77484 Measuring and Explaining the Impact of Productive Efficiency on Economic Development Ruwan Jayasuriya and Quentin Wodon A limitation of most empirical cross-country studies that focus on determinants of GDP or GDP growth is that they fail to distinguish explicitly between inputs used in produc- tion and conditions that facilitate production. For example, physical capital, human capital, and labor are production inputs, whereas the quality of institutions, macro- economic stability, and market quality are conditions that facilitate production. This article takes this distinction seriously and uses a stochastic frontier approach to study factors affecting economic performance. A panel data set of 71 countries for the 1980–98 period is used to estimate a production frontier with physical capital, human capital, and labor as inputs. The article also analyzes what drives productive efficiency, using the institutional framework, macroeconomic stability, market quality, and urbaniza- tion as possible explanatory factors. Urbanization turns out to be an important deter- minant, with the rule of law, inflation rate, and market quality also affecting productive efficiency. Measuring economic performance is an issue not only of academic interest but also of practical concern. Numerous cross-country studies that use gross domes- tic product (GDP) levels or growth rate as a yardstick for economic performance have found that conventional factors used to determine output, such as physical and human capital and labor force size, do not fully explain production. Although the results are somewhat sensitive to the model specification, mea- sures of market distortion, macroeconomic environment, political stability, research and development, and the depth of financial markets have all been found to affect economic development (for reviews, see among others Barro and Sala-i-Martin 1995; Sala-i-Martin 1997; Solow 2000; Aron 2000; Easterly 2001). Ruwan Jayasuriya is a senior associate at PricewaterhouseCoopers LLP; his email address is ruwan. jayasuriya@us.pwc.com. Quentin Wodon is lead poverty specialist in the Poverty Reduction and Eco- nomic Management Unit of the Africa Region at the World Bank; his email address is qwodon@world- bank.org. This study was funded by the Research Support Budget at the World Bank. The article benefited from discussions with Christine Fallert Kessides, comments from referees, and suggestions from Alan Winters. THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1, pp. 121–140 doi:10.1093/wber/lhi006 Ó The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oupjournals.org. 121 122 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 I. THE LITERATURE ON PRODUCTIVE EFFICIENCY The focus in the literature has recently shifted to the quality of public and private institutions and of markets in explaining economic performance in cross-country analyses (Brunetti and others 1998; Hall and Jones 1999; Knack and Keefer 1995; Keefer and Knack 1997).1 Although the institutional framework and market structure of a country measure different aspects, they overlap to a considerable degree. Both can be measured by such factors as the quality of bureaucracy, pervasiveness of corruption, rule of law, risk of appropriation, contract repudia- tion, political environment, and civil liberties, and both would be expected to have an impact on production and allocation decisions. Market and institutional defi- ciencies may distort public and private decisionmaking and lead entrepreneurs to undertake wasteful rent-seeking activities that divert time and resources from productive activities, thereby preventing firms from adjusting effectively to techno- logical change. Weak institutions and market structures may result in suboptimal selection and use of inputs. In developing economies, where the potential for industrialization and the potential gains from industrialization are higher, the inability of firms to fully benefit from low-cost access to advanced technology from overseas and better returns to scale (relative to developed economies) may be especially damaging to development. The macroeconomic environment has also received much attention in studies of economic performance. The inflation rate, and to a lesser extent the black market premium, are widely used as proxies for macroeconomic conditions. Numerous theoretical studies have also focused on the costs of inflation (for surveys, see Briault 1995; Temple 2000). These analyses have shown that businesses and households perform poorly when inflation is high and unpre- dictable. Although empirical studies have found some support for the harmful effects of inflation, the evidence is not overwhelming. Inflation rates of 100 percent a year or higher have been found to inhibit economic development, but the impact of moderate inflation is less clear. Urbanization has largely been omitted in models of economic performance, yet the results reported here show it to have a key positive impact on productive efficiency. Urbanization likely influences productive efficiency through a variety of channels (for a review of the role of cities in development, see World Bank 2002, chap. 6). With the presence of universities, research centers, and many firms, cities thrive on learning and innovation, thereby facilitating spillover 1. Brunetti and others (1998), using firm-level data from a private sector survey in 73 countries to gauge the environment faced by local businesses, find that the institutional framework is crucial in explaining differences in economic performance. Hall and Jones (1999) also find that good institutions and sound policies help for economic development by supporting entrepreneurial activities, capital accumulation, invention, skill acquisition, and technology transfers. Aiming to explain why poor coun- tries are falling behind rather than catching up with wealthy nations, Keefer and Knack (1997) also conclude that deficient institutions and government policies lead to poor performance. Jayasuriya and Wodon 123 effects (Glaeser and others 1992; Adams 2001). Personal contacts remain important in the digital age, and they are easier to maintain in cities (Wheeler and others 2000; Glaeser 1998; Lall and Ghosh 2002). Cities lead to economies of scale, encourage the division of labor, and provide a better environment for matching skills with needs (Quigley 1998; Mills 2000; Ciccone and Hall 1996). Cities also make access to education, health services, and infrastructure easier because costs tend to be lower and competition is