝑡 − 𝑃𝑡−𝑘 ) − � ∆ 𝜃𝑖𝑡−𝑘 (�𝑖𝑡−𝑘 − 𝑃𝑡−𝑘 ) 𝐸𝑛𝑡𝑟𝑖𝑒𝑠 𝐸𝑥𝑖𝑡𝑠 where Δ is changes over the k-years’ interval between the first year (t-k) and the last year (t); θit is the share of firm i in the given industry at time t (expressed in output or employment); pi is the productivity of firm i; and Pt is the aggregate (weighted average) productivity level of the firm. The components of the equation can be defined as follows: • The first term is the within effect, which is productivity growth within each firm, weighted by initial output shares. It also reflects the impact of productivity growth in individual sectors on aggregate productivity in the economy. 4 • The second term, the between effect, captures the gains in aggregate productivity arising from the expanding market of high-productivity firms or from shrinking shares of low-productivity firms weighted by the initial shares. Hence a positive between effect means that aggregate productivity rises because the sector displays higher than average productivity and factor inputs are moving into it. • The third term is the cross effect or covariance, which is positive if the firms that are gaining market shares (and increasing factor usage) are also those with above- average productivity growth or negative if the firms that are downsizing are the more productive ones. • The fourth term is the entry effect, which is the sum of the differences between each entering firm’s productivity and the initial productivity in the industry, weighted by the market share of the entering firm. • The fifth and final term is the exit effect, which is the sum of differences between each exiting firm’s productivity and the initial productivity in the industry weighted by the market share of the exiting firm. 2 This breakdown of the five components can also be presented differently as three components (see Bartelsmann and Scarpetta, 2007): • the within component; • a reallocation component, which is the sum of the between-effect and cross-effect components described earlier. It captures the productivity gains from reallocating factor inputs across firms; and, • the net entry component (also known as firm turnover or firm churning) reflects the productivity gains resulting from the creation of new, more productive firms and the exit of obsolete firms, calculated as the sum of the entries and exits terms. For the purposes of this study, we analyze growth between two time periods: 2003 and 2008. Therefore, we define continuing firms as firms with data in both years; new entrants began operation after 2003; and exiting firms are firms that left the dataset after 2003. A caveat on productivity decomposition: One potential problem with our methodologies is that the presence of measurement errors in assessing market shares and relative productivity will yield spurious decomposition. This needs to be given careful consideration here, since we excluded firms in the top (bottom) one percentile of TFP to minimize the impact of outliers. As a result, our calculations of market shares and 2 One potential problem with this methodology is that the presence of measurement errors in assessing market shares and relative productivity will yield spurious decomposition. This problem is carefully considered here, since we exclude firms in the top (bottom) one percentile of TFP to minimize the impact of outliers. As a result, our calculations of market shares may be distorted, especially as many industries in Russia consist of large market-dominant firms. As a result, we present two results on decomposition, the first using the full sample (i.e. including outliers), and the second without outliers. 5 aggregate productivity in each sector may be distorted, to the extent that many industries in Russia consist of large market-dominant firms which in many cases recorded large TFPs. Consequently, while we discuss results of the decomposition that excludes outliers, we also present results using the full sample (i.e. including outliers) in the Annex. (iii) Productivity and Firm Characteristics Finally, to investigate the basic correlates of firm productivity, we regress the productivity of manufacturing firms on a set of core firm characteristics, as follows: 𝐿𝑛𝑇𝐹𝑃 = 𝛼 + � 𝜑𝑛 𝑋𝑖𝑛 ,𝑡 + 𝜀𝑖,𝑡 𝑛 where TFPi,t is the TFP of firm i operating at time t, which is calculated according to the semi-parametric estimation technique developed by Levinsohn and Petrin (2003); and Xni.t is a vector of n firm characteristics, each with coefficient 𝜑 . The vector of firm characteristics includes a rich set of variables relevant for productivity. These include size (according to the official