POLICY RESEARCH WORKING PAPER 2770 Downsizing and Productivity Gains in the Public and Private Sectors of Colombia Martin Rama Constance Newman The World Bank Development Research Group Public Services January 2002 POLICY RESEARCH WORKING PAPER 2770 Abstract Public sector restructuring, including labor downsizing, added per worker is similar in both cases, state-owned has been one of the main areas of policy activism in enterprises experience an increase in total value added, developing countries and transition economies. But little and in value added per unit of capital, whereas both is known about its actual effects. Rama and Newman use indicators decline in private enterprises. The difference, panel data on Colombian enterprises spanning more than which could simply reflect the larger extent of initial one decade to assess the impact on several productivity inefficiency in state-owned enterprises, does not appear indicators. The results suggest that the productivity gains to depend on the degree of competition in product from downsizing are larger in state-owned enterprises markets. than in private enterprises. While the increase in value This paper-a product of Public Services, Development Research Group-is part of a larger effort in the group to understand employment and pay issues in the public sector. Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Hedy Sladovich, mail stop MC3-311, telephone 202-473-7698, fax 202-522-1154, email address hsladovich@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at mrama@worldbank.org or cnewman@ers.usda.gov. January 2002. (32 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Research Advisory Staff Downsizing and Productivity Gains in the Public and Private Sectors of Colombia Martin Rama Development Research Group World Bank mrarna(worldbank.org Constance Newman cnewman@ers.usda.gov This paper was written for the research project on Public Sector Downsizing, supported by the World Bank Research Committee (RPO 683-69). We are grateful to James Tybout for access to the data, and to Tilahun Temesgen for excellent research assistance. Correspondence related to the paper should be addressed to cnewman(ers.usda.gov or mrama(worldbank.org. 1. Introduction Can state-owned enterprises be made more efficient? Many would argue that this is a hopeless endeavor in developing countries and transition economies. In their view, political considerations and insider power will always prevail. As a result, overstaffing, an inadequate skill mix, and ineffective work practices are unavoidable. From this perspective, the only prospect to increase the productivity of state-owned enterprises is to transfer them to the private sector. For others, however, the key to higher productivity is not ownership, but rather a competitive environment. Public sector restructuring would thus yield some promise if it were accompanied by increased competition in product markets and a harder budget constraint. Regardless of the merits of these two views, it is clear that privatization is not always an option, at least not in the short run. Governments often want to retain control over industries that are allegedly "strategic." They can also be reluctant to divest because of the threat of political resistance by vocal stakeholders, including trade unions. Regulatory weaknesses and the potential for corruption and asset stripping in the privatization process may be powerful deterrents too. And even when governments are committed to divest, some restructuring may be needed in preparation for privatization. In many cases, dealing with labor redundancies is seen as prerequisite to attract private investors. The restructuring of state-owned enterprises has been an important area of policy activism in developing countries and transition economies. This restructuring has often involved substantial employment cuts. In 1991-93 alone, Haltiwanger and Singh (1999) 1 identified 41 World Bank loans and credits indirectly supporting public sector downsizing operations in client countries. Of them, 15 were aimed at restructuring or privatizing state- owned enterprises. By a conservative estimate, a similar number of downsizing operations was supported by the World Bank during the rest of the 1990s. A World Bank memorandum issued in 1996 paved the way for the direct funding of separation packages for redundant workers, as opposed to the indirect support of the early 1990s, which was channeled through the government's budget. The 1996 memorandum justified lending for separation packages if the case could be made that labor downsizing would increase productivity. However, measuring the productivity impact of enterprise restructuring in general, and of labor downsizing in particular, is not an easy task. In the private sector, the cheerful response of stock markets to restructuring or downsizing announcements suggest that this impact is positive. But