A Firm-level Productivity Diagnostic for Kenya’s Manufacturing And Services Sector A Firm-Level Productivity Diagnostic for Kenya’s Manufacturing and Services Sector Ana Cusolito and Xavier Cirera1 1 Ana Cusolito and Xavier Cirera are Senior Economists at the Innovation and Entrepreneurship Unit of the Trade & Competitiveness Global Practice of the World Bank. TABLE OF CONTENTS ABSTRACT.............................................................................................................................................................................. i 1. WHY IS PRODUCTIVITY SO IMPORTANT FOR KENYA’S FUTURE?.................................................... 1 2. WHY IS A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC A USEFUL TOOL FOR POLICYMAKERS?. 3 3. THE PRODUCTIVITY DIAGNOSTIC.......................................................................................................... 7 4. IMPLICATIONS FOR POLICY..................................................................................................................... 29 References .............................................................................................................................................................................. 31 Annex ....................................................................................................................................................................................... 33 LIST OF FIGURES Figure 1: Value added (Ksh) per worker, 1969-2010 ............................................................................................. 2 Figures 2-5: Productivity distributions.......................................................................................................................... 11 Figures 6-7: Productivity dispersion.............................................................................................................................. 13 Figure 8: Skewness............................................................................................................................................................. 15 Figures 9-12: Productivity and firm size....................................................................................................................... 17 Figures 13-16: Productivity and firm age .................................................................................................................... 18 Figures 17-20: Cumulative Distribution Function (CDF) and export status.................................................... 19 Figures 21-25: Relative Cumulative Distribution Function (CDF) and firm age............................................. 21 Figure 26: Box plot TFPR ................................................................................................................................................. 22 Figure 27: Misallocation of labor across sector........................................................................................................ 24 Figures 28-29: Misallocation of labor across sectors in the top 10 manufacturing and services sectors. 24 Figures 30-31: OP covariance in the manufacturing and services sector........................................................ 25 LIST OF BOXES Box 1: Data sources .......................................................................................................................................................... 9 Box 2: Indirect measure of allocative efficiency..................................................................................................... 22 Box A1: Productivity variables used in the analysis................................................................................................. 33 LIST OF TABLES Table 1: Correlation between productivity measures........................................................................................... 12 Table 2: Productivity dispersion (logs) ....................................................................................................................... 12 Table 3. Distance to the national productivity frontier ........................................................................................ 15 Table 4: Kolmogorov–Smirnov test results............................................................................................................... 20 Table 8: Olley-Pakes (OP) decomposition, measuring productivity through TFPQ ................................... 25 Table 9: OP decomposition (TFPQ, manufacturing top 10 only) ..................................................................... 26 Table 10: Capital and output distortion averages.................................................................................................... 27 ABSTRACT T his technical note implements a firm-level productivity diagnostic using the census of manufacturing firms and a large services survey in Kenya. By using a number of stylized productivity indicators, we aim to identify the ability of Kenyan firms to grow. The information presented in this diagnostic will help to conduct evidence-based policy-making. Specifically, implementing firm-level productivity diagnostics provide the necessary information for (i) improving the targeting of economic policies, (ii) enhancing their effectiveness, (iii) making more accurate predictions of the effects of industry shocks and policy reforms on the economy, and (iv) understanding the behavior of macroeconomic variables by tracking the evolution of variables at the firm-level. This note shows that there is a lot of heterogeneity in firms’ attributes and performance,and this can potentially be attributed to the presence of economic distortions that affect the efficient allocation of resources across firms, with the manufacturing sector showing a lackluster performance compared to the services sector. Overall, the findings highlight the importance of locating productivity at the center of the competitiveness agenda as a key instrument for employment creation and poverty reduction. 2 Age available for the manufacturing sector. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR i ii ii A FIRM-LEVEL A FIRM-LEVEL PRODUCTIVITY PRODUCTIVITY DIAGNOSTIC DIAGNOSTIC 1. WHY IS PRODUCTIVITY SO IMPORTANT FOR KENYA’S FUTURE? A ccelerating the process of economic development in Kenya and achieving the ambitious targets laid out in the Vision Comparing Kenya with similar countries in other regions suggests some relative underperformance in the Kenyan economy. 2030 will require a substantial increase in Kenya’s economic growth model is driven firm-level productivity. Kenya’s economy has primarily by the services sector and an undergone a significant process of structural over-reliance on the domestic market. transformation over the last decade. The This indicates the significant lack of economy showed an accelerating trend after competitiveness of the private sector, which 2002 with GDP growth increasing steadily prevents firms from entering and surviving in from below 1 percent in 2002 to 7 percent in international markets. It also acts to constrain 2007. The economy has been hit by several the potential of the economy in terms of shocks since 2007, starting with the post- future growth and employment creation. election violence in January 2008, which led to This, combined with accelerating population a slowdown in GDP growth to 1.5 percent and growth, explains the high levels of youth 2.7 percent in 2008 and 2009, respectively. unemployment in Kenya. New entrants to the Nevertheless, economic growth started to labor market–around 20 years of age–face an rebound in 2010, and recent predictions unemployment rate of around 35 percent. suggest higher growth rates during the Importantly, while investment has continued period 2014-2018, exceeding the growth to accelerate over the last decade, driven rates achieved before 2008. mainly by public investment, aggregate productivity growth turned negative during Amidst this positive growth context, the the 2008-09 economic crisis and during Kenyan Government launched the Second the macroeconomic instability period in Medium-Term Plan (MTP-2) to the Vision 2011, and it has remained stagnant since 2030 in October 2013. The aim of Kenya’s the 1970s. More importantly, as Figure 1 Vision 2030 is to create “a globally competitive suggests, aggregate productivity remains and prosperous country with a high-quality of around 1980 levels and although the long- life by 2030” and to shift the country’s status term decline experienced in manufacturing to the upper-middle income level. While has been halted, it has not been reversed the improvement in economic performance over the last decade. Therefore, only by in the past decade is remarkable, there are increasing firm productivity can Kenyan firms indications that achieving such ambitious become globally competitive and generate targets might be difficult, especially given the quantity of high quality jobs required to the slow rate of job creation. boost incomes and achieve shared prosperity. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 1 . 