definition of the Russian government contained in Federal Law #209-FZ, 2007), 3 market share (the percentage of the firm’s revenue relative to total revenue of the sector (NACE level 4 classification)); industry structure (based on the Herfindahl index of market concentration––NACE level 4 classification); and dummy variables for each of the following: location in the capital city (Moscow); new entrants (ages 0 or 1); exiting firms (firms that cease commercial operations); a quadratic age function (defined as the square of firm age divided by 1000, similar to Alam et al 2007), industry (NACE level 4 classification), and year dummies. The model is estimated with White’s (1980, 1982) heteroscedasticity-consistent GLS estimators. 3. The Data 3 The following are the definition of firm sizes according to the Russian Law: Micro–firms with the number of employees less than 15, maximum annual revenue from sales of RUB 60 million (or EUR 1.64 million); Small––between 15 and 100 employees, with annual revenue from sales max RUB 400 million (EUR 10.94 million); Medium-sized firms: between 101 and 250 employees, and RUB 1 billion (EUR 27.64 million); There are no specific thresholds for large firms but they are implicitly defined as all those above the largest medium-size threshold. 6 The data on key variables required to compute firm-level TFP, are obtained from the Amadeus database. 4 The sample covers 78,704 manufacturing firms (NACE 10 – 35 i.e. 25 manufacturing sectors divided into 270 four-digit manufacturing sub-sectors) for which data are available for the Russian Federation, over the period 2003 to 2008. Firms’ output, capital stock, materials, and labor, are proxied by operating revenue (turnover), tangible fixed assets, cost of goods sold, and the number of employees, respectively. This provides an unbalanced panel of about 204,000 observations. This number of observations is the result of data cleaning procedures similar to those used in Alam et al (2008) that excludes: (i) missing observations on any of the key variables for estimated productivity functions; (ii) observations including a positive number of subsidiaries in order to prevent double counting of firms and also to avoid the inclusion of shell companies which are not directly involved in the production process; (iii) observations involving ratios greater than 1 between material and output; and (iv) observations with TFP (post-estimation) greater (less) than one percent standard deviation to minimize the impact of outliers. Full description of the characteristics of the sample is presented in Annex 1. 4. Results and Discussion (i) Productivity Dynamics and Output Growth The Levensohn and Petrin (2003) estimates of the Cobb-Douglas production function for Russian manufacturing firms over the period 2003 to 2008 indicate diminishing returns to scale (results are summarized in table 1). All factor inputs are statistically significant. However, the null hypothesis of constant returns to scale is rejected overall, even at the one percent level of statistical significance i.e., increasing all three factor inputs by 100 percent increases output by less than 100 percent (94 percent in this case). 4 Amadeus is a database of comparable financial information (in a standardized format) for public and private companies, on over 18 million companies across 43 Western and Eastern European countries. It includes comprehensive information on all main dimensions of firm operations and performance (e.g., the profit and loss account, and the balance sheet). For more detail on this database, please see http://bvdinfo.com/Products/Company-Information/International/AMADEUS.aspx. 7 Table 1: Levinsohn-Petrin (2003) productivity estimator (dependent variable: ln TFP) Coefficients z-statistic Log of Labour (no. of workers) 0.09*** 24.83 Log of Capital Stock (tangible fixed 0.07*** 8.89 assets) Log of Materials (cost of goods sold) 0.78*** 39.11 Wald test of constant returns to scale: Chi2 = 29.89 (p = 0.0000) *** indicates significance at the 1 percent level. Average productivity of manufacturing firms increased steadily from 2003 to 2008, albeit at an uneven annual growth rate: TFP grew by about 3 percent in 2004, and by nearly 6 percent in 2008 (Figure 1). This results in an average annual TFP growth of 4 percent over the whole period, similar to what was found in Alam et al (2007) for the period 1999-2003. Contrary to conjectures of some analysts attributing Russia’s growth mostly to high oil prices, there is no sign of an obvious slowdown in TFP growth in Russia’s manufacturing over the entire decade 1999-2008. The result of rising