Saving and investment rates in the BRICS countries

Abstract

The first decade of the new millennium witnessed the arrival of five large, fast-growing emerging economies on the global stage. These countries, Brazil, Russia, India, China and South Africa, collectively known as the BRICS countries, have achieved robust growth and steadily outpaced the advanced economies in the last quarter of a century or so. In spite of their recent slowdown, on the basis of their demographics, vast natural resources, physical and human capital accumulation and overall progress, they are likely to remain the engines of global economic growth and development for some time in the future. In the light of the rising role of BRICS countries in the world economy, this paper aims to study the relationship between their domestic saving and investment rates in the vein of Feldstein–Horioka but on a country-by-country basis using time-series econometric techniques. The results suggest that the basic Feldstein–Horioka regression is misspecified for each BRICS country. The preferred autoregressive-distributed-lag specifications imply that capital is not perfectly mobile internationally in any of the BRICS countries, but it is more mobile in South Africa and Russia than in India, Brazil and China.

Keywords:

• BRICS countries
• saving and investment
• capital mobility
• Feldstein–Horioka puzzle

JEL Classifications:

• C32
• E21
• E22
• F32
• F41
• O16

Acknowledgements

I am grateful for the helpful comments of two anonymous referees of this journal and of the participants of the Conference of the Business & Economics Society International, Monte Carlo, 6–9 July 2013. Of course, any remaining error is mine.

Notes

1. In order to check whether domestic investment or its components are equally responsive to the different types of savings, Feldstein and Horioka (1980) also estimated variants of equation (1) where overall saving was replaced by household saving, corporate saving and government saving, or/and total investment was replaced by private investment or corporate investment for nine OECD countries. The results suggested no major difference among the three types of saving in their contribution to total investment or total private investment.

2. Feldstein and Horioka (1980) also considered a model of four simultaneous equations to deal with the potential endogeneity of the saving ratio but their two-stage least squares (TSLS) parameter estimates proved to be very similar to the original OLS estimates.

3. About present value models and cointegration, see e.g. Campbell and Shiller (1987).

4. All calculations were performed with EViews 8.

5. These conclusions are based on two simple but occasionally contradicting rules of thumb: (1) in any test, the rejection of the null hypothesis at the usual significance levels is the stronger outcome and (2) the conclusion supported by the majority of the tests is the more plausible.

6. It is important to remember though that whenever the ZA, P or LP test rejects the null hypothesis of a unit root, the outcome is somewhat ambiguous because a rejected null might imply stationarity, structural break or both.

7. By rearranging the terms on the right side of (8), we obtain the generalized EC version of equation (6): This version was popularized by Jansen and Schulze (1996).

8. See e.g. Patterson (2000, 352–355).

9. It is also evident in Table 3 that some slope estimates are below while some others are above the corresponding correlation coefficients, illustrating that it can be indeed misleading to compare directly estimates of β obtained for different countries or over different time periods.

10. For each country, all possible specifications were originally estimated over a common sample period, but subsequently the preferred specifications were re-estimated from the largest available samples. These latter results are displayed in Table 6. The Bai-Perron multiple breakpoint test is not available for Russia and China because the partitions lead to a singular matrix in at least one of the sub-samples.

11. In the case of South Africa, model (6*) still suffers from heteroscedasticity, but still seems to be better than model (4).