Poilcy, Research, and External Affairs WORKING PAPERS Trade Policy Country Economics Department I ,ie 6vnrgo BanK September 1991 WPS 769 Entry-Exit, Learning, and Productivity Change Evidence from Chile Lili Liu The effects of plant turnover and learning on productivitv growth are economctrically i measur-ed usilln a large panel of- Chilean estahlishnimuti cox\erinu the period 197(Q-86. I hc Po..x, Rcwca J., dn! I u ,.i \ mr;, ( , , c - t.-.rtcs Ic V r lp Io Os. 1,:c: ' I ig , I. 1.)~ ~ ~ ~ ~ ~~~o- Cn.fUrPF 1!.tKh,S . d'ean sldi: :4d her, )n:c1:C>C IA ., dc.:.T;p-er.: :-.1wV I h-c~, prDise. d 'h- J-' 1`. the z.oiorv, rcfllc, o,rl! thcj .r s. -. : d l:. r..: 11 w.R ard c.:1 - ..d g; 1 hc mieItcs. :.:>r :L"n 1 e I',.: aa' Le e. <.ua.m >nr 2eFsv: ,. I:II L iD: t\ A\ 3 ,k.. al x P1t,-, es,vr-d J ot e [ -n< nM , Policy, Research, and External Affairs _7160 i . le -:i I U Trade Policy WPS 769 This paper-- a product olflhu'l'iade Poli c Divisiorn Counltr Elconioilics Deparlmientil -is part ot'the PRE research projeCt on Industrial Compelition. Productive Efficiency. and t'heir Relation to Trade Regimes (RPO674-46). Copiesare available l'iee romii the World Bank, 1818 H Street N\W, Washinoton, DC 2433 P'iease contact lDawn Ballanivne. room N 10-023, extension 37947 (41 pages) Septemiiher 199 1. Do compe!.tive pressuLIrs really force inCelicicint planis. These dif'ferenices in productivity across producers to shut down') Does tile enlitainec of colhors are bodt s\stenlati: and peisistent over new pro(dueers typically increase or worsen time. Th.e egap in productivity between incum- industrNwide cficicIsc \.) Is tliere evidence ol benrs and exilino planis, and between entering svstenatic learnino processes' Ilfllereis (lo aiid exitin plants, has x; ideiie( o er tinmc sMiile these processes diflfer across plilat cohorts? How tIhC gap beteeen incunibents zatnd entering planits do, 11he el'tects of' plant turnover comiibine to shapie has shrunk over timiie. 'I'llis is bvecause exitinLe ovcral iates of industrial produclivit\ growth? plants lia\ c dcclinin- productivity over timic. \k hile entering plants LradalUlylU spee 11p tllthir L.iu addresses these largely unecXplorcd productivity gromv th. Moreover, competitive questions by app. ing econometric tcClhn1i(qUCS pressures hav e driven botth incumilbents and from lhe efficiency f'rontiers and tIhe icpanel dataci einitanits to improve tIheir productivitv literature. Shc colistructs plant-specific tlime- variant technical ef'licienc\ indices for surviving. * The ratio o' skilled labor to uniskilled labor exiting, and entering pi.; cohorts. She Iheln is higher and increasing m oie rapidly amriong uses these to compare productivit\ gro\\ il IratIes inucuitmbes at(d entrants than amlom, exitino across plant coh1orts and to exaiminC Ihe neCIt Cl f'ct p:mtns, proN iding an imlpOr;tant SOliurCe of' tean1inle of plant turnover an(d learning patterns on and productivit\ gro\ith. manulacluring-kkide prOductivity growth. Alithouglh the conolnomv ide recession The analysis is based on plant-level paniel afl`eted lhe produc!"i"i v' e b h cohort to diata for all ChlilCan m ianulacturing plan11ts with at dif'ierenit degrees. there are stead\. increases in least 10 workers, covering eigiht years in the productivity over tlhe sample period, reflecting post-ref'orm adjusimenit period 1979-86. both Ihc r-cplaccimCInt ofl inilicienit produceis b\ el'flicient ones and tlhe improvement ot produci - LI-U ILIUi LiIL iiiI u itti iLc of plam turn\ er, it\ b! Incumbents andi entrants. lu4 erent !earnino palterns across cohoni in driving the Chilean nmniUlacturing-widc prodUc- T'lies e l icienc gain1s have niot becen iso- livity clhangcs. She I inds that: lated by traditional total lactor productivilN studies hased on secioral data. These gains I'lhe cvidence sUppont5 the hypolltcsis thiait suggest that in icroeconiomiiic rclorni - - including cOmpCtitiVC pl-cssUrcs lorce 1css 'elicienit piroduc- trade liberalization, privati/ation, aid ihc dc- cr so fail morc oflCIn than otheis Average rculalion ol markets - have been effective in technical efficiencv levels are higher amnmg promoting efficiency impro\ emnctis in thc suzA\ivingt and en ici-ing plants than aimn'n en ili, (ing Chileian rlnuliFacturing sector. 1T1, 1"1: I:.{. l,.::| X;- hl'vliw; - :k,:ll U'i''10"T: ' k: j\ :.'1:.1 I01; lik" md F\;|j Th,~~~~~~~~~~|8t.ti! '. 'Il,ti,: ll Entry-Exit, Learning, and Productivity Changc: Evidencc From Chile By Lili Liu Table of Contents 1. Introductioi. I II. Review of Productivity and Entry-Exit Studies 4 III. Empirical Methodology 9 V. Entry-Exit and Productivity Changc: Application to Chilean Industries 17 V. Conclusions 27 Tables and Figures 29 Bibliography 39 This paper was prepared for the World Bank research project, Industrial Competition, Productive Efficiency, and 'Their Relations to Trade Regimes (RPO 674-46). I would like to thank Robert Dernberger, James Levinsohn, Hal Varian, and especially James Tybout for helpful comments and stixrestions. I. Introduction Do competitive pressures really force inefficient producers to shut d. wn? Docs the entry of new producers typically improve or worsen industry-wide efficiency? Is there cvidence of systematic learning processes, and if so, do these processes differ across plant cohorts? How do plant turnover effects combine to sLape the overall rate of industrial productivity growth? Although theoretical studies on entry-exit and productivity growth have recently emerged (Jovanovic, 1982; Pakes and Ericson, 1988). micro empirical evidence on the linkage between entry-exit patterns and productivity in manufacturing sectors is very scant. Most studies of industrial productivity change have been done at the sectoral level' and thus have been unable to capture the effects of entry, exit, and heterogeneity on productiviry growth. Some micro studies of productivity change have been done, but they have been either cross-sectional or limited to selected industries or to a small sample of firms.2 Hence they too have been unable to systematically address the issues mentioned above. Recently, as comprehensive micro data have become available, studies have emerged on the actual entry-exit patterns in manufacturing sectors (Dunne, Roberts, and Samuelson. 1989, on the U.S. manufacturing sector; Tybout, 1989, on the Chilean manufacturing sector). The substantial degree of heterogeneity in plant size, market share, and failure rates first revealed by these studies has motivated the present research to address the largely unexplored questions raised above. Much of this literature is surveyed in Chenery, Robinson, and Syrguin (1986), the World Bank (1987), Pack (1988), and Havyrlyshyn (1990). 2 See, for example, Cornwell, Schmidt, and Sickles (1990), Handoussa, Nishimizu, and Page (1986), Page (1984), and Tybout, Corbo, and De Melo (1990). 3 This paper applies econometric techniques from the efficiency frontiers literature and the panel data literature to construct plant-specific time-variant technical efficiency indices for surviving, exiting, and entering cohorts. The-se are then used to compare productivity growth rates across plant cohorts and to examine the net effect of plant turnover and lcarning patterns on manufacturing-wide productivity growth. The analysis is based on plant-level panel data from Chile covering the period 1979-86. F:or several reasons, these data provide an excellent basis for inference. First, they include all Chilean manufacturing plants with at least 10 workers. This allows one to identify entering, exiting, and surviving plants and to look at their relative importance in driving manufacturing aggregates. Second, from 1974 to 1979 Chile underwent sweeping reform programs to liberalize its trade regime, privatize state firms, and deregulate markets. The data cover eight adjustment years following the reform, a period during which intense foreigin competition, rising interest rates, and other shocks to the economy led to plant turnover rates much higher than in either developed countries (Dunne, Roberts, and Samuelson, 1989) or in other developing countries with less extensive liberalization and external shocks (Tybout, 1989; Roberts and Tybout, 1990). Finally, the removal of market distortions excludes the possible bias in estimating productivity gains because almost all prices are determined by the world market. The findings support the hypothesis that competitive pressures force less efficient producers to fail more frequently than others. Average technical efficiency levels are higher among suiviving and entering plants than among exiting plants. These differences in productivity across plant cohorts are both systematic and persistent over time. The gap in productivity between