productivity is robust to the effects of “survivorship bias� i.e., both firms within the panel (14,067 firms) (mainly new and exiting firms) and those outside the panel have seen an increase in TFP, although non-panel firms have experienced a faster growth rate averaging 4.5 percent, compared to 2.7 percent for panel firms (Figure 2). Figure 1: Average Productivity of Russian Figure 2: Average Productivity - Panel Vs. Non- Manufacturing Firms. Panel Firms Non-Panel Firms Panel Firms Non-Panel (growth rate) Panel (growth rate) TFP Mean Growth Rate Growth TFP Growth Rate Levels Rate % 13.5 10% 13.5 6% 13.0 8% 13.0 5% 12.5 12.5 6% 4% 12.0 12.0 3% 4% 11.5 11.5 2% 2% 11.0 11.0 10.5 1% 10.5 0% 10.0 0% 10.0 -2% 2003 2004 2005 2006 2007 2008 2003 2004 2005 2006 2007 2008 8 The result of the growth accounting decomposition shows that output growth within the manufacturing sector has been driven primarily by TFP gains: TFP accounted for an average 63 percent of the annual growth in output (value-added) from 2004 to 2008 as shown in Figure 3. The contribution of labor and capital accumulation to Russia’s manufacturing output remains low, in line with evidence from Alam et al (2008). Figure 3: Decomposition of Average Value Figure 4: Aggregate and Average Productivity in Added Growth in Russia's Manufacturing Sector: Russia's Manufacturing Sector 2004 to 2008. Labour Capital TFP Aggregate TFP Average TFP Efficiency 40% TFP Efficiency 35% 25 8 30% 7 20 6 25% 15 5 20% 4 15% 10 3 10% 2 5% 5 1 0% 0 0 2004 2005 2006 2007 2008 2003 2004 2005 2006 2007 2008 (ii) Productivity Decomposition The most productive manufacturing firms are found to have a larger market share. The result of the Olley-Pakes decomposition shows, in the first instance, that efficiency is positive in the aggregate (see Figure 4). Moreover, the efficiency increased during the period of observation, slowly at first (between 2003 and 2005), then sharply from 2005, peaking in 2007 – meaning that there was a persistent increase in the share of output from the most productive firms in the entire manufacturing sector until the economic crisis that started in 2008. However, at the industry level (2-digit), the efficiency dynamics varied slightly (see Table 2). While efficiency estimates in all industries are positive through the years, indicating a higher productivity of market-dominant firms, the scale of efficiency in a number of industries – 9 Table 2: Olley-Pakes (1996) Decomposition of Allocative Efficiency (i.e. Aggregate TFP minus Average TFP) by NACE 2-digit Industry Classification of Manufacturing Firms. NACE 2 Industry 2003 2004 2005 2006 2007 2008 Food Product 4.64 3.66 4.43 5.21 4.27 5.19 Beverages 6.23 7.02 7.79 8.44 8.39 5.83 Tobacco 5.96 3.80 4.66 5.94 6.38 4.76 Textiles 1.26 1.49 1.59 2.10 2.89 3.93 Apparel 3.21 2.56 2.75 3.52 3.77 6.55 Leather 2.49 2.52 2.52 3.40 3.54 3.37 Wood 3.08 2.91 2.92 4.39 4.52 5.21 Paper 4.24 4.52 4.92 4.62 3.97 4.85 Printing 2.76 2.82 3.16 3.15 3.12 3.66 Coke & Refined Petrol Products 6.52 5.80 8.92 10.37 12.68 11.87 Chemicals 4.43 5.81 6.13 5.90 6.87 4.79 Basic Pharmaceutical 9.27 9.48 9.94 11.74 10.62 9.80 Rubber 4.18 3.69 3.29 4.20 4.48 5.84 Non-metallic 3.23 3.07 3.11 4.46 4.77 4.92 Basic Metal 10.52 12.52 9.40 10.72 9.63 6.29 Fabricated Metals 3.62 3.38 2.99 3.66 4.58 4.43 Computer 2.37 2.24 2.24 3.07 3.51 3.35 Electrical 3.20 3.06 2.84 4.38 4.12 4.25 Machinery & Equipment 2.12 2.92 2.90 4.01 4.27 4.72 Motor Vehicles 3.89 4.03 4.72 6.70 8.77 7.71 Other Transport 4.14 2.36 3.12 3.50 2.77 3.26 Furniture 2.20 2.26 2.17 3.06 3.00 3.32 NEC 4.74 3.38 3.65 5.04 5.80 6.00 Machine Repair & Installation 1.55 1.25 1.90 2.94 2.45 3.37 Utilities 4.30 4.39 4.21 5.60 7.89 6.97 Beverages, Tobacco, Basic Metal, and other Transport – had declined through 2008, by 6.4%, 20.2%, 40.2%, and 21.1%, respectively, suggesting a decrease in the share of output of the most productive firms in these sectors. On the other hand, efficiency in some sectors, like Textiles, Apparels, Machinery and Equipment (including Repair and Installation), more than doubled over the period. 5 5 See the Annex for results using the full sample. 10 Figure 5: Decomposition of the Sources of Total Factor Productivity Growth in Russia's Manufacturing Sector between 2003 and 2008 (no outliers) Within Reallocation Net Entry 50% 40% 30% 20% 10% 0% -10% -20% 11 Interestingly, TFP growth in most industries is attributable to new entrants. This is indicated by the results of the Foster et al (2001) decomposition method (see Figure 5), which shows that with the exception of a few industries (i.e. beverages, coke and refined petroleum, chemicals, basic pharmaceuticals, and motor vehicles manufacture) reallocation of capital and labor play a key role in productivity growth (which, given the structure of the dataset, also includes old firms i.e. registered before 2003, but for whom data became available between 2003 and 2008). 