1. Why is productivity so important for Kenya’s future? Figure 1: Value added (Ksh) per worker, 1969-2010 400 350 300 250 200 150 100 50 0 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Total Economy Manufacturing Trade services Source: Authors’ own elaboration from data from de Vries et al. (2013) 2 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 2. WHY IS A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC A USEFUL TOOL FOR POLICYMAKERS? Productivity is the best predictor of income poverty reduction as not all transitions per capita from poverty require a change in the type of work undertaken. F or several years, policymakers and researchers have emphasized the role of factor accumulation to foster economic In Bangladesh and Vietnam, for example, growth and progress. However, after a poverty transitions have been dominated decade of empirical research on economic not by changes in income sources from farm growth, economists concluded that, although to nonfarm income but by higher income physical and human capital accumulation within the same sector (Dang and Lanjouw play a crucial role in accounting for economic 2012). This is also the case in Sub-Saharan progress in some countries, total factor Africa where poverty reduction in rural areas productivity (TFP)—a measure of efficiency is more closely associated with increases in and technological change-explains the bulk farm productivity (Christiaensen, Demery, of cross-country differences in both the level Kuhl 2011). China offers additional insight and growth rate of per capita GDP (Easterly into this, as increasing labor productivity and Levine 2001). Indeed, at the macro-level, in agriculture has been a key factor to TFP growth in the average country accounts understanding poverty reduction in lagging for more than half of output per worker Chinese provinces (Christiaensen et al. 2009). growth. Further, recent empirical evidence In other words, productivity is essential to shows that the typical Latin American country create better jobs and help households would have increased income per capita by escape poverty, regardless of the sector in 54 percent since 1960 if its TFP would have which those households work. grown at the same pace as its counterparts Using a firm-level productivity diagnostic in the rest of the world (Pages 2010). At the improves policymaking micro-level, evidence has shown that large and persistent differences in productivity If productivity is one of the main drivers levels across businesses are ubiquitous of competitiveness and a crucial factor for (Syverson 2011). economic growth and poverty reduction, then implementing a sound productivity As Paul Krugman (2014) famously claimed, diagnostic is critical to guide the design "Productivity isn't everything, but in the long and implementation of economic policies. run it is almost everything. A country's ability While the initial analysis on productivity was to improve its standard of living over time conducted at the industry and/or country depends almost entirely on its ability to raise level, over the past decade the focus has its output per worker.”Indeed, productivity shifted dramatically towards the firm as the is also crucial for shared prosperity and unit of analysis. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 3 2. Why is a Firm-level Productivity Diagnostic a Useful Tool for Policymakers? Firm-level productivity diagnostics provide efficiency responses to industry shocks granular information for evidence-based such as trade liberalization, showing policy-making. Specifically, implementing that studies that rely on oversimplifying firm-level productivity diagnostics provide assumptions to estimate firm-level TFP the necessary information for (i) improving the can overestimate the effects of trade targeting of economic policies, (ii) enhancing reforms on aggregate productivity. their effectiveness, (iii) making more accurate • Fourth, firm-level productivity predictions of the effects of industry shocks diagnostics complement and improve and policy reforms on the economy, and more aggregate analyses as evidence (iv) for better understanding the behavior shows that changes in macroeconomic of macroeconomic variables by tracking variables, both at the cyclical and the evolution of variables at the firm-level. secular frequencies, are certainly best More specifically: understood by tracking the evolution • First, firm-productivity diagnostics of economic variables at the firm-level improve the targeting of economic (Haltiwanger 2007). policies, especially those aimed at enhancing firms’ internal capabilities, Therefore, when designing or evaluating fostering firms’ productivity growth, policies to foster productivity growth and and facilitating access to international trade, it is very important to determine markets. which firms are going to benefit the most • Second, firm-productivity diagnostics from the proposed policy interventions and are important to enhance the which ones may need further assistance. effectiveness of economic policies as For example, some empirical studies show evidence shows that there is a lot of that productivity-enhancing policies, such heterogeneity in firms’ attributes and as fostering product market competition performance even within very narrowly to ignite innovation, may not have a defined industries. This suggests that significant impact on laggard firms if the policies that target the average firm,as productivity gap between the top and the was traditionally done in the past,may worst performer is large (Aghion et al. 2005). not end up having the expected impact Another clear example is export-promotion if the firm productivity distribution is policies oriented to reduce the fixed cost significantly dispersed. of exporting. These interventions may • Third, empirically sound firm- not have a significant effect on aggregate productivity diagnostics allow exports if productivity dispersion is high and policymakers to make more accurate the productivity cutoff, which reflects the predictions about the effects of industry minimum efficiency level required to have shocks and/or policy reforms on positive benefits for entering into the export economic performance. For example, a market (e.g., Melitz 2003), is close to the top recent study by De Loecker (2011) begs of the productivity distribution, indicating a for a serious re-evaluation of a long list large efficiency gap between the median and of empirical diagnostics that document best firm. 4 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 2. Why is a Firm-level Productivity Diagnostic a Useful Tool for Policymakers? A guide for policymakers cross-sectoral comparisons within a country. More importantly, it aims at helping the Kenyan The current productivity diagnostic prepared for Kenya is geared to provide policymakers Government better target economic policy, with useful information about the productivity more accurately predict the expected impacts distribution, identifying which are the of select productive sector policies, and better most productive and unproductive firms. link macro policies to the real microeconomic Further, the analysis explores the effects of dynamics of the productive sector. economic distortions in the prices of factors of production — such as land, labor, and capital Before we start with the analysis, the following -and products on the allocation of labor and sections present a description of the capital across firms, as well as their impact on indicators used to implement the diagnostic aggregate productivity. The analysis facilitates and their relevance for economic policy. cross-firm comparisons within a sector and FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 5 3. THE PRODUCTIVITY DIAGNOSTIC 3.1 Some key definitions On the supply-side, these determinants might include adjustment-costs, mark-ups, Labor productivity or TFP? and policy distortions that impact the price of different factors of production (such as land, T here are different measures that can be used to evaluate the efficiency with which firms transform labor, capital, labor, and capital), which ultimately affect the relative marginal cost of production and therefore, the final price of products. and intermediate inputs into production. On the demand-side, those determinants The two most commonly used measures might involve mark-ups, quality-upgrading, are (i) labor productivity and (ii) total factor and product price distortions (e.g., taxes), productivity (TFP). which affect final product prices. The risk is, therefore, that one measures as physical Although it is reasonable to think that the productivity parameters that are not strictly above mentioned measures must be highly related to technical efficiency but to the correlated, as efficiency gains obtained characteristics of the market structure in through technological change (TFP) will which firms operate or to the economic policy definitively make labor more productive, distortions that can affect both the relative labor productivity and TFP may not be price of inputs and the final price of products. correlated if productivity gains are obtained through capital accumulation instead If data on prices is available at the plant of innovation or the adoption of new or firm-level, then the best measures to technologies. Thus, identifying the pattern capture labor productivity are (i) deflated of productivity growth and economic growth sales (output) per worker or value-added per is crucial for policy-making. worker, and/or (ii) deflated sales (output) per hour worked or value-added per hour worked. Challenges when measuring productivity The best measure to capture TFP is “physical Measuring productivity, strictly defined as TFP or TFPQ”, defined as the deflated value the technical efficiency with which firms of sales (output) minus the contribution of transform inputs into production, is a labor and capital (it could also exclude the challenging task. The most challenging issue contribution of materials, too). is disentangling technical efficiency from other supply and demand-side determinants In practice, since firm prices are often not that affect firm sales—the variable commonly available in the data, sales per worker, sales used to measure production when product per hour worked, revenue TFP or TFPR, prices are not available. and physical TFP are the most commonly FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 7 3. The Productivity Diagnostic used measures to capture productivity. prices can introduce a bias in the Although labor productivity and TFPR will not estimated coefficients if the market exclusively reflect a firm’s technical efficiency, structure in which firms operate is not they are good indicators of the level of perfectly competitive, meaning that competitiveness of a firm in a market. firm prices will deviate from industry prices or GDP deflators, which are Even if one observes output prices, in usually the variables used to deflate order to obtain a good measure of physical nominal values; productivity, we also need to control for 4. “Omitted Input Price Bias” as lack of differences in input prices. In the absence of data on factor prices can introduce a bias input prices, TFP estimates using revenue as in the estimated coefficients because in output can provide more accurate estimates the presence of imperfect competition in than estimates obtained when only output is input markets, input prices are likely to be correctly deflated since the impact of prices firm-specific; on outputs and inputs tend to cancel out. In 5. “Endogeneity of the product-mix” as order to compare the different productivity ignoring the multi-product dimension measures in this analysis, we use one labor of some firms implies an assumption of productivity indicator (value added per the same technology for all products worker) and three proxies of TFP representing produced by a firm. three different methods. (Box A1 in the Appendix provides their definition.) The second and third measures of TFP The first TFP measure (TFP-est) represents employed are based on the Hsieh and Klenow the residual of estimating a traditional Cobb- (2014) framework implemented in Cirera et Douglas production function of sales on al. (2015).Under some specific and strong factors costs. In essence, this is the portion assumptions about demand (e.g., constant of output not explained by the contribution mark-ups for all firms) and production (e.g., of intermediate inputs, land, labor, and constant returns to scale and absence of capital used in production. As such, its level adjustment costs), these measures allow is determined by how efficiently those inputs firms’ optimal input choices to be used and factors of production are utilized. This to calculate revenue productivity (TFPR), measure presents several caveats: which is de facto a measure of firm-specific distortions and physical efficiency (TFPQ). 