surviving and exiting plants and between entering and exiting plants has widened over time, while the gap between 4 surviving and entering plants has narrowed. This occurred because the productivity of exiting plants declined over time while that of entering plants increased. Moreover, competitive pressures have driven both surviving plants and entrants to improve their productivity. The ratio of skilled labor to unskilled labor is higher and increasing more rapidly among incumbents and entrants than among exiting plants, providing an important source of learning and productivity growth. Although the economy-wide recession affected the productivity of each cohort to different degrees, productivity increased steadily over the sample period, reflecting both the replacement of inetficicnt producers by efficient ones and productivity improvements among surviving and entering plants. These efficiency gains have not been isolated by traditional total factor productivity studies based on sectoral data. These gains suggest that microeconomic reforms--including trade liberalization, privatization, and market deregulation--have been effective in promoting efficiency improvements in the manufacturing sector. The rest of the paper is organized as follows. Section II reviews the literature on productivity and entry-exit studies. Section III specifies the models to be estimatcd, and section IV analyzes both descriptive evidence and the results from fitting the econometric models- Section V concludes the paper by pointing out the potential problems in the estimation models and possible corrections. II. Review of Productivity and Entry-Exit Studies The issue of industrial productivity growth has long interested development economists since cfficient resource use helps te promote successful industrialization. Numerous empirical studies 5 of productivity in developing countries are summarized by Chenery, Robinson, and Syiquin (1986), the World Bank (1987), Pack (1988), and Havyrlyshyn (1990). However, empirical research on industrial productivity has suffered from two major shortcomings. First, the majority of the studies employ the traditional measure of productivity: total factor productivity (TFP).3 This measure is based on strong assumptions, such as constant returns to scale and competitive markets, yet most studies completely ignore the possible bias of cstimates if those assumptions are violated. Second, even if these problems are solved, the problem of aggregation remains. Most studies have been at the macro or sectoral levels, and have been unable to capture the effects of entry, exit and heterogeneity on productivity growth. Tybout (1990a), after discussing possible approaches for dcaling with violated assumptions, concludes that the aggregate studies assume a well-defined production technology for all plants within the industrial, sectoral, or country analysis, completely ignoring plant heterogeneity: "if tcchnological inno' ition takes place through a gradual process of efficient plants displacing inefficient ones, and/cr through the diffusion of new knowledge, approaches to productivity measurement based on 'representative plant' behavior are at best misleading. At worst, they fail to capture what is imporiant about pruducLivity growth altogether, as Nelson (e.g., 1981) has long argued" (pp. 28-29). Despite the recent advances in theoretical work on entry-exit, learning, and productivit, (Jovanovic, 1982; Pakes and Ericson, 1987), micro empirical evidence on the linkage between entry- exit and plant productivity in manufacturing sectors is still very scant. This study draws on two lines 3 See, for example, Nish;i7zu and Robinson (1984), and Nishimizu and Page (1982). 6 o. recent niicro empirical research, each of which has a different focus: entry-exit analysis and efficiency froaties for panel data. A. EnryExit- A.4nalysis One line of micro empirical research a.tempts to identify actual patterns of entry-exit in manufacturing sectors. To delineate manufacturing-wide patterns of entry-exit, plant-specific time series data covering all plants in the manufacturing sector are needed. Plant identification codes and Standard Industrial Classification (SIC) codes for each observation allow files to be merged into a single panel-data base sorted by plant, year, and product type. The inter-temporal patterns of missing values for each plant can thus be used to identify entering, exiting, and surviving plants. As comprehensive micro-level panel data have become available, studies have been done on actual entry-exit patterns in manufacturing sectors. For example, Baldwin and Gorecki (1987) provide a summary of entry-exit patterns in Canadian industries; Dunne, Roberts, and Samuelson (1989) analyze the actual patterns of entry-exit in the U.S. manufacturing sector; and Tybout (1989) does the same for the Chilean manufacturing sector. These studies were the first to reveal the substantial degree of heterogeneity in plant size, market share, and failure rates, which was largely .~ui!rset *n nrt'vinii thenretical and empirical work. The performance measures used in these studies are typically output share and relative plant size. What is entirely missing in the entry-exit literature is an examination of how plant cohorts differ in efficiency levels and total factor productivity and how this heterogeneity, together with turnover effzcts, systematically helps to shape overall productivity performance and growth. 7 B. Traditional Efficiency Frontier Analysis Another line of micro empirical research focuses on the measurement and estimation of firm-specific technical efficiency, entirely omitting the issue of entry-exit and learning. The common approach used in these studies is based on the framework of production frontier and technical efficiency models first proposed by Farrell (1957). The production function f(x) defines the maximum possible output a firm can produce given input bundles x, constituting the efficiency frontier or the best-practice frontier. Technical inefficiency is the amount by which a firm's actual output falls short of the efficiency frontier, reflecting non-minimized costs due to excessive use of inputs. Once the frontier (or the "best-practice") production. rarely known a priori, is estimated, an efficiency index for an economic unit can be derived fron, the deviation of iis acAual output from the frontier. Various estimation techniques, and their strengths and weaknesses are summarized in Forsund, Lovell, and Schmidt (1980), Schmidt (1985), and Bauer (1990). Most studies on efficiency frontiers have been cross-sectional, imposing limitations on the econometric estimation of efficiency itself. Consider the following econometric model of a Cobb-Douglas efficiency frontier, yi = ' x,+v.-u,, u>0 i=1,...,N where y, and x; are the logarithm of output and the vector of inputs, respectively, v; is the random error term and ui represents tcchnical inefficiency. There are several shortcomings with the cstimation of the efficiency frontier using cross-sectional data. First, to obtain estimates of p:ant- specific technical efficiency, one must specify probability distributions for both statistical random errors and tcchnical inefficiency terms, and it is unclear how robust the results are for those 8 assumptions. Second, technical efficiency has to be assumed to be independent of inputs, Icading to biased estimates if inefficiency is known a priori by firm managers. Finally, firm-specific inefficiency indices can be estimated, but they cannot be estimated consistently. When panel data are available, there is a potentially better alternative to estimation of the efficiency frontier. The repeated observations over time for a given firm provide information on its efficiency that is unavailable from cross-sectional data. Hence the estimation of firm-specific technical efficiency does not require strong distributional assumptions about composed error terms. In addition, the assumption that technical efficiency is independe-t of factor inputs does not have to be imposed. The panel data estimation of efficiency frontiers was introduced by Pitt and Lee (1981) and Schmidt and Sickles (1984). The issue that concerns us here is whether a competitive environment is conducive to higher efficiency. Limited evidence suggests a positive linkage between competitive pressures and higher productivity. Using cross-sectional data, Tybout, Corbo. and de Melo (1990) derive industry- specific technical efficiency indices for Chilean industries for 1967, when an import-substitution regime was in place, and for 1979, when trade liberalization policies had been implemented. Although overall industrial efficiency did not improve between the two census years, the industries (at the three-digit classification level) that experienced relatively large reductions in protection improved relative to others. Cornwell, Schmidt, and Sickles (1990) found higher productivity in U.S. airlines during the deregulated period than during the regulated period. This paper attempts to bring together the two bodies of literatures on entry-exit and productivity and thereby to shed light on the largely unexplored issues raised in the introduction. 