6 (iii) Productivity and Firm Characteristics The results of the empirical analysis of firm characteristics are statistically strongly significant; they are also broadly consistent with earlier analyses of theoretical and empirical literature using the Russian data (e.g., Alam et al. 2007). All independent variables, except measure of the industry structure, are statistically significant at the 1 percent level). Results are summarized in Table 3. The following principal findings are of particular interest. • First, firm size matters for productivity, with all size variables statistically and strongly significant (relative to micro-sized firms) and large firms being the most efficient, followed by medium-sized firms, indicating the existence of economies of scale and scope. • Second, the “primary city effect� of Moscow, reflecting agglomeration effects of Moscow city on productivity, is statistically and quantitatively significant although the effect is weaker for medium-size and large firms. In other words, it matters more for productivity if a firm has grown over a threshold size than that it is located in Moscow per se (Figure 6). Intuitively, this finding means that once a firm grows to a certain size, locating in the capital—while certainly an advantage––becomes less critical to its productivity. • Third, younger firms tend to be more efficient than older ones, suggesting that firm old age makes knowledge, abilities, and skills obsolete and induces organizational decay (as in Agarwal and Gort, 1996 and 2002). There is a quadratic effect in the relationship between age and productivity i.e., there is a decline in average TFP from the first quintile of the age quadratic until the fourth quintile (the threshold), then rising sharply to the fifth quintile. 6 Results when outliers are included are slightly different (See Figure A2 in the Annex). 12 Table 3: Manufacturing Productivity and Firm Characteristics (Dependent variable: Ln TFP). Variables (1) (2) 0.13 *** 0.13 *** Small Firm (0.001) (0.001) 0.21 *** 0.21 *** Medium Firm (0.003) (0.003) 0.30 *** 0.30 *** Large Firm (0.004) (0.004) 0.12 *** 0.11 *** Moscow (0.003) (0.003) -0.006 *** -0.006 *** Firm Age (0.000) (0.000) 0.03 *** 0.03 *** Age Squared/1000 (0.001) (0.001) -0.02 *** -0.02 *** New Entrants (age 0 -1) (0.002) (0.002) -0.065 *** -0.06 *** Exiting Firms (0.003) (0.003) -0.22 *** -0.22 *** Federal/State-Owned (0.011) (0.011) -0.28 *** -0.25 *** Municipal-owned (0.008) (0.009) 0.98 *** 0.98 *** Market Share (of NACE level-4 subsector total) (0.054) (0.054) -0.003 -0.01 Industry Structure Herfindhal Index (0.009) (0.01) Year Dummies (2004 – 2008) Yes Yes Industry Dummies No Yes 2.3*** 2.2 *** Constant (0.002) (0.000) Number of Observations 170,554 170,554 Number of Firms 524,87 52,487 Adjusted R sq 0.20 0.23 Note: Column (1) excludes sector dummies, whereas column (2) includes them. dependent variable calculated is ln (TFP) calculated according to the technique of Levinsohn and Petrin(2003). Standard errors adjusted for clustering on firms are noted in parentheses. The industry dummies are defined at 2-digit NACE level for manufacturing (10 – 35). *** significant at the 1 percent level, ** significant at the 5 percent level, * significant at the 10 percent level. • Fourth, state and municipal ownership shows strong negative and statistically significant relationships with productivity, perhaps reflecting the weak commercial incentives in state and municipal firms compared with private firms (see Figure 7). 13 Figure 6: Firm Productivity in Figure 7: Firm Productivity By Ownership Moscow Vs Other Cities Type Other Cities Moscow State-Owned Municipal-Owned TFP Private-Owned 16 TFP 15 15 14 13 13 11 12 11 9 10 7 2003 2004 2005 2006 2007 2008 2003 2004 2005 2006 2007 2008 • Fifth, productivity is negatively correlated with both very young firms (under the age of one year) and exiting firms. This suggests that existing (or incumbent) firms enjoy certain advantages, which may be due to either weak competition or other productivity and competitiveness advantages. But with many existing firms consisting of state and municipal companies with lower productivity than in private firms, it appears that firm survival is not correlated with high productivity. Internal managerial incentives in incumbent firms may not be