it would be hazardous to assume that it is positive in the public sector as well. If political pressures and insider power lead to "wrong" recruitment decisions, they could also lead to "wrong" downsizing decisions (Rama 1999). There are not many empirical studies on the productivity impact of downsizing, and most of them refer to the private sector. The pioneering paper in this literature, by Baily, Bartelsman, and Haltiwanger (1996), refers to manufacturing in the United States. It shows that plants that increased employment contributed as much to overall productivity growth as plants that reduced employment. Similar conclusions are reached by Lach (1999) for the manufacturing sector of Israel. Studies relying on more focused samples identify stronger effects. In a paper examining the performance of 118 U.S. firms, Espahbodi, John, and Vasudevan (2000) find that operating performance improves significantly following downsizing. Something similar happens in motor vehicle manufacturing in the United Kingdom, according to a study by Collins and Harris (1999). However, this study also finds that "unsuccessful" downsizers tend to have among the worst productivity growth rates. The evidence is even thinner concerning downsizing in the public sector. In a study dealing with 281 privatization episodes in Mexico, La Porta and L6pez-de-Silanes (1999) found that downsizing efforts prior to privatization did raise privatization prices, suggesting that the productivity impact was positive. But the estimated effect was barely significant. Another paper by Sheehan, Morris, and Hassard (2000), more descriptive in nature, dealt with redundancies in Chinese state-owned enterprises. It found that the potential political repercussions of job losses hindered the freedom of management to adjust employment levels in the interest of efficiency, which suggests that downsizing could increase productivity. But this is a conjecture. Another line of research has emphasized the impact of competitive pressures on public sector productivity. State-owned enterprises would not be able to keep outdated work practices, or poor monitoring, if their survival was at stake. Inefficiencies of this sort could only last in enterprises that enjoy some monopoly power, or are subject to a soft budget constraint. While many state-owned enterprises are sheltered from competitive forces, those that are not should exhibit productivity levels that are close to those of the private sector, despite being state owned. Bartel and Harrison (2000) provide some evidence supporting this view in the case of Indonesia. Based on their findings, it would be tempting to conclude that public sector downsizing could increase productivity if it were to take place in a competitive environment. But again, this is a conjecture. In the absence of more systematic results, estimates of the impact of labor downsizing on public sector productivity usually rely on more or less arbitrary assumptions. One common, extreme assumption, is that the marginal productivity of all separated workers is zero. At the other extreme, it can be assumed that state-owned enterprises operate on their technological frontier, their only source of inefficiency being the excessive number of workers per unit of capital. In this case, the marginal productivity of redundant workers would be lower than their marginal cost to the enterprise, but it would still be positive. Assumptions of this sort cover too broad a range to provide reliable estimates. In an assessment of public sector downsizing in Algeria, Ruppert (1999) showed that the economic returns to labor downsizing could vary from strongly positive in one extreme case to strongly negative in the other one, hence the need for more accurate estimates. This paper exploits a unique plant-level data set to estimate the actual impact of labor downsizing on public sector productivity in a developing country, namely Colombia. This data set, which includes almost 80,000 observations, has a panel structure and spans more than a decade (1977-1991). Over that period, Colombia experienced sustained economic growth but did not embark in a privatization program. Many enterprises undertook a restructuring of their activities, sometimes involving substantial employment cuts. The extent and persistence of these cuts is used in the paper to identify several hundred downsizing episodes, many of which affected state-owned enterprises. Taking advantage of the panel nature of the data set, the paper then estimates the impact of downsizing on several productivity indicators. This impact is systematically compared across privately- and state-owned enterprises, taking into account the extent of competition in product markets. 