The Structure of Growth in the Westen Cape Manufacturing Sector, 1970-1996

Abstract

This paper examines growth patterns in the Western Cape over the 1970-1996 period by means of primal growth accounting decompositions. Evidence by magisterial district, by statistical region and by SIC 3-digit manufacturing sector is presented. We find that the Western Cape differs from national growth patterns. Manufacturing in the Western Cape has relied consistently on capital accumulation for growth, while labour has contributed more to manufacturing growth than nationally. TFP growth has not been an important contributor to growth in Western Cape manufacturing, with a relatively minor exception in the 1980's.

Notes

1 See for example Fedderke (2002a) and Arora and Bhundia (2003).

2 See Fedderke (2004), and Fielding (1997, 2000).

3 See Aron and Muellbauer (2000) and Romm (2005).

4 See Fedderke (2002b) and Fedderke and Liu (2002).

5 See Kularatne (2002).

6 See Fedderke and Romm (2006).

7 See Perkins, Fedderke and Luiz (2005), and Fedderke, Perkins and Luiz (2006).

8 See Mariotti (2002) and Koch et al. (2005).

9 See Fedderke, De Kadt and Luiz (1999, 2003), and Fedderke and Luiz (2002, 2007c).

10 See Fedderke (2005).

11 See Fedderke, De Kadt and Luiz (2001a, 2001b), and Fedderke and Luiz (2007a, 2007b, 2007).

12 There are two potential exceptions. Naude and Krugell (2003) and Krugell and Naude (2005) both engage geographically disaggregated GDP growth rates for South Africa. While the write up in the two contributions is somewhat vague on the data employed, working paper versions of these papers used data from PIMSS. Global Insight was the source of the data, and continues to be the only source for geographically disaggregated data for South Africa over the post-1996 period. Should the two papers have been premised on the Global Insight data, serious concerns about the accuracy of the associated results arise. As Cameron (2005) reports, the published margin of error between Global Insight estimates of regionally disaggregated average annual GDP growth rates and StatsSA estimates varies between 50% and 260%. Over the space of a decade, the compounding effect of such errors would generate dramatic divergence. Given that Naude and Krugell (2003) and Krugell and Naude (2005) provide no explanation of how such margins of error were dealt with in their work, we find it impossible to engage with their results. In any event, in South Africa institutional distortions explicitly prevented convergence in many ways and over extended periods of time. This renders interpretation of results from a convergence framework considerably more difficult than in the established US precursors, for instance. For these reasons we prefer the simpler procedure constituted by growth accounting, at least as a first cut at the problem.

13 StatsSA does not make available 3-digit manufacturing sector data at the magisterial district level, for reasons of data confidentiality.

14 A fuller discussion of these and related issues can be found in Fedderke (2002b).

15 See also the discussion in Hulten (1986) and Jorgenson and Grilliches (1967). For a useful overview of the developments see Barro (1998), which provides a more elaborate treatment of the condensed material that follows here.

16 Note that under Hicks-neutrality the term for technological progress reduces to See Solow (1957).

17 See Diewert (1976).

18 Mutatis mutandis for the dual approach.

19 Unfortunately reliable instrumentation is particularly fraught in this context, making instrumental variable estimation difficult.

20 The degree to which variation in capacity utilization is important is a matter of some dispute. Hall (1988), Caballero and Lyons (1992) argue for its unimportance. Basu (1995) dissents. Oliveira Martins and Scarpetta (1999) provides an extension to the debate and methodology. Burnside, Eichenbaum and Rebelo (1993) extend the argument to labour hoarding over the business cycle. One should also bear in mind that one strand of the debate emphasizes that less than full capacity utilization is itself a sign of inefficiency. Fluctuations in TFP measurement due to fluctuations in capcity utilization would thus be interpretable as changes in efficiency. See for instance Domar (1961: 715 fn1).

21 For a South African application see the discussion in Thirtle, Van Zyl and Vink (2001).

22 Malmqvist indexes decompose productivity changes into changes in technical efficiency and an index of technical change. Change in technical efficiency is meant to capture relative efficiency (whether a sector is moving closer to or further away from best practice) while technical change is meant to measure changes in best practice. In effect, it distinguishes "catch-up" from "true" technological advance. Reliable implementation does require the identification of best practice, however, with both parametric (econometric) and nonparametric (programming) approaches being used in the literature. Results depend on the assumption that (some) observed data points reflect best practice. In parametric approaches results are sensitive to assumptions concerning the functional form of technology. In programming approaches, results are sensitive to measurement error while the absence of assumptions regarding functional form precludes the use of diagnostic tests to evaluate results. Both approaches are also unable to identify the contribution of factor inputs to production, information that is valuable in its own right. Discussions of Malmqvist indexes can be found in Charnes, Cooper and Rhodes (1978), Seiford and Thrall (1990), Fried, Lovell and Schmidt (1993), and Ten Raa and Mohnen (2000).

23 See Hall (1990), Roeger (1995), and Oliveira Martins and Scarpetta (1999) for discussions of this approach.

24 The now standard references are Griliches (1979), Romer (1986) and Lucas (1988).

25 See Romer (1990), Grossman and Helpman (1991: ch3), Aghion and Howitt (1992) and Grossman and Helpman (1991: ch4).

26 A full discussion of the detail can be found in Barro (1998).

27 See Fedderke, Kularatne and Mariotti (2007).

28 See Fedderke (2005).

29 The use of decompositions analogous to those used in this paper continues in the literature, though ideally the distinction between types of factor inputs is taken into account. Examples from the literature include Young (1995) for East Asian countries, Christenson, Cummings and Jorgenson (1980) for the OECD, Elias (1990) for Latin America. For a more encompassing view see Maddison (1987), and see also the discussion in Jorgenson (1988) and Fagerberg (1994).