strongly favoring productivity in the presence of weak competitive pressures. Managers of existing firms can maintain sub-optimal use of factor inputs. By contrast, intensified competition pressures force managers to speed up the adoption of new technologies in order to survive. • And sixth, market share (at the NACE industry level four) matters and is correlated with productivity. This is corroborating empirical evidence in Szegedi et al (2009) that firms can increase productivity by raising the market share. In fact, market share has the largest coefficient and in our model is the single most important driver of manufacturing productivity in Russia. But the estimated effect of the broader measure of industry concentration, the Herfindahl index, while having the theoretically correct negative sign, is not significant. 14 (iv) Productivity and Infrastructure In the next stage of this analysis, we shift the attention to the infrastructure- productivity link as infrastructure has emerged as a major constraint to Russia’s firms in many recent firm surveys. The 2008 BEEPS survey highlights the severity of the infrastructure challenge, relative to other countries within the region. About 14.2 percent of Russian firms identify insufficient water as a major or severe constraint, compared to a regional average of 9.6 percent. Likewise, more firms in Russia complain about electricity and telecommunication facilities than in Ukraine, Romania, and Belarus (see Figures 8 to 11). Recent analyses of infrastructure and productivity growth, especially those based on detailed firm-level data and a wider menu of infrastructure measures (e.g., Fedderke and Bogetic 2009) have shown strong association between infrastructure and productivity that has sometimes proven elusive in more aggregated studies. Figure 8: % of firms experiencing Figure 9: % of firms indentifying insufficient water electricity as major or severe constraint 18 60 16 55 14 50 12 45 10 8 40 6 35 4 30 2 25 0 20 Source: BEEPs 2008. 15 Figure 10: % of firms indentifying Figure 11: % of firms indentifying transport as major or severe telecoms as major or severe constraint constraint 60 40 50 35 40 30 30 25 20 20 15 10 10 0 Source: BEEPs 2008. Both Russian and other international data suggest a major deterioration of transportation infrastructure (World Bank 2011a, b). The state of Russia’s transport infrastructure is generally poor, especially for a high-income country and has been declining because of underinvestment in maintenance and rehabilitation. According to the 2010 Enabling Trade Index (World Economic Forum), Russia ranks 48th on the availability and quality of transport infrastructure (Figure 12). The index measures the state of transport infrastructure across all modes of transport in each country, as demonstrated by the density of airports, the percentage of paved roads, and the extent to which they are congested, as well as the transshipment connections available to shippers from each country. 16 Figure 12: The quality of road infrastructure index (from 0 (worst) to 7 (the best)) in select countries of Europe and Central Asia 7 6 5 4 3 2 1 0 Source: Enabling Trade Index, 2010, The World Economic Forum While the quality of transport infrastructure in Russia varies significantly across different modes of transport––and different parts of the country––the road infrastructure is estimated to have deteriorated the most. According to the World Economic Forum’s 2010 Enabling Trade Index (ETI), Russia’s railway infrastructure quality ranks 33rd in the world, relatively good but still significantly behind the Western European levels to which the country aspires. But its ranking on the quality of road infrastructure is 111th, among the world’s worst; this––in a vast country that is heavily resource-intensive and that depends on transport of goods and services for its exports. The quality of port and air infrastructure is also comparatively poor––Russia ranks 82th and 87th, respectively. These differences are also evident in comparing the asset deterioration in recent years, by transport modes. According to Rosstat, the asset deterioration rate reached 48 percent for air transport, roads and maritime port infrastructure in 2008. The deterioration in railways was much lower––24 percent. A recent survey of quality of the federal road network indicates that the majority of roads do not meet the minimum riding quality requirements (Table 4). 