3 2. The Analytical Framework Most studies on the impact of restructuring on productivity focus on the entry and exit of enterprises or plants. This turnover process is often labeled "external" restructuring, as opposed to the "internal" restructuring that takes place in continuing enterprises. Productivity gains from "internal" restructuring are often computed as a residual, much the same as total factor productivity gains in macroeconomic growth accounting. "Internal" restructuring is thus treated as a continuous process, rather than as a discrete change. Decompositions of productivity growth along these lines can be found in the studies by Davis, Haltiwanger, and Schuh (1996) for the United States, and by Disney, Haskel, and Heden (2000) for the United Kingdom. A similar approach is applied in the studies collected by Roberts and Tybout (1996) to a series of developing countries, including Colombia. The approach in this paper, on the other hand, identifies restructuring efforts based on changes in employment at the plant level. This approach has some similarity with the one applied by Baily, Bartelsman, and Haltiwanger (1996) to the analysis of productivity in U.S. manufacturing. Baily and others compare plants whose employment level increases to plants whose employment level decreases between 1977 and 1987. The former are identified as "downsizers" and the latter as "upsizers." However, minor changes in employment over a decade may not reflect any major restructuring effort. This is why our paper relies on a critical downsizing threshold. Only plants whose employment reduction exceeds this threshold in any given year, and is at least partially sustained during the following year, are considered downsizers. The focus on employment cuts that are not only large, but also sustained, is warranted to deal with measurement error bias. Productivity is often measured in units of output (say, value added) per worker. But employment is measured imperfectly. Consider the case where the number of workers reported for a specific plant in a specific year falls below the actual level. For instance, one digit could be missing in the reported data. Assume also that value added is correctly measured. This plant could be considered a downsizer, and it would also appear to experience a substantial increase in value added per worker. However, measurement error is presumably uncorrelated over time (missing digits are more or less randomly distributed in the sample). If this is so, the apparent reduction in employment would not be sustained over time, and the plant should not be considered as a downsizer. The potential effects of downsizing oil productivity are illustrated by Figure 1. The bold solid line in this figure is a standard production function, linking the labor input L to the output level Y for a given capital stock K0. The initial equilibrium, represented by point A, is one where the enterprise is overstaffed. At the prevailing wage level w, profits are maximized when 4 employment is equal to LI. In graphical terms, the optimum for the enterprise is represented by point B, where the marginal productivity of labor is equal to its cost. Moving from A to B entails a reduction in total output (from Yo to YI). If the capital stock remains unchanged, there is also a proportionate decline in output per unit of capital. But output per worker increases. This is reflected in the steeper slope of OB, compared to OA. Figure 1. Downsizing and Productivity Output (Y) Y = F(Ko,L) YO ......... Y , .. ... ... ........ . ..-*..... . .. . . ..-..... .. .... Y F(K1,L) A' zo .... ;4.............A l /Li' L 0 L1' L1 Lo Labor (L) This simple analysis requires two qualifications. First, it assumes that the capital stock remains unchanged, whereas actual restructuring efforts may also involve discarding outdated equipment, or introducing new technologies. Consider the simplest example, where an obsolete line of production is shut down. In this example, the reduction in employment is associated with a reduction in the capital stock from Ko to KI. Because of the smaller capital stock, the production function shifl downwards, as reflected by the bold broken line in Figure 1. The new equilibrium could be represented by a point like B'. Because of the smaller capital stock, the optimal employment level is lower than in the previous example, and the decline in total output is larger. The increase in output per worker can be either larger or smaller depending on technology. Figure 1 corresponds to the case where output per worker is the same in B and B'. The second qualification refers to the nature of the initial inefficiency. In point A, the enterprise is overstaffed but makes the best possible use of its personnel. Point A lies on the production function indeed. However, the initial situation could be one where the enterprise is not only overstaffed, but also fails to exploit its technological possibilities, as in point A'. By shedding its excess labor, and reorganizing production so as to take full advantage of its resources, the enterprise could move from point A' to