9 III. Empirical Methodology A. Defining Entering. Exiting. and Surviving Plants The data cover all plants in the Chilean *nanufacturing sector with at least 10 workers. Plamt identification codes and SIC codes for each observation allow us to merge files into a single panel-data base sorted by plant, year, and product type. The intertemporal patterns of missing values for each plant can thus be used to identify entering, exiting, and surviving plants. Plants are divided into cohorts in three categories: surviving plants, exiting plants, and entrants. Surviving plants stay in the sample for the entire 1979-1986 period, so there is no chu ge in their sample size. There are six exiting cohorts and each exits the data base consecutively between 1978+i and 1979+i (i=1,...,7), respectively. For example, the 1979 exiting cohort produced only in 1979, the 1980 exiting cohort produced only in 1979 and 1980, the 1981 exiting cohort produced only from 1979 through 1981, and so forth. Entrants enter the data base between 1979+i and 1980+i (i=1...., 7), respectively. For example, 1980 entrants entered the data base in 1980 and stayed through the rest of the sample years, 1981 entrants entered the data base in 1981 and stayed through the remaining sample years, and so on.4 4 Since our X a cover only plants with 10 or more workers, entry-exit may also reflect an adjustment in labor around the cut-off point. However, this problem can be minimized by excluding plants which entered and exited repeatedly during the sample period (in addition, capital stock variables cannot be constructed from the perpetual inventory method for these plants). Since the data cover the cyclical period (growth- recession-growth), fluctuation in labor adjustment will likely be reflected by these plants. They account for 10% of plants but only 3% of total output. 10 Unfortunately, capital stock variables were reported on'v in 1980 and 1981, so capital stock variables derived from the perpetual inientory method could not be constructed for entrants after 1981 or for exiters in 1979. Those plants are, therefore, excluded ii the econometric estimation of total factor productivity'. These plants do report data on other factor inputs and on output, so they can be included in the analysis of simple performance measures like labor prodLctivity. This will be done to augment the efficiency frontier analysis. B. Efficiency Analysis We begin with a Cobb-Douglas representation of technology relating factor inputs and output for a given industry6: yi, =a+Ff Y, +vit -u, (1) = 1, 2,..., N u; > 0 A tiny portion of plants reported all data other than capital stock in 1980 or 1981. They are also excluded from the estimation. 6 The estimation model with balanced data design is based on Schmidt and Sickles (1984). The Cobb-Douglas technology is chosen because iL deals better with industrial census data, as indicated by Griliches and Ringstad (1971). 11 where, y,, is the observed output for plant i at time t, expressed in logarithms; x, is a vector of K inputs, also expressed in logarithms; the industry subscript is suppressed; and a and p are unknown parameters to be estimated. The disturbance is composed of two different types of errors. The first, v1,, represents random errors in the production process. The second, ui, represents technical inefficiency. Its distribution is one sided, reflecting the fact that output must lie on or below the frontier, a + F,f7t + va. The random error vi, is assumei to be distributed identically and independently across plants and time with identical zero mean and constant variance. It is also assumed to bc uncorrelated with factor inputs. This assumption holds if the realized values of v,, are unanticipated by managers when they choose factor inputs (Zellner, Kmenta, and Dreze 1966). If data are sorted by plant and year, the error vector v = ( vil, ..., vi,, vi2...) has a covariance matrix ao/I, where I is an identity matrix with an order of NTxNT. The other error component. ui, is assumed to be independently and identically distributed across plants with mean p and variance q.2. Equation (1) can easily be modified to fit into the standard variable-intercept model in the panel data literature7. We maay rewrite the equation as: y,, = ai' + x, + v;, (2) where a, = a - u;. Schmidt and Sickles (1984) discuss the application of panel data techniques to stochastic frontier models. 12 1. Fixed-Effect Models Equation (2) is a fixed-effect model if a, is treated as fixed, and the least-square dummy variable (LSDV) approach or covariance estimation can be applied. If the a, are correlated with X., in an unknown manlier, the OLS estimator is BLUE. It is known from the panel data literature8 that A A fl, is unbiased and consistent when either N or T goes to infinity. f, is also called the "within estimator" since it is based exclusively on deviations of plant output and factor inputs from their own time series means. The associated plant-specific intercepts, a,,,, are the difference between a plant's actual level of output averaged over time and the predicted level of output given the plant's factor 'A inputs averaged over time. The a<, estimator, are also BLUE, but consistent only when 1' goes to infinity. Differences in the intercepts across plants reveal relative efficiency differences. To derive a measure of technical efficiency relative to the production frontier, we follow Schmidt and Sickles (1984) and define: a = max (a,) (3) u,= a -a, (4) As N goes to infinity, the efficiency of the most-efficient plant will approach 100%. 8 See, for example, Hsiao (1986). 13 2. Random-Effect Models Altcmatively, the technical etticiency indices can be treated as random variables. In that case, equation (2) fits into a random-effect model and can be estimated using the variance components approach. In treating the u, terms as random variables, all the assumptions about the random vector v remain unchanged. In addition, v and u are assumed to be uncorrelated, and the u vector has elements of iid random components with constant variance au'. But, in contrast to the fixcd-cffect model, technical efficiency and factor inputs must be uncorrelated to yield consistent estimators. Some of the well-known results from the panel data literature can be summarized. As either N or T goes to infinity, the GLS estimators of a and f6, with known au' and a,2, are consistent and more efficient than are the within estimators. For fixed T as N goes to infinity, the efficiency property remains. But as T goes to infinity, the GLS estimator is equivalent to the LSDV estimator. In the more realistic case of unknown variance of u and v, N approaching infinity is required to obtain a consistent estimator of au2. Thus the strongest case for GLS is when N is large and T is small, and input and technical efficiency are uncorrelated. To derive a plant-specific measure of technical efficiency, we can follow Schmidt and Sickles (1984). Define, 1 T, a, =_ E s (5) T t=1 14 where the ej, terms are the residuals from GLS estimation. Given P(,, a, is consistent as T goes to infinity. Following equations 3.3 and 3.4, u, can be separated, which requires that N go to infinity. Thus the consistent estimator of technical inefficiency requires that both N and T go to infinity. 3. Generalizing to Open Panel The preceding discussion on estimation techniques is based on balanced data design; it assumes that each cross-sectional unit has T periods of observations. Unbalanced panel data do not create problems for the fixed-effect model since only within-group variations are relevant. But GLS estimation of the random-effect model has to be modified. Hsiao (1986) discusses the case where the total number of observations remains constant for t=1, ..., T, with the number of observations dropped equaling the number of observations added in each period. In our analysis of entry-exit, the total number of observations is unlikely to remain constant if the number of entrants does not match the number of exiting plants. The extension of Hsiao is straightforward. I. 1ixtU-EL&kIA UT Lfl*ULiA*hLAf9.) The obvious advantage of the LSDV approach is that it does not require that cfficiency and the regressors be uncorrelated. The disadvantages are that the LSDV estimators are less efficient than the GLS ones and that time-invariant plant-specific attributes other than technical efficiency cannot be included as regrcssors because of the problem of perfect collinearity. Examples 15 of such attributes might be type of ownership and firm location. One way to solve the problem is to regress the estimated plant-specific technical inefficiency on those attributes. This of course assumes that those effects are observable and that they are uncorrelated with technical efficiency. If input choices are not correlated with technical efficiency, the random-effect model will be better since GLS estimators are more efficient. To see whether the random-effect model can be used, a Hausman test can be used. Hausman (1978) noted that under the null hypothesis that ac are uncorrelated with x, the GLS achieves the Cramer-Rao lower bound, but under the alternative hypothesis, GLS is a biased estimator. l,, is consistent under both the null and the alternative hypothesis. lience, the Hausman test asks whether f and f, are significantly different. 