Imports and economic growth in Africa: Testing for granger causality in the frequency domain

ABSTRACT

In this paper, we test for causality between imports and economic growth in 41 African countries. We differ from earlier studies by testing for causality at high and low frequency levels, that is, in the short and long run, respectively. In doing so, we employ a frequency domain Granger causality test robust to pretest biases relating to unit root and cointegration tests. We document that there is: (i) unidirectional causality running from imports and economic growth in 7 countries in the short run and 5 countries in the long run, (ii) unidirectional causality running from economic growth to imports in 4 countries in the short run and 10 countries in the long run, (iii) bidirectional causality in only one country in the short run and 3 countries in the long run, and (iv) no causality in 29 countries in the short run and 23 countries in the long run. Our findings suggest that, for the most part, the neutrality hypothesis is valid in the short- and long-run periods. We imply from our findings that possible changes in the causality dynamics between imports and economic growth over time should be taken into consideration before designing policies.

KEYWORDS:

• Imports
• economic growth
• frequency domain
• granger causality test
• Africa

JEL CLASSIFICATIONS:

• F10
• F43
• O11

Data availability statement

The data used in this study are available on request.

Disclosure statement

No potential conflict of interest was reported by the author(s).

ORCID

Olufemi Adewale Aluko http://orcid.org/0000-0001-5628-766X

Patrick Olufemi Adeyeye http://orcid.org/0000-0001-6211-1564

Notes

1 Despite its good sample size and power properties, the BC Granger causality test is not without its own limitation. Tastan (2015) notes that the Wald statistics produced by the BC Granger causality test are based on complex nonlinear restrictions on the autoregressive parameters of the VAR model, which complicates the statistical inference for the feedback measures because numerical approximation methods which are often not easy to compute must be applied.

2 Recently, Troster (2018) propose the Granger causality test in quantiles which can also be used to detect nonlinearities in Granger causality relationships. However, unlike Granger causality test in quantiles, the frequency domain Granger causality test is a spectral Granger causality test which allows us to clearly distinguish temporary (short-run) and permanent (long-run) causality.

3 The BC Granger causality test is implemented in Stata using the ‘bcgcausality’ command developed by Tastan (2015).

4 Toda and Yamamoto (1995) Granger causality test overcomes the shortcomings of traditional time domain Granger causality tests. We refer readers to Zapata and Rambaldi (1997), and Toda and Yamamoto (1995) for shortcomings of the traditional time domain Granger causality tests. However, the Toda and Yamamoto Granger causality test suffers from small sample bias but still performs better than the traditional time domain Granger causality tests when the sample size is small (see Kurozumi and Yamamoto 2000).

5 The countries are Algeria, Cameroon, Central African Republic, Chad, Cote d’Ivoire, Gabon, Guinea, Guinea-Bissau, Kenya, Malawi, Mozambique, Namibia, Nigeria, Senegal, Togo, and Tunisia.