point B. This move could be associated with an increase in total output (from Zo to YI), and a corresponding increase in output per unit of capital. There would also be a substantial increase in output per worker, because productivity in the initial equilibrium was abnormally low (the line OA' was flatter than the line OA). The discussion in the previous paragraphs has implications for the measurement of productivity gains. Many studies focus on total factor productivity growth, i.e., on the change in output that is not accounted for by changes in capital and labor inputs. But this calculation requires that all changes be measured in comparable units. Typically, changes in inputs are multiplied by indicators of their marginal productivity. Thus, for instance, the change in employrnent is multiplied by the average labor cost per worker. However, this approach implicitly assumes that resources are been used efficiently, which is inconsistent with the need for restructuring. Point A in Figure 1 is characterized by a marginal productivity of labor below the labor cost w. And the gap is even larger in point A'. Under these circumstances, attaching weights to the changes in capital and labor inputs involves some arbitrariness. Rather than trying to compute total factor productivity, this paper relies on the three indicators considered in the discussion of Figure 1. These are total output (Y), output per unit of labor (Y/L), and output per unit of capital (Y/K). Enterprise restructuring in general, and labor downsizing in particular, may lead to changes in all three. Using lowercase letters for logs, productivity gains in plant "i" and year "t" can be defined as follows: 6 dy. = Log Yi,t - Log Yi,t-l (1) d(y/l)l,t (LogY.t -LogY. - (Log L. -Log. L 1) (2) d( )i, t Log Y.,t Log Y., -1 L-otgKi t Log Kit1 (3) By construction, these three indicators are expressed in relative terms. For relatively small values they can be interpreted as percentages. The paper compares productivity changes across plants, after classifying them along two dimensions: state owned versus privately owned, and downsizers versus nondownsizers. The simplest analyses describe the distribution of dy, d(y/l) and d(y/k) in each of the four groups resulting from this two-dimensional classification. More elaborate analyses control for other characteristics of the plants, such as their initial "size," the taxes and subsidies they are subject to, their sector of activity, or the province they are located in. These variables, as well as the year considered, are summarized by the vector X1,t. Other important variables refer to the degree of competition in product markets, identified as Mi,t. The basic specification used in the econometric analysis has the following forn: dy,t = a + al Si + a2 D It +a3 SI Dit + aX Xi,t aM i,t i,t In this equation, Si is a dummy variable equal to one if enterprise "i" is owned by the state, whereas Di,t is a dummy variable equal to one if enterprise "i" downsizes its workforce in year "t." The default case corresponds to a privately owned enterprise that does not downsize. Parameter al measures the gap in productivity growth rates between state owned and privately owned enterprises in the absence of any downsizing. Parameter a2 reflects the impact of downsizing on the productivity of privately owned enterprises, whereas parameter a3 assesses whether this impact is different in the public and the private sectors. Similar equations can be estimated replacing the explained variable dy by d(y/l), or by d(y/k). However, not all the variables in vector X can be retained in these other equations. This is because a spurious correlation could emerge between plant "size" indicators and productivity indicators. Plant size can be measured by Log Ki,tI and Log Li,t . But the latter variable is used to compute d(y/l) (see equation 2), whereas the former is used to calculate d(y/k) (see equation 7 3). In the presence of measurement error, having the same variable in the left-hand side and the right-hand side of the equation can bias the estimates. Consequently Log K,-1l is dropped from vector X when the left-hand side variable is d(y/k), whereas Log Li,t-l is dropped when the left- hand side variable is d(y/l). A shortcoming of the specification in equation (4) is to ignore the potential impact of unmeasured plant characteristics on productivity growth. A common finding across studies is the importance of idiosyncratic factors in explaining differences in productivity across plants. If those idiosyncratic factors were correlated with any of the explanatory variables, estimating equation (4) by ordinary least squares would yield biased results. The availability of panel data makes it possible to overcome this problem, by letting the constant term in equation (4) be plant specific. This is the same as introducing a dummy variable for each enterprise. The drawback of the panel data