University of Cape Town

Transcription

1 Africa Region Working Paper Series No. 96 Industry Concentration in South African Manufacturing: Trends and Consequences, by Johannes Fedderke and Gábor Szalontai School of Economics University of Cape Town DECEMBER 2005

2 Africa Region Working Paper Series No. 96 December 2005 Abstract This paper is a result of a wider policy research and knowledge work on growth and jobs issues in South Africa, which the Bank promotes in collaboration with leading South African researchers. The objective is to contribute to major economic and policy issues facing South Africa as it embarks on the second decade of its democratic transition. These issues include growth and jobs, export competitiveness, service delivery, small and medium-size enterprise development and investment climate, industrial concentration, infrastructure and growth, municipal and financial management, land reform, regional integration, trade and poverty, HIV/AIDS, and especially important service delivery. This paper examines industry concentration for the South African manufacturing sector over the period (including the last year of the last manufacturing industry census data available) for the three-digit industry classification. The paper documents both the high level of industry concentration in South African manufacturing, and a rising trend in concentration across a wide range of industries. The paper further explores the impact that industry concentration has on a wide range of indicators of industry performance. It finds that increased concentration serves to lower output growth, raise unit labour costs, and to lower labour productivity. The impact of concentration on employment, total factor productivity and investment rates, based on bivariate examinations of the data, is ambiguous. The paper concludes by examining the impact of concentration on employment and investment rates using a dynamic heterogeneous panel estimation methodology. It finds that increased concentration unambiguously lowers employment. The Africa Region Working Paper Series expedites dissemination of applied research and policy studies with potential for improving economic performance and social conditions in Sub-Saharan Africa. The series publishes papers at preliminary stages to stimulate timely discussions within the Region and among client countries, donors, and the policy research community. The editorial board for the series consists of representatives from professional families appointed by the Region s Sector Directors. For additional information, please contact Momar Gueye, (82220), mgueye@worldbank.org or visit the Web Site: The findings, interpretations, and conclusions in this paper are those of the authors. They do not necessarily represent the views of the World Bank, its Executive Directors, or the countries that they represent and should not be attributed to them

3 Authors Affiliation and Sponsorship Johannes Wolfgang Fedderke, Professor of Economics, Cape Town University, South Africa, Consultant at The World Bank Gábor Szalontai, School of Economics, Cape Town University, South Africa, Consultant, The World Bank. Sponsor and Editor: Željko Bogetić, Lead Economist, AFTP1, The World Bank. FOREWORD This paper is a result of a wider policy research and knowledge work on growth and jobs issues in South Africa, which the Bank promotes in collaboration with leading South African researchers. The objective is to contribute to major economic and policy issues facing South Africa as it embarks on the second decade of its remarkable democratic transition. These issues include growth and jobs, export competitiveness, service delivery, small and medium-size enterprise development and investment climate, industrial concentration, infrastructure and growth, municipal and financial management, land reform, regional integration, trade and poverty, HIV/AIDS, and especially important service delivery. It is hoped that dissemination of papers such as this will contribute to a wider exchange of ideas on policy and development experiences both within South Africa and across the African countries. Such knowledge work is key to understanding complex development issues and dilemmas confronting the policymakers. It is also a necessary ingredient in promoting sound policies and economic growth in the region. Ritva Reinikka Country Director Botswana, Lesotho, Namibia, South Africa, Swaziland The World Bank

4 Industry Concentration in South African Manufacturing: Trends and Consequences, by Johannes Fedderke and Gábor Szalontai School of Economics, University of Cape Town December 2005

5 Industry Concentration in South African Manufacturing: Trends and Consequences, Summary This paper examines industry concentration for the South African manufacturing sector over the period (including the last year of the last manufacturing industry census data available) for the three-digit industry classification. The paper documents both the high level of industry concentration in South African manufacturing, and a rising trend in concentration across a wide range of industries. The paper further explores the impact that industry concentration has on a wide range of indicators of industry performance. We find that increased concentration serves to lower output growth, raise unit labour costs, and to lower labour productivity. The impact on employment, total factor productivity and investment rates based on bivariate examinations of the data are ambiguous. The paper concludes by examining the impact of concentration on employment and investment rates using a dynamic heterogeneous panel estimation methodology. We find that increased concentration unambiguously lowers employment. For investment rates, increased inequality of market share serves to raise investment rates, while falling firm numbers for any given inequality of market share lowers investment rates. The difference can be interpreted as the distinct impact that the pursuit of managerial objectives (large market share promoting further productive capacity expansion) and market contestability (under falling numbers of firms in an industry, monopolistic incentives to curtail productive capacity rise) have on investment. JEL Classification: Keywords: L11, L5, L6, D29, D49 Concentration, South Africa, Manufacturing industry, Investment, Employment.