greater. One limitation of most cross-country studies is that they lump together all the independent variables in regressions that focus on the determinants of GDP levels or growth rates. Yet not all independent variables are the same. Variables such as physical capital, human capital, and labor are production inputs, but others such as the quality of institutions, market structures, or macroeconomic management are conditions that facilitate production, not inputs. This article takes this distinction seriously. It estimates a production function that depicts optimal output levels given input use and measures economic performance using the productive efficiency of reaching optimal output. This framework is used to analyze the determinants of efficiency that facilitate the production process. A range of institutional, macroeco- nomic, and market quality variables as well as the level of urbanization are explored. The work on analyzing productive efficiency dates back to empirical work by Farrell (1957). Over time two broad approaches have been used in production frontier estimation: deterministic methods and stochastic techniques.2 The deterministic methods, data envelopment analysis and the free disposal hull, apply linear programming techniques to construct a frontier by using a piece- wise linear envelope that connects best performers.3 The main advantage of the deterministic methods is that none or few restrictions are imposed on the production technology, but the disadvantage is the inability to disentangle white noise from the inefficiency measures. In the stochastic techniques random shocks are incorporated that account for some of the deviations from the production frontier. Following Aigner and others (1977) and Meeusen and van den Broeck (1977), the first technique (the error components model) assumes that the error term has two components: one for white noise and one a nonnegative component for inefficiency (Battese and Coelli 1992, 1995). The second technique applies fixed-effects and random-effects methods to measure efficiency, with the effects being nonnegative (Cornwell and others 1990; Kumbhakar 1990). 2. As noted by an anonymous referee, in the 1960s there was much activity estimating production functions and using the results to assess relative levels of efficiency (see, for example, Arrow and others 1961). Economists were reluctant to introduce noninput variables directly into the production function but would compute the residual and try to explain it through various factors (see, for example, Denison 1964). Covariance analysis, or panel estimation, was used to allow for unobserved country-specific effects, and some researchers regarded the dummy variables as measures of efficiency to be explained. 3. On data envelopment analysis, see Charnes and others (1978), Fa ¨ re and others (1994), Coelli (1995), Tulkens and Vanden Eeckhaut (1995), and Kumar and Russell (2002). On free disposal hull analysis, see Deprins and others (1984) and Tulkens (1993). 124 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 II. METHODOLOGY This article estimates a production frontier using an extension to panel data of the error components model of Aigner and others (1977) proposed by Battese and Coelli (1992, 1995). Similar to the augmented neoclassical model, the model uses physical capital, human capital, and labor force size as production inputs. The production frontier, given input use, depicts the optimal output level, whereas country-level productive efficiency is measured by comparing actual outcome and optimal outcome. The model estimates the impact on productive efficiency of the institutional framework, macroeconomic stability, market quality index (reliance on market mechanisms in the production process and allocation of resources), and level of urbanization. Other efforts have been made recently to analyze the role of various factors in determining productive efficiency. For example, Kumar and Russell (2002) use the data envelopment analysis estimation approach and focus on labor produc- tivity growth for 57 countries over 1965–90. They construct a world production frontier and decompose labor productivity growth into technological change, improvements in efficiency, and capital accumulation. This article uses the stochastic frontier approach instead and focuses on a different outcome, namely, GDP. It calculates productive efficiency by estimating a world produc- tion frontier and attempts to explain what drives productive efficiency. Rather than explicitly discussing the impact of technological change, improvements in efficiency, and capital accumulation on country-level production, it focuses primarily on factors driving improvements in efficiency. The stochastic frontier approach proposed in Battese and Coelli (1992, 1995) and panel data are used to estimate a production possibilities frontier to deter- mine optimal GDP outcomes given input use. The model estimated in this study is discussed in Kumbhakar and Lovell (2000), and a generalized production frontier approach to estimating inefficiency can be found in Kumbhakar and others (1991). Comparing a country’s actual GDP outcome with the optimal GDP outcome derived from the production frontier yields a measure of economic performance—productive efficiency. This estimation framework can be used to quantify the impact of the institutional structure, macroeconomic environment, market quality, and urbanization on a country’s economic performance in reaching optimal GDP outcomes. Let Yit represent real GDP for country i at time period t, and let Xit depict the inputs used in production: physical capital, human capital (years of schooling), and number of workers. The log-log specification is used in the estimation. Incorporating the period variable (t) captures the impact of technology improve- ments. Over time, technology is expected to improve and cause an outward shift in the production frontier. As a result, the parameter corresponding to the period variable is expected to be positive (and is also used as a robustness test for the stability of the impact of other variables). To enable the production frontier to vary by region, regional dummy variables (DRegion) are used for Asia, Jayasuriya and Wodon 125 Latin America and the Caribbean, Middle East and North Africa, and North America and Europe, with Africa as the omitted region (see appendix table A.1 for the list of countries by region).4 These regional dummy variables enable testing for the robustness of the findings with and without their inclusion in the specification. The