approach is that parameter al cannot be estimated anymore. This is because the dummy variable Si is the sum of all the dummy variables for state-owned enterprises, so that there is perfect colinearity. The specification used to exploit the panel nature of the data is the following: dyt D=,t 83 S, Di,t + 8X Xi t +IJM Mi t + Vi t (5) This equation can be estimated using fixed effects. Again, similar equations can be estimated for d(y/l) and d(y/k), dropping from vector X the "size" indicators Log Lt-l and Log Kt-I respectively. Finally, the paper also assesses whether a competitive environment affects the impact of downsizing on productivity. This is achieved by interacting the market characteristics M with the dummy variables S and D. Equation (4) can thus be rewritten as: dy. =a ħa S.+a D. +ia S.D. +a X. +a M. + (6) i, t O 1 Si +2 i t 3 Si Di t + XXi t + aMM + 6 4 i Mi t 5 D ,t 3i t 6 i i,t i, t i, t whereas equation (5) becomes: 8 dy i t = Oi +42 D t +183 SiDi t +8XXi t + 1Mi t + (7) +4 S.M. +fl D. M. +t S. D M. +v 4 1 1, 5 I,t l,t 6 O , s , The key parameters to assess the impact of product market competition on the productivity gains from restructuring are a4 to a6 and 114 to 16 3. The Manufacturing Sector in Colombia Colombia is one of the few developing countries where the data needed to estimate equations (4) to (7) are available. The original source of these data is the census of manufacturing plants conducted by the Departamento Administrativo Nacional de Estadistica (DANES). The census reports information on dozens of state-owned enterprises. Mark J. Roberts and James R. Tybout transformed the individual cross-section data sets into a panel, by matching plant records across survey rounds for 1977 to 1991. The matching was based on stable characteristics of firms, such as their initial year of operation and their location. It also involved information on inventories at the beginning and the end of the year (see Roberts 1996). As a result, between 87 and 92 percent of all enterprises were matched in any given year. During the period covered by the data, the manufacturing sector enjoyed a stable macroeconomic environment and moderate growth. Unlike other Latin American countries, Colombia did not experience high levels of inflation or serious aggregate imbalances, and was not forced to implement fiscal reforms. Its approach to macroeconomic policy was gradualist and fiscally conservative, in accordance with the country's tradition. Public expenditures grew during the 1980s, but the level of taxation grew sufficiently to cover the growth in public expenditures. Public sector debt was relatively small, and average deficits were estimated as being only 1.5 percent of GDP (Carrasquilla 1996, and Galat 1998). Microeconomic policies were relatively stable too. Trade policy went from a period of liberalization in the late 1970s to a protectionist period through most of the 1 980s. The late 1970s witnessed a series of reductions in quantitative restrictions and nominal tariffs. But many of the restrictions were reintroduced in 1981 when a declining real exchange rate hurt exports and led to pressure to protect the import-competing domestic market. Export promotion stayed at levels that had been introduced in 1967 (Roberts 1996), so trade policy was focused on the import side. More significant trade liberalization did not occur until the early 1 990s. 9 The size and scope of the public sector was diverse, including large industry and finance. The largest nonfinancial components were the nation's social security system, the state-owned petroleum company, the electricity and telecommunication sectors, the state-owned coal company, a large public transportation project in Medellin, and the national coffee fund. Most provincial governments also had a print shop and a rum brewery. There were some privatizations in the banking sector in the mid- 1 980s as a result of government intervention in an earlier banking crisis. But there was not a systematic attempt to divest state-owned enterprises (Zuleta 1993). Lack of privatization warrants the use of the time-invariant dummy variable Si to identify state-owned enterprises in equations (4) to (7). The total number of observations used in the analysis is 79,149. Among them, 12,761 correspond to enterprises appearing only once in the data. However, multiple observations of the same plant are common. The average number of times that an enterprise appears in the data is 9.5, and the median is 10. In a study covering the first nine years of the panel (1977 to 1985), Roberts (1996) found that the overall patterns of entry and exit were similar to those of industrialized countries. Most of the observations in the data set are located in the major metropolitan areas of Bogota, Cali, and Medellin. Enterprises are relatively small, as two-thirds of them employ less than 50 workers and only 8 percent have more than 200 