6 Industry Concentration in South African Manufacturing: Trends and Consequences, by Johannes Fedderke and Gábor Szalontai* I. Introduction The study of industrial concentration in the South African manufacturing sector has been largely neglected in the past. While notable exceptions exist, 1 concentration has featured less in the debate on South African industry characteristics and performance than the exceptionally high degrees of concentration might suggest it should. The lack of availability of concentration ratio data series has surely hindered research in this field. This paper is aimed at providing an initial exploratory analysis of concentration within South African manufacturing, by extending data on concentration to the period. We explore trends in concentration over this time, and extend the analysis to a preliminary consideration of its impact on industrial performance. We focus on two measures of concentration, namely the Gini coefficient and the Rosenbluth index, both of which document high degree of industrial concentration. We analyse the changes in these measures over the period and investigate patterns of association between the concentration measures and our indicators of industry performance. We find that for the 24 industries considered, in almost all sectors, less than 5% of firms account for over half the industry output. We also find that there has been considerable change in the relative rankings of industry concentration over the period under study. These high levels of concentration and their dynamic nature clearly warrant further investigation. *School of Economics, Cape Town University. Note: The paper was edited for publication in the Africa Region Working Paper series by Željko Bogetić (Lead Economist, AFTP1). 1 See for example Du Plessis (1989), Fourie and Smith (1989), and Leach (1992).

7 The paper also considers the direction of association between the concentration measures and a range of indicators of industry performance. The initial focus of the analysis is on bivariate associations. The results show that there are numerous interactions between our two concentration measures and other economic variables. What is also evident is that the concentration measures generally drive the measures of economic performance and not vice versa market structure influences conduct and, therefore, performance. ARDL cointegration analysis further suggests that higher industry concentration serves to lower output growth and labour productivity, to raise relative unit labour cost, while the impact on employment, investment rates and total factor productivity are ambiguous. The paper therefore also considers the relationship between concentration and both employment and investment rates in multivariate specifications, employing panel estimation techniques. The general findings are that increased concentration lowers employment in South African manufacturing industry. For investment, the implication is that rising market share increases investment if sufficient rival firms remain in the market. By contrast, falling market contestation as measured by the number of firms in the industry serves to strongly decrease the investment rate. One interpretation of this contrasting evidence is that increasing market share promoting investment rates is evidence of the presence of managerial objectives in South African manufacturing industry. By contrast, where small firm numbers have already been obtained, and hence market contestation is diminished, monopolistic incentives to reduce productive capacity emerge. Section 2 of the paper introduces the data and methodology employed in the study. Section 3 presents the descriptive evidence on concentration. Section 4 considers the bivariate associations between concentration and a range of industry performance indicators. Section 5 presents econometric evidence on the impact of concentration on employment and investment rates. Section 6 concludes. II. Data and Methodology Consistent set of concentration indexes over the period data had to be compiled for a number of years from raw data. The raw data were provided by Statistics South Africa (StatsSA) on request.the compilations are based on tables published by StatsSA in the Census of Manufacturing. For 3-digit SIC manufacturing industries, Table 6 of the Census of Manufacturing shows the cumulative percentages of gross output, accounted for by cumulative percentages of firms. These tables were available in the 1976, 1979 and 1985 manufacturing censuses. The tables for the 1988, 1991, 1993 and 1996 censuses were not published in the original censuses, but were compiled by StatsSA on request. The required data for the 1972 census was taken from Leach (1992). While our sample period extends the one previously available, the limitation of the sample is that it covers only the years until 1996 because no more recent manufacturing census data are available. However, with slow-changing concentration 2