5. Time-Variant Productivyy The models outlincd above assume that plant-specific technical efficiency is time- invariant. Relaxing that assumption and allowing productivity to change over time enables one to identify time paths of technical efficiency for various plant cohorts. Cornwell, Schmidt, and Sickles (1990) introduce a parametric function of time into the production function to replace the coefficient uf plaiji-NpeciLIc technical efficiency. The functional fLULn II. y,= x,' + a,, + v, (6) where a, = w"'O1 W", = (1, t, t2) 0, = (6,.1 6,2, 063)' 16 and other variables are defined as before. The measurement of productivity growth focuses on temporal variation, and the model allows the rate of productivity growth to vary over plants (cross-sectional variation). Efficiency levels can be derived from the residuals based on either the GLS estimators or the within estimators, as in A Cornwell, Schmidt, and Sickles (1990). That is, O, is estimated by regressing (y,,-x,t') for plant i on w;,, where f are GLS estimators if factor inputs are assumed to be uncorrelated with plant-specific time-variant effects. However, a, is not consistent as T goes to infinity if factor inputs are correlated with firm and time specific effects. Under these conditions, the consistent estimators of a,, as T goes to infinity, can be derived by estimating equation (6) using OLS directly (i.e. the within estimation). T1he dimensionality problem in this regression can be avoided by proceeding in steps. First, regressing A A y,i and each factor input x*, on wi, plant by plant, to get predicted values y,t and x.. Second, pooling A A A all the data and regressin. (yi, - yi,) on (xY, - x;,) to get ,B. Finally, the residuals (yi, - x,,'P) can be used to derive plant-specific time-variant technical efficiency. To derive technical efficiency relative to the frontier, the analogy (Cornwell, Schmidt, and Sickles, 1990) to equations (3) and (4) is as follows: The frontier intercept at time t is: A A a, = max ( a,,) (7) and the plant-specific technical efficiency of plant i at time t is: u= at - ait (8) 17 IV. Entry-Exit and Productivity Change: Application to Chilean Industries This sectior, first provides some background on the liberalization policies implemented in Chile during 1974-79 and summarizes descriptive statistics on the entry-exit pattern and plant adjustment during the post-reform period. It then analyzes the results from fitting the econometric models developed above. A. Liberalization Policies and Plant Adiustment in Chile Before 1974, Chile had one of the most protected manufacturing sectors in developing countries. It also had heavily regulated factor and output markets, extensive price controls, widespread black markets, a highly controlled credit market, and a segmented and highly unionized labor market. As a result, there were de facto administrative barriers to entry, and productive inefficiency was rampant. The Chilean reforms of 1974-79 reform was swift and comprehensive. Micro reforms included removal of protective trade barriers, privatization, and market deregulation. By June 1979, Chile had a uniform 10% tariff rate (except for motor vehicles) and all nontariff barriers were removed (the rapid reduction in trade protection is apparent in table 1). All but one bank and most firms were privatized and almost all prices were decontrolled. By April 1980 both the domestic and external financial markets were liberalized. New labor legislation greatly reduced the power of labor unions and prohibited strikes, --the policy reform most strongly applauded by firm managers (Corbo and Sanchez, 1985). 18 The contractionary macroeconomic policies aimed at fighting the fiscal deficit and inflation, together with the first oil shocks of 1973, plunged the economy into recession during 1973- 76. Industries started to recover in 1976, and the recovery lasted through 1981 as firms dropped redundant labor. At the same time, backward wage indexation was introduced. However, the 1981 debt crisis in Latin America, overvalued exchange rates, and highly leveraged and undiversified financial markets in Chile led to another major recession in 1982-83. Industrial output and employrnent fell sharply, accompanied by a deepened "deindustrialization" process as import substitution policies were abandoned in the mid-1970s'°. But industrial output picked up briskly in 1984, stalled temporarily in 1985, and resumed rapid growth in 1986. The structural adjustment program in 1985 aimed at expanding nontraditional export through devaluation of exchange rate, stabilization of copper prices and assistance to export producers; at encouraging public saving and private investment; at strengthening regulations of financial systems; and at reducing external debt. As a result of effective Inicroeconomic policies implemented by the end of 1980 and the structural program in 1985, the period 1986-89 observed rapid growth under a favorable macroeconomic environment: the industrial sector has been a leading sector in growth; employment has expanded; .nd tho -n-uft^.A1 trAe b2!anice has improved. The Chilean indistries are now one of the most efficient in Latin America." High rates ot plant turnover and improvements in eticiency are two ot the most important adjustments made by manufacturing firms in response to the changed policy incentives. ;° To help the adjustment reduce unemployment, the fixed exchange rate and wage indexation were abandoned For detailed analysis of the import substitution strategies prior to 1974. liberalization policies implemented during 1974-1979 and post-reform performance of the economy, see Corbo, de Melo and Tybout (1985). Condon. Corbo and de Melo (1989), and Edwards and Edwards (1987). 19 This result was first casually suggested by a 1982 qualitative survey conducted by Corbo and Sanchez (1985) covering 10 manufacturing firms. Facing inc-zased import competition, the surveyed firms closed their inefficient plants and reduced production lines. They also strived to increase efficiency by expanding investment, improving management and product quality, and, especially, reducing labor force. Compared with 1976, all the firms in the sample had cut employment by 1982-- by 50% in the largest firm and 20% in the smallest one. More comprehensive descriptive statistics further rcflect the importance of entry-exit and the heterogeneity of plant efficiency. Aggregate exit rates were much higher than entry rates during the sample period. The net result was a reduction in the total number of plants (table 2). Compared the end of sample years 1986 with the beginning of the sample year 1979, 25 out of 28 industries had net decline in the total number of plants: at least one third decline in 50% industries. Although the 1982 recession, touched off by the financial market crisis, may largely explain the high exit rates in 1982 and 1983, the equally high exit rates during the period of rapid growth in 1979-81 suggest the importance of turnover effects in industry restructuring in response to the rapidly changed incentives. Before going through the more rigorous econometric analysis of efficiency frontiers and total factor productivity, iooking at a simple measure labor productivity may help to reveal some of batt=s Of h*e o ' pka(rAU LIIA rtc;cy jaUJUTdblu li. idUiC ) plrebnus labor productivity across plant cohorts. Efficiency labor units'2 were derived using a weighted averagc of labor inputs to take into account the heterogeneity of labor productivity. Plants were divided into three cohorts, as discussed in section III.A: surviving plants, exiting plants, and entrants. Four interesting observations emerge from table 3. First, average labor productivity is higher among 12 The efficiency labor units is based on Grilichcs and Ringstad (1971). 