regional dummy variables are also important because, even after controlling for inputs and efficiency, other regional (or country-level) factors may affect GDP.5 For example, as noted by Nelson and Pack (1999), rapid growth in some Asian economies was accompanied by ‘‘productive assim- ilation’’ or shifts in the size of firms and sectors of specialization, which led to changes in economic structure and higher growth. The production frontier is estimated as follows: X ð1Þ ln Yit ¼ þ ln Xit þ t Period þ gRegion DRegion þ ðvit À uit Þ i = Country 1, . . . , Country N and t = 1, . . . , T. The technical inefficiency effects are estimated as: X uit ¼ 0 þ jInstitutinal ZInstitutional it þINF Inflationit ð2Þ þ MKT Marketit þ URB Urbanit þ wit uit ¼ Ã Zit  þ wit Estimating equations 1 and 2 separately leads to biased results (Wang and Schmidt 2002), and thus a one-step procedure is used in the estimation. The error term in the production frontier presented in equation 1 consists of two components: the random noise term (vit) that accounts for random shocks and measurement errors, and the nonnegative term (uit) used to measure ineffi- ciency. The vit and the uit terms are assumed to be independent. The vit term is assumed to be iid N(0,sv2). The nonnegative uit term that depicts deviation from the optimal (best practice) outcome is assumed to be independently dis- tributed of the factor inputs (X), and as modeled in equation 2 is a function of country-specific variables that vary over time. The uit term is obtained from a truncated (at zero) normal distribution with a variance of su2, but with means that are a linear function of the observable country-specific variables.6 The production frontier presented in equation 1 is in terms of known input and output variables, whereas the inefficiency terms are assumed to be a function of 4. Any geographic grouping is somewhat arbitrary, but the regional groups probably capture some relevant commonalities across countries. 5. The authors are grateful to an anonymous referee for pointing this out. 6. The uit term is assumed to be independently distributed and is obtained from a truncated normal distribution with a mean of *Zitd and a variance of su2 (Battese and Coelli 1995). The derivation of the log-likelihood function using the distributional assumptions on the uit and vit terms and the maximum likelihood estimation approach used in the estimation are discussed in Battese and Coelli (1995). 126 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 an unknown vector of coefficients, d, and a known set of the institutional, macroeconomic, and market quality variables along with urbanization. Indices of bureaucratic quality, prevalence of corruption, and rule of law are used as institutional variables. The inflation rate is used as a proxy for macroeconomic stability. The reliance of a country on market mechanisms in the production process and allocation of resources is proxied using a market quality index. The productive efficiency measure of country i at time period t is defined as:7 À � E Yit jXit ; DRegion ; uit ð3Þ Efficiencyit ¼ À � E Yit jXit ; DRegion ; uit ¼ 0 i = Country 1, . . . , Country N and t = 1, . . . , T The numerator depicts the observed country i outcome in period t at a given level of input use, Xit. The denominator represents the corresponding optimal (or best practice) outcome for country i in period t, which implies no ineffi- ciency (uit = 0). The maximum likelihood estimation method is used to simultaneously esti- mate parameters of the stochastic frontier (equation 1) and the model for technical inefficiency effects (equation 2).8 III. DATA The study uses data for 71 countries for 1980–98. All variables are averaged over five-year intervals (1980–84, 1985–89, 1990–94, and 1995–98) to reduce the impact of short-run fluctuations on the estimated parameters (to capture long-term effects). There are two groups of variables: one for estimating the production frontiers and one for explaining country efficiency in producing output. The first group of variables consists of real GDP, real domestic capital stock, average years of schooling (a proxy for a country’s stock of human capital), and total number of workers. Data on real GDP and total number of workers (a country’s employment base) are from Penn World Table 6.0, compiled by Heston and Summers (1996). Real GDP is in constant U.S. dollars at purchasing power parity (PPP) terms (chain index; expressed in international prices, base 1996). The Heston and Summers real GDP measures account for and assign suitable weights for cross-country price differences of various components of GDP, which enables meaningful cross-country comparisons. Real domestic capital stock data are from Kraay and others (2001) in constant U.S. dollars 7. The conditional expectation of the uit term, conditional on the observed value of vit À uit, is used in calculating the efficiency measures (Battese and Coelli 1992, 1995). 8. FRONTIER version 4.1 is used in the estimation (Coelli 1996). Jayasuriya and Wodon 127 in PPP terms.9 The human capital data are from the educational attainment database compiled by Barro and Lee (2000). The second group of variables consists of country-level data on the institu- tional framework, macroeconomic stability, market quality, and urbanization. Indices of bureaucratic quality, prevalence of corruption, and rule of law are used to proxy a country’s institutional framework. Data on these indices are from the International Country Risk Guide (ICRG) published regularly by Poli- tical Risk Services (2004; see www.prsgroup.com/icrg/icrg.html). Data on the structure of the economy and use of markets variable used to measure a country’s market quality are from the Economic Freedom of the World: 2001 Annual Report (Gwartney and Lawson 2001). Data on the inflation rate, used as a proxy for a country’s macroeconomic stability, and urbanization are from the World Development Indicators database (World Bank 2001). The ICRG indices are subjective assessments by a worldwide network of experts. To ensure coherence and cross-country comparability, the indices are subject to peer review. The bureaucratic quality index measures the strength and expertise of bureaucrats and their ability to manage political alterations without drastic interruptions in government services or policy changes. Higher values of this index indicate greater bureaucratic quality. The rule of law index assesses the strength and impartiality of the legal system and popular observance of the law. Higher values of