8 ratios, the findings of this study are likely to be relevant for policy at least for several years after the last year of the sample period. Discrete measures of concentration consider only a limited number of firms in an industry. 2 These frequently used measures, such as the percentage of industry size accounted for by a certain number of firms in the industry, have the advantage that they are easy to calculate as they depend on only one point on the concentration curve. If concentration curves of industries under study intersect, however, there will be ambiguity in determining which industry is more concentrated relative to another. Summary concentration measures in contrast to the discrete measures take into account all the firms in the industry. Two of the more prominent summary indices of concentration are the Herfindahl and Rosenbluth indices. The basic differences between various summary measures of concentration lie in the weights assigned to the market shares of firms in calculating the respective indices (Needham, 1978). Changes in the size distribution of firms will produce changes of different magnitudes in the various indices. The Herfindahl Index will be relatively insensitive to changes in small firms shares, while the Rosenbluth index will react strongly, due to the different weights assigned to small firms by each index. Another class of concentration indices is given by relative concentration measures. These are in contrast to the absolute measures such as the Herfindahl or Rosenbluth measures. Relative concentration measures can be depicted using Lorenz curves and they focus attention on the degree of inequality of firm sizes. Gini s concentration measure which we use in the present study is the ratio of the area between the Lorenz curve and the line of absolute inequality. This measure is bounded by 0 for equal sized firms and 1 for an industry in which there is only one firm. Again, the basic difference between relative and absolute measures of concentration lies in their weighting schemes. Absolute concentration measures are weighted sums of the firms shares, while relative measures are weighted averages (Needham 1978). In the international literature concentration measures such as the Herfindahl index have become the most widely used but could not be used in South Africa for this purpose. Unfortunately, for South African industry, the Herfindahl index while currently published by StatsSA, could not be computed on the basis of the data available for earlier manufacturing indexes. The Herfindahl index is defined as: n HI = ( ) 2 ms i where ms i is the market share of firm i, and n denotes the number of firms i= 1 in the industry. This measure is bounded by 1/n under equal firm size and 1 when only 1 firm is present in an industry. The unavailability of the market shares contributed by firm i in the historical manufacturing census precludes the computation of the Herfindahl index for South Africa. Previous authors interested in concentration in South African 2 Typical examples of such measures are given by the occupancy count the number of firms producing a given percentage of output - and the four firm concentration measure. 3

9 industry have, therefore, resorted to the use of the Rosenbluth and Gini indexes, the route we also chose to follow. 3 In the present study, the Gini coefficient was calculated as the area between the Lorenz curve and the line of equal distribution of industry production. Since the cumulative percentages [0, 5, 10,, 100] of firms given in the manufacturing censuses are evenly spaced, a technique of numerical integration known as the Simpson s onethird rule was used to approximate the Lorenz curve. Simpson s one-third rule is based on the fact that any three points determine a unique quadratic polynomial. Thus the Lorenz curve can be approximated by successive arcs of quadratic polynomials through successive sets of 3 known and equidistant points on the curve. This provides one quadratic estimating the curve over the range 0 to 10, another from 10 to 20, with the 10 th quadratic estimating the curve from 90 th to the 100 th percentile of firms. The area under the Lorenz curve can then be estimated by summing the areas under each quadratic, and from this the area between the Lorenz and the line of equal distribution follows trivially. Formally Simpson s one-third rule can be written as: x2 h f x) dx = ( f 0 + f1 + 3 x 0 ( f 2 ) (1) where h represents the interval width (5% or 0.05 in the case of the manufacturing census data), and f i is the percentage contribution to output of firms at the 3 points at which each successive arc is being estimated. The rule is applied separately to each 10 percent interval and the area under the curve is obtained by summation of the individual results. Linear interpolation was used to calculate the Gini for the years in which a manufacturing census was not published. The Rosenbluth index is defined as: 1 n R= 2 ( ims. i ) 1 (2) i= 1 where ms i is the market share of the ith-ranked firm and n the number of firms (see the discussion in Leach 1992). The measure is bounded by 1/n when firms are of equal size and tends to 1 the more unequal are firm sizes. As noted by Leach (1992) the Rosenbluth index of concentration can also be expressed as a positive function of the Gini coefficient and a negative function of the number of firms: 1 R = { n(1 G)} (3) 3 Du Plessis (1989) and Fourie and Smit (1989) employ the Gini coefficient as a measure, and Leach (1992) uses both the Gini measure and the Rosenbluth measure of concentration. 4

10 where G is the Gini coefficient and n is the number of firms in the industry. In the present study the Rosenbluth index was computed using (3). Figure 1 demonstrates the relationship between the Rosenbluth and Gini measures of concentration. A number of points are worth noting about the two indexes. First, while there is a clear functional relationship between them, it is a non-linear one. A rising level of production inequality as measured by the Gini results in increases in the Rosenbluth index, but at an increasing rate (Rosenbluth is convex in the Gini). Second, the Rosenbluth index responds strongly to changes in the number of firms present in the industry, which the Gini does not. Thus for a Gini coefficient of 0.9, a tenfold increase in the number of firms present in production will lower the Rosenbluth index tenfold, even though production is equally concentrated under the computations. 4 Mapping of Gini into Rosenbluth Index Rosenbluth 0.4 Rosenbluth10 Rosenbluth40 Rosenbluth Gini Figure 1: Relationship between Gini coefficient and Rosenbluth concentration index. In Rosenbluthx, x denotes the number of firms involved in the comparison. Due to differences between the 4 th and 5 th editions of the Standard Industrial Classification for all economic activities (SIC) the approach of Kayemba (2000) was adopted in combining industry codes, to make comparisons between sectors feasible. The South African classification convention follows SIC which first appeared in 1970 (Kayemba 2000). Revisions in the classification system have led to several discrepancies 4 Leach (1992: 395) does not consider the Gini coefficient a measure of concentration, merely one of inequality of the distribution production - since the measure does not adjust for the number of firms that produce the concentrated output. Of course this argument assumes that unequally distributed production across firms is contestable no matter how small the firms mounting the contestation. This may or may not be true. 5

