Industry Concentration in South African Manufacturing Industry: Trends and Consequences, 1972-96¶
Johannes Fedderke; Gabor Szalontai (2003). Industry Concentration in South African Manufacturing Industry: Trends and Consequences, 1972-96. School of Economics, University of Cape Town (author affiliation; working paper).
| Type | working paper |
| Scope | Economics and economic history of apartheid |
| Method | econometric |
| Period studied | 1972–1996 |
| Regions | South Africa |
| Themes | manufacturing industrial policy, business and state, labour markets wages, methods data sources, macro fiscal finance debt |
| Citation key | fedderke2003 |
Argument¶
South African manufacturing was highly concentrated throughout 1972-96, and concentration rose in most three-digit industries on the Gini measure, with a more mixed picture on the Rosenbluth index. Bivariate PSS/ARDL bounds tests suggest the direction of association mostly runs from concentration to performance. Higher concentration is linked to lower output growth, higher relative unit labour costs and lower labour productivity. Panel (Pooled Mean Group) estimates for 24 industries show concentration unambiguously lowers employment. For investment, unequal market share raises it but falling firm numbers lower it. The authors read this as managerial-objective behaviour combined with weakening contestability, and they conclude that competition policy should stay on the agenda.
Evidence and method¶
Census of Manufacturing (StatsSA) cumulative output-by-cumulative-firm-share tables for 1972 (from Leach 1992), 1976, 1979, 1985, 1988, 1991, 1993 and 1996, with the later tables compiled by StatsSA on request. Gini coefficients are computed by Simpson's rule with linear interpolation between censuses. Rosenbluth indices are derived from the Gini and firm numbers. Sectoral data on output, employment, capital stock, investment, TFP, relative unit labour costs, GOS/GDP and labour productivity are drawn on, including data from Edwards and Golub (2002) and a political instability index from Fedderke, De Kadt and Luiz (2001).
Place in the debates¶
- Concentration, efficiency and performance in South African manufacturing (structure-conduct-performance versus efficiency-gains view) (see Industrialisation, protection and industrial structure). Finds concentration does not raise efficiency. It is linked to lower output growth, lower labour productivity and higher unit labour costs, which the authors attribute to managerial objectives pursued at the expense of efficiency and profitability.
- Which concentration measure is appropriate (Gini as inequality of firm size versus Rosenbluth/Herfindahl as concentration adjusted for firm numbers) (see Data reliability and econometric method). Uses both because the Herfindahl cannot be computed from the historical census data. Notes the Gini shows clearly rising concentration while the Rosenbluth is more mixed and may understate concentration because it is sensitive to small firms.
- Competition policy and market contestability versus regulation of market share (see Industrialisation, protection and industrial structure). Argues the strongest returns come from keeping the number of firms in a market up (even to 5-10), rather than from regulating market share.
Findings¶
-
Only two industries became less unequal on the Gini between 1972 and 1996, so concentration rose almost across the board. (p. 8)
22 sectors show increased inequality between 1972 and 1996. Only 2 sectors have shown a decrease in their Gini
-
The Rosenbluth index shows a more mixed trend, with most industries less concentrated in 1996 than in 1972. (p. 8)
The remaining 17 sectors have become less concentrated over the same time period.
-
The authors caution that the Rosenbluth index may understate concentration because of its sensitivity to small firms. (p. 9)
-
Bivariate ARDL results link higher concentration with higher relative unit labour costs. (p. 13)
On relative real unit labour cost, both concentration measures predict a higher unit relative labour cost under increased concentration.
-
Panel estimates show that concentration (Rosenbluth) lowers manufacturing employment, with an elasticity of roughly -0.10 to -0.27. (p. 20)
with a 1% increase in the Rosenbluth concentration index leading to a decrease in employment between 0.10% and 0.27%.
-
Policy implication: ensuring market contestability matters more than regulating market shares. (p. 21)
the policy implication to follow is that the strongest potential returns for competition policy lies with ensuring market contestability, rather than by the regulation of market share.
-
Investment responds in opposite directions to the two concentration measures (Gini positive, Rosenbluth negative). (p. 22)
results are more nuanced, since the Gini and the Rosenbluth indexes carry opposite signs (Gini positive, Rosenbluth negative)
-
The downside investment effect of monopoly-like conditions is large. (p. 23)
the investment rate could be decreased by as much as 3 percentage points.