this index indicate more effective enforcement and greater adherence to the law. The corruption index measures actual or potential corrup- tion within the political system, which distorts the economic and financial environment, reduces government and business efficiency by enabling indivi- duals to assume positions of power through patronage rather than ability, and introduces inherent instability in the political system. Higher values indicate a decreased prevalence of corruption. The three indices, which use different rating systems, have been normalized to take values between 0 and 100, with higher values indicating better outcomes. Inflation as measured by the consumer price index reflects the annual percen- tage change in the cost to the average consumer of acquiring a fixed basket of goods and services. A country’s market power is proxied by the structure of the economy and use of markets variable. The share of the public sector in industry and investment, use of price controls, and top marginal tax rates are incorpo- rated in this index. This index has been normalized to take values between 0 and 100, with higher values indicating the existence of more effective market struc- tures. Urbanization data refer to the urban population as a share of the total population. 9. As described in Kraay and others (2001), initial estimates of domestic capital stock are obtained from the Penn World Table data set. Flow data on gross domestic investments are then used to construct time series of capital stock measures valued in constant U.S. dollars at PPP (base year 1990). Although it could be interesting to test for the sensitivity of the results here to alternative choices for the measurement of capital (say, not using PPP-based data), such alternative measures are not available. 128 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 Summary statistics indicate that in general, input use increased over the 1980–98 period, although with differences across regions (table 1). The Africa region is the least endowed in physical capital (the stock of physical capital even declined), human capital, and number of workers and thus has the lowest optimal output among all regions. In contrast, the North America and Europe region has high and increasing endowment levels and is thus able to reach the highest optimal output levels. The Asia, Latin America and Caribbean, and Middle East and North Africa regions in general have also experienced a steady increase in input endowments over time. Summary statistics indicate that the variables used for the determinants of efficiency improved over the 1980–98 period, although again with differences across regions (table 2). As was the case for input endowments, the Africa region has the smallest magnitudes for the efficiency determinant variables, implying greater potential for efficiency enhancements in the region from improvements in the institutional framework, macroeconomic stability, and market quality and from greater urbanization. The North America and Europe region has high magnitudes for the efficiency-determinant variables and has had steady improvements over time. The Asia, Latin America and Caribbean, and Middle East and North Africa regions generally also experienced steady improvements in the determinants of efficiency variables during this period. IV. RESULTS The estimation results can be divided into two broad categories: production frontier estimates and determinants of efficiency. Results are presented for four different specifications to provide tests for robustness. The specifications vary by whether a time trend is used in the production frontier or in the determinants of inefficiency and whether regional dummy variables are included in the production frontier. Regional differences as captured by regional dummy variables could themselves reflect differences in conditions that facilitate production, such as the bureaucratic quality index or the market quality index.10 For example, if African countries have a lower efficiency level, this may not be because of ‘‘Africanness’’ but possibly because of factors influencing efficiency (for example, lower rule of law index or a higher black market premium). At the same time there may still be real differences in efficiency related to geographic location. To deal with this issue, the model is estimated with and without regional dummy variables. The parameter estimates for the production frontiers show that a country’s physical capital stock and number of workers have a positive and statistically significant affect on GDP levels (table 3). Given the log-log specification for key inputs, the associated parameters represent elasticities. A 1 percentage point increase in the level of capital stock leads to a 0.38–0.43 percentage point increase in GDP. A 1 percentage point increase in the number of workers leads 10. The authors are grateful to an anonymous referee for pointing this out. T A B L E 1 . Summary Statistics for Production Function Variables, 1980–98 1980–84 1985–89 1990–94 1995–98 Africa region Number of observations 5 10 13 7 GDP (constant 1996 US$ at PPP; billions) Mean 16.71 16.10 14.57 19.50 Capital stock (constant 1990 US$ at PPP; billions) Mean 18.21 12.47 10.52 13.32 Years of schooling Mean 2.29 2.47 2.87 3.75 Workers (thousands) Mean 5,859 5,340 5,869 6,129 Asia region Number of observations 11 13 12 14 GDP (constant 1996 US$ at PPP; billions) Mean 392.33 568.48 786.71 876.24 Capital stock (constant 1990 US$ at PPP; billions) Mean 745.60 1,044.20 1,518.83 1,755.54 Years of schooling Mean 5.46 5.70 6.46 6.69 Workers (thousands) Mean 48,481 95,543 110,871 103,011 Latin America and Caribbean region Number of observations 15 19 18 19 GDP (constant 1996 US$ at PPP; billions) Mean 136.15 126.15 149.14 166.30 129 Capital stock (constant 1990 US$ at PPP; billions) Mean 200.38 184.84 242.77 278.42 Years of schooling Mean 4.39 4.80 5.17 5.39 Workers (thousands) Mean 7,392 7,026 8,439 9,088 Middle East and North Africa region Number of observations 3 3 5 5 GDP (constant 1996 US$ at PPP; billions) Mean 72.17 86.11 119.44 144.66 Capital stock (constant 1990 US$ at PPP; billions) Mean 55.11 77.99 313.12 115.60 Years of schooling Mean 4.42 4.81 4.87 5.46 Workers (thousands) Mean 5,100 5,856 7,612 8,325 North America and Europe region Number of observations 14 17 17 18 GDP (constant 1996 US$ at PPP; billions) Mean 645.23 678.10 760.07 819.12 Capital stock (constant 1990 US$ at PPP; billions) Mean 1,453.91 1,467.97 1,657.27 1,674.34 Years of schooling Mean 7.56 8.06 8.63 9.01 Workers (thousands) Mean 16,392 15,316 16,104 15,937 Source: Heston and Summers (1996); Barro and Lee (2000); Kraay and others (2001); World Bank (2001). 