11 emerging across the different manufacturing censuses. The new classification was used for the 1993 and 1996 censuses. The results for the petroleum industry are problematic. Until the 1993 manufacturing census the Petroleum Industry (SIC 353) was combined with Other Manufacturing (SIC 390). This was done for political reasons and the decision to disaggregate the results was taken and implemented only from the 1993 Census of Manufacturing onwards. Data on the Tobacco industry (SIC code 314) was only available for two of the censuses, and the sector had to be dropped from analysis due to data unavailability. Thus we are left with 24 industries that we examine in detail. III. Changes in Concentration, We report results for the Rosenbluth absolute summary measure of concentration, the Gini relative summary measure of concentration, as well as the discrete measure of concentration provided by the proportion of production contributed by specified percentages of all firms in an industry. The evidence from the discrete concentration measure and the Gini coefficient is strictly consistent, indicating a high and rising level of industry concentration. The evidence from the Rosenbluth measure is more muted in indicating rising concentration though nevertheless provides evidence in favour of high and rising concentration. We begin with the discrete measures of concentration, by considering the proportion of output accounted for by some given percentages of firms. Table 2 reports the results. In the interests of parsimony we report results only for 3 censuses, and we report the percentage of output produced by 5, 10 and 15 percent of all firms in a specific industry. What is notable is that in all three census years, in almost all sectors the largest 5% of firms produce more than 50% of output. By the 1990 s, there is no manufacturing sector in which 5% of the firms do not produce at least 50% of total output. And the largest 10% of firms produce at least 50% of output for all three manufacturing censuses under consideration. For a number of sectors, the largest 5% of firms have been able to increase the proportion of output they contribute strongly from the 1970 s to the 1990 s. Figure 2 shows that in Rubber & rubber products, Beverages, Glass & glass products, and Other industries, the largest 5% of firms have been particularly successful in raising the proportion of output over the sample period. But the upward trend has been general across most manufacturing industry, so that by 1996, 4 sectors had the largest 5% of firms contribute more than 80% of output. 5 And only 6 sectors had the largest 5% of firms contribute less than 60% of output. 6 Independently of any summary concentration measure, therefore, it is clear that South African manufacturing industry maintains a high 5 Motor vehicles, Rubber, Beverages, Other industries. 6 Electrical machinery, Furniture, Textiles, Clothing, Footwear, Plastics. 6

12 level of concentration, with a small proportion of firms dominating production in almost all sectors. Summary measures serve to confirm the evidence though more unambiguously in the case of the Gini than the Rosenbluth measure. Table 3 shows the trends in the Gini coefficient as a measure of concentration both in level terms and industry rank. Here and for the Rosenbluth parsimony restricts results to the years 1972, 1980, 1990 and The evidence here further supports the result from the discrete concentration measure that suggests increased concentration in the South African manufacturing industry, by demonstrating increasing inequality in the distribution of productive activity over the sample period. The total of 22 sectors show increased inequality between 1972 and Only 2 sectors have shown a decrease in their Gini coefficient over the same time period. 7 Looking at the relative rankings, there has also been a noticeable shift within the sectors. Only 2 of the 24 sectors, 8 showed no change in their relative rankings. Of all the sectors, 9 have had a negative change in their rank, indicating an increase in inequality between 1972 and 1996 relative to other manufacturing sectors. 9 The positive change for the remaining 12 sectors indicates decreased inequality over this time period. The results from the Rosenbluth measure are more mixed. Table 4 reports the trends in the Rosenbluth index in level terms and in industry rank. The evidence confirms an increase in the level of concentration in some manufacturing sectors, though there is also countervailing evidence. Seven sectors show a positive change between 1972 and 1996, 10 an indication of increased concentration. The remaining 17 sectors have become less concentrated over the same time period. It is notable that the increases in concentration as measured by the Rosenbluth have been most prevalent in the already concentrated industries. Figure 3 demonstrates that 3 of the 5 most concentrated sectors in South African manufacturing were amongst those that reported an increase in concentration over , and 6 of the 7 sectors reporting an increase in concentration lie in the 50% of sectors above the mean level of concentration. 7 Electric machinery, Other transport equipment. 8 Other Chemicals and Motor Vehicles. 9 Food, Leather & leather products, Wood & wood products, Paper & paper products, Basic chemicals, Plastics, Glass & glass products, Basic non-ferrous metals, Other industries. 10 Food, Beverages, Leather & leather products, Basic chemicals, Glass & glass products, Basic iron & steel, and Basic non-ferrous metals. 7