-
Data construction: the concentration series for the later censuses had to be specially compiled by StatsSA, since the data were not published. (p. 3)
The tables for the 1988, 1991, 1993 and 1996 censuses were not published in the original censuses, but were compiled by StatsSA on request.
-
Sample coverage: 24 three-digit industries remain after tobacco was dropped and classification problems were handled. (p. 6)
Thus we are left with 24 industries that we examine in detail.
Evidence ledger¶
Sources used¶
| Source | Kind | Period | How used |
|---|---|---|---|
| Census of Manufacturing, Table 6 (cumulative percentages of gross output by cumulative percentages of firms), 3-digit SIC; 1976, 1979 and 1985 censuses | industrial mining census | 1976, 1979, 1985 | Supplies the Lorenz-curve points from which the Gini and Rosenbluth concentration indices and the discrete concentration shares are computed. |
| Unpublished Census of Manufacturing concentration tables for 1988, 1991, 1993 and 1996, compiled by StatsSA on request | industrial mining census | 1988, 1991, 1993, 1996 | Extends the concentration series to the 1990s using raw data supplied specially by StatsSA. |
| Leach (1992) concentration data for the 1972 census | secondary literature compilation | 1972 | Provides the 1972 baseline concentration data, and the Rosenbluth-Gini relationship used to compute the Rosenbluth index. |
| Relative real unit labour cost data (Edwards and Golub 2002) | secondary literature compilation | approx. 1970-97 | Supplies the relative real unit labour cost indicator used in the bivariate PSS tests. |
| Index of political instability from Fedderke, De Kadt and Luiz (2001) | secondary literature compilation | 1970-97 | Serves as the systemic uncertainty regressor in the investment equation. |
| Kayemba (2000) combined manufacturing data series / SIC concordance | secondary literature compilation | 1970-96 | Provides the approach for combining 4th and 5th edition SIC industry codes into comparable industries. |
New data¶
A consistent 1972-96 series of Gini and Rosenbluth concentration indices, plus discrete concentration shares, for 24 three-digit SIC manufacturing industries. It was built from raw StatsSA census tables (1976, 1979, 1985 published; 1988-96 compiled on request; 1972 from Leach), with Gini computed by Simpson's rule and linear interpolation between census years.
Inference¶
Quantification: Econometric (causal design).
Descriptive computation of concentration indices from census cumulative-output tables, followed by sector-by-sector bivariate PSS bounds F-tests and ARDL cointegration for direction and sign of association. The final step is Pooled Mean Group dynamic heterogeneous panel estimation of labour-usage and investment equations with concentration regressors, interpreted through industrial-organisation theory (managerial objectives, contestability).
Key claims¶
| Claim | Basis | Strength |
|---|---|---|
| Increased concentration lowers output growth and labour productivity and raises relative unit labour costs; the direction of association mostly runs from concentration to performance. (p. 13) | Sector-by-sector bivariate PSS F-tests and ARDL signs (Tables 4a-7), using the sectoral performance series. | Partial data |
| Concentration unambiguously lowers manufacturing employment: a 1% rise in the Rosenbluth index reduces employment by 0.10-0.27%, with larger effects when firm numbers are small. (p. 21) | PMG panel estimates of the labour usage equation (Table 8) across 24 sectors, with Figure 4 combining the Gini and Rosenbluth effects. | Systematic data |
The author's caveats¶
The authors note that the Herfindahl index cannot be computed because historical censuses lack firm market shares, and that the Rosenbluth index may understate concentration because of its sensitivity to small firms. The tobacco industry was dropped for lack of data, and the petroleum industry was combined with Other Manufacturing until 1993. The Gini for non-census years is linearly interpolated, and the bivariate specifications are acknowledged to be underspecified and atheoretical.
Gaps in the evidence¶
Firm-level market shares, which would allow Herfindahl indices and direct tests of contestability and managerial behaviour, are not used; the paper relies on grouped census output distributions and interpolated years. Causal identification of concentration effects is weak given small industry panels, many bivariate tests, and possible endogeneity; instruments or firm-level panel data would test the argument better. Source documentation for the sectoral performance series (capital stock, TFP, employment) is thin, though the underlying StatsSA data existed then.
Replicability: Data described but not public.