ON THE RELATIONSHIP BETWEEN MILITARY EXPENDITURE, THREAT, AND ECONOMIC GROWTH: A NONLINEAR APPROACH

Abstract

The main objective of the paper is to decipher the military expenditure–economic growth relationship, taking the level of economic development (income) into consideration. Our findings suggest the following: (i) military expenditure has a significantly negative relationship to economic growth for the 23 countries with initial incomes (threshold variable) less than or equal to $475.93; (ii) when the threat level is heightened, economic growth (23 countries) is expected to decrease. However, military expenditure in the presence of sufficiently large threats increases growth; (iii) for the remaining 69 countries whose initial incomes (real GDP per capita in 1992 price) exceed $475.93, no significant relationship exists whether the threat variable is taken into consideration or not.

Keywords:

• Military expenditure
• Economic growth
• Threat
• Threshold

JEL Codes:

• H56
• C22

ACKNOWLEDGEMENTS

The valuable comments from the editor and two referees are greatly appreciated. Any errors are our own. Financial support from the National Science Council (NSC96‐2415‐H‐194‐003‐MY2) is gratefully acknowledged. A preliminary draft was presented at the 83rd WEA International Conference in Hawaii, 2008.

Notes

¹Defense spending and military spending is used interchangeably in this paper.

²This variable was constructed from data on the Militarized Interstate Dispute collected by the Correlates of War Project (COW) at the University of Michigan (http://www.correlatesofwar.org/). Threat is defined as the number of years a country was at war with each of its adversaries during the period of 1992 to 2002 summed over the set of its adversaries. Refer to Aizenman and Glick (2003: 3) for the information on how to construct this variable and source of the data.

³See Hansen (1997: 6) for the bootstrap procedure.

⁴The grid‐searching procedure for the optimal c* requires upper and lower 15% observations be truncated for sample estimation. However, for a large sample only the upper and lower 10% of observations need to be removed. The a priori assumption is that optimal c will not occur in the truncated region.