130 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 T A B L E 2 . Summary Statistics for Determinants of Inefficiency, 1980–98 1980–84 1985–89 1990–94 1995–98 Africa region Bureaucratic quality index Number of observations 5 10 13 7 Mean 33.33 47.00 46.67 47.62 Minimum 16.67 20.00 26.67 16.67 Maximum 50.00 66.67 66.67 62.50 Corruption index Number of observations 5 10 13 7 Mean 36.67 43.00 48.46 45.24 Minimum 0.00 0.00 0.00 33.33 Maximum 66.67 66.67 63.33 50.00 Rule of law index Number of observations 5 10 13 7 Mean 23.33 40.00 43.07 61.31 Minimum 16.67 16.67 13.33 37.50 Maximum 33.33 83.33 83.33 75.00 Inflation Number of observations 5 10 13 7 Mean 32.32 37.37 29.05 20.33 Minimum 13.56 2.73 4.35 5.68 Maximum 70.28 155.25 122.19 37.13 Market quality index Number of observations 5 10 13 7 Mean 23.80 22.60 19.77 46.64 Minimum 17.00 13.00 0.00 34.50 Maximum 30.00 31.00 43.00 60.50 Urbanization Number of observations 5 10 13 7 Mean 22.14 27.82 31.86 34.04 Minimum 9.62 10.42 11.72 13.35 Maximum 31.64 39.58 55.40 49.00 Asia region Bureaucratic quality index Number of observations 11 13 12 14 Mean 59.09 59.74 63.33 69.35 Minimum 16.67 16.67 30.00 37.50 Maximum 100.00 100.00 100.00 100.00 Corruption index Number of observations 11 13 12 14 Mean 48.48 50.00 59.44 58.63 Minimum 0.00 0.00 33.33 33.33 Maximum 100.00 100.00 96.67 91.67 Rule of law index Number of observations 11 13 12 14 Mean 56.06 50.26 61.39 78.57 Minimum 16.67 13.33 16.67 50.00 Maximum 100.00 100.00 100.00 100.00 Inflation Number of observations 11 13 12 14 Mean 10.41 7.18 7.25 6.60 Jayasuriya and Wodon 131 Minimum 3.91 1.15 2.00 0.60 Maximum 18.44 14.76 13.05 20.44 Market quality index Number of observations 11 13 12 14 Mean 34.09 35.23 46.58 51.18 Minimum 17.00 13.00 19.00 20.00 Maximum 53.00 56.00 79.00 92.00 Urbanization Number of observations 11 13 12 14 Mean 45.26 44.08 48.83 52.84 Minimum 15.44 17.92 19.22 20.80 Maximum 85.68 85.34 84.94 100.00 Latin America and Caribbean region Bureaucratic quality index Number of observations 15 19 18 19 Mean 40.00 38.95 43.15 48.68 Minimum 16.67 16.67 16.67 12.50 Maximum 66.67 66.67 66.67 70.83 Corruption index Number of observations 15 19 18 19 Mean 46.67 46.49 50.93 50.66 Minimum 16.67 3.33 26.67 33.33 Maximum 83.33 83.33 80.00 79.17 Rule of law index Number of observations 15 19 18 19 Mean 41.11 42.28 49.07 57.90 Minimum 16.67 16.67 23.33 33.33 Maximum 66.67 66.67 70.00 83.33 Inflation Number of observations 15 19 18 19 Mean 52.82 62.23 56.99 17.20 Minimum 5.81 0.49 8.19 1.25 Maximum 178.46 219.47 432.78 61.41 Market quality index Number of observations 15 19 18 19 Mean 41.47 38.47 46.00 61.08 Minimum 17.00 19.00 0.00 27.50 Maximum 83.00 80.00 74.00 89.50 Urbanization Number of observations 15 19 18 19 Mean 58.73 60.92 63.76 64.58 Minimum 36.02 37.92 38.30 34.15 Maximum 86.00 87.80 89.22 90.65 Middle East and North Africa region Bureaucratic quality index Number of observations 3 3 5 5 Mean 44.44 55.56 55.33 63.33 Minimum 33.33 50.00 43.33 50.00 Maximum 50.00 66.67 80.00 95.83 Corruption index Number of observations 3 3 5 5 Mean 55.55 55.55 56.67 59.17 Minimum 33.33 33.33 46.67 41.67 Maximum 83.33 83.33 80.00 70.83 (Continued) 132 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 TABLE 2. Continued 1980–84 1985–89 1990–94 1995–98 Rule of law index Number of observations 3 3 5 5 Mean 38.89 38.89 52.00 78.33 Minimum 33.33 33.33 43.33 66.67 Maximum 50.00 43.33 60.00 83.33 Inflation Number of observations 3 3 5 5 Mean 67.40 36.02 12.38 10.79 Minimum 8.90 7.32 5.85 4.08 Maximum 177.53 81.82 20.63 28.78 Market quality index Number of observations 3 3 5 5 Mean 11.67 16.33 20.00 27.80 Minimum 10.00 11.00 12.00 21.00 Maximum 14.00 21.00 30.00 39.00 Urbanization Number of observations 3 3 5 5 Mean 61.78 63.14 64.18 66.51 Minimum 43.84 43.98 44.22 44.80 Maximum 89.08 90.00 90.46 90.95 North America and Europe region Bureaucratic quality index Number of observations 14 17 17 18 Mean 86.91 86.47 89.41 93.29 Minimum 50.00 50.00 56.67 70.83 Maximum 100.00 100.00 100.00 100.00 Corruption index Number of observations 14 17 17 18 Mean 86.90 87.26 85.88 83.33 Minimum 50.00 66.67 56.67 62.50 Maximum 100.00 100.00 100.00 100.00 Rule of law index Number of observations 14 17 17 18 Mean 88.10 87.25 91.18 96.53 Minimum 50.00 46.67 70.00 87.50 Maximum 100.00 100.00 100.00 100.00 Inflation Number of observations 14 17 17 18 Mean 12.12 5.42 4.66 2.26 Minimum 5.05 0.69 2.08 0.79 Maximum 22.76 17.19 16.21 6.86 Market quality index Number of observations 14 17 17 18 Mean 31.29 36.24 48.00 58.00 Minimum 10.00 14.00 17.00 40.00 Maximum 53.00 72.00 79.00 83.00 Urbanization Number of observations 14 17 17 18 Mean 71.74 71.73 73.01 75.55 Minimum 32.52 41.00 50.58 58.45 Maximum 95.64 96.20 96.70 97.15 Source: PRS (2004); Gwartney and Lawson (2001); World Bank (2001). T A B L E 3 . Joint Estimation for the Production Frontier and the Determinants of Inefficiency Model I Model II Model III Model IV Production frontier Constant 2.8794 (7.70) 3.3083 (8.05) 2.6540 (6.90) 3.0903 (7.65) Log(capital stock) 0.4173 (16.12) 0.3813 (14.46) 0.4283 (15.69) 0.3929 (15.24) Log(years of schooling) 0.0223 (0.36) 0.0538 (0.97) 0.0009 (0.01) 0.0358 (0.66) Log(workers) 0.5949 (19.71) 0.6137 (21.61) 0.5860 (17.97) 0.6015 (21.40) Period 0.0209 (1.55) 0.0148 (1.20) Dummy variables (Sub-Saharan Africa omitted) Asia 0.1723 (2.38) 0.1885 (2.59) Latin America and Caribbean 0.1316 (2.14) 0.1485 (2.49) Middle East and North Africa 0.4583 (5.51) 0.4543 (5.67) North America and Europe 0.3606 (4.62) 0.3736 (4.81) 133 Determinants of inefficiency Constant 1.5358 (9.96) 1.3380 (8.98) 1.6090 (10.10) 1.3689 (8.78) Bureaucratic quality index 0.0828 (0.32) 0.0230 (0.09) 0.0362 (0.14) À0.0328 (À0.14) Corruption index 0.0172 (0.08) 0.2045 (0.94) À0.0438 (À0.18) 0.1777 (0.88) Rule of law index À0.6373 (À2.51) À0.4953 (À2.10) À0.4226 (À1.75) À0.3338 (À1.48) Inflation 0.2016 (3.04) 0.1150 (1.87) 0.2060 (3.09) 0.1176 (1.82) Market quality index À0.3851 (À2.01) À0.5895 (À3.11) À0.2437 (À1.25) À0.4613 (À2.45) Urbanization À1.8223 (À7.62) À1.5690 (À5.48) À1.8814 (À7.28) À1.6037 (À5.28) Period 0.0650 (1.77) 0.0489 (1.52) Log likelihood function 107.48 108.51 100.80 101.68 Number of observations 238 238 238 238 Note: Numbers in parentheses are t-statistics. Source: Authors’ estimation based on data described in text. 134 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 to a 0.59–0.61 percentage point increase in