Trade and Poverty in South Africa: Exploring the Trade-Labour Linkages

Abstract

An important mechanism through which trade affects poverty is through its impact on employment. This paper explores the relationship between trade, employment and technological change in South Africa using correlation analysis, a decomposition analysis and the estimation of an induced labour demand model using industry panel data. We find little correlation between employment changes and changes in protection or trade flows, but find a bias in tariff reductions towards labour-intensive sectors. Manufacturing trade flows have also been biased against labour-intensive sectors, but the net effect of trade on employment is close to zero or positive once indirect effects are included. The majority of employment change is attributed to technological change. We find little evidence that this technological change has been induced by increased openness.

Notes

1 An overview of many of these studies and an application of the various methodologies used in the debate is presented in Edwards (2006).

2 We draw on Quantec (2004) for employment data. There are, however, some concerns over the reliability of South African statistical series. Bhorat and Oosthuizen (2005) use household survey data and estimate a rise in employment of 200 000 individuals within manufacturing from 1995-2002. Quantec (2004) data indicates a decline in employment in manufacturing of 10% over the same period.

3 This is not a direct application of the Stolper Samuelson theorem (see Feenstra and Hanson, 1999) but rather a ‘consistency check' similar to that used internationally by Lawrence and Slaughter (1993) and locally by Fedderke et al. (2003), Behar and Edwards (2005) and Edwards (2006).

4 Edwards (2006), for example, estimates that bias of liberalization against labour-intensive sectors mandated a decline in real wages of 19% to 17% and a real increase in the return to capital of 5% to 20% between 1994 and 2003. Liberalisation also mandated a 40% - 47% decline in wages of semi- and unskilled labour between 1994 and 2003, although Edwards and Behar (2006) use firm level data to show that the mandated decline in wages is concentrated amongst semi-skilled labour.

5 Similar methodologies have already been applied to South Africa by Edwards (2001a, 2001b, 2006) and Jenkins (2002), but the analysis here updates and extends these studies by considering the labour impact of regional variations in the composition of trade.

6 Much controversy surrounds the reliability of South African statistical series, particularly those dealing with employment numbers. The Quantec data is compiled by combining a set of industry and national account indicators with a consistent input-output framework spanning three decades. In particular, the data are manipulated to ensure consistency with the Statistics SA, national accounts data and the Input-output structure of the Supply-Use tables prepared by Statistics South Africa. Sector level data for the years between the available Input-Output (IO) tables are mostly interpolated. This may induce significant errors into the data, particularly during the period subsequent to 1996, when the last official manufacturing Census was conducted.

7 There were of course other changes in the global and domestic environment that will have affected trade flows, so in the empirical analysis we can only assess the consistency of the results with those expected under liberalisation and are not able to test the relationship directly.

8 These sectors are the 2nd, 7th and 4th most important downstream industries for textiles, the 1st being clothing and footwear.

9 Jenkins (2002), for example, estimates that rising import penetration led firms to rationalise their use of labour leading to an estimated reduction in total employment in manufacturing of 100 000 between 1990 and 2001. Edwards (2003) uses firm level data and finds some evidence that trade-induced technological change reduced employment, but the effect was small.

10 Values are measured in 1995 prices. The limitations regarding this data, as discussed in footnote 6, remain valid and some caution is required in analysing the results.

11 The fixed effects model was developed for static models and we are estimating a dynamic model. In fact this is only a problem if we have a short time series as the lagged dependent variable will introduce bias, but this will get smaller as T, the number of time periods, gets larger. So for this study the fixed effects results should be reasonable.

12 Dunne and Edwards (2006) provide more details on the derivation of the model and the long run coefficients. It is interesting to note that the returns to scale, which can be computed from the coefficients is close to 1.

13 This entailed re-estimating the dynamic model using the Arellano and Bond (1991) estimation procedure to deal with potential problems associated with lagged dependent variable and endogeneity biases.