13 % 10% 15% 5% 10% 15% 5% 10% 15% Food and food products Beverages Textiles Clothing, except footwear Leather and products from leather Footwear Wood and wood and cork products Furniture Paper and paper products Printing, publishing and allied industries Basic chemicals Other Chemicals Rubber products Plastic products Glass and glass products Other Non-metals Basic iron and steel industries Non-ferrous metal basic industries Metal products, except machinery and equipment Machinery, except electrical Electrical machinery apparatus Motor vehicles, parts and accessories Transport equipment Other manufacturing industries Table 1. Contribution to output, by given percentage of firms. 8

14 Plastic products Footwear Clothing, except footwear Leather and products from leather Textiles Wood and wood and cork products Non-ferrous metal basic industries Furniture Motor vehicles, parts and accessories Basic iron and steel industries Other Chemicals Other Non-metals Basic chemicals Transport equipment Food and food products Printing, publishing and allied industries Electrical machinery apparatus Metal products, except machinery and equipment Machinery, except electrical Rubber products Beverages Glass and glass products Paper and paper products Other manufacturing industries Figure 2: Proportion of Output Contributed by Largest 5% of Firms

15 Low rank indicates high concentration Change Δ rank Food and food products Beverages Textiles Clothing, except footwear Leather and products from leather Footwear Wood and wood and cork products Furniture Paper and paper products Printing, publishing and allied industries Basic chemicals Other Chemicals Rubber products Plastic products Glass and glass products Other Non-metals Basic iron and steel industries Non-ferrous metal basic industries Metal products, except machinery and equipment Machinery, except electrical Electrical machinery apparatus, appliances and supplies Motor vehicles, parts and accessories Transport equipment Other manufacturing industries Table 3. Gini Coefficients. Levels and rankings. 10

16 Ranks: Low rank indicates low concentration Change: Food and food products Beverages Textiles Clothing, except footwear Leather and products from leather Footwear Wood and wood and cork products Furniture Paper and paper products Printing, publishing etc Basic chemicals Other Chemicals Rubber products Plastic products Glass and glass products Other Non-metals Basic iron and steel industries Basic non-ferrous metal industries Metal products, Machinery, except electrical Electrical machinery Motor vehicles Transport equipment Other industries AVERAGE FOR ALL INDUSTRIES Table 2. Rosenbluth Concentration Index. Level and ranking. Rank 1972 Rank 1980 Rank 1990 Rank 1996 Δ rank

17 Clothing, except footwear Food and food products Furniture Rosenbluth Concentration Index: Rosenbluth Index Glass and glass products Rubber products Transport equipment Basic iron and steel industries Non-ferrous metal basic industries Basic chemicals Paper and paper products Beverages Footwear Leather and products from leather Other manufacturing industries Motor vehicles, parts and accessories Other Non-metals Plastic products Other Chemicals Electrical machinery apparatus, appliances and supplies Textiles Wood and wood and cork products Printing, publishing and allied industries Machinery, except electrical Metal products, except machinery and equipment Figure 3: Concentration by industry, , Rosenbluth Index

