Ersa Workshop Infrastructure and Growth – Theory, Empirical Evidence and Policy Lessons

1. Ersa Workshop Infrastructure and Growth – Theory, Empirical Evidence and Policy Lessons Cape Town, 29 – 31 May 2006 Infrastructural Investment in Long-run Economic Growth: South Africa 1875-2001 J. Fedderke (UCT), P. Perkins (Wits), J. Luiz (Wits)

2. 1. Introduction Renewed interest amongst South African policy-makers in economic infrastructural investment following an extended period of decline (from mid-1970s to 2002) The decline coincided with poor economic growth in SA In the literature, the empirical evidence on the infrastructure – growth relationship is relatively mixed This paper provides a long-run, time-series investigation of the infrastructure – growth relationship in SA

3. 2. The Role of Infrastructure in Economic Development Labour-intensive Cobb-Douglas production function, from Barro model (1990): y = Agak1-a (0 < a < 1) where y = output A (> 0) = level of technology g = government spending on productive services (e.g. infrastructure) k = private capital

4. Marginal product of g: Marginal product of k: As g/y rises, g/k rises: As g/y rises, dy/dk rises but dy/dg falls

5. Maximize utility: where Then steady state growth is given by:

6. The impact on g of raising g/y depends on whether dy/dg is greater or less than 1: When dy/dg > 1, dg/d(g/y) > 0 When dy/dg = 1, dg/d(g/y) = 0 When dy/dg < 1, dg/d(g/y) < 0 The core rationale for infrastructural investment that emerges is that it raises the marginal product of other capital… …which in turn raises the rate of economic growth, but within limits

11. 3. Estimation of the Structural Model The Barro model and descriptive evidence suggest the following framework as a basis for empirical investigation [equation numbers correspond with those in the paper]: y = y(k, g) [6’] k = k(y, g) [7’] gi = g(y, gj), i ≠ j [8]

12. Employ standard VECM model in which: • Stationarity characteristics of the data: standard augmented Dickey-Fuller test statistics • All variables found to be I(1), except for total capital stock and public-sector infrastructural capital stock: both I(2)

13. Parsimonious specification of [6’] – [8]: [9] Possibility of multiple relationships between different forms of infrastructure, which may render identification of the system difficult Choice of public infrastructure, roads and electricity rests on PSS-F tests and prior literature

14. Eskom’s generating capacity: uncomfortably close to winter peak demand Canning et al. (growth); Pereira (growth & inv.)

15. Maximal Eigenvalue and Trace Statistics

16. Imputed elasticities (of output, capital investment and public infrastructural investment) at variable mean values

17. Replace LNTORD with LNRLIN, LNRCOA, LNRPAS, LNRFRT, LNPARD, LNPASV, LNGDSV, LNPORT, LNFTEL, LNRGDS, LNRCAP CV1: Elasticity of output wrt capital investment: 0.03 – 0.09 (prev 0.06), except for LNRGDS (0.15) Elasticity of output wrt electricity: 0.07 – 0.24 (prev 0.2), except for LNRGDS and LNRCAP (both negative)

18. CV2: Elasticity of capital investment wrt output: 2.03 – 6.70 (prev 2.44), except for LNRGDS (11) and LNRCAP (11.2) Elasticity of capital investment wrt public infrastructural investment: 0.68 – 1.54 (prev 1.37), except for LNRGDS and LNRCAP (both negative) CV3: Elasticity of public infrastructural investment wrt output: 2.73 – 24.77 (prev 3.93) Elasticity of public infrastructure wrt gj: all negative (prev 87.72 for total roads)

19. Considering the fragility of the previous results, we estimate a more parsimonious system: [10] Trace statistic indicates r = 2

20. Imputed elasticities (of output and capital investment) at variable mean values Results reasonably close to initial results: CV1: 0.06 (DLNKPC) and 0.20 (LNELEC) CV2: 2.44 (LNYPC) and 1.37 (DLNIFPC)

21. An obvious concern with [10] is that the results may be sensitive to the inclusion of additional regressors Consequently, we test this by including property rights and political instability: [11]

22. Two alternative specifications of [11]: Exclusion of property rights α51 = α52 = 0 = β15 = β25 Same identification structure with weak exogeneity restrictions β15 = β16 = β24 = 0 α31 = α32 = α41 = α51 = 0

23. Imputed elasticities at variable mean values

24. 4. Main findings Investment in infrastructure appears to have led economic growth in South Africa The impact of infrastructure is direct and indirect, the latter occurring by raising the marginal productivity of other capital This result is robust both to the use of a parsimonious growth model and to a fuller specification incorporating institutional determinants of economic development There is weak evidence of feedback from output to infrastructure; this is not robust

26. Empirical studies using US data: wide range of estimates for the elasticity of output with respect to public capital or some type of public infrastructure Aschauer (1989): 0.39; supported by Munnell (1990): 0.15-0.34 (smaller at state level) Holtz-Eakin (1994): elasticities ≈ 0; supported by Garcia-Milà et al. (1996) Lau & Sin (1997): 0.11; Pereira (2000): 0.04 Econometric methodologies have generated much controversy

28. SA’s economic infrastructure has developed in phases, in some cases closely linked to the development of the mining industry