GDP. Human capital (measured by the number of years of schooling) has a positive, but insignificant impact on GDP, which is somewhat surprising.11 The regional dummy variables are statistically significant, with each region having a higher production frontier than Africa, the excluded region.12 These results indicate that increasing input use (or factor endowments) is one of the means of reaching higher optimal output levels. The time period variable, when included in the production possibilities frontier, is positive with a significance level (p-value) of 0.1229 in model III and 0.2325 in model IV. These results (especially for model III) depict improvements in technol- ogy during the 1980–98 period and indicate that with each additional period (and controlling for other factors) the production possibilities frontier shifts outward by 1.5–2.1 percentage points over the level reached in the previous period. The analysis of the impact of the institutional framework, macroeconomic stabi- lity, market quality, and urbanization on a countries’ productive efficiency does not incorporate differences in industry mix across countries because of a lack of reliable data.13 Improvements in the rule of law index and the market quality index and a decrease in inflation lead to decreases in inefficiency (see table 3). Inefficiency declines by 1.57–1.88 percentage points for a 1 percentage point increase in urba- nization, by 0.42–0.64 percentage point for a 1 percentage point increase in the rule of law index, by 0.38–0.59 percentage point for a 1 percentage point increase in the market quality index, and by 0.12–0.21 percentage points for a 1 percentage point decrease in the inflation rate. These results suggest that improvements in urbaniza- tion, institutional framework, market quality, and macroeconomic stability that lead to better productive efficiency outcomes are another way to boost output levels.14 In recent years indices of bureaucratic quality, corruption, rule of law, and market quality have improved in many countries. The level of urbanization has also increased, and inflation rates have declined. This implies that many countries 11. An anonymous referee suggested including the education variable instead as a determinant of efficiency, but this study follows the literature in keeping education as an input to the production function. 12. Note that the Middle East and North Africa dummy variable is higher than the North America and Europe dummy variable, which would imply that more can be produced in the Middle East and North Africa than in North America and Europe with the same level of inputs. One potential explanation for this apparently counterintuitive finding could be the impact on GDP of oil and tourism in the countries included in the Middle East and North Africa region (although one could argue that these industries also require high levels of inputs). 13. As indicated by an anonymous referee, industry mix may have an impact on productive efficiency because of differences in productivity across sectors. For example, countries with a higher share of GDP in the primary sector may be less efficient, and this will not be captured in the estimation of the frontier. However, the use of urbanization in the analysis of the determinants of efficiency may be considered a proxy for sectoral shares of GDP, because a higher level of urbanization is typically a sign of an economy with a lower emphasis on agriculture (but as mentioned earlier, there may also be other reasons for more efficient production in countries that are more urbanized). 14. As pointed out by an anonymous referee, productivity (and hence efficiency) and per capita output may themselves have an impact on urbanization. Unfortunately, the authors are unaware of any mechanism to incorporate appropriate instruments to correct for such potential endogeneity using the stochastic frontier approach. This must bet left for future work. Jayasuriya and Wodon 135 have achieved greater productive efficiency. Indeed, summary statistics at the region level on the estimated efficiency measures for models I and II show an increase in efficiency during the period under review (table 4).15 Yet levels of productive efficiency, like input use, vary by region, with the lowest levels in the Africa region and the highest in North America and Europe over the period under review. These results suggests that both high input endowments and greater produc- tive efficiency have played a key role in North America and Europe’s success in achieving high output levels, whereas lack of input endowments and compara- tively low productive efficiency have contributed to the Africa region’s poorer performance. It is heartening to note, however, that the Africa region has experienced strong improvements in productive efficiency over the period under review, with this result being robust to the choice of specification. Although the Africa region has the smallest magnitudes for the efficiency-determinant variables (see table 2), it experienced relatively greater improvements in these variables, with greater impacts on efficiency (see table 3). As a result, the Africa region experienced strong improvements in productive efficiency during the 1980–98 period. While the sample of countries used in the estimation is unbalanced, that does not seem to affect the efficiency trend results. A general increase in the levels of efficiency is observed for a majority of the countries in the sample, including those in Africa, for all four periods. For countries that are not in the sample for all four periods, there is no evidence of more omissions because of lack of data for countries with high levels of efficiency in the early periods or for countries with low levels of efficiency in the later periods. If there had been such bias, the improvements in efficiency could have been due to the unbalanced nature of the panel. Also, although the use of regional dummy variables in some of the specifica- tions means that the standard for comparing efficiency levels is region-specific, this standard is increasing over time because the regional dummy variable is invariant over time and the time trend in the production function is positive. Therefore, the improvement in productive efficiency in Africa is not just a reflection of a scenario in which even the best-performing countries in the region would be doing poorly. It is worth emphasizing again that the stochastic frontier approach used here requires that the inefficiency terms (uit) be nonnegative, as the terms measure the deviation from the optimal (best practice) outcome. An alternative estimation method would be to substitute equation 2 into equation 1 and then estimate a single regression using traditional estimation techniques (ordinary least squares). This method would not permit imposing efficiency measures, so any comparison of the results with those presented here would need to be done with care. A traditional approach would provide less information because it could not mea- sure efficiency as well. Still, it is worth noting that in the traditional approach, the capital stock and labor force variables remain statistically significant, whereas 15. The estimated efficiency measures for models III and IV are available on request. 