workers. The distribution of observations by region, size, and number of appearances can be found in Table 1. The characteristics of state-owned enterprises differ markedly from the overall sample. The corresponding number of observations is 494, of which 78 are from enterprises appearing only once in the data set. State-owned enterprises are located in more disperse areas of the country, with only one-third in the three major metropolitan areas. They are also much larger than private firms, with more than half of them employing 200 workers or more. The distribution of observations by sector is shown in Table 2. State-owned enterprises are clustered in a few major activities: beverages, printing and publishing, food, petroleum derivatives, and transportation. Private enterprises are fairly evenly distributed across all sectors, with slightly heavier concentrations in food and clothing and shoes. On average, value added per worker is higher for state-owned enterprises, and value added per unit of capital is lower, but there are important differences by sector. Some of the sectors in which state-owned enterprises exist are among the more capital intensive and more concentrated. Beverages and petroleum derivatives are also industries where the state has a high proportion of market share. 10 Table 1. Sample Characteristics All enterprises State-owned enterprises Geographic Location (percent) (percent) Bogota D.E., Soacha 34.15 13.56 Cali, Yumbo 11.05 9.72 Medellin, Valle de Aburra 21.40 9.11 Manizales, Villamaria 1.60 4.66 Barranquilla, Soledad 6.79 1.82 Bucaramanga, Gir6n, Floridablanca 5.57 3.04 Pereira, Santa Rosa de Cabal, Dosquebra 2.74 Cartegena 1.58 5.06 Rest of the country 15.13 53.04 Total 100.00 100.00 Plant size (percent) (percent) Less than 20 workers 33.33 7.69 Between 20 and 50 workers 33.40 18.42 Between 50 and 100 workers 15.76 11.54 Between 100 and 200 workers 9.09 10.53 More than 200 workers 8.42 51.82 Total 100.00 100.00 Plants by number of appearances in data I year 1,891 18 2 years 3,476 18 3 years 3,837 9 4 years 4,476 16 5 years 5,085 35 6 years 4,818 12 7 years 4,473 63 8 years 4,288 8 9 years 3,987 18 10 years 3,500 20 11 years 3,905 33 12 years 3,156 36 13 years 4,355 26 14 years 27,902 182 Total 79,149 494 Source: Constructed by the authors using data from the 1977 to 1991 rounds of the Colombian manufacturing census. Table 2. Summary Statistics for Enterprises State-owned enterprises Private enterprises Value Value Value Value Number added added Number added added of obser- Workers per per of obser- Workers per per Sector of activity vations perfirm worker capital vations perfirm worker capital Food 62 133.9 65.1 5.5 13,507 72.2 82.1 11.4 Beverages 219 275.5 320.6 11.4 1,467 193.2 185.5 15.1 Tobacco 200 205.9 146.4 23.9 Textiles 12 35.3 45.7 1.2 5,479 136.0 46.6 16.8 Clothing and shoes 13,055 58.1 38.5 15.6 Leather goods except shoes 1,128 85.2 33.3 17.1 Wood and cork products 1,999 39.4 38.4 10.6 Wood furmiture 2,186 44.1 27.0 13.4 Paper and paper products 1,780 84.3 104.9 17.4 Printing andpublishing 86 73.7 16.0 2.9 4,019 65.6 28.0 7.7 Chemicals 3 23.3 53.6 2.1 5,158 104.7 145.8 24.5 Petroleum derivatives 41 1383.7 196.5 4.8 291 40.2 187.6 13.3 Rubber products 4,404 75.7 50.1 9.7 Nonmetallic minerals 9 29.6 14.6 6.5 4,676 94.3 51.0 144.7 Metals 1,111 172.0 98.7 6.4 Metal products 6,716 58.3 48.2 7.7 Nonelectrical machinery 1 30.0 433.0 18.8 3,790 52.8 100.3 11.0 Electrical machinery 2,478 94.9 85.7 8.0 Transport equipment 46 242.7 37.6 0.6 2,662 97.9 52.3 8.0 Other manufacturing 15 144.5 503.2 3.2 2,549 56.1 88.1 31.4 All sectors 494 295.2 190.8 7.0 78,655 79.3 67.2 21.3 Source: Constructed by the authors using data from the 1977 to 1991 rounds of the Colombian manufacturing census. Value added per worker is measured in thousands of pesos per year. Value added per unit of capital is measured as a fraction, over a one-year period. 4. Downsizing Episodes For a reduction in employment to reflect a restructuring effort, it has to affect a substantial fraction of the plant's workforce, and a substantial number of workers, within a short period. To determine how substantial is substantial enough, some critical threshold has to be set. On the other hand, for the reduction to be considered durable, employment has to stay below its initial level for some time. Again, a critical threshold is needed to determine how far from the initial level is far enough. Rather than using two independent thresholds, this paper relies on one critical value, identified in what follows as N. It assumes that a plant experiences a downsizing episode if employment falls by at least N percent, and by no less than N workers, between one year and the next. In addition, employment has to recover by less than N- I percent, and by less than N- I workers, during the following year. 12 The dummy variable Dit is thus defined as follows: D =I1tifLogLi -LogLi -