18 The ranking of the sectors (according to Rosenbluth) also reveals the presence of considerable shuffling of sectors in their relative rank ordering of concentration. Only 3 of the 24 industries under study showed no change in their relative ranking of concentration, 11 while a number of sectors exhibit strong changes. Of all the sectors 9 have had a negative change in their rank, indicating an increase in their relative concentration. 12 The remaining 12 sectors exhibit positive changes in their rank, or a decrease in their level of concentration relative to other industries. Two factors explain the differences between the results to emerge from the Rosenbluth and the Gini measures. 13 First, recall that only the Rosenbluth measure takes into account the number of firms in an industry. Second, and critically, the Rosenbluth concentration measure is noted for its sensitivity to the presence of small firms with little market share a factor that has favoured the use of the Herfindahl index internationally. Thus, the Rosenbluth index may well serve to understate the degree of concentration though it is useful in capturing the effect of market entry. On balance, evidence suggests that South African manufacturing industry has historically been highly concentrated, and that the trend has been to increase, not decrease concentration. It remains to establish what, if anything, can serve to explain the rising concentration, and what impact such rising concentration has had on manufacturing sector performance. IV. Patterns of Association An immediate question that presents itself is whether the two measures of concentration presented in this paper are of significance to other dimensions of the economy. Only where concentration exercises a significant impact on economic performance or where it is itself the outcome of other aspects of economic activity is it of general analytical interest. In this section of the paper we, therefore, extend the analysis in two additional directions. The procedure followed is explicitly atheoretical. We examine test statistics that provide some insight into the nature of the association between variables, both in terms of direction of association, and the sign of the association. Throughout, the analysis is conducted at the three digit sectoral level. Examination is of the association between each of the Rosenbluth and Gini concentration measures, and output (as measured by value added), employment, total capital stock, investment, growth of total factor productivity, relative real unit labour cost, 14 the ratio of gross operating surplus to 11 Metal products, Glass & glass products, Wood & wood products. 12 Footwear, Furniture, paper & paper products, Printing, Basic chemicals, Rubber, Plastics, Other nonmetal industries, Machinery, Electrical machinery, Other transport equipment, Other industries. 13 Note that the differences also extend to industry rankings on the two measures. For instance, while the most concentrated industry as measured by the Rosenbluth index in 1996 is Glass and Glass products, the Gini coefficient consistently identifies the Motor vehicles sector as the one subject to the most unequal production (except in 1980 when the Basic Iron and Steel industry is most unequal). 14 This data was kindly supplied by Edwards and Golub (2002). 13

19 value added, and labor productivity (as measured by the ratio of value added to total employment). In order to explore the directions of association between the variables included in this study, we consider the use of the test statistic proposed by Pesaran, Shin and Smith (1996, 2001) (hereafter PSS) F-statistics. Results are reported in Table 4: our tests show that there are numerous and significant interactions between our concentration measures and output growth, employment growth, investment, TFP, relative real unit labour cost, GOS/GDP and output per worker. For only one sector (Printing, publishing and allied industries) do we find no pattern of association. GDP Growth Empl. Invest Total Invest Mach. TFP RULC GOS/ Y/L Growth GDP Food Xi ** 10.47** ** Rosen 5.57* ** 5.50* * Xi ** ** Gini Beverages Xi 8.49** ** 9.62** 5.85** 8.79** Rosen * Xi 6.14** ** Gini * Textiles Xi ** 10.84** 11.21** Rosen ** 6.17** 2.29 Xi * 5.06* 5.11* * Gini ** 2.20 Wearing apparel Xi ** * See also the discussion in Pesaran (1997) and Pesaran and Shin (1995a, 1995b). 16 Suppose that the question is whether there exists a long run relationship between the set of variables yt, x1,t, xn,t. Univariate time series characteristics of the data are not known for certain. The PSS approach to testing for the presence of a long run relationship proceeds by estimating the error correction specification given by: p n p n+ 1 Δ yt = α0 + βiδ yt i+ γ ji, Δ xjt, i+ δ1yt 1 + δkxkt, 1 + εt i= 1 j= 1 i= 1 k= 2 (4) The order of augmentation, p, is determined by the need to render the error term white noise, and is chosen from the set of all feasible lag structure combinations by means of an information criterion.16 The test proceeds by computing the standard F-statistic for the joint significance of δ1= δ2= = δn+1 = 0. The distribution of the test statistic is non-standard, and influenced by whether the xi,t are I(0) or I(1). The critical values are tabulated by Pesaran, Shin and Smith (1996), with xi,t I(0) i providing a lower bound value, xi,t I(1) i an upper bound value to the test statistic. The test statistic is computed with each of the yt, x1,t, xn,t as the dependent variable. Where the estimated test statistic exceeds the upper bound value, we reject the null of δ1= δ2= = δn+1 = 0, and infer the presence of a long run equilibrium relationship. Where the estimated test statistic lies below the lower bound value, we accept the null, and infer the absence of a long run equilibrium relationship. The test is indeterminate either where the computed test statistic lies between the upper and lower bound values (in which case it is not clear whether a long run relationship between the variables is present), or where more than one variable is confirmed as the outcome variable of a long run equilibrium relationship (in which case the long run relationships between the variables would not be unique).16 14