1. Assuming the same and Hicks-neutral The criticism of this framework relies on the technology for broadly defined sectors; strong assumptions mentioned. The main 2. “Simultaneity Bias”as input choices implication of this approach is that dispersion may not be independent of firm in TFPR is strictly attributed to economic characteristics, including efficiency, distortions at the firm-level, while the truth which is assumed using Ordinary Least is that heterogeneity of firm performance or Squares to estimate TFP-est; marginal products of labor and capital can 3. “Omitted Output Price Bias” as be perfectly explained by mark-ups, demand absence of information on product shocks, and/or adjustment costs. 8 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic The advantage of using this set of four effects on aggregate productivity at the productivity variables is that it allows us to sectoral and economy level (Restuccia and compare them when building the productivity Rogerson 2008; Hsieh and Klenow 2009; diagnostics for the economy. Alfaro, Charlton, Kanczuk 2008; Midrigan and Yi Xu 2010). The static productivity analysis The static productivity analysis is geared 3.2 Kenya productivity diagnostic to evaluate the within and the between component of aggregate productivity. The 3.2.1 INDICATORS TO ANALYZE WITHIN- within component is related to individual FIRM PRODUCTIVITY firms becoming more productive; that We start this section by providing a set is, increasing the amount of output they of indicators that are relevant for both (i) produce with a constant amount of inputs. targeting economic policies in order to The between component is associated with enhance firms’ capabilities, improve firm the reallocation of factors of production and productivity, and foster economic growth, economic activity toward more productive and (ii) predicting the impact of economic firms. The latter component is associated policies. with static allocative efficiency, which is captured by the fact that firms with higher The data used for the analysis corresponds than average productivity in a particular to the 2010 Census of Industrial production sector or economy should have a larger than and the 2011 Integrated Survey of Services average size in the sector or the economy (see Box 1). These are the two more extensive (usually measured through employment or sources of information for firms in the Kenyan sales share). Lack of static allocative efficiency economy, with around 5,000 observations suggests misallocation of resources across including both datasets. firms, and it can have important negative Box 1: Data sources The 2010 Census of Industrial Production The 2010 Census of Industrial Production is a survey jointly realized by KNBS and the Ministry of Industrialization. Reporting information for 2,252 firms (1,814 firms responded and information for 438 others have been imputed on other regular surveys), this survey gathered information relating to the calendar year 2009, although some information is available for 2010. Targeted firms belong to ISIC Rev 4 sections B (Mining and quarrying), C (Manufacturing), D (Electricity, gas, steam and air conditioning supply) and E (Water supply, sewerage, waste management and remediation activities). We focus the analysis on firms belonging to the manufacturing sector. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 9 3. The Productivity Diagnostic The 2011 Integrated Survey of Services The 2011 Integrated Survey of Services focuses on businesses in the service sector in 2009/2010 fiscal year. KNBS collected data on 3191 formal services firms (over 4464 targeted), spread across 13 different service subsectors (following ISIC rev.4). Some data cleaning has been performed in order to cope with the low quality of the data. Firms reporting an activity which does not belong to the usual services category have been excluded from the sample. Also, firms reporting total employment levels (including the owners of the firms) equal to 0 have been excluded from the sample. a) Aggregate indicators • Log of TFP Est is the proportion of output not explained by the estimated i. The productivity distribution contribution of factors of production The starting point is to summarize the entire such as labor and capital set of firms’ productivity data points, that • Log of TFPR is total factor productivity of is, the productivity distribution. The Kernel revenue, measured based on firm sales density presented in Figures 2-5 represents • Log of TFPQ is total factor productivity the productivity distribution function and of physical efficiency, measured based serves to have an overall view of the degree of on the deflated value of sales minus the productivity heterogeneity and/or dispersion contribution of labor and capital. across firms. Each figure uses a different measure of productivity to compare the Information about the distribution of productivity of firms in the Kenyan services firms’ productivity such as the maximum sector, the manufacturing sector, and in the and minimum productivity, the degree of aggregate. It is useful to briefly review the dispersion among firms’ productivity, or different measures of productivity: comparisons between the bottom 25 percent of firms and the bottom 50 percent, the • Log of Va/L is labor productivity, median firm, or the bottom 75 percent of measured as the value-added of labor firms can be very useful as a first summary of the data. 10 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic Figures 2-5: Productivity distributions Aggregate Kernel Density Aggregate Kernel Density 0.35 0.35 0.30 0.30 0.25 0.25 0.20 0.20 Density Density 0.15 0.15 0.10 0.10 0.05 0.05 0.00 0.00 0 5 10 12 14 15 18 20 0 5 10 15 Log of VA/L Log of TFP Est. Aggregate Manufacturing Services Aggregate Manufacturing Services Aggregate Kernel Density Aggregate Kernel Density 0.35 0.35 0.30 0.30 0.25 0.25 0.20 0.20 Density Density 0.15 0.15 0.10 0.10 0.05 0.05 0.00 0.00 -6 -4 -2 0 2 4 6 0 5 10 Log of TFPR Log of TFPQ Aggregate Manufacturing Services Aggregate Manufacturing Services Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services Services firms appear to be more productive years was disappointing, with manufacturing than manufacturing growth (3.1 percent) significantly lagging behind overall economic growth (5.0 percent). The main conclusion that can be drawn from Figures 2-5 is that firm-level productivity in The left tail of the productivity distributions, Kenya is lower in the manufacturing sector which is far thicker in the manufacturing than in the services sector, providing micro- sector than in the services sector, may economic foundations that rationalize be indicative of the existence of policies why Kenya’s manufacturing sector is small favoring the survival of inefficient firms in relative to the services sectors. The relatively the manufacturing sector relative to the poor productivity performance of firms in the services sector. Indeed, there are indications manufacturing sector is one of the factors that firms in the manufacturing sector are explaining why the overall activity of the operating well below their full capacity manufacturing sector over the past seven (Cirera, 2015). FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 11 3. The Productivity Diagnostic Table 1 shows the correlation among the This is important for policy-making for at different measures of productivity. The least two reasons. First, as we will see later, high correlation between labor productivity dispersion in revenue TFP (TFPR) may reflect (VA/L)and physical TFP(TFPQ), for example, the existence of economic distortions in factor suggests that labor efficiency gains in prices and product prices that negatively Kenya are most likely obtained through affect aggregate productivity at the sectoral technological change—including innovation level. It may also capture market-power and technology adoption—rather than and/or adjustment costs, which prevent the accumulation of capital. Indeed, a recent growth of small firms once they enter the World Bank Group report by Cirera (2015) market. Second, dispersion indicates the highlights that firm-level innovation rates need to design tailored policies for different in Kenya are relatively high as compared segments of the productivity distribution, to international standards, and they also as targeting the average firm may not generate significant revenues. Nevertheless, have the desired impact if there is a lot of the innovation activity is often related to heterogeneity in performance across firms. marginal improvements in the quality of existing goods rather than radical innovations Table 2 takes the four measures of such as the creation of new products that productivity, and within that the measures are breakthroughs to the national and/or among the manufacturing firms, the services international markets. firms, and the firms in aggregate, and shows their degree of productivity dispersion Table 1: Correlation between productivity measures across three different indicators: standard VA/L TFPQ TFPR TFP Est. deviation, bottom 75 percent vs. bottom VA/L 1.000 0.766 0.616 0.754 25 percent, and bottom 90 percent vs. 10 TFPQ 0.766 1.000 0.878 0.723 percent. The larger the resulting number, TFPR 0.616 0.878 1.000 0.716 the larger the dispersion. TFP Est. 0.754 0.723 0.716 1.000 Note: correlations are statistically significant at the 5% level. Table 2: Productivity dispersion (logs) Log productivity SD 75/25 90/10 ii. Stretched vs. squeezed distributions VA/L Aggregate 1.34 1.141 1.298 VA/L Manufacturing 1.385 1.157 1.346 The second indicator is productivity VA/L Service 1.282 1.134 1.277 dispersion, which can be defined as (i) TFPQ Aggregate 1.761 1.272 1.658 the standard deviation of the productivity TFPQ Manufacturing 2.019 1.387 1.935 distribution, (ii) the inter-quartile ratio TFPQ Service 1.636 1.244 1.556 (bottom 75 percent of firms versus the bottom TFPR Aggregate 1.208 0.576 1.394 25 percent), and (iii) the ratio between the TFPR Manufacturing 1.359 0.813 1.609 bottom 90 percent of firms and the bottom TFPR Service 1.09 0.536 1.227 10 percent, after removing outliers. These TFP Est. Aggregate 1.615 1.224 1.541 indicators provide information about how TFP Est. Manufacturing 1.451 1.26 1.507 stretched or squeezed the productivity TFP Est. Service 1.54 1.213 1.473 distribution is. Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services 12 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic Worse functioning and more distorted Figures 6 and 7 show productivity markets in the manufacturing sector dispersion among the top 10 manufacturing Table 2 shows that productivity dispersion and services subsectors, ranked by their size measured by the contribution to total is larger in the manufacturing sector than value-added. They show both the standard in the services sector. Further, evidence for deviation (SD) within each subsector and the Kenya is in line with previous productivity inter-quartile range, which is also another studies, both for developing and developed measure of statistical dispersion, equal to countries, which show that the dispersion the difference between the upper and lower of physical TFP (TFPQ), a true measure of quartiles, IQR = Q3 − Q1. Dispersion is high efficiency, is larger than the dispersion of in sectors such as manufacturing of electrical revenue TFP (TFPR), a measure of market equipment, chemicals, and fabricated metal performance. This suggests that more products (Figure 6). The same occurs with efficient firms in Kenya set lower product travel agency activities, financial services, and prices than inefficient firms, implying that social work activities in the services sectors productivity gains can be welfare improving. (Figure 7). Figure 6-7: Productivity dispersion Productivity Dispersion TFPR (Manufacturing) Manufacture of electrical equipment + Manufacture of chemicals and chemical products + Manufacture of fabricated metal products + except machinery and equipment Manufacture of food products + Industry Manufacture of basic metals + Manufacture of other non-metallica mineral products + + Manufacture of rubber and plastic products + Manufacture of beverages + Manufacture of paper and paper products + -6 -4 -2 0 2 Productivity Dispersion TRPR (Services) Travel agency, tour operator, reservation service and related services + Financial service activities, except insurance + and pension funding Social work activities without accommodation + Warehousing and support activities for transportation + Insurance, reinsurance and pension funding, + except compulsory social security Telecommunications (Services) + Advertising and marketing research (Services) + Publishing activities + Security and investigation activities (Services) + Land transport and transport via pipelines + 0 10 20 30 40 Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 13 3. The Productivity Diagnostic These large and high dispersion sectors iii. Symmetric vs. asymmetric distributions provide a first list of priority sectors A third aggregate indicator is the skewness for tailored policy interventions, since of the distribution, which quantifies how understanding and removing the distortions symmetric the distribution is. For a unimodal that create this wedge in firms’ performance distribution, a negative skew indicates that could have important benefits in terms of the tail on the left side of the probability productivity and employment growth. density function is longer than the right-hand side, showing a large proportion of firms with Technological gaps are larger than labor efficiency gaps low productivity levels surviving in the market, compared to the proportion of firms with high Further, dispersion of labor productivity is productivity. In this situation, the mean (or smaller than dispersion of TFPQ, suggesting average) is less than the mode (or maximum that the technology gap between leading point of the curve), compared with the case and laggard firms is larger than the labor of a normal distribution, where the mean and efficiency gap. The immediate implication mode are equal. A normal distribution, which of this empirical regularity, which applies is symmetric, has zero skewness and the tails to both the manufacturing and services on either side of the curve are exact mirror sectors, is that policies geared to foster images of each other. Conversely, a positive technical change, such as innovation skewed curve indicates that the tail on the policies (e.g., tax incentives, matching right-hand side is longer than the tail on the grants, royalties) or policies oriented left-hand side, and the mean is greater than to facilitate technology adoption the mode, showing a small proportion of (e.g., subsidies for the adoption of new firms with low productivity levels surviving in technologies), should be more firm-specific the market, compared to the proportion of than policies aimed at increasing value- firms with high productivity. added per worker through the accumulation of capital (e.g., subsidies to credit). Figure 8 shows the skewness of the productivity distribution for both Productivity dispersion is much larger in the manufacturing and services, across the four Kenyan manufacturing sector than in other measures of TFP. countries A comparison of dispersions in TFP with More efficient firms tend to coexist in the other countries such as China and India, services sectors or other African countries such as Ethiopia Figure 8 shows that the productivity and Ghana (Cirera et al. 2015), shows a much skewness in the services sector for measures larger dispersion of technical efficiency in like TFPQ and TFPR is positive, indicating Kenya than in the rest of the countries for both a longer tail to the right-hand side than TFPQ and TFPR. This implies, as we will see to the left-hand side, with more efficient below, that distortions in the manufacturing firms surviving in the Kenyan services sector in Kenya are pervasive. sector. The negative skewness measure 14 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic in the manufacturing sector for almost all Table 3. Distance to the national productivity frontier productivity measures suggests a longer Median distance 
to lower productivity tail (left-hand side) than frontier the high productivity tail (right-hand side), Aggregate 3.865 another indication of potential misallocation VA/L Manufacturing 4.357 in the sector. Service 3.547 Aggregate 4.536 TFPQ Manufacturing 4.738 Figure 8: Skewness Service 4.48 Skewness, by Aggregation Level Aggregate 3.362 TFPR Manufacturing 3.711 0.1 Service 3.254 Aggregate 3.774 0.0 TFP Est. Manufacturing 3.826 Skew Service 3.462 -0.1 Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services -0.2 There is a large productivity gap in Kenyan TFP Est. TFPQ TFPR VA/L firms, especially in the manufacturing TFP sector Manufacturing Service Source: Authors’ own elaboration from Census of Industrial In the case of Kenya, Table 3 shows two Production and Integrated Survey of Services important things. First, that the labor productivity gap between leading and laggard iv. Productivity gap (leaders vs. laggards) firms is smaller than the technical efficiency gap. This may be explained by a combination The last indicator we explore is the median distance to the frontier, measured as the of factors, including the highly concentrated median value of the ratio between the rates of investments in innovation, which was productivity level of each firm and the recently documented by Cirera (2014), and productivity level of the top performer. the random nature of the innovation process, This measure provides crucial information which makes some firms luckier than others about the productivity gap between the in creating new products and/or processes median firm and the best performer. Table or improving the quality of existing goods. In 3 shows this indicator across the four other words, this fact shows that labor is more productivity measures, and within each, or less similar in their levels of productivity across manufacturing firms, services firms, across different types of firms, but the real and firms in the aggregate. The higher the difference comes in the efficiency with which number, the higher the productivity gap. The firms use their labor force, which is an issue gap appears to be relatively high, especially related to innovation and access to new in the manufacturing sector. technologies. This fact is in line with the issue that economic growth in Kenya is related to FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 15 3. The Productivity Diagnostic innovation and technology adoption rather b) Who are the most productive and than accumulation of capital. Second, Table unproductive firms? 