17 Table 4: Length and percentage of federal road network not meeting standards (2009) Length (km) Percentage Failing to meet minimum riding quality requirements 28,500 57.1 Failing to meet minimum grip requirements 12,200 24.4 Failing to meet minimum strength requirements 24,900 49.9 Failing to meet minimum defects requirements 35,100 70.3 Source: Avtodor, Long-Term Program of the Russian Motor Roads Public Company (2010-2015), December 31, 2009. Given these findings based on descriptive statistics, the question arises whether they can be corroborated by econometric analysis and to what extent infrastructure deficiencies affect firm productivity. This is the question we take up by explicitly modeling changes in TFP productivity growth using a vector of infrastructure variables from the BEEPs surveys, combined with firm characteristics. Since BEEPs surveys are infrequent, and in this case were conducted in 2004 and 2008 only, results in this section use data for 2004 and 2008 only. Following Escribano and Gausch (2004) and Alam et al (2008), we postulate the general regression model, with changes in productivity being the dependent variable, and include controls for firm characteristics. ∆lnTFPijt=α + β1∆infrastructuret-1 + ∑ⱷn Xni.t + εi,t For each of Russia’s regions, and for each sector (2-digit NACE classification), we calculate firm size (micro, small, medium, and large) and sector responses to each of the following infrastructure variables: (a) Electricity: number of days of power outage, percentage losses from power outage, and response to the question: Did you experience power outage? (b) Water: frequency of insufficient water supply, and a dummy on whether the firm experienced insufficient water supply. (c) Transportation: The proportion of firms reporting that transportation is a major or severe constraint, derived from the question: “How much of an obstacle is transportation?� 18 (d) Telecommunications: The proportion of firms reporting that telecommunications is a major or severe constraint, derived from the question: “How much of an obstacle is telecommunication�? (e) The Internet: Firm response to the question: “Do you have high-speed broadband internet?� We then calculate the regional, sectoral, and firm-size averages for each variable. The model is estimated using the principal component analysis. Results are presented in the table 5 below. We highlight two main findings. • First, access to good infrastructure such as telecommunications, transportation, broad band services, and water supply are important drivers of productivity in the manufacturing sector. Conversely, an increase in the proportion of firms with difficulty in accessing infrastructure services is significantly correlated with lower productivity. • Second, water supply deficiencies exert the largest negative impact on firm productivity. This is not a variable that is often associated with firm productivity in middle-income countries where basic water supply issues are resolved although it has come up in a recent analysis of South Africa and SACU countres (Bogetic and Fedderke 2006); nor has it been highlighted in recent analyses for Russia. But, interestingly, this finding is consistent with an independent result in an ongoing analysis of the Human Opportunity Index (HOI) for Russia, which identifies water supply as one of the infrastructure sectors where, despite broad coverage, there remain problems of quality and equity of access across Russia’s vast territory and income groups. 19 Table 5: Regression Results with BEEPs Variables (years 2004 and 2008 only) Variables (1) Beeps Variable (2) Including Firm Only Characteristics ∆teleobs -0.08 ** -0.09 ** (0.035) (0.037) ∆ transobs -0.06 * -0.07 * (0.037) (0.039) ∆broadband 0.06 * -0.06* (0.032) (0.032) ∆waterfreq -0.104 * -0.12 ** (0.058) (0.059) Constant 9.17 *** 10.2 *** (3.287) (3.563) Observations 19,385 19,385 R-squared 0.03 0.04 *** significant at the 1 percent level, ** significant at the 5 percent level, * significant at the 10 percent level. ∆teleobs = How much of an obstacle is telecommunications?, ∆ transobs= How much of an obstacle is transportation? ∆broadband= do you have high-speed broadband internet? ∆waterfreq= frequency of insufficient water supply? 