20 surviving plants and entrants than among exiting plants. Second, all cohorts display increasing trends in productivity over the sample period, although three cohorts had various degrees of decline in productivity in the recession and in 1985. Third. the increase in productivity for the ex ting plants as a whole reflects both within-group efficiency improvement over time and changes in the sample sizes over time. This is because each cohort of exiting plants improves its productivity over time, and the exiting of the least efficient plants also contributes to efficiency gains. This applies to entrants as well. Fourth, as a result of within-group improvements and the dropping out of the least efficient plants, labor productivity for the whole manufacturing sector increased steadily over the sample period. Although the results are suggestive, they reflect only single factor productivity. The following section applies the econometric models developed in section III to derive and examine the distributions of cohort-specific technical efficiency. B. Econometric Analysis of Cohort-Specific Productivity The variables used in estimation are derived as follows. Plant output was deflated by three-digit industry-specific output price indices. Intermediate material inputs were deflated by their own indices. which was constructed from sectoral output prices using the 1977 Chilean input-output table. Each energy input was deflated by its own deflator, which was constructed from reported physical volumes and values. The perpetual inventory method was applied to derive capital stocks, with each of four capital goods categories deflated by its own deflator."3 Unfortunately, capital 13 A complete description of the data preparation is in Appendix II which is available upon request. 21 stock was reported only in 1980 and 1981, so capital stock variables derived from the perpetual inventory method could not be constructed for entrants after 1981 and exiters in 1979. Those plants are, therefore, excluded from the estimation.'4 As indicated in section III, the key question in choosing the fixed or the random-effect model is whether input choices are correlated with technical efficiency. The Hausman test is first applied to balanced data in 27 industries. Only surviving plants (accounting for over '0% of plants in all industries) are included in the test because computation of the Hausman statistic is more cumbersome with unbalanced panei. If the null hypothesis is rejected using the balanced data, there is no need to go over the unbalanced data. The test statistic indicates rejection of the null hypothesis for 23 industries with 3 degrees of freedom and a significance level of 0.025. The test was then applied again for the remaining four industries to the unbalanced data wihich included surviving plants, entrants, and exiters. The null hypothesis was rejected with 3 degrees of freedoms and a significance level of 0.025 for two out of these four industries. To sum up, the null hypothesis that inputs and technical efficiency are uncorrelated is rejected for 25 industries. Given that the assumptions of the error components framework are not satisfied by our data, we first fit the fixed-effect model to equation (2) where plant-specific technical efficiency is time-invariant and time dummies are added to the equation 2: yi, = aj + i x;, + vi; 14 In addition, industry 3114 (tobacco) has only three to four plants in the sample period, so there are not enough degrees of freedom for estimation. 22 This will give us a general idea of productivity differentials across cohorts, although it will force all plants to exhibit tiie same rate of productivity change through time. The estimated coefficients are presented in Table 4. The overall fit reflected by adjusted R2 looks reasonable. The estimated labor elasticities are positive for all industries, significant at the 0.05 level for 24 of 27 industries and at the 0.10 level for 2 more industries. But the elasticities exhibit considerable variation across industries, from 0.021 to 0.329, with most industries averaging 0.2. The estimated elasticities of intermediate inputs are all statistically significant at the 0.05 level. Although capital elasticities are significant at the 0.05 level for 11 industries and at the 0.10 level for 1, the elasticities appear small, and 5 industries have implausible negative elasticities, suggesting possible measurement errors.'5 The estimated low returns to scale do not support the hypothesis of constant returns to scale when plant fixed effects are controlled for. Experiments with the largest industry (SIC 312) based on cross-sectional estimation indicate increasing returns to scale when individual effects are not controlled for. This finding suggests possible bias in the estimated elasticities due to cross-sectional data, measurement errors, inappropriate functional form, or simultaneity. IN The plant-specific intercept a, measures relative technical efficiency among plants. We could follow Schmidt and Sickles (1984) to obtain plant-specific technical efficiency indices measured relative to the frontier (equations 3, 4). However, we would gain little insight from doing so since we are mainly interested in the evolution of cohort-specific technical efficiency. The transformation of fe:ativc cfficiency only shifts all cohorts by some common unit, leavring relative patterns unchanged. i5 Using techniques developed by Griliches and Hausman (1986), Westbrook and Tybout (in progress) estimated returns to scale by specifically dealing with measurement errors. Their approach is being adapted to an additional paper. 23 Table 5 reports average cohort-specific technical efficiency over time, which is the weighted average of rclative plant-specific indices (a,'s) within each cohort. Since thc time trends are already controlled for, the differences in technical efficiency among cohorts reflect their average deviations from the timc trend. The most notable result is that technical efficiency is higher on avcrage among surviving plants and entrants than among exiting plants. The distribution of technical efficiency for surviving plants versus entrants does not show any cbviously uniform pattern. Entrants have higher technical efficiency than surviving plants in some industries and lower technical e'ficiency in others. Since the time dummies force all plants to follow a common productivity growth path, they obscure plant-specific productivity changes over time. To relax this, we apply OLS to equation (6) and derive a,, which is the predicted value of plant-specific time-variant technical efficiency. After deriving the a,, values, we average them over surviving plants, exiting plants, and entrants in the manufacturing sector, respectively for each time period. The average technical efficiency by plant cohorts in the manufacturing sector is plotted in figure 1. (The 1979 exiting plants and 1982-86 entrants are excluded from figure 1 because of missing capital stock data, and the 1980 exiting plants are excluded because of lack of degrees of freedom to estimate a,,.) Recall the findings in Table 5 (where plant-specific technical efficiency is assumed to be time-invarian;) that tcchnical efficiency is on average higher among surviving plants and entrants than among exiting plants. Figure I reintorces this finding by showing that in every time period technical efficiency is on average higher among surviving plants and entrants than among exiting plants. In addition, figure 1 indicates the time path of technical efficiency change. Surviving plants show slightly declining productivity from 1979 to 1985, but with productivity stabilizing from 1985 to 1986. Entrants started with lower technical efficiency than surviving plants and show a slight decline in productivity from 1980 to 1983 24 similar to that of surviving plants, but their productivity increased faster than that of surviving plants from 1983 to 1986. In contrast to the trends of surviving plants and entrants, exiting plants have declining productivity throughout 1980-85. As a result, the net gaps in technical efficiency between exiters and entrants and between exiters and surviving plants increase over time. Figure 2a shows the time paths of average productivity for each exiting cohort. Three patterns are obvious. First, cohorts with the lowest technical efficiency exited rirst in every year. (Although we were not able to estimate technical efficiency for the 1979 and 1980 exiting cohorts, table 3 indicates that they have low labor productivity.) Second, productivity improvement occurred only from 1979 to 1980, a high growth period, and only for two exiting cohorts (1982 and 1983). But the two still had much