136 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 T A B L E 4 . Summary Statistics on the Estimated Efficiency Measures (models I and II) Number of Standard Observations Mean Minimum Maximum Deviation Model I Africa region 1980–84 5 32.24 21.62 41.48 8.64 1985–89 10 39.90 21.62 71.51 15.41 1990–94 13 36.88 21.30 81.61 15.52 1995–98 7 46.65 23.01 83.79 19.54 Asia region 1980–84 11 55.81 29.43 91.45 20.46 1985–89 13 55.13 27.96 90.63 19.87 1990–94 12 59.36 30.75 91.20 20.23 1995–98 14 64.70 35.20 94.40 19.36 Latin America and Caribbean region 1980–84 15 65.64 36.06 88.90 15.35 1985–89 19 66.63 37.45 89.62 15.40 1990–94 18 68.66 41.38 87.22 15.17 1995–98 19 68.52 40.78 91.22 15.90 Middle East and North Africa region 1980–84 3 86.19 79.69 94.35 7.47 1985–89 3 83.04 79.48 89.49 5.60 1990–94 5 81.36 40.88 94.43 22.85 1995–98 5 91.52 83.77 95.69 4.78 North America and Europe region 1980–84 14 85.23 74.28 93.81 6.54 1985–89 17 87.86 74.55 95.34 5.86 1990–94 17 89.07 74.50 96.12 4.93 1995–98 18 92.10 81.49 96.72 4.05 Model II Africa region 1980–84 5 39.92 25.90 51.12 10.89 1985–89 10 48.89 25.83 86.76 18.99 1990–94 13 44.34 25.58 93.95 17.71 1995–98 7 55.30 27.75 95.03 21.61 Asia region 1980–84 11 61.12 33.06 94.76 21.31 1985–89 13 60.43 31.23 95.38 20.79 1990–94 12 65.51 34.34 95.74 21.35 1995–98 14 70.96 40.08 97.24 19.48 Latin America and Caribbean region 1980–84 15 73.60 40.20 94.02 16.64 1985–89 19 74.36 41.71 95.16 16.37 1990–94 18 76.37 46.02 93.12 16.25 1995–98 19 75.67 45.30 96.31 16.62 Jayasuriya and Wodon 137 Middle East and North Africa region 1980–84 3 75.80 67.82 87.42 10.29 1985–89 3 71.10 67.56 78.07 6.04 1990–94 5 72.63 36.98 87.58 20.75 1995–98 5 82.83 72.40 91.45 8.38 North America and Europe region 1980–84 14 81.82 69.25 93.28 8.05 1985–89 17 85.32 70.33 96.19 7.36 1990–94 17 87.05 70.02 96.96 6.22 1995–98 18 91.09 78.40 97.55 5.21 Source: Authors’ estimation based on data described in text. education (years of schooling) remains statistically insignificant.16 The ranking of the regional dummy variables also remains the same. The rule of law index, the market quality index, and the urbanization rate are again statistically significant, and inflation is statistically significant only in some of the specifications. The results are thus broadly similar, but the advantage of the approach taken here is the ability to explicitly separate the impact of production inputs and that of efficiency in using the inputs. V. CONCLUSION There is an extensive literature on identifying and measuring factors that improve economic performance, as measured by GDP levels and growth rates, using cross-country analyses. In contrast to previous studies, the approach used here estimates a production possibilities frontier that depicts optimal output for different levels of input use and calculates efficiency by comparing actual output levels with optimal outputs. This framework permits studying not only how greater input use increases the optimal output levels that can be reached (according to the production frontier) but also how better conditions that facilitate production can help in reaching these optimal output levels. Similar to previous growth studies, the results indicate statistically significant positive relationships between production and levels of physical capital and number of workers employed. The impact of years of schooling is positive in all cases but lacks statistical significance. The production frontier estimation framework shows an impact of the institutional framework, macroeconomic stability, quality of markets, and level of urbanization on productive efficiency. Finally, the results also indicate that average world productive efficiency levels have improved during 1980–98. High input endowments and greater productive efficiency played a key role in North America and Europe’s success, 16. The results of the ordinary least squares estimation are available on request. 138 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 T A B L E A . 1 . Countries by Region Sub-Saharan Latin Middle East and North America Africa Asia America North Africa and Europe Botswana Australia Argentina Egypt, Arab Rep. Austria Cameroon Bangladesh Bolivia Iran, Islamic Rep. Belgium Congo, Rep. China Brazil Israel Canada Ghana Indonesia Chile Jordan Switzerland Kenya India Colombia Tunisia Denmark Malawi Japan Costa Rica Spain Niger Korea, Rep. Ecuador Finland Senegal Sri Lanka Guatemala France Sierra Leone Malaysia Honduras United Kingdom Togo New Zealand Haiti Greece Uganda Pakistan Jamaica Ireland Congo, Dem. Rep. Philippines Mexico Iceland Zambia Singapore Nicaragua Italy Zimbabwe Thailand Panama Netherlands Peru Norway Paraguay Portugal El Salvador Sweden Trinidad and Tobago United States Uruguay Venezuela whereas the Africa region performed poorly due to a lack of input endowments and low productive efficiency. The highest improvement in productive efficiency over time was observed in the Africa region, however, which is promising for the future. REFERENCES Adams, J. D. 2001. Comparative Localization of Academic and Industrial Spillovers. NBER Working Paper 8292. Cambridge, Mass.: National Bureau of Economic Research. Aigner, D. J., C. A. K. Lovell, and P. Schmidt. 1977. ‘‘Formulation and Estimation of Stochastic Frontier Production Function Models.’’ Journal of Econometrics 6(1):21–37. Aron, J. 2000. ‘‘Growth and Institutions: A Review of the Evidence.’’ World Bank Research Observer 15(1):99–135. Arrow, K. J., H. B. Chenery, B. S. Minhas, and R. M. Solow. 1961. ‘‘Capital-Labor Substitution and Economic Efficiency.’’ Review of Economics and Statistics 43(3):225–50. Barro, Robert J., and Jong-Wha Lee. 2000. ‘‘International Data on Educational Attainment: Updates and Implications.’’ Harvard University, Department of Economics, Cambridge, Mass. Barro, Robert, and Xavier Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hill. Battese, G. E., and T. J. Coelli. 