20 Rosen ** 0.05 Xi * 4.99* ** Gini ** Leather Xi 13.35** ** ** 7.31** Rosen Xi 11.10** ** Gini Footwear Xi 11.56** 5.02* 6.48** ** Rosen Xi * * ** 3.78 Gini Wood Xi 6.40** ** Rosen Xi 5.68* 6.07** * 7.77** Gini 6.38** ** 5.28* 9.26** Paper Xi Rosen ** Xi 6.14** ** Gini Printing Xi * Rosen Xi 5.04* * Gini Basic chemicals Xi ** 24.64** 21.87** 9.77** ** Rosen 9.92** 6.94** 5.01* * Xi ** Gini 7.26** 6.68** 8.87** 6.76** 7.57** ** 6.43** Other Chemicals Xi ** Rosen ** 6.63** Xi Gini * 6.02** 6.31** 5.15* ** 4.66 Rubber Xi 5.48* ** 3.49 Rosen ** Xi ** 7.04** 5.01* ** 2.08 Gini Xi 8.25** 6.39** 7.03** ** ** GDP Growth Empl. Invest Total Invest Mach. TFP RULC GOS/ Y/L Growth GDP Xi 6.68** 18.52** Gini Glass Xi ** ** 0.28 Rosen ** Xi ** 2.73 Gini Other Non-metals Xi 5.84** ** 5.05* Rosen ** 9.61** ** 4.40 Xi 6.99** ** 3.11 Gini Basic iron & steel Xi 10.53** ** 8.67** 6.47** 14.42** * Rosen Xi 8.65** 6.47** ** 7.91**

21 Gini ** Basic non ferrous metals Xi ** 1.41 Rosen ** 0.23 Xi ** 6.05** 6.76** 2.18 Gini 6.54** 6.73** * 6.12** Metal products Xi * * Rosen * Xi ** * 1.47 Gini * Machinery Xi ** 6.28** Rosen Xi Gini ** Electrical machinery Xi ** 5.28* 7.26** 5.25* ** 0.83 Rosen Xi * Gini ** Motor Xi Rosen Xi ** 5.49* 5.40* Gini Other transport Xi Rosen ** Xi * Gini ** Furniture Xi ** ** 5.47* Rosen Xi ** ** 0.64 Gini ** 5.01 Other manufacturing Xi ** * Rosen Xi ** Gini * Table 4: PSS ARDL F-Tests for Direction of Association. Rosen denotes the Rosenbluth index. Gini denotes the Gini coefficient. X i denotes the relevant economic development indicator specified. * Indicates indeterminate outcome. ** Indicates presence of a long run relationship. Table 4 allows us to analyse the patterns of association between the 2 concentration measures and our indicators. We will consider each of our concentration measures and indicator variables in turn. Looking at the Gini measure, for output growth we find that for 6 sectors it is the Gini that drives output growth, for 3 sectors the direction of association is reversed and for 2 sectors the association could be simultaneous. For employment growth, the Gini forces employment growth for 9 sectors, for 2 sectors employment growth forces the Gini and for 3 sectors the results are ambiguous. For total investment we find that in 1 sector the Gini forces total investment, for 4 sectors it is total investment that forces the Gini and for 4 sectors the results are ambiguous. For investment in machinery, in 2 sectors Gini is the forcing variable, in 2 sectors Gini is the outcome variable and in 1 sector the result is ambiguous. For TFP the direction of association runs from Gini to TFP in 7 sectors, while for 2 sectors the direction of association is reversed and for 4 sectors it is ambiguous. For relative real 16

22 unit labour costs, there are 4 sectors in which the Gini is the forcing variable, 2 in which the Gini is the outcome variable and none in which the direction of association is ambiguous. For GOS/GDP we have 5 sectors where GOS/GDP is the outcome variable and 4 in which the Gini is the outcome, while for a remaining 1 sector the results are ambiguous. Finally for output per labourer we have 1 sector in which the Gini is the forcing variable, 4 sectors in which the Gini is the outcome variable and a further 4 sectors for which the results are ambiguous. Looking at the Rosenbluth measure, for output growth we find that for 7 sectors it is Rosenbluth that drives output growth, for 1 sector the direction of association is reversed and for 2 sectors the association could be simultaneous. For employment growth, Rosenbluth forces employment growth for 7 sectors, for 2 sectors employment growth forces Rosenbluth. For total investment we find that in 9 sectors Rosenbluth forces total investment, for 2 sectors it is total investment that forces Rosenbluth and for 2 sectors the results are ambiguous. For investment in machinery, in 5 sectors Rosenbluth is the forcing variable. For TFP the direction of association runs from Rosenbluth to TFP in 7 sectors, while for 2 sectors the direction of association is reversed and for 4 sectors it is ambiguous. For relative real unit labour costs, there are 4 sectors in which the Rosenbluth is the forcing variable, 5 in which Rosenbluth is the outcome variable and 1 sector in which the direction of association is ambiguous. For GOS/GDP we have 8 sectors where GOS/GDP is the outcome variable and 4 in which Rosenbluth is the outcome. Finally for output per labourer we have 6 sectors in which Rosenbluth is the forcing variable, 1 sector in which Rosenbluth is the outcome variable and a further 6 sectors for which the results are ambiguous. While we find evidence of interaction between a number of industries, the degree and sign of the interactions vary widely across the different industries. To analyse the degree of interaction we follow the approach of Fedderke and Schirmer (2003). Degree of interaction is defined as the number of our indicators that interact with either of the concentration measures employed. Thus the strength of the interaction is not taken into account. Table 6 reports the high impact sectors, for which we find an interaction between the concentration measures and 8 or more of the indicators as well as low impact sectors, which exhibit interactions with 7 or less indicators. We see from the table that for the majority of sectors are classified as low impact, indicating that concentration is typically associated with a few of the indicator variables. From Table 6 we also see that the channels of association are industry specific. The general finding is that concentration forces the indicators of industry development though in a few instances industry concentration appears to be the outcome of industry development see also Table 7. For example for Beverages, concentration seems to have an impact on output growth, while for the Basic Chemicals secto, the direction of association is reversed. As a final step in our descriptive analysis we consider the signs of the association implied by the PSS F-tests. The signs of association are obtained from the associated ARDL cointegration analysis. Tables 7 and 8 present the findings, summarising the direction of association as well as the sign of association. Since the 17