5. Conclusion This note presents the results of an empirical analysis of firm-level productivity growth in Russia’s manufacturing sector during the period 2003-08 using a rich Amadeus database as well as the recent BEEPs surveys. The principal conclusions are as follows: First, productivity grew steadily between 2003 and 2008, with an annual growth rate averaging 4 percent over the period, showing no signs of a slowdown from the previous period after the 1998 crisis. Second, firm characteristics like size, location, age and the structure of firm ownership are important determinants of productivity. Specifically • Firm size matters: large firms tend to be most productive, followed by medium- size firms, and small firms, reflecting the economies of scale and scope. • Location is important: the “primary city effect� of Moscow––capturing the Moscow city agglomeration effects on productivity––is statistically and quantitatively significant although it is weaker for medium and large firms. In other words, while location matters, size matters even more. 20 • Younger firms tend to be more productive: over time, a firm’s old age may make its knowledge, capacities, and skills obsolete and could result organizational decay in (for related evidence, see Agarwal and Gort, 1996 and 2002). But after a certain threshold of age (fourth quintiles in the firm sample), firm productivity rises sharply, suggesting that the oldest and most established firms tend to maintain a productive edge despite the potential downsides of firm age. • State and municipal ownership show strong, negative, and statistically significant relationships with productivity, perhaps reflecting the weak commercial incentives in state and municipal firms compared with private firms. At the same time, state firms tend to be more productive than municipal firms. • A firm’s dominance in an industry is key to productivity in Russia. A firm’s market share of an industry’s sales exerts the most significant influence on firm productivity. Third, based on the analysis of BEEPs data, firms that complain about access to and/or the quality of infrastructure– particularly water, electricity, transport, and the internet– tend to show lower productivity; this corroborates other recent analyses showing rising infrastructure gaps in Russia and their impact on firm investment and economic activity. Fourth, water supply deficiencies exert the largest negative impact on firm productivity. 21 References Alam, Asad, Casero, F. A., Khan, F., Udomsaph, C. (2008). Unleashing Prosperity: Productivity Growth in Eastern Europe and the Former Soviet Union. The World Bank, Washington DC. Agarwal, R., and M. Gort, 1996, The evolution of markets and entry, exit and survival of firms, The Review of Economics and Statistics, 78 (3), 489-498; Agarwal, R., and M. Gort, 2002, Firm and product life cycles and firm survival, American Economic Review, Papers and Proceedings, 92, 184-190. Bartelsman, Eric, and Stefano Scarpetta (2007). Reallocation and Productivity in Transition and Integration Phases. Background Paper, World Bank, Washington DC. Bogetić, Željko and Johannes Fedderke (2006). “International Benchmarking of Infrastructure Performance in Southern African Customs Union Countries,� World Bank Policy Research Paper Series No. 3987 (August). Escribano Alvaro and J. Luis Guasch (2004). “Econometric Methodology for Investment Climate Assessments (ICA) on Productivity using Firm Level Data: The Case of Guatemala, Honduras and Nicaragua�. Mimeo World Bank, June. Fedderke, Johannes W. and Željko Bogetić (2009). “Infrastructure and Growth in South Africa: Direct and Indirect Productivity Impacts of 19 Infrastructure Measures,� World Development, Vol. 39, No. 9, pp. 1522-1539. Foster, L., J.C. Haltiwanger, and C.J. Krizan. 2001. “Aggregate Productivity Growth: Lessons from Microeconomic Evidence.� In E. Dean, M. Harper, and C. Hulten, eds., New Developments in Productivity Analysis. Chicago: University of Chicago Press. Levinsohn, J. and A. Petrin. (2003). Estimating production functions using inputs to control for unobservables. Review of Economic Studies 70(2): 317–342. Olley, G. Steven, and Ariel Pakes (1996). The Dynamics of Productivity in the Telecommunications Equipment Industry. Econometrica 64 (6), pp. 1263 – 1297. Szegedi, Zoltan and Korom, Erik (2009). Market share increase: What happens to productivity and profitability? International Journal of Procurement Management, Volume 3, Number 1, 2 December 2009 , pp. 72-90(19). White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48, pp. 817–838. 22 White, H. (1982). Maximum likelihood estimation of misspecified models. Econometrica 50, pp. 1–25. 