1992. ‘‘Frontier Production Functions, Technical Efficiency, and Panel Data: With Applications to Paddy Farmers in India.’’ Journal of Productivity Analysis 3(1/2):153–69. ———. 1995. ‘‘A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function for Panel Data.’’ Empirical Economics 20(2):325–32. Briault, C. 1995. ‘‘The Costs of Inflation.’’ Bank of England Quarterly Bulletin 35(1):33–45. Jayasuriya and Wodon 139 Brunetti, A., G. Kisunko, and B. Weder. 1998. ‘‘Credibility of Rules and Economic Growth: Evidence from a Worldwide Survey of the Private Sector.’’ World Bank Economic Review 12(3): 353–84. Charnes, A., W. W. Cooper, and E. Rhodes. 1978. ‘‘Measuring the Efficiency of Decision Making Units.’’ European Journal of Operational Research 2(6):429–44. Ciccone, C., and R. E. Hall. 1996. ‘‘Productivity and the Density of Economic Activity.’’ American Economic Review 86(1):54–70. Coelli, T. J. 1995. ‘‘Recent Developments in Frontier Modeling and Efficiency Measurement.’’ Journal of Agricultural Economics 39(3):219–45. ———. 1996. ‘‘A Guide to FRONTIER Version 4.1: A Computer Program for Stochastic Frontier Production and Cost Function Estimation.’’ CEPA Working Paper 96/07. The Centre for Efficiency and Productivity Analysis, Armidale, Australia. Cornwell, C., P. Schmidt, and R. C. Sickles. 1990. ‘‘Production Frontiers with Cross-Sectional and Time- Series Variation in Efficiency Levels.’’ Journal of Econometrics 46(1/2):185–200. Denison, E. F. 1964. ‘‘Measuring the Contribution of Education (and the Residual) to Economic Growth’’ and ‘‘Reply.’’ In The Residual Factor in Economic Growth. Paris: Organization for Economic Cooperation and Development, Study Group in the Economics of Education. Deprins, D., L. Simar, and H. Tulkens. 1984. ‘‘Measuring Labor-Efficiency in Post Offices.’’ In M. Marchand, P. Pestieau, and H. Tulkens, eds., The Performance of Public Enterprises: Concepts and Measurement. North Holland: Amsterdam. Easterly, William R. 2001. The Elusive Quest for Growth: Economists’ Adventures and Misadventures in the Tropics. London: MIT Press. ¨ re, R., S. Grosskopf, M. Norris, and Z. Zhang. 1994. ‘‘Productivity Growth, Technical Fa Progress, and Efficiency Change in Industrialized Countries.’’ American Economic Review 84(1): 66–83. Farrell, M. J. 1957. ‘‘The Measurement of Productive Efficiency.’’ Journal of the Royal Statistical Society (A) 120(3):253–82. Glaeser, E. L. 1998. ‘‘Are Cities Dying?’’ Journal of Economic Perspectives 12(2):139–60. Glaeser, E. L., H. D. Kallal, J. A. Scheinkman, and A. Shleifer. 1992. ‘‘Growth in Cities.’’ Journal of Political Economy 100(6):1126–52. Gwartney, James, and Robert Lawson, with Walter Park and Charles Skipton. 2001. Economic Freedom of the World 2001 Annual Report. Vancouver, Canada: Fraser Institute. Available online at www.freetheworld.com. Hall, R. E., and C. I. Jones. 1999. ‘‘Why Do Some Countries Produce So Much More Output per Worker than Others?’’ Quarterly Journal of Economics 114(1):83–116. Heston, A., and R. Summers. 1996. ‘‘International Price and Quantity Comparisons: Potentials and Pitfalls.’’ American Economic Review 86(2):20–24. Keefer, P., and S. Knack. 1997. ‘‘Why Don’t Poor Countries Catch Up? A Cross-National Test of an Institutional Explanation.’’ Economic Inquiry 35(3):590–602. Knack, S., and P. Keefer. 1995. ‘‘Institutions and Economic Performance: Cross-Country Tests Using Alternative Institutional Measures.’’ Economics and Politics 7(3):207–27. Kraay, A., N. Loayza, L. Serven, and J. Ventura. 2001. ‘‘Country Portfolios.’’ CEPR Discussion Paper 2974. Centre for Economic Policy Research, London. Kumar, S., and R. R. Russell. 2002. ‘‘Technological Change, Technological Catch-up, and Capital Deepening: Relative Contributions to Growth and Convergence.’’ American Economic Review 92(3):527–48. Kumbhakar, S. C. 1990. ‘‘Production Frontiers, Panel Data, and Time-Varying Technical Inefficiency.’’ Journal of Econometrics 46(1/2):201–12. Kumbhakar, S. C., and C. A. K. Lovell. 2000. Stochastic Frontier Analysis. Cambridge: Cambridge University Press. 140 THE WORLD BANK ECONOMIC REVIEW, VOL. 19, NO. 1 Kumbhakar, S. C., S. Ghosh, and T. McGuckin. 1991. ‘‘A Generalized Production Frontier Approach for Estimating Determinants of Inefficiency in U.S. Dairy Farms.’’ Journal of Business and Economic Statistics 9(3):279–86. Lall, S. V., and S. Ghosh. 2002. ‘‘Learning by Dining: Informal Networks and Productivity in Mexican Industry.’’ Policy Research Working Paper 2789. World Bank, Washington, D.C. Meeusen, W., and J. van den Broeck. 1977. ‘‘Efficiency Estimation from Cobb-Douglas Production Functions with Composed Error.’’ International Economic Review 18(2):435–44. Mills, E. S. 2000. ‘‘The Importance of Large Urban Areas—and Governments’ Roles in Fostering Them.’’ In S. Yusuf, W. Wu, and S. Evenett, eds., Local Dynamics in an Era of Globalization. Washington, D.C.: World Bank. Nelson, R. R., and H. Pack. 1999. ‘‘The Asian Miracle and Modern Growth Theory.’’ Economic Journal 109(457):416–36. Political Risk Services. 2004. International Country Risk Guide. East Syracuse, N.Y. Available online at www.prsgroup.com/icrg/icrg.html. Quigley, J. M. 1998. ‘‘Urban Diversity and Economic Growth.’’ Journal of Economic Perspectives 12(2):127–38. Sala-i-Martin, X. 1997. ‘‘I Just Ran 2 Million Regressions.’’ American Economic Review 87(2):178–83. Solow, R. M. 2000. Growth Theory: An Exposition, 2nd ed. New York: Oxford University Press. Temple, J. 2000. ‘‘Inflation and Growth: Stories Short and Tall.’’ Journal of Economic Surveys 14(4):395–426. Tulkens, H. 1993. ‘‘On FDH Analysis: Some Methodological Issues and Applications to Retail Banking, Courts and Urbana Transit.’’ Journal of Productivity Analysis 4:183–210. Tulkens, H., and P. Vanden Eeckhaut, 1995. ‘‘Non-Parametric Efficiency, Progress and Regress Measures for Panel Data: Methodological Aspects.’’ European Journal of Operational Research 80:474–99. Wang, H.-J., and P. Schmidt. 2002. ‘‘One-Step and Two-Step Estimation of the Effects of Exogenous Variables on Technical Efficiency Levels.’’ Journal of Productivity Analysis 18(2):129–44. Wheeler, J. O., Y. Aoyama, and B. Wolf, eds. 2000. Cities in the Telecommunications Age: The Fracturing of Geography. New York: Routledge. World Bank. 2001. World Development Indicators Database. Washington, D.C. Available online at www.worldbank.org/data. ———. 2002. World Development Report 2003: Dynamic Development in a Sustainable World. New York: Oxford University Press.