Structural Transformation in Africa: Static Gains, Dynamic Losses

Abstract

This paper places recent growth and structural transformation in 11 Sub-Saharan African countries in historical and international perspective. During the early post-independence period, resources were reallocated to manufacturing activities with high productivity growth. Structural change stalled in the mid-1970s. When it resumed in the 1990s, workers mainly relocated to distributive trade services. Productivity levels in these activities were higher than in agriculture, enhancing overall economy performance. But services productivity growth was sluggish and increasingly falling behind the world frontier. These patterns are also observed in Latin America, but not in Asia.

Acknowledgements

The research on which this paper is based is part of the ‘Structural Change and Productivity Growth in Africa’ project. This project is financed by the Economic and Social Research Council (ESRC) and the UK government’s Department for International Development (DFID) as part of the DFID/ESRC Growth programme, grant agreement ES/J00960/1. We thank Louise Fox, Douglas Gollin, Margaret McMillan, James Thurlow, and Dani Rodrik for comments and suggestions as well as participants at the ACET/IFPRI conference ‘The Quest for Economic Transformation in Africa’ in Nairobi, Kenya, and seminar participants at the University of Groningen and Wageningen University. The database is available at www.ggdc.net/asd.

Notes

1. McMillan et al. (2014) argue that there has been a turnaround in Africa from growth-reducing structural change during the 1990s to growth-enhancing structural change since 2000. This will be discussed in section 4.

2. The upward revision of Ghana’s GDP by about 60 per cent in 2006 and the 89 per cent increase in Nigeria’s GDP in 2014 are notable examples. Changes for Ghana and Nigeria partly relate to the inclusion of services activities in GDP that were previously not covered. For Ghana, several communication and personal services, recreation services, and professional services were not included in GDP before the revision. Nigeria did not include many telecommunication services and media activities in its GDP before the revision.

3. In most developing countries a fixed-base Laspeyres volume index is used.

4. Because of the application of fixed-base Laspeyres volume indexes by most statistical offices, additive consistency is lost and linked sectoral GDP therefore do not add up to total GDP for earlier periods.

5. The preferred age boundary is 15 years and older, however for some countries the age boundary differs, see de Vries et al. (2013b).

6. In this paper we do not directly observe marginal productivity, but measure average productivity across sectors and over time. If a production function is Cobb Douglas, the marginal productivity of labour is average productivity times the labour share in value added. If labour shares differ across sectors, an analysis based on average productivity may be misleading. For example, high average productivity in a capital-intensive sector may simply reflect a low labour share. We assume that marginal and average productivities have a strong positive correlation. Gollin et al. (2014) found that differences in average productivity in agriculture and manufacturing are related to large gaps in marginal productivity, giving some credibility to the approach adopted here. However, there is a clear need for further research in this area.

7. Distribution services include wholesale and retail trade, hotels and restaurants, and transport services.

8. We thank one of the referees for pointing this out to us.

9. One might argue that when workers move towards market services, it becomes more labour-intensive and as a result the capital-labour ratio and also labour productivity decline. Typically the reallocation of workers requires additional investment in order for average products in expanding sectors not to change (Chenery et al., 1986).

10. For example, McMillan and Rodrik (2011) state ‘When changes in employment shares are positively correlated with productivity levels, this term will be positive, and structural change will increase economy-wide productivity growth’ (p. 13).

11. One might argue that the interaction term does not fully reflect structural change, as it interacts changing employment shares with productivity changes. However, the structural change term in Equation (1), used by McMillan and Rodrik (2011) can be split into the second and third term of Equation (2). Hence, the structural change term used by McMillan and Rodrik (2011) combines the static and dynamic reallocation effect. We will show that it is important to make this distinction in order to understand differences in the role of structural change for growth between the period 1960 to 1975 and the period 1990 to 2010.

12. Typically the dynamic reallocation effect is negative (but small) as it is difficult to absorb additional labourers at the same rate of marginal productivity, for example due to adjustment frictions or organisational restructuring, see van Ark (1996) for Europe, and Timmer (2000) for Asia.

13. In Appendix B of de Vries et al. (2013a) we describe the main differences in data sources compared to McMillan and Rodrik (2011). The most notable differences are the inclusion of updated national account data for Ghana, Ethiopia, Botswana, and Zambia, and the use of new population census or labour force survey data for Nigeria and Malawi.

14. These numbers correspond closely to the unweighted average reported in McMillan et al. (2014).

15. It should be noted that we assume that PPPs for gross output are representative for value added. More specifically, we are not able to correct for differences in prices of intermediate inputs. If intermediate input goods were heavily subsidised value added might be over estimated. This was common practice in various African countries during the period of import substitution industrialisation (Szirmai, 2005).

16. Faster productivity growth in US market services is partly related to differences in accounting for price changes in retail output (Inklaar & Timmer, 2008). The United States statistical office uses a quality-adjusted price deflator, especially for the consumption of information and communication technology goods. Measured sales volumes are smaller in most African countries, partly because they do not make use of hedonic price deflators.

17. This is, for example, observable from the low levels of spending on information and communication technologies by wholesale and retail firms in African economies (WITSA, 2010). The opportunities from the use of ICT in wholesale and retail firms are substantial; see Triplett and Bosworth (2004).

18. Ndulu, O’Connell, Bates, Collier, and Soludo (2008) is a major exception; although different in focus.