Armed Conflict, Military Expenses and FDI Inflow to Developing Countries

Abstract

This paper investigates the relationship between military expenditure and FDI inflow conditioning on the exposure of a country to armed conflict in the long run. We apply the band spectrum regression estimator, and the maximal overlap discrete wavelet transform, to a panel of 60 developing countries, for the years 1990 to 2013. The estimated results indicate that military expenditure, in the absence of armed conflict, reduces FDI inflow. However, the negative effect is mitigated by increased military expenditure, in the presence of armed conflict. We also show that the effect of military expenditure on FDI is time sensitive, in that it takes time for military expenditure to affect FDI inflow. FDI inflow in response to higher military expenditure is higher for the country that faces higher armed conflict than the country that faces lower armed conflict. The findings are robust in the case of overall as well as internal conflict. These results are also robust to the alternative specification, subsample analysis with different armed conflict thresholds, and the estimation using the time variant long-run models.

Keywords:

• Conflict
• military expenses
• FDI
• wavelet

JEL Codes:

• F21
• H56
• D74
• F51

Notes

1. FDI net inflows are the value of inward direct investment made by non-resident investors in the reporting economy.

2. For more information about the MODWT see Percival and Walden (2006), Crowley (2007) and Andersson (2008); Although it is possible to decompose data using several different methods, such as the Hodrick-Prescott filter or Fourier transformation, wavelet transformation offers considerable advantages. Wavelet decomposition combines time and frequency domains. It is localized both in time and in frequency, which preserves the time domain and frequency domain information of the original series (Maslova, Onder, and Sanghi 2013). Thus, it does not introduce phase shifts that change the location of events in time and allows for the observation of structural breaks, outliers and nonlinearities in the data series (Ramsey 1999; Percival and Walden 2006). Unlike the Hodrick-Prescott filter, in which data are decomposed into short and long runs, the filtered time horizons are known. This data-driven technique for separating short and long runs ensures that short-term fluctuations arising from measurement errors in the data do not affect the results.

3. This definition of long run is commonly used in the literature (see e.g. Andersson and Karpestam 2013; Khalid 2016).

4. We find similar results using eight years as our definition of long-run.

5. The test results are available on request.

6. The effect of external conflict has not been examined (separately) due to lack of enough observations.