23 ANNEX 1 Data Description Although data on 78,708 firms were obtained from Amadeus, the number of firms ranges from 25,561 in 2003 to 41,516 in 2008. Only 14,067 firms have data in all six years i.e. panel firms. Noteworthy to mention however, that this does not only reflect a high exit or entry rate of firms into the manufacturing sector, but also an improvement in the data collection processes of the data provider (Amadeus). Micro-sized, small, and medium- sized enterprises (MSMEs) form the bulk of the firms, although they account for only 18 percent of total output of the sample, and 32 percent of total workforce. Year No. of firms per year Exits New Entrants Moscow 2003 25,561 498 1644 3573 2004 30,794 1347 2517 5643 2005 31,069 1322 3231 5890 2006 38,553 7064 4298 7498 2007 37,319 8292 4402 6825 2008 41,516 n/a 3605 7191 Total 204,812 Firm Size 2003 2004 2005 2006 2007 2008 Total Micro 11,749 14,129 11,351 13,330 13,026 9,946 73,531 Small 6,914 9,682 12,695 17,193 16,798 22,271 85,553 Medium 3,163 3,281 3,505 3,989 3,602 4,264 21,804 Large 3,735 3,702 3,518 4,041 3,893 5,035 23,924 Total 25,561 30,794 31,069 38,553 37,319 41,516 204,812 Ownership: Year Federal/State Municipal Private firms 2003 568 812 24,181 2004 595 910 29,289 2005 535 876 29,658 2006 579 918 37,056 2007 470 844 36,005 2008 456 887 40,173 24 The list of Manufacturing sectors (NACE 2-digit classification) provided is below. These are further broken down into 270 sub-sectors, used in computing market shares. 2-Digit NACE 2003 2004 2005 2006 2007 2008 Total Food Product 3,924 4,461 4,408 5,145 4,741 5,181 27,860 Beverages 658 779 723 885 839 902 4,786 Tobacco 23 21 23 25 24 33 149 Textiles 587 714 746 908 841 902 4,698 Apparel 1,340 1,517 1,328 1,686 1,620 1,504 8,995 Leather 261 288 287 327 288 283 1,734 Wood 1,182 1,460 1,427 1,927 1,921 1,964 9,881 Paper 428 524 553 664 659 718 3,546 Printing 1,101 1,426 1,365 1,798 1,749 1,783 9,222 Coke Refine Petrol 114 126 146 185 170 217 958 Chemicals 940 1,172 1,209 1,469 1,428 1,615 7,833 Basic Pharmaceutical 334 405 400 463 397 445 2,444 Rubber 1,263 1,619 1,743 2,222 2,259 2,622 11,728 Non-metallic 1,648 1,952 2,010 2,511 2,534 3,148 13,803 Basic Metal 348 424 445 551 523 681 2,972 Fabricated Metals 2,130 2,639 2,722 3,511 3,532 4,081 18,615 Computer 1,524 1,889 1,856 2,386 2,195 2,407 12,257 Electrical 1,056 1,259 1,327 1,612 1,605 1,838 8,697 Machinery 3,020 3,670 3,720 4,587 4,466 4,892 24,355 Motor 444 501 527 621 610 729 3,432 Other Transport 371 446 451 563 537 681 3,049 Furniture 773 956 989 1,268 1,238 1,341 6,565 NEC 402 510 519 728 664 690 3,513 Repair Machine 41 50 54 60 66 80 351 Utilities 1,649 1,986 2,091 2,451 2,413 2,779 13,369 Total 25,561 30,794 31,069 38,553 37,319 41,516 204,812 25 ANNEX 2: Productivity Decomposition, using full sample i.e., including outliers. Relative to the results reported in the text, the impact of outliers is quite substantial. Both aggregate productivity and average productivity are not only higher than those reported previously in Section 4, but also efficiency is lower on aggregate (negative in some cases), particularly between 2004 and 2007 (see Figure A1). At the sectoral level, results are mixed (see Table A2). Figure A1: Aggregate and Average Productivity in Russia's Manufacturing Sector - (including outliers) Aggregate TFP Average TFP Efficiency TFP Efficiency 25 8 7 20 6 5 15 4 3 10 2 1 5 0 -1 0 -2 2003 2004 2005 2006 2007 2008 Table A2: Olley-Pakes (1996) Decomposition of Allocative Efficiency (i.e. Aggregate TFP minus Average TFP) by NACE 2-digit Industry Classification of Manufacturing Firms – Including Outliers Sector 2003 2004 2005 2006 2007 2008 Food Product 3.04 0.00 -0.82 -0.85 -2.20 4.91 Beverages 3.71 -0.73 -0.78 0.84 0.36 13.14 Tobacco 5.96 -4.98 -4.98 -5.68 -5.38 10.92 Textiles -3.24 -5.52 -2.50 -2.32 -1.18 5.27 Apparel 0.89 -2.18 -1.28 -0.78 -1.79 9.20 Leather -0.06 1.44 1.47 -2.37 -0.48 1.10 Wood 1.00 -0.54 -1.61 0.42 -2.23 2.55 Paper 2.29 2.49 1.92 0.94 -5.85 -0.71 Printing -3.32 -1.45 -2.40 -5.24 -6.55 5.47 Coke & Refined Petrol Products -4.76 1.70 -16.15 3.65 5.76 16.18 Chemicals -63.89 1.56 2.07 -3.57 1.11 10.52 Basic Pharmaceutical 7.42 6.94 4.57 9.93 0.47 -3.65 Rubber 3.16 -0.88 -1.22 -3.02 -1.09 5.63 26 Non-metallic 2.63 -0.52 -1.75 -0.32 -0.46 4.02 Basic Metal 10.28 9.71 6.88 15.40 6.43 8.83 Fabricated Metals 1.80 -0.63 -3.20 -1.62 -2.48 3.30 Computer -0.97 -5.09 -3.35 -5.75 -5.55 28.01 Electrical 0.06 -2.08 -5.32 -4.11 -4.74 -1.66 Machinery -1.87 -2.97 -3.23 -1.71 -7.45 0.86 Motor 3.39 0.30 -1.84 -0.26 0.01 5.52 Other Transport 1.36 -3.02 -2.56 -4.54 -6.01 1.89 Furniture -2.28 -2.17 -0.91 -2.40 -3.83 2.16 NEC 3.25 -4.49 -5.36 -24.85 -5.23 4.56 Repair Machine -1.52 -0.29 1.41 1.29 -0.51 2.24 Utilities 4.09 2.23 -4.80 1.44 2.14 7.06 Figure A2: Decomposition of the Sources of Total Factor Productivity Growth in Russia's Manufacturing Sector between 2003 and 2008 Within Reallocation Net Entrants 350% 300% 250% 200% 150% 100% 50% 0% -50% -100% When outliers are included in the sample, it appears that much of the growth in TFP can be linked to within-firm effects. 27