lower efficiency than did surviving plants. Finally, after 1980, all exiting cohorts showed declining productivity. In contrast to entrants, exiting plants were not able to catch up when the economy started to recover in 1984. Figure 2b plots average technical efficiency levels for entering cohorts. Although the 1980 entrants had declining technical efficiency initiallv. thev were able to bounce hack after the recession. The 1981 entrants showed much faster and steadier growth, leading to a steadily increasing trend for entrants as a whole and suggesting that learning processes were taking place. The time path of each exiting (entering) cohort in Figure 2a (Figure 2b) illustrates that turnover effects and within-group efficiency changes combined to shape the general time path of productivity for exiting (entering) plants as a whole (figure 1). Therefore, the trend of exiting (entering) plants over time in tigure 1 reflects both within-group efficiency changes and turnover effects. Take the example of productivity change for exiting plants from 1982 to 1983. The average 25 technical efficiency of all exiting plants in 1983 (figure 1) is the weighted average of efficiency levels of each exiting plant cohort in 1983 (figure 2a). When the 1982 cohort exits, the lowest technical efficiency cohort in 1982 has dropped out, so the market share of the remaining plants (which had higher technical efficiency) increased, causing a positive turnover effect. However, from 1982 to 1983, each remaining exking cohort (i.e., plants that continued production from 1982 to 1983 but dropped out later) failed to improve its productivity. Their declining productivity outweighs the positive turnover effect, leading to a net decline in total productivity from 1982 to 1983 (figure 1). As figure 1 shows, overall productivity levels among entering, exiting, and surviving plants do not increase significantly, probably because of the economy-wide recession. Despite this, it is possible that industry-wide productivity improved as the less efficient producers exited and incumbents and entrants increased their market share. This appears to have been the case. Figure 3 plots average technical efficiency in manufacturing over the sample period. Changes in technical efficiency within each cohort, variations across cohorts, and plant turnover have combined to inicrcase the industry-wide average productivity from 1982 to 1986. A significant portion of the efficiency gain is due to distributional effects, i.e., replacing inefficient producers by efficient and/or improving plants. Such results suggest that competitive pressures have been significant and that micro reform policies have been effective in discriminating between inefficient and efficient producers. Note that the estimated trend in figure 3 may have underestimated the improvement in productivity for two reasons. First, as mentioned before, we excluded the 1979 and the 1980 exiting cohorts. These two cohorts have lower labor productivity (table 3), so their exit would probably have increased average productivity from 1979 to 1981. Second, all entrants after 1981 were excluded, as mentioned earlier. Table 3 shows that average labor productivity among entrants after 1981 increased 26 over time. We might have obtained a different trend for the period of 1983 to 1986) had these entering cohorts been included. Sustained learning at an industry-wide level should be reflected in the growth of measured technical efficiency not accounted for by measurable factor inputs (Pack, 1990). But capacity utilization may significantly influence the measured residuals, and cyclical demand may also exert an impact on residual changes. This measurement problem is particularly acute in the present study because the data cover periods of growth (1979-80), recession (1982-84), and recovery (1985- 86). The change in plant-specific technical efficiency levcls will thus reflect both plant-levcl effects and industrial or macro fluctuations over time. Two bits of evidence suggest that micro efficiency improvement took place despite fluctuations in capacity utilization. First is the ratio of skilled to unskilled labor by cohorts, a simple yet revealing statistic, suggesting the important impact of education on learning and productivity growth (table 6). The ratios are higher and increasing faster among surviving plants and entering plants, reinforcing the labor productivity findings presented in table 3. The increase in the ratio of skilled to unskilled labor contributed to the rise of labor productivity. Second, even if we make the cxtreme assumption that the trends presented in figure 1 reflect purely capacity utilization changes (which is unlikely), figure 3 would still reflect efficiency gains from replacing less efficient plants with more efficient ones. 27 V. Conclusion The findings support hie hypothesis that the forces of competition discriminate against less efficient producers. Average technical efficiency levels are higher among surviving and entering plants than among exiting plants. The gap in productivity between surviving and exiting plants and between exiting and entering plants has widened over time; while the gap between surviving ano entering plants has narrowed. Moreover, both surviving and entering plants have improved their productivity. The ratio of skilled labor to unskilled labor is higher and increasing more rapidly among surviving plants and entrants than among exiting plants, suggesting that learning is a source of productivity growth. Although the economy-wide recession in 1982-83 affected the productivity of each cohort to different dcgrces, .hcrc were steady increases in productivity over the sample period, reflecting both the replacement of inefficient producers by efficient ones and the improvement of productivity by surviving plants and e. -nts. This suggests that microeconomic policies that removed all distortions are effective in pushing efficiency improvement in the manufacturing sector. The study suggests several areas for additional research which is currently in progress. First, the model specifications have assumed away measurement errors which are shown to be empirically important (Westbrook and Tybout, in progress). If returns to scale were underestimated, the estimated technical efficiency based on residuals would also include scale effects. Second, open panel data also bring up the question of selectivity bias. Finally, it would be of great interest to compare the econometric estimation of efficiency frontiers with the mathematical programming 28 method"6. Efforts are being made to test whether the results ar4. sensitive to different model specifications. 16 The mathematical programming model is also called data envelopmcnt analysis (DEA). The theoretical motivation can be found in Farrell (1957) and Varian (1984a, 1984b). Banker. Charnes, and Cooper are among the major contributors to empirical models. Recent developments in DEA arc summarized by Seitord and Thrall (1990). 29 Table 1 Effective Protecteon in Chile, 1961-79 Effective protection (percent) Sector L961 1967 1974 1976 1978 1979 Foods products 2,884 365 161 48 16 12 Beverages 609 -23 203 47 19 13 Tobacco products 141 -13 114 29 11 11 Textiles 672 492 239 74 28 14 Footwear and clothing 386 16 264 71 27 14 Wood and cork 21 -4 157 45 16 1S Furniture 209 -5 95 28 11 11 Paper and paper products 41 95 184 62 22 17 Printing and pubLishing 82 -15 140 40 20 12 Leather and Leather products. 714 18 181 46 21 13 Rubber products 109 304 49 54 26 15 Chemicals products 89- 64 80 45 16 13 Petroleum and coal products 45 1,140 265 17 12 13 Nonmetallic mineral products 227 1 128 55 20 14 Basic metals 198 35 127 64 25 17 Metal products 43 92 147 77 27 15 Nonelectrical machinery 85 76 96 58 19 13 ElectricaL machinery 111 649 96 58 19 13 Transportation equipment 101 271 -- -- -- - Other manufacturing 164 -- -- - - -cU, ';hted we:i ha:.ic ::l 3 . ; . '' .0 e* ,^ 4.7 i3.* Standard deviation 618 279.0 60.4 15.70 5.3 1.7 Variability coefficient 1.78 1.57 0.399 0.31 0.27 0.124 Range 2.863 1.163 216 60 17 6 Source: Corbo and Sanchez (1985) Table 2 Net Reduction In the Number of l'lants by the Three-Digit Industry 1979-86 …========= =====s==3s============e= ======= t_m:w …-l == = …= .