3 shows that the distance to the frontier is significantly larger when measured through Once productivity measures have been physical TFP (TFPQ) than through revenue calculated and the productivity distribution TFP (TFPR). The latter is logical because examined, the second step in the analysis adjustments in prices due to efficiency involves examining the links between gains reduce the revenue performance gap productivity and firm characteristics or between leaders and laggards. attributes, such as size, age, export status, import status, R&D intensity, and type of These results are very important for policy- ownership. Identifying the most productive making for several reasons highlighted and unproductive firms is crucial to target below: economic policies aimed at boosting economic growth. • First, evidence shows that productivity- enhancing policies, such as fostering i. Average of log-productivity by percentile product market competition to ignite of firm characteristic innovation, are unlikely to have a significant impact on laggard firms if the Figures 9-12 compare the productivity of productivity gap between leaders and firms according to their size, across the four laggards is large. In the case of Kenya, measures of productivity. The horizontal axis this is likely to be the case, especially in measures size by deciles (the smallest 10 the manufacturing sector. percent of firms, the smallest 20 percent, the • Second, if the productivity gap is a smallest 30 percent, etc.), while the vertical good proxy of the gap in the absorptive axis shows average productivity. An upward capacity of firms, then policies oriented sloping line means productivity increases to increase laggard firms’ productivity with firm size, while a downward sloping line may be inefficient to boost the means it decreases with size. efficiency level of low productivity firms. In other words, laggard firms The most important message provided by in Kenya are unlikely to have enough Figures 9-12 is that productivity performance capacity to benefit from policies to varies across firms, depending on their size. support innovation and could benefit However, the direction of the difference more from policies that support basic between size groups depends on the type of management and organizational skills. productivity measure being considered. First, • Third, export-promotion policies the figures show that there are not significant oriented to facilitate access to differences in labor productivity across firms international markets by reducing the of different size, although small firms tend fixed cost of exporting may not have to produce more value-added per worker a significant impact on aggregate than large firms. However, large firms tend exports if the minimum efficiency level to display a better performance than small required to cover the fixed cost of firms in terms of technical efficiency, which exporting is high. may be explained by the fact that large firms 16 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic Figures 9-12: Productivity and firm size Aggregate Ave VA/L by Size (by Deciles) Aggregate Ave TFPQ by Size (by Deciles) 12 4 10 Avc Log VA/L Avc Log VA/L 0 8 6 -4 4 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Size (by Deciles), % Size (by Deciles), % Aggregate Ave TFPR by Size (by Deciles) Aggregate Ave TFP Est. by Size (by Deciles) 14 2 1 12 Avc Log VA/L Avc Log VA/L 0 10 -1 8 -2 6 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Size (by Deciles), % Size (by Deciles), % Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services are more capable of overcoming the large that economic distortions disproportionately fixed costs of innovating as they can usually and favorably affect small firms, although have better access to external sources of these companies are not the most efficient. innovation finance than small firms. Figures 13-16 compare the productivity of A comparison of innovation activities at the manufacturing firms according to their age, firm-level in Kenya shows that medium-sized across the four measures of productivity. and large firms are more innovative than These figures show that mature firms are small companies (Cirera 2014). Interestingly, more productive in terms of labor and small firms display larger revenue TFP than technical efficiency than young firms in the large companies. However, this does not manufacturing sector.3 However, revenue necessarily mean that they are more efficient, TFP is larger for young firms, suggesting as this variable may be contaminated with that distortions have a larger and favorable mark-ups, adjustment costs, and product and impact on these firms. factor price distortions. The result may reflect 3 Age available for the manufacturing sector. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 17 3. The Productivity Diagnostic Figures 13-16: Productivity and firm age Manufacturing Ave VA/L by Size (by Deciles) Manufacturing Ave TFPQ by Age (by Deciles) 12 2 10 Avc Log VA/L Avc Log VA/L 0 8 -2 6 4 -4 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Age (by Deciles), % Size (by Deciles), % Manufacturing Ave TFPR by Age (by Deciles) Manufacturing Ave TFP Est. by Age (by Deciles) 2 14 1 12 Avc Log VA/L Avc Log VA/L 0 10 -1 8 -2 6 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Age (by Deciles), % Size (by Deciles), % Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services ii. Cumulative distribution functions by the closer the curves are to each other, the group smaller the differences. Exporters are more productive than non- exporters Figures 17-20 show that the productivity performance of exporting firms is larger In addition to the previous graphs that allow than that of non-exporting firms. This fact us to explore productivity heterogeneity can be the result of both static and dynamic across firms’ groups, plotting cumulative distribution functions by type of firms is very efficiency gains. Evidence shows that most informative to heterogeneity. Figures 17- productive firms are the ones able to cover 20 show the cumulative distribution of firm the fixed cost of exporting and entering into productivity by firms’ export status (exporters international markets. At the same time, and non-exporters), according to the four access to international markets can create measures of productivity. The further learning-by-doing effects that make firms apart the two curves are, the greater the more efficient and/or allow them to absorb difference in productivity levels between knowledge spillovers, which help them to exporters and non-exporters. Conversely, become more competitive. 18 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic Figures 17-20: Cumulative Distribution Function (CDF) and export status Manufacturing CDF VA/L by Export Status Manufacturing CDF TFPQ by Export Status 1.00 1.00 0.75 0.75 0.50 0.50 0.25 0.25 0.00 0.00 -1.0 -5 0 5 0 5 10 15 Log VA/L Exporter Non-Exporter Exporter Non-Exporter Manufacturing CDF TFPR by Export Status Manufacturing CDF TFP Est. by Export Status 1.00 1.00 0.75 0.75 0.50 0.50 0.25 0.25 0.00 0.00 -4 0 4 -5 10 Log TFPR Log TFP Est. Exporter Non-Exporter Exporter Non-Exporter Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services The productivity gap between exporters is high. Alternatively, the lack of differences and non-exporters is larger when both between both distributions may reflect the groups are compared according to their fact that non-exporting firms face relatively technical efficiency, suggesting that more distortions than exporting firms. investments in technological change (e.g., innovation and technology adoption) can be The Kolmogorov–Smirnov Test (KST) is a more effective than investments in capital non-parametrical test of the equality of to help firms achieve the minimum level of continuous and one-dimensional probability efficiency required to overcome the fixed distributions that can be used to compare costs of exporting. Interestingly, there are two samples. In the particular case of Figures no significant differences between exporters 17-20, the KST is used to compare how equal and non-exporters when measuring the exporter and non-exporter curves are productivity through TFPR, suggesting that to each other. The Kolmogorov–Smirnov the pass-through of efficiency gains to final statistic quantifies the distance between the prices obtained by exporting firms, probably empirical distribution functions of exporters ex-ante to enter into international markets, and non-exporters. The null hypothesis is FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 19 3. The Productivity Diagnostic that the samples are drawn from the same productivity distributions across the two distribution, which means that the productivity groups are statistically equal. Column (5) distribution of exporters and non-exporters provides information of whether one group are not statistically different from each other. first-stochastically dominates the other. Table 4 confirms that the productivity of exporters The information provided by the KST can is larger than that of non-exporters, except also be complemented with a First-order when productivity is measured with TFPR. Stochastic Dominance Test (FSDT), which evaluates differences across groups (e.g., iii. Relative productivity cumulative exporters and non-exporters) by examining distributions functions between groups differences in the cumulative distribution Mature firms are not necessarily more functions of the variable of interest (e.g., efficient than young firms productivity). According to the definition, productivity distribution A (e.g. exporter) Productivity differences can also be “first-order stochastically dominates” examined by drawing the productivity productivity distribution B (e.g., non-exporter) relative cumulative distribution function if and only if for any productivity value “x”, the (PRCDF) and comparing it with the 45 degree cumulative distribution function FB(x)≥FA(x). line. Figures 21 to 25 show the relative CDFs Visually, A dominates stochastically B if the of young vs. mature firms, across the four productivity cumulative distribution function measures of productivity. To infer conclusions of A is, for all “x” values, to the right of B, about the figures, there are a couple of simple meaning that the proportion of firms with rules that can be applied: productivity values “x” in A is always no • First, the farther the distance between smaller than the proportion of firms in B the PRCDF curve and the 45 degree line, (e.g., Prob(A≥x)≥(Prob(B≥x)), which reflects the farther the productivity differences that overall, firms in group A have a better between both groups. productivity performance than firms in group B. • Second, if the PRCDF curve is below the 45 degree line, then the group Table 4 shows the results of those tests. represented