23 specifications examined here are clearly underspecified, we note, but attach less importance to the significance of regressors, than to the sign implied by the bivariate association. High Impact Sectors Interaction with 8 Indicators Basic chemicals Basic iron and steel industries Non-ferrous metal basic industries Low Impact Sectors Interaction with 7 Indicators Food and food products Beverages Textiles Clothing, except footwear Leather and products from leather Footwear Wood and wood and cork products Furniture Paper and paper products Printing, publishing and allied industries Other Chemicals Rubber products Plastic products Glass and glass products Other Non-metals Metal products, except machinery and equipment Machinery, except electrical Electrical machinery apparatus, appliances and supplies Motor vehicles, parts and accessories Transport equipment Other manufacturing industries Table 5: Assessing the Extent of Interaction between Industry Concentration and Indicators of Industry Development Outcome Variable <0 >0 Output Growth Beverages (R*&G) Leather (R&G*) Wood (R*) Other non metals (R) Paper (G) Basic Chemicals (G) Basic Iron & Steel (G) Employment Growth Furniture (R&G) Other Manufacturing (R&G) Food (G*) Wood (G) Rubber (G) Plastic (G) Footwear (R*) Basic Iron & Steel (R) Other non metals (G) Textiles (R*) Basic Chemicals (R) Plastic (R) Machinery (R) Electrical Machinery (R) 18

24 Investment (Total) Investment (Machinery) TFP Metal Products (G) Motor (G*) Food (R*) Plastic (R) Rubber (G) Textiles (R*) Electrical Machinery (R) Beverages (R*&G) Leather (R&G) Footwear (R) Wood (R&G) Paper (G) Textiles (R) Wearing apparel (R) Footwear (R) Basic Iron & Steel (R) Machinery (R) Food (R&G) Basic Chemicals (R) Basic Iron & Steel (R*) Basic Chemicals (R) Plastic (R) Basic Iron & Steel (R&G) Basic non-ferrous metals (G) Rel. Unit Lab. Cost Beverages (R) Basic Chemicals (R) Basic Iron & Steel (R*&G) Wearing apparel (G) Basic non-ferrous metals (G) GOS/GDP Output/Labour Furniture (R) Electrical Machinery (R) Rubber (G*) Glass (G) Food (R) Leather (R*) Basic Chemicals (R*) Glass (G) Leather (R*) Glass (R) Plastic (R) Table 6: Summary of F-tests and ARDL Cointegration Findings. <0,>0 denotes the sign of the association. (R) denotes the Rosenbluth result, (G) the result for the Gini concentration index. * denotes statistical significance at the 10% level. Full estimation results available from the authors. Outcome Variable: Forcing Variable: Outcome Variable: Forcing Variable: Output Growth Output Growth Rosenbluth Basic Chemicals: >0 Gini Basic Non-ferrous metals: <0 Employment Growth Employment Growth Rosenbluth Basic Chemicals: >0* Gini Wood: >0 Other non metals: >0 Basic Non-ferrous metals: <0 Investment (total) Investment (total) Rosenbluth Other non metals: >0 Gini Wood: <0* Investment (mach.) Investment (mach.) Rosenbluth Gini Other Chemicals: >0 TFP TFP Rosenbluth Food: >0 Gini Wood: <0 19