==-=== ISIC 1979 1980 1981 1982 1983 1984 1985 1986 (1986-1979)1979 312 1610 1507 1421 1383 1354 1401 1397 1343 -0.166 313 211 188 158 151 148 138 127 111 -0.474 314 3 4 4 4 3 4 4 4 0.333 321 503 445 403 350 327 336 337 331 -0.342 322 442 398 346 305 265 294 275 280 -0.367 323 90 76 68 59 53 '51 50 46 -0.489 324 185 156 137 127 127 133 128 137 -0.259 331 524 449 406 358 335 339 342 313 -0.403 332 211 192 171 143 116 117 115 104 -0.507 341 70 67 60 56 52 60 5s 57 -0.X86 o 342 242 227 206 196 177 167 164 163 -0.326 351 65 59 59 56 51 58 60 S1 -0.0 2 352 171 166 159 148 145 151 149 153 -0.105 353 .10 9 9 9 10 10 2 2 -0.8 354 8 7 9 9 B 9 16 16 1 355 63 67 59 53 52 56 55 48 -0.238 356 170 163 149 142 142 161 161 166 -0.024 361 13 13 10 15 11 12 12 11 -0.154 362 33 28 29 24 20 22 22 18 -0A55 369 135 139 128 110 104 108 115 113 -0.163 371 64 48 42 35 36 32 31 32 -0.5 372 34 31 28 27 21 24 25 25 -0.265 381 459 447 413 365 322 358 351 347 -0.244 382 169 140 145 18 125 133 127 115 -0.32 383 87 72 64 57 55 59 56 59 -0.322 384 150 124 112 94 87 83 86 86 -0.427 385 15 20 14 15 15 14 16 15 0 :390 77 66 63 55 44 48 52 49 -0.364 _ 39 77 =49 = == == = = = = = = = = = = = = TABLE 3 Labour Productivity in the Manufacturing Sector 1979 1980 1981 1982 1983 1984 1985 1986 Surviving Plants 1101 1161 1231 1201 1223 1264 1241 1422 Exiters Average 717 765 8K7 831 784 950 898 Exiters Decaqonsition: 1979 629 1980 689 718 1981 667 664 706 1982 798 826 913 793 1983 743 778 806 698 638 1984 999 929 1099 1076 943 836 198S 767 840 894 897 150 1011 898 Entrants Average 866 1094 1119 1081 1000 1005 1013 Entrants Decomposition: 1980 866 926 828 946 955 942 1040 1981 1370 1641 1554 1326 1423 1129 1982 1073 1142 1024 1053 1142 1983 958 1064 1171 1145 1984 931 949 993 1985 827 953 1986 939 Manufacture Avera5e 906 971 1069 1071 1100 1139 1139 1288 Note: Labour Is dkfined as efficiency labour units. Output Is the gross value of output. Table 4 - Fixed-Effect Model Regression Coefficients Dependent Variable Ln (Y) (Standard Efror in P; rentheses. * implies significance at a = 0.05 level, ** implies a = 0.10 level) Industry Ln(L) Ln(K) Ln(M) RTS 12 j2 F-Stat N 312 0.142' (0.008) 0.4107"s (0.005) 0.74* (0.006) 0.89 0.98 0.036 2622.7 8500 313 0.095' (0.026) 0.03* (0.004) 0.58- (0.007) 0.705 0.96 0.093 126.37 693 321 0.200* (0.02) 0.029' (0.009) 0.664' (0.013) 0.89 0.96 0.06 499.39 2383 322 0.244* (0.023) 0.004 (0.011) 0 644* (0.015) 0.89 0.96 0.067 387.7 2019 323 0.292* (0.061) 0.057* (0.027) 0.558* (0.032) 0.91 0.97 0.063 66.95 361 324 0.201* (0.028) 0.031' (0.016) 0.715* (0.021) 0.95 0.98 0.04 249.94 862 331 0.231' (0.022) 0.030' (0.014) 0.652* (0.013) 0.91 0.95 0.082 420.64 1929 332 0.201' (0.04) 0.05' (0.020) 0.678' (0.022) 0.93 0.96 0.064 194.83 860 341 0.058* (0.024) 0.112' (0.033) 0.654' (0.03) 0.82 0.98 0.047 69.326 351 342 0.088' (0.017) 0.081' (0.018) 0.538' (0.018) 0.71 0.97 0.06 235.79 1223 351 0.021 (0.065) 0.004 (9.017) 0.658' (0.047) 0.683 0.95 0.115 39.56 338 352 0.132* (0.02) 0.005 (0.009) 0.603' (0.018) 0.74 0.97 0.05 158.51 1064 353 0.052" (0.04) 0.012' (0.007) 0.683' (0.006) 0.75 0.99 0.026 61.42 64 354 0.24 (0.091) .4.178 (0.098) 0.81' (0.08) 0.87 0.99 0.02 86.2 48 355 0.181' (0.038) 0.141* (0.036) 0.541* (0.034) 0.86 0.97 0.054 81.229 372 356 0.188* (0.03) 0 007 (0.014) 0.638- (0.019) 0.83 0.96 0.054 216.17 871 361 0.271 * (0.21) 0.D97' (0.051) 0.3'8* (0.109) 0.71 0.97 0.075 30.2 51 362 0.329' (0.078) 0040 (0.062) 0.585S (0.05) 0.95 0.9; 0.079 33.21 175 369 0.283* (0.04) -0.001 (0.014) 0.63* (0.031) 0.91 0.97 0.09 113.11 646 371 0.164' (0.058) C.053 (0.06) 0.605' (0.042) 0.82 0.97 0.09 39.64 344 372 0.228* (0.06) _0.028 (0.056) 0.694* (0.042) 0.89 0.99 0.076 42.34 168 381 0.209* (0.018) O.D18' (0.008) 0.6310 (0.013) 0.86 0.97 0.065 452.99 2193 382 0.069' (0.025) C.002 (0.01) 0.638' (0.023) 0.71 0.94 0.087 129.53 691 383 0.216* (0.049) 01.032 (0.03) - 0.687* (0.03) 0.94 0.97 0.086 92.3 413 384 0.208' (0.038) 40.0005 (0.017) 0.682' (0.025) 0.89 0.96 0.094 162 675 385 0.287' (0.109) 0.069 (0.13) 0.462* (0.069) 0.82 0.92 0.063 50.75 87 390 0.291* (0.051) -0.004 (0.024) 0.615' (0.035) 0.9 0.94 0.076 75.38 366 TAOLE 5 Ave,age Technical Efficiency by Pl.ant-Cohorts Surviving Plants Exiting Plants Entrants In kstry Mean SE STD N Mean SE SI'D N Mean SE STD N 312 2.384 0.010 0.286 836 2.304 0.01? 0.305 310 2.418 0.046 0.321 49 313 4.369 0.068 0.560 67 4.049 0.100 0.546 30 5.308 0.211 0.366 3 321 2.8 0.018 0.274 232 2.731 0 028 0.297 111 2.702 0.063 0.22? 13 322 3.048 0.019 0.251 169 2.897 0.022 0.270 152 3.029 0.163 0.399 6 323 3.489 0.049 0.269 30 3.147 0.063 0.328 2? 3.120 1 324 2.321 0.025 0.221 78 2.132 0.036 0.241 45 2.218 0.063 0.126 4 331 2.788 0.020 0.261 163 2.638 0.032 0.363 129 2.766 0.110 0.396 13 332 2.456 0.038 0.300 62 2.298 0.033 0.302 84 2.237 0.059 .1T 4 341 2.916 0.077 0.449 34 2.494 0.148 0.592 16 2.906 1 342 4.304 0.040 0.434 115 3.84a 0.048 0.380 62 4.015 0.181 0.313 3 351 4.045 0.089 0.536 36 3.633 0.124 0.44T 13 352 4.407 0.046 0.489 115 3.971 0.105 0.546 27 4.067 0.168 0.377 4 353 4.296 0.257 0.726 8 354 3.143 0.196 0.480 6 355 3.213 0.047 0.287 38 3.144 0.086 0.320 14 3.053 1 356 3.523 0.028 0.264 91 3.295 0.OS8 0.303 2? 3.365 0.151 0.301 4 361 5.174 Q.185 0.414 S 4.707 0.185 0.261 2 362 3.158 0.078 0.340 19 2.M 0.138 0.338 6 369 3.33? 0.051 0.400 61 2.902 0.062 0.363 34 3.108 0.04 0.065 2 371 3.549 0.059 0.355 36 3.260 0.090 0.348 15 372 3.275 0.074 0.323 19 3.059 0.100 0.201 4 381 3.363 0.022 0.321 209 3.084 0.031 0.349 126 3.366 0.224 O.44 4 382 3.893 0.049 0.399 67 3.925 0.015 0.429 33 3.828 0.342 0.484 2 383 2.717 0.043 0.2 2.425 0.094 0.363 1S 384 2.983 0.040 0.313 62 2.862 0.054 0.356 43 2.681 0.209 0.295 2 385 4.456 O.06 0.216 8 4.051 0.122 0.211 3 4.092 1 390 3.421 0.04S 0.260 34 3.172 0.083 0.373 20 3.782 1 Ueighted Avarage 2.998 0.015 0.734 2471 2.806 0.017 0.618 1292 2.821 0.065 0.704 l1 Simple Average 3.345 0.157 0.683 19 3.102 0.142 0.620 19 3.266 0.177 0.772 19 Note: N=Nuber of Observatiors. SExStandard Error of Mean. SlDzStmdsrd Error. Simple average: average over industry man. Weighted *verage: averse over the entire sample. lndustry 351, 353, 354, 371, 372, and 383 were excluded from mean calculation. Figure 1: Technical Efficiency Change By Three Cohorts average technical efficiency 3.75 - 3.65- 3.55 - 3.45 _ 3.35 I ,__,__,--_-e 1979 1980 1981 1982 1983 1984 1985 1986 year - surviving plants I exiting plants entering plants 27 Industries Figure 2a: Technical Efficiency Change Exiting Cohorts average technical efficiency 3.61 -_--- 7 3.56 ' 3.51 3.46- 3.41 - 3.36 _ 3.31 3.26 ' I - 1979 1980 1981 1982 1983 1984 1985 1986 year - 1981 exiters 1 1982 exiters 1983 exiters 9 1984 exiters 1985 exiters 27 industries Figure 2b: Technical Efficiency Change Entering Cohorts average technical efficiency 3.71 3.66 _ 3.61 . 3.56 3.51 3.46 - / 3.41 __ _L_ _ .L._. _ 1 .... __I_11 _! 1979 1980 1981 1982 1983 1984 1985 1986 year - 1980 entrants - I 1981 entrants 27 industries Figure 3: Technical Efficiency Change Manufacturing Average average technical efficiency 3.75 3.73 - 3.71 / 3.69 - 7' 3.69 3.67 5 3.65 --_l_ |_ __ __ 1979 1980 1981 1982 1983 1984 1985 1986 year - all plants 27 industries TABLE 6 The Ratio of Skiltld over Unskilled Labour: Manufacture Sector Weighted Average over the Entire Sample 1979 1980 1981 1982 1983 1984 1985 1986 Surviving Plants 0.277 0.285 0.285 0.298 0.345 0.337 0.330 0.419 Exiters Average 0.2 2 0.248 0.241 0.268 0.262 0.266 0.268 Exiters Decomposition: 1979 0.262 1980 0.238 0.240 1981 0.224 0.229 0.234 1982 0.248 0.247 0.251 0.265 1983 0.242 0.315 0.242 0.273 0.258 1984 0.231 C.237 0.221 0.264 0.257 0.268 1985 0.226 0.235 0.249 0.270 0.270 0.265 0.268 00 Entrants Average 0.239 0.234 0.261 0.256 0.269 0.268 0.339 Entrants Decompositior,: 1980 0.239 0.215 0.243 0.240 0.244 0.243 0.561 1981 0.266 0.289 0.282 0.291 0.300 0.358 1982 0.264 0.263 0.279 0.331 0.431 1983 0.252 0.244 0.246 0.294 1984 0.281 0.283 0.326 1985 0.225 0.275 1986 0.315 Manufacture Average 0.261 0.270 0.278 0.291 0.324 0.319 0.314 0.398 Note: Labour is defined as efficiency labour units. Output is the gross value of output. BIBLIOGRAPHY Aigner, D., C Lovell, and P. Schmidt (1977), "Formulation and Estimation of Stochastic Frontier Production Function Models," Journal of Econometrics, 6: 21-37. Baldwin, J. and P. GorecWi (1987), "Plant Creation Versus Plant Acquisition: The Entry Process in Canadian Manufacturing," International Journal of Industrial Organization 5 March). Bauer, P (1990), 'Recent Developments in the Econometric Estimation of Frontiers," Journal of Econometrics. Vol. 46, October/November Condon, T., V. Corbo and J. de Melo (1989), "Exchange Rate-Based Disinflation, Wage Rigidity, and Capital Inflows: Tradeoffs for Chile, 1977-1981," World Bank working paper, 146, Washington D.C. Corbo, V. and M. Sanchez (1985), "Adjustment by Industrial Firms in Chile During 1974-82," in Scrambling for Survival--How Firms Adjusted to the Recent Reforms in Argentine Chile. and Uruguay, World Bank Staff Working Papers, No. 764. Corbo. V., J. de Melo and J. Tybout (1985), "What Went Wrong with the Recent Reforms in the Southern Cone?" World Bank working paper, DRD 128. Cornwell, C., P. Schmidt, and R. Sickles (1990), "Production Frontiers with Cross-Sectional and Time-Series Variation in Efficiency Levels," Journal of Econometrics, Vol. 46, October/November. Dunne, T., M. Roberts, and L. Samuelson (1989), "Patterns of Firm Entry and Exit in U.S. Manufacturing Industries," The Rand Journal of Economics Winter. Edwards, S and A. Edwards (1987), Monetarism and Liberalization: The Chilean Experiment. Cambridge, Mass.: Ballinger Publishing Company. Farrell, M. (1957), "The Measurement of Productive Efficiency," Journal of the Royal Statistical Society. Series, A. 120. Ferrier, G. and C. Lovell (1990), "Measuring Cost Efficiency in Banking: Linear Programming Evidence," Journal of Econometrics Vol. 46, October/November Forsund, F., C. Lovt.l, and P. Schmidt (1980), "A Survey of Frontier Production Functions and of Their Relationship to Efficiency Measurement," Journal of Econometrics. 