on the Y-axis (mature firms) Column (4) provides the statistical p-value has a better productivity performance associated with the null-hypothesis of the than the group represented on the KST. P-values lower than 0.05 mean that X-axis (young firms), as a smaller the KST rejects the null hypothesis that the percentile in Y has the same productivity Table 4: Kolmogorov–Smirnov test results Which group P-Value null: CDFs are Group 1 Group 2 stochastically equal dominatesthe other? (1) (2) (3) (4) (5) Manufacturing, VA_L Exporter Non-Exporter 0.00 Group 1 Manufacturing, TFPQ Exporter Non-Exporter 0.00 Group 1 Manufacturing, TFPR Exporter Non-Exporter 0.61 Neither Manufacturing, TFP2 Exporter Non-Exporter 0.01 Group 1 Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services 20 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic level than a larger percentile in X. The dominates the PCDF of the group reverse applies when the PRCDF curve represented in the X-axis (young firms). is above the 45 degree line: the group represented on the X-axis (young firms) The Figures show that mature firms are more has a better productivity performance efficient than young firms in the case of than the group represented on the value added per worker (VA/L) and TFPQ, Y-axis (mature firms). but only in the middle of the distribution. • Finally, if the PRCDF is below the 45 In the case of TFPR and TFP estimated, degree line and never cross it, then young firms are more efficient in some parts the PCDF of the group represented in of the distribution. the Y-axis (mature firms) stochastically Figures 21-25: Relative Cumulative Distribution Function (CDF) and firm age Manufacturing Relative CDF of VA/L by Age Manufacturing Relative CDF of TFPQ by Age 1.0 1.0 0.8 0.8 0.6 0.6 Mature Mature 0.4 0.4 0.2 0.2 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Young Young Manufacturing Relative CDF of TFPR by Age Manufacturing Relative CDF of TFP Est. by Age 1.0 1.0 0.8 0.8 0.6 0.6 Mature Mature 0.4 0.4 0.2 0.2 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 Young Young Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 21 3. The Productivity Diagnostic 3.2.2 EXPLORING ALLOCATIVE EFFICIENCY in marginal products of these factors across We start this section by dividing indicators in firms. As such, the marginal product of labor measures the marginal change in two categories: indirect and direct measures output per marginal change in labor, while of allocative efficiency. Recall that an conversely, the marginal product of capital economy displays allocative efficiency when measures the marginal change in output per firms that show a productivity performance marginal change in capital. This framework above the average/median productivity has been used extensively. However, recent in a particular industry are larger than the research shows that in dynamic settings, average/median productivity firm in the other reasons like volatility in sales coupled same industry. By contrast, misallocation of with adjustment costs in capital could also resources occurs when productive firms are explain productivity dispersion (Asker, smaller in terms of factor and market shares Collard-Wexler, De Loecker 2014). than unproductive firms. A simple way of examining the dispersion Exploring allocative efficiency is important of revenue TFP is to make a box-plot because misallocation can have negative representation of the variable as per Figure economic effects at the aggregate level. 26. The box represents the productivity Empirically, economies with low allocative levels between the bottom 75 percent of efficiency tend to suffer from the following firms and the bottom 25 percent of firms, characteristics: (i) low rates of firm entry with the top part of the box representing and exit; (ii) low post-entry growth rates of the 75 percentile, the bottom part the 25 efficient firms; (iii) high firm-level productivity percentile, and the line inside the box the dispersion, even within very narrowly defined median productivity level. The dots reflect industries; (iv) low correlation between size the productivity outliers, that is, firms with an (measured through employment shares or extraordinary good productivity performance market shares) and firm-level productivity; (if they are above the box) or an extraordinary and (v) higher employment growth in mature bad performance (if they are below the box). firms than in young firms. Figure 26: Box Plot TFPR Indirect measures of allocative efficiency TFPR Boxplot The first indirect measure of misallocation 6 is the dispersion of revenue TFP (physical productivity multiplied by a firm’s output 3 price). As suggested above, Hsieh and Klenow (2009) have presented a framework TFPR 0 where optimal input choices are directly informative about the efficiency with which labor and capital are allocated across firms. -3 The analytical framework used by the authors identifies the level of misallocation of labor Aggregate Manufacturing Services and capital in an industry from the variation Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services 22 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic Box 2: Indirect measure of allocative efficiency The authors show that in the absence of distortions in product prices (e.g., taxes) and factor prices (e.g., labor market regulations, subsidized credit), that is, in the absence of heterogeneous policy treatment of firms in the same industry, optimal input choices and marginal product of capital and labor should equalize across firms. This is true if and only if the production function displays constant returns to scale, consumers are assumed to prefer and love varieties (a la Dixit-Stiglitz), and product mark-ups are constant. Needless to say, the latter are two strong assumptions. In this case, more capital and labor is allocated to firms with high physical TFP (TFPQ). The reallocation of inputs occurs until factor and product markets are in equilibrium (e.g., demand equals supply). By contrast, sizable gaps in marginal products of labor and capital across firms within narrowly defined industries are key signs of misallocation of resources under the identifying assumptions made by the authors. Hsieh and Klenow also show that when A (TFPQ) and TFPR are jointly log-normally distributed (a specific assumption about the distribution of TFPQ and TFPR and its linear combination), there is a simple closed-form expression for aggregate TFP,4 showing that the variance or dispersion in revenue TFP (TFPR) has a negative impact on aggregate sectoral productivity. In other words, if both TFPQ and TFPR fulfill the statistical assumption used by HK (jointly log-normally distributed), then the extent of misallocation is worse when there is greater dispersion of marginal products. However, it is worth mentioning that dispersion of physical TFP is not necessarily a measure of misallocation, as this variation can be explained by the randomness of the technological process, which can generate differences in innovation outputs and outcomes across firms, even when optimal input choices in terms of factors of production and R&D are identical across firms. 4 Figure 26 shows more dispersion for the Firms are not growing as they should manufacturing sector (because of the The second indirect way to measure bigger box) with a lower median and a lot misallocation is to plot the ratio of of values concentrated between the 25th and employment between the bottom 10 50th percentile. This suggests the presence percent of firms in terms of productivity of distortions and allocative inefficiency in and the bottom 50 percent of firms, and the manufacturing sector more than in the compare that to the ratio of employment services sector, which has a smaller box between the bottom 90 percent and the higher up on the vertical axis. bottom 50 percent of firms in terms of productivity. The idea is that if labor is FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 23 3. The Productivity Diagnostic efficiently allocated across firms within an Figures 28 and 29 show the misallocation industry, then employment in the latter of labor by manufacturing and services should be much higher than employment in subsectors. They show that allocative the former. efficiency varies significantly across different subsectors. In manufacturing, there is Figure 27 shows this comparison for significant misallocation in metal products, manufacturing firms, services firms, and firms where low productivity firms are larger, in the aggregate, using the TFPQ measure or in the leather sector, where some low of productivity. The larger the bar, the higher productivity firms are significantly large. the concentration of employment. Figure 27 Misallocation in the services sector appears effectively shows that while employment is to be less prevalent, although in a significant concentrated in firms with higher productivity number of sectors the middle productivity in both the manufacturing and services sector, firms are the largest, larger than some highly the difference is much larger in the latter. efficient firms. Figure 27: Misallocation of labor across sector Direct measures of allocative efficiency 3 Relative Employment Share by TFPQ Decile Ratios The first direct measure of allocative efficiency is the Olley and Pakes (OP) decomposition (Olley and Pakes 1996), 2 which uses the difference between size— measured through a firm’s employment share or market share in the industry—and 1 productivity to infer the efficiency with which labor or output is allocated across firms, by sector and subsector (Bartelsman, 0 Aggregate Manufacturing Services Haltiwanger, Scarpetta 2013). The idea Relative Employment behind this measure is that in a well- 10% / 50% 90% / 50% Source: Authors’ own elaboration from Census of Industrial Production and functioning economy, where there are no Integrated Survey of Services policy distortions and no differences in the Figures 28-29: Misallocation of labor across sectors in the top 10 manufacturing and services sectors Manufacture of Relative Employment Share by TFPQ Decile Ratios Service Relative Employment Share by TFPQ Decile Ratios Manufacture of food products Financial service activities, except insurance and pension funding Manufacture of beverages Telecommunications Manufacture of