30 See Leontieff (1953).

31 See for instance Harberger (1998), Roeger (1995), Oliveira Martins and Scarpetta (1999).

32 Domar (1961:724f) shows that the TFP computed on value added will be a multiple of the "true" TFP. Domar also points out that the magnitude of the TFP measured on gross output recognizing the impact of intermediate inputs, may simply reflect what he terms the "thinness" or "thickness" of the industry, viz. the extent to which inputs are transformed within the production processes of the industry. Use of the gross output TFP measure would therefore introduce another source of cross-industry variation in TFP not reflecting technical change properly understood.

33 Since RCR is generated on the additional value added generated in each industry, its attraction is that it enables additive aggregation. The process of aggregation avoids the problems highlighted by Domar (1961:717ff), since the concern is not with the computation of an aggregate growth rate of TFP, but with the aggregate gain in output due to TFP growth industry by industry.

34 Detailed description of all relevant data manipulations are available from the authors.

35 It is unfortunate that the most recent data on the manufacturing sector is eight years old. More recent data would throw light on the fruits of government policies in the ten years of democracy. It will be interesting to update the study when such data does become available. The more recent firm survey of 2001 does not provide ready geographic disaggregation.

36 The Pottery, China and Earthenware sector has been included in the Non Metallic Mineral Products sector as in the earlier years of the sample Pottery, China and Earthenware was not distinguished from Glass and Other Non-metallic Products. The data would suggest that during the 1970's and again during the 1990's many categories of manufacturing output were simply grouped under the sector Other Manufacturing Industries. It is clear that if this is the case the classification will affect results both by under-reporting changes in sectors to which activity has not been allocated, and over-reporting Other Manufacturing Industries activity.

37 We classify the growth performance of the magisterial districts into three categories, fast, intermediate and slow. The fast growing districts are defined as such simply by virtue of being the 11 top ranked districts in the relevant period. Slow growers by contrast are the 11 districts that grow most slowly, and the intermediate districts are the 11 sectors distributed between fast and slow growers.

38 The only exception to this finding is that nominal growth of the fastest growing districts rises in the 1980's, before declining substantially during the 1990's. However, this is in part an artefact of the introduction of the Moorreesburg district, with associated initial high growth rates.

39 See the discussion in Barro and Sala-i-Martin (1995).

40 See famously Romer (1986).

41 The attribution of the developed world's accelration has been to the role of technology. In the case of the developing countries of East Asia there is some dispute as to whether the acceleration is due to factor accumulation, or efficiency gains. See the discussion in Young (1995), and Lim (1994) for instance.

42 See Fedderke (2005) for an extensive discussion of the detailed evidence.

43 It is important to emphasise that the regression evidence requires careful interpretation. The evidence requires interpretation as identifying summary characteristics across magisterial districts, rather than in causal terms.

44 We consider the relative contribution of the two factor inputs, and technological progress to total manufacturing growth.

45 The 1990's report the same pattern in the Western Cape that Fedderke et al. (2001) report for South Africa as a whole. In particular, output growth in the manufacturing sector comes to be led heavily by capital investment.

46 Note that we have excluded Mossel Bay from the TFP evidence during the 1990's. The very large scale of investment in the magisterial district distorts the evidence significantly -- and completely dominates the evidence to emerge from all other magisterial districts. As a consequence we have suppressed the Mossel Bay data in the graphical representation of the data, in order to allow insight into the development in the province as a whole.

47 The study remarks repeatedly on the likelihood that this is a reflection of problems of data classification.

48 By diversification we mean that the total manufacturing sector output of the statistical region is not dominated by one or two three digit manufacturing sectors.

49 A distinct possibility is that new manufacturing activity was simply classified under "other" rather than receiving proper classification in relevant industry groupings.

50 This is true regardless of how we treat the data issues presented by the apparent industry start-up.

51 The concern voiced in preceding sections raised by the Other Manufacturing Industry performance resurfaces again in the growth accounting exercise. The very dramatic growth rates implied by capital accumulation in this sector raises the prospect that increasing manufacturing activity was inaccurately classified in the OMI sector, rather than appropriately allocated to industry grouping by Statistics South Africa.

52 It is important to emphasise that the regression evidence requires careful interpretation. The evidence requires interpretation as identifying summary characteristics across manufacturing sectors, rather than in causal terms.

53 The evidence presented excludes the Other Manufacturing Industries sector because the 1980's and 1990's distort the findings substantially, due to very strong capital and TFP growth. The likely reason for these findings are the classificatory problems related to the OMI sector that have been noted a number of times in the preceding discussion. The strength of the effect in the current context is such that to all intents and purposes only the OMI sector comes to contribute to the growth of manufacturing value added in the Western Cape in these two categories. Once again, we caution that significant classificatory problems brought about by the inclusion of new manufacturing activity in the Western Cape over this period in OMI even where inappropriate, will have skewed the data and our results. Finally, we note that the strength of the effect also points to the likely candidacy of the Mossgas projects as driving the strength of the OMI capital and TFP growth.

54 The Food sector contributed positively to output growth through job creation during the 1970's and 1980's, though job losses during the 1990's led to a negative contribution to output growth from labour in the this important sector. Clothing again proves to consistently contribute positively to output growth through job creation, over all three sub-periods of the sample. By contrast, the Textiles sector has positive contributions to output growth from labour inputs during the 1970's and 1990's, but a negative contribution during the 1980's.

55 In particular, this is true for Textiles, Fabricated Metal Products and Printing.

56 Note that results exclude Other Manufacturing Industry throughout.