13. Griliches, Z. and V. Ringstad (1971), Economies of Scale and the Form of the Production Function: An Economic Study of Norwegian Manufacturing Establishment Data, Amstcrdam: North-Holland. Handoussa, H., M. Nishimizu, and J. Page (1986), "Productivity Change in Egyptian Public Sector Industries after the 'Opening'," Journal of Development Economics, 20. 40 Hausman, J. (1978), "Specification Tests in Econometrics," Econometrica, 46. Havyrlyshyn, 0. (1990), 'Trade Policy and Productivity Gains in Developing Countries: A Survey of theJ Literature," The World Bank Research Obsenrer, 5: pp 1 -24. Hsiao, C. (1986), Panel Data Analysis, Cambridge University Press. Jovanovic, B. (1982), "Selection and the Evolution of Industry," Econometrica, 50, (May). Liu, L. (1990), "Chilean Data Preparation," CEC1TP, the World Bank. Kmenta, J. (1971), Elements of Econometrics, New York: Macmillan Publishing Co., Inc. Meeusen, W. and J. van den Broeck (1977), "Efficiency Estimation from Cobb-Douglas Production Functions with Composed Error," International Economic Review 18:435- 444. Nelson, R. (1981), "Research on Productivity Growth and Productivity Differences: Dead Ends and New Departures," Journal of Economic Literature, 19 (September). Nishimizu, M. and J. Page (1982), "Total Factor Productivity Growth, Technological Progress,and Technical Efficiency Change: Dimensions of Productivity Change in Yugoslavia, 1965-1978," Economic Journal 92. Nishimizu, M. and S. Robinson (1984), "Trade Policies and Productivity Change in Semi- industrialized Coun.ries," Journal of Developnment Economics 16 (September/October). Page, J. (1984), "Firm Size and Technical Efficiency: Application of Production Frontiers to Indian Survey Data," Oxford Economic Papers 32. Pack, H. (1988), "Industrialization and Trade," in H. Chenery and T.N. Srinivasan, eds., Handbook of Development Economics. vol.1, Amsterdam: North-Holland. Pack, H. (1990), "Learning and Productivity Change in Developing Countries," forthcoming in G. Helleriner, ed., Trade Policy. Industrialization and Development: New Perspectives. manuscript. Pakes, A. and R. Ericson (1987), "An Alternative Theory of Firm Dynamics," Department of Economics, Columbia University. Pitt, M. and L. Lee (1981), "The Measurement and Sources of Technical Inefficiency in the Indonesian Weaving Industry," Journal of Development Economics, 9. Roberts, M. and J. Tybout (1990), "Size Rationalization and Trade Exposure in Developing Countries," forthcoming in R. Baldwin, ed., Empirical Studies of Commercial Policy, University of Chicago Press for NBER. 41 Scherer, F. (1980), Industrial Market Structure and Economic Performance, 2nd ed. Boston: Houghton Mifflin Company. Schmidt, P. (1985), "Frontier Production Functions: A Survey," Econometric Reviews 4. Schmidt, P. and R. Sickles (1984), "Production Frontiers and Panel Data," Joumal of Business and Economic Statistics 2. Seiford, L and R. Thrall (1990), "Recent Developments in DEA: The Mathematical Programming Approach to Frontier Analysis," Journal of Econometrics. Vol. 46, October/November. Theil, H. (1971), Principles of Econometrics. New York: Wiley. Tybout, r. (1989), "Entry, Exdt, Competition, and Productivity in the Chilean Industrial Sector," unpublished paper, the World Bank, Washington, D.C. Tybout, J. (1990a), "Researching the Trade/Productivity Link: New Directions," The World Bank PRE Workine Paner. No. 638. Tybout, J. (1990b), "Making Noisy Data Sing," Journal of Econometrics forthcoming. Tybout, J., V. Corbo, and J. de Melo (1990), "The Effects of Trade Reforms on Scale and Technical Efficiency: New Evidence from Chile," forthcoming in Journal of Intemational Economics. Varian, H. (1984a), "The Nonparametric Approach to Production Analysis," Econometrica. 52: 579-597. Varian, H. (1984b), "Nonparametric Analysis of Optimizing Behavior with Measurement Error," working paper, University of Michigan. Varian, Hal (1978), Microeconomic Analysis New York: W. W. Norton & Company. Westbrook, D. and J. Tybout (1991), "Estimating Returns to Scale with Large Imperfect Panels: An Application to Chilean Manufacturing Industries," Georgetown T Tnivurditv Zellner, A., J. Kmenta, and J. Dreze (1966), "Specification and Estimation of Cobb-Douglas Production Function Model," Econometrica, 34 (October). The World Bank (1987), The World Development Report 1987, New York: Oxford University Press. PRE Working Paper Series Contact RLeAuthor tDsle for pa,oer WPS742 The Cost of the District Hospital: A. J. Mills A Case Study from Malawi WPS743 Antidumping Enforcement in the Angelika Eymann August 1991 N. Artis European Community Ludger Schuknecht 37947 WPS744 Stainless Steel in Sweden: Anti- Gunnar Fors August1991 N. Arlis dumping Attacks Good International 37947 Citizenship WPS745 The Meaning of "Unfair in U.S. J. Michael Finger August 1991 N. Artis Import Policy 37947 WPS746 The Impact of Regulation on Dimitri Vittas August 1991 W Pitayatonakarn Financial Intermediation 37666 WPS747 Credit Policies in Japan and Korea: Dimitri Vittas August 1991 W. Pitayatonakarn A Review of the Literature 37666 WPS748 European Trade Patterns After the Oleh Havrylyshyn August 1991 N. Castillo Transition Lant Pritchett 37947 WPS749 Hedging Commodity Price Risks in Stiln Claessens August 1991 S. Lipscomb Papua New Guinea Jonathan Coleman 33718 WPS750 Reforming and Privatizing Poland's Esra Bennathan August 1991 B. Gregory Road Freight Industry Jeffrey Gutman 33744 Louis Thompson WPS751 A Consumption-Basad Direct Tax for Charles E. McLure, Jr. August 1991 CECSE Countries in Transition from Socialism 371RR WPS752 Inflation and Stabilization in Roberto de Rezende August 1991 L. Ly Yugoslavia Rocha 37352 WPS753 The CMEA System of Trade and Martin Schrenk August 1991 CECSE Pavment- The [ egacy and the 37188 Aftermath of Its Termination WPS754 Estimating Returns to Scale with James R. Tybout August 1991 S. Fallon Large Imperfect Panels M. Daniel Westbrook 37947 WPS7.55 Hedging Crude Oil Imports in Stijn Claessens August 1991 D. Gustafson Developing Countries Panos Varangis 33714 WPS756 Taxes Versus Quotas. Tne Case Arvind Panagariya August 1991 S. Fallon of Cocoa Exports Maurice Schiff 37947 WPS757 Managing the Trans'ion: Enhancing Eric Rice August 1991 P. Infante the Efficiency of Eastern European 37642 Governments PRE Working Papel Seroes Contact ILe AAh r foL12,per WPS758 Is 1There Excess Co-Movement of Theodosios B. Palaskas August 1991 D. Gustafson Primary Comnmodity Prices? A Panos N. Varangis 33714 Co-Integration Test WPS759 The Profamilia Family Planning Jesus Amadeo August 1991 0. Nadora Program, Co!umb!-v An Economic Dov Chernichovsky 31091 Perspective Gabriel Ojoda WPS760 How Coniicting Definitions of Alexander J. Yeats September 1991 J. Jacobson "Manufactures" Distort Output and 33710 Trade Statistics WPS761 Uncertairty and the Discrepancy Gerhard Pohl September 1991 P. ILee between Rate-of-Return Estimates Dubravko Mihaljek 81950 at Prolect Appraisal and Project Completion WrS762 Debt, Debt Relief, and Giowth: Daniel Cohen September 1991 S. King-Watson A Baigaining Approach Thierry Verdier 31047 WPS763 A Valuation Formula for LDC Debt Daniel Cohen September 1991 S. King-Watson 31047 WPS764 African Financing Nseds in the 1990s Jorge Culagovski Victor Gabor Maria Cristina Germany Charles P. Humphreys WPS765 Withholding Taxes and International Harry Huizinga September 1991 S. King-Watson -,&' T IU Ierms 3 'i n47 WPS766 Economic Crisis, Structural Francois Diop September 1991 0. Nadora Adjustment, and Health in Africa Kenneth Hill 31091 tsmail Sirageldin v .Of IvIacroeconomic `:in1 A ~c). -t-e~L Cf4 I Analysis (Applied to Kenya) David Ndil 34046 WPS768 Going to Market: Privatization in Manuel Hinds September 1991 L. R. Hovsepian Central and Eastern Europe Gerhard Pohl 37297 vvir7379 Eritry-ir.xii Ledrililly, adui Wi t; iu x VIeIl Iel L;. Gialla.dlyfl Pioductivity Change- Evidence from 37947 Chile