fabricated metal products, Insurance, reinsurance and pension funding, except machinery and equipment except compulsory social security Manufacture of other non-metalic Wholesale trade, except of motor vehicles mineral products and motorcycles Manufacture of chemicals and Retail trade, except of motor vehicles chemical products and motorcycles Manufacture of rubber and plastic products Wholesale and retail trade and repair of motor vehicles and motorcycles Warehousing and support activities Manufacture of leather and related products for transportation Manufacture of basic metals Publishing activities Repair and installation of machinery and equipment Social work activities without accomodation Relative Employment Travel agency, tour operator, reservation service Relative Employment Manufacture of paper and paper products 10% / 50% 90% / 50% and related activities 10% / 50% 90% / 50% Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services 24 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic way firms are treated by laws and regulations, for that of the services sector, indicating that firms that are more productive are able to labor reallocates more efficiently in the latter. hire more labor, use more capital, and earn a larger market share than unproductive firms.5 Figures 30 and 31 organize manufacturing and services subsectors by their International Table 8 shows the Olley-Pakes decomposition Standard Industrial Classification of All in the manufacturing sector, the services Economic Activities (ISIC) codes, and plot sector, and in the aggregate. A high and their OP covariance term to compare positive value of the covariance term is them. The figures show that most of the associated with high allocative efficiency. manufacturing and services subsectors display a positive covariance. Table 8: Olley-Pakes (OP) decomposition, measuring productivity through TFPQ Simple Share of Table 9 shows in table format what is Industry Covariance ISIC2 Mean employment displayed in Figures 30 and 31. As before, Aggregate 8.93 1.42 NA a high and positive value of the covariance Manufacturing 8.56 1.14 43.26 NA term is associated with high allocative Services 9.22 1.63 56.74 NA efficiency. The table shows that the highest Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services covariance appears in the most important manufacturing sector, which is food The results in Table 8 show a positive products, while the lowest displays in basic covariance in the aggregate, manufacturing, metals. Large misallocation in the services and services sectors. This suggests that sector is not concentrated in the top-10 despite the distortions, labor may reallocate sectors. Only a few top 10 services sectors, to more efficient firms, and in the absence such as tourism and publishing activities, of distortions this positive reallocation could show negative covariance. Since the values be much higher. Importantly, the covariance are close to zero, a random allocation of for the manufacturing sector is lower than labor across firms is likely. Figure 30-31: OP Covariance in the manufacturing and services sector Manufacturing TFPQ OP Covariance Term Service TFPQ OP Covariance Term 1.5 3 2 1.0 OP Covariance Term OP Covariance Term 1 0.5 0 -1 0.0 -2 21 28 27 24 19 29 31 13 23 17 14 33 22 18 32 25 20 11 15 16 10 72 41 79 10 13 57 56 20 69 92 77 47 93 29 46 71 28 50 68 65 49 62 52 59 85 55 51 70 75 74 95 68 73 91 64 80 45 01 81 60 86 96 61 63 56 18 78 88 ISIC Code ISIC Code Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services The OP measure decomposes an index of industry-level productivity, defined as the weighted average of firm-level (log) 5 productivity, into a moment of the firm productivity distribution, such as the unweighted firm-level average, and a moment of the joint distribution of productivity and employment share or market share, such as the covariance between both variables. A high and positive value of the covariance term is associated with high allocative efficiency. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 25 3. The Productivity Diagnostic Table 9: OP decomposition (TFPQ, manufacturing top 10 only) Share of Industry Simple Mean Covariance ISIC2 employment Manufacture of basic metals (Manufacturing) 10.35 0.26 4.23 24 Manufacture of other non-metallic mineral products (Manufacturing) 8.58 0.71 2.86 23 Manufacture of paper and paper products (Manufacturing) 8.78 0.77 3.24 17 Repair and installation of machinery and equipment (Manufacturing) 9.07 0.84 0.94 33 Manufacture of rubber and plastics products (Manufacturing) 9.08 0.91 4.82 22 Manufacture of fabricated metal products, except machinery and equipment (Manufacturing) 9.15 1 2.92 25 Manufacture of chemicals and chemical products (Manufacturing) 9.39 1.03 3.63 20 Manufacture of beverages (Manufacturing) 8.82 1.08 3.66 11 Manufacture of leather and related products (Manufacturing) 9.69 1.27 3.24 15 Manufacture of food products (Manufacturing) 8.44 1.48 42.76 10 Services (Top 10 only) Travel agency, tour operator, reservation service and related activities (Services) 9.74 -0.35 2.77 79 Publishing activities (Services) 10.11 -0.05 0.67 58 Retail trade, except of motor vehicles and motorcycles (Services) 9.25 0.3 5.91 47 Wholesale trade, except of motor vehicles and motorcycles (Services) 9.49 0.38 6.72 46 Insurance, reinsurance and pension funding, except compulsory social security (Services) 10.97 0.67 1.3 65 Warehousing and support activities for transportation (Services) 10.05 0.68 1.78 52 Financial service activities, except insurance and pension funding (Services) 10.04 1 3.46 64 Wholesale and retail trade and repair of motor vehicles and motorcycles (Services) 9.23 1 2.37 45 Telecommunications (Services) 10.53 1.23 2.24 61 Social work activities without accommodation (Services) 10.19 2.89 1.01 88 Source: Authors’ own elaboration from Census of Industrial Production and Integrated Survey of Services 26 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 3. The Productivity Diagnostic Distortions and misallocation products to the US are between 163 percent and 195 percent for the manufacturing sector The second direct measure of allocative and between 143 percent and 163 percent efficiency involves the calculation of firm for the services sector (assuming a product distortions and the productivity gains a elasticity of substitution of 3 and 5).6 country can obtain by removing those distortions. Cirera et al. (2015) directly These are very large numbers and suggest estimate these distortions, and Table 10 shows that understanding the nature of these the average value of distortions in factor distortions and effectively removing them is prices and product prices, both in an absolute likely to have very large returns. Identifying sense and weighted by employment. The the nature of these distortions is not an easy higher the number, the higher the distortion. task, however, since different regulations and Table 10 shows that capital distortions are market failures affect the allocation of factors higher than output distortions. Capital of production. Cirera et al. (2015) suggest distortions are higher in the services sector that the relative costs from corruption and than manufacturing. However, the reverse lack of access to finance is likely to explain occurs in terms of output distortions, which a significant part of these distortions, are almost negligible in the services sector. although the data doesn’t allow the authors to empirically confront this theory. There are large efficiency gains from removing distortions Key takeaways from the overall Kenyan Once the distortions have been calculated, productivity analysis one can conduct two counterfactual Before moving into the final section, exercises: (i) what would have been the size which presents the main conclusions and of a firm under the absence of distortions; policy recommendations, we would like and (ii) how would this affect the allocation to summarize the key takeaways from the of labor and capital across firms and sectors? overall productivity analysis for Kenya. The following bullet points present the main Estimations by Cirera et al. (2015) show that conclusions for the cross-sectors and within- efficiency gains from equalizing marginal sector analysis. Table 10: Capital and output distortion averages Weighted ave. Weighted ave. Ave. capital Ave. output capital output distortion distortion distortion* distortion* Aggregate 1.79 0.05 0.44 -0.35 Manufacturing 1.52 0.27 1.54 0.37 Services 1.90 -0.03 -0.39 -0.90 Source: Cirera et al. (2015) *Weighted by Employment 6 This is an assumption that needs to be made in order to calculate the efficiency gains. In the analytical framework used by HK, there is a constant elasticity of substitution (CES) between goods, which comes from consumers’ preferences, and it’s reflected in the demand for each good. The CES assumption for consumers’ preferences means that the utility function has a constant percentage change in goods proportions (e.g., x and y) due to a percentage change in marginal rate of substitution. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 27 3. The Productivity Diagnostic 1. Cross-sector analysis: Large firms tend to display a better a. Overall, labor efficiency gains in Kenya performance than small firms in terms seem to be most likely obtained through of technical efficiency, which may be technological change, including explained by the fact that large firms innovation and technology adoption, are more capable of overcoming the rather than through the accumulation large fixed costs of innovating. of capital. • Age: (i) Mature firms are more b. Services firms appear to be more productive in terms of labor and productive than manufacturing firms. technical efficiency than young c. Misallocation of labor is observed firms in the manufacturing sector; both in the services and manufacturing (ii) however, revenue TFP is larger sectors, although it is less prevalent in for young firms, suggesting that the services sector. distortions have a larger and more favorable impact on these firms than d. Allocative efficiency varies significantly on larger firms. across sub-sectors, both within the services and manufacturing sectors. • Export-status: (i) Exporters display a better productivity-performance e. There are very large efficiency gains than non-exporter firms; (ii) The from removing policy distortions both in productivity gap between exporters the services and manufacturing sectors. and non-exporters is larger when 2. Within-sector analysis: both groups are compared according a. There is a lot of dispersion in firm to their technical efficiency instead productivity-performance within the of labor efficiency, suggesting that services and manufacturing sectors. investments in technological change This dispersion may be a consequence (e.g., innovation and technology of several factors, including: (i) policy adoption) can be more effective than distortions; (ii) different product mark- investments in capital to help firms ups due to differences in market power; achieve the minimum level of efficiency (iii) different intermediate inputs/factors required to overcome the fixed costs of production costs due to differences of exporting; (iii) interestingly, there in firms’ bargaining power; and/or (iv) are no significant differences between different adjustments costs. exporters and non-exporters when measuring productivity through b. Technological gaps are larger than TFPR, suggesting that the pass- labor efficiency gaps. through of efficiency gains to final c. There is a lot of heterogeneity prices obtained by exporting firms, in firm productivity-performance probably ex-ante to enter into across different types of firms in the international markets, is high; (iv) manufacturing industry: Alternatively, the lack of differences • Size: (i) Small firms tend to produce between both TFPR distributions may more value-added per worker reflect the fact that non-exporting than large firms but differences in firms face relatively more distortions labor productivity across different than exporting firms. size groups are not significant; (ii) 28 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC 4. IMPLICATIONS FOR POLICY T his note has provided a detailed description of firm-level productivity in the manufacturing and services sectors in an efficient allocation of labor and capital across firms as more than 160 percent (Cirera, Fattal-Jaef, Maemir 2015). This could have a Kenya. The results of the analysis portrays very large impact on economic growth, and a picture of firm dynamics in Kenya that is more importantly, on the creation of high significantly dysfunctional and results in quality jobs that the country needs to absorb the fact that there are significant market a rapidly increasing labor force. distortions that make firms in Kenya co- exist at very different productivity levels, The findings in this report also have even within very narrowly defined industries. important implications for the design and The dispersion of revenue TFP may capture implementation of existing policies. First, the presence of economic distortions policies oriented to promote economic both in factor and product markets and/or growth should concentrate on removing heterogeneity of mark-ups across firms. In distortions and leveling the playing field addition, the dispersion in physical TFP may for all types of firms. Although there data reflect the random nature of the innovation limitations do not allow us to identify process and/or the heterogeneity in firms’ the type of distortions that explain the attributes and capabilities, suggesting observed misallocation, there is evidence that economic policies geared to foster that entrepreneurs in Kenya continue to face productivity at the aggregate level should burdensome institutional and regulatory depart from the standard assumption of a barriers that not only affect their ability to representative firm. cover the fixed costs of entering into the domestic market, but also their capacity to More importantly, the results put policies generate post-entry growth. This is a common that enhance productivity and remove problem in several developing countries. distortions at the center stage of the These barriers affect incumbents’ ability to economic policy agenda in Kenya. The increase profits, create new jobs, and supply observed misallocation, reflected for high value-added exports. There is a need instance in the negative covariance between to improve the business environment across productivity and employment, is likely to the board and to ensure a non-discriminatory result in lower aggregate productivity, implementation of business regulations and especially in the manufacturing sectors where administrative requirements across all firms. this misallocation is more prevalent and significantly larger than in some other peer Second, policy targeting should take into countries in Africa. Some estimates calculate consideration the large heterogeneity of the potential productivity increase from attributes and performance across Kenyan removing these distortions and achieving firms. Laggard firms in Kenya are very far away FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 29 4. Implications for Policy from the productivity frontier and therefore, patterns of economic activity at the firm-level. generic policies, targeting very different A clear example is the need to examine firms’ firms with the same instrument, may fail to behavior to rationalize the relatively lackluster achieve the expected objectives. By contrast, performance of the Kenyan manufacturing tailoring the interventions to the specific sector. A significant number of aggregate needs of a particular group of firms can studies have documented a continuous loss be more effective in enhancing aggregate of competitiveness of Kenyan firms, both in efficiency. For example, export promotion regional and international markets, coupled policies that aimed at reducing the large with the declining role of the manufacturing fixed costs of entering into international sector in the overall economy. The analysis markets may fail to boost aggregate exports presented in this microeconomic diagnostic if low productivity firms are targeted, given provides foundations to explain the observed the fact that the distribution of efficiency is aggregate results. highly dispersed. Overall, the findings highlight the importance Third, one implication of the analysis is the of locating productivity at the center of fact that policies aimed at changing the the competitiveness and economic growth behavior of macroeconomic variables should agenda as a key instrument for employment rely on granular diagnostics that explore creation and poverty reduction. 30 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC REFERENCES Aghion,Philippe, Nick Bloom, Richard Blundell, Rachel Griffith, and Peter Howitt. (2005). “Competition and Innovation: An Inverted-U Relationship.” The Quarterly Journal of Economics 120 (2): 701-728. Alfaro, Laura, Andrew Charlton, and Fabio Kanczuk. (2008). “Firm-Size Distribution and Cross-Country Income Differences.” National Bureau of Economic Research Working Paper 14060. Asker, John, Allan Collard-Wexler, and Jan De Loecker. (2014). “Dynamic Inputs and Resource (Mis) Allocation.”The Journal of Political Economy 122 (5). Bartelsman, Eric, John Haltiwanger, and Stefano Scarpetta. (2013). “Cross-country Differences in Productivity: The role of Allocation and Selection.” American Economic Review 103 (1): 305–334. Christiaensen, Luc, Lionel Demery, and Jesper Kuhl. (2011). “The (Evolving) Role of Agriculture in Poverty Reduction—An Empirical Perspective.” Journal of Development Economics 96 (2): 239-54. Christiaensen, Luc, Ruchira Bhattamishra, Lei Pan, and Sangui Wang. (2009). “Pathways Out of Poverty in Lagging Regions: Evidence from Rural Western China.” World Bank, Washington, DC. Cirera, Xavier (2015). "Catching Up to the Technological Frontier?," World Bank Other Operational Studies 21684, The World Bank. Cirera, Xavier, Roberto N. Fattal-Jaef, and Hibret B. Maemir. (2015). “Taxing the good? Distortions, Misallocation, and Productivity in Sub-Saharan Africa.” The World Bank, Mimeo. Dang, Hai-Anh, and Peter Lanjouw. (2012). “Measuring Poverty Dynamics and Labor Transitions with Synthetic Panels Based on Cross-Sections.” World Bank Policy Research Working Paper Series #6504. De Loecker, J. (2011). “Product Differentiation, Multi-Product Firms and Estimating the Impact of Trade Liberalization on Productivity “, Econometrica, Vol. 79, No. 5 (September), 1407–1451. Easterly, William, and Ross Levine. (2001). ”What have we learned from a decade of empirical research on growth? It's Not Factor Accumulation: Stylized Facts and Growth Models.“ World Bank Econ Review15 (2): 177-219. Haltiwanger, John. (2007). “Measuring and Analyzing Aggregate Fluctuations: The Importance of Building from Microeconomic Evidence.” Federal Reserve Bank of St. Louis Review 79 (3): 55–78. Hsieh, Chang-Tai, and Peter Klenow. (2009). “Misallocation and Manufacturing TFP in China and India.” Quarterly Journal of Economics 124 (4): 1403–48. Krugman, Paul. (1994). “The Age of Diminished Expectations.” Cambridge: MIT Press. Melitz, Marc. (2003). "The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity." Econometrica, Econometric Society 71 (6): 1695-1725, November. Midrigan, Virgiliu, and Daniel Yi Xu. (2010). “Finance and Misallocation: Evidence from Plant-level Data.” National Bureau of Economic Research Working Paper 15647. Olley, G. Steven, and Ariel Pakes. (1996). “The Dynamics of Productivity in the Telecommunications Equipment Industry.” Econometrica 64 (6): 1263–97. FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 31 Pages, Carmen. (2010). “The Age of Productivity: Transforming Economies from the Bottom Up.” Inter- American Development Bank. Washington, DC: Palgrave Macmillan. Restuccia, Diego, and Richard Rogerson. (2008). “Policy Distortions and Aggregate Productivity with Heterogeneous Establishments.” Review of Economic Dynamics 11 (4): 707–20. Syverson, Chad. (2011). “What Determines Productivity?”Journal of Economic Literature 49:2, 326-365. Vries, Gaaitzen de, Marcel Timmer, and Klass de Vries. (2013). “Structural Transformation in Africa: Static Gains, Dynamic Losses.” GGDC Research Memorandum 136. 32 A FIRM-LEVEL PRODUCTIVITY DIAGNOSTIC ANNEX 1 Box A1: Productivity variables used in the analysis FOR KENYA’S MANUFAC TURING AND SERVICES SEC TOR 33 World B nk Group D lt C nt r M n n i Ro d, Upp r Hill P. O. Box 30577 – 00100 N irobi, K n T l phon : +254 20 2936000