RESULTS AND DISCUSSION

Based on the related literature covered above our main hypotheses in this paper are as follows:

4. RESULTS AND DISCUSSION

We employ System GMM analysis based on an unbalanced data-set, in order to investigate the relationship between military expenditures and inequality in the context of different political and welfare regimes.7 Our dynamic panel approach uses System-GMM based on Roodman.8 and Roodman (2009). We used an AR(1) and an AR(2) model to capture the persistence in our data. In addition, AR(1) and AR(2) models are desirable based on the Arellano-Bond test for AR(2). To consider any cross-sectional dependence, we included time dummies as instruments in some regressions. Since there may be an endogeneity problem for most of our explanatory variables, we set THEIL, AF, ARMIMP, RGDP, GROWTH, and first and second lags of the dependent variable (MECGE) as endogenous. In order to avoid over-identification we used the collapse option, hence the GMM instrument is constructed by creating one instrument for each variable and lag distance (rather than one for each time period, variable, and lag distance). The other independent variables are instrumented as suggested by Roodman (2009). In other words, the other explanatory variables are treated as typical instrumental variables instead of GMM because they are assumed to be exogenous. All estimations were conducted with two-step efficient GMM to fix any non-spherical errors, and small sample corrections (Windmeijer-corrected standard errors) to the covariance matrix estimate (Windmeijer, 2005). The estimated models pass the specification tests. According to Arellano-Bond test

statistics for AR(1) and AR(2), the consistency of the GMM estimators is verified, as there is no evidence of a second-order serial correlation in the differenced residuals of the models. The Hansen test statistics approve the validity of the GMM instruments. Finally, the Difference-Hansen test statistics provide no evidence to reject the null hypothesis of the validity of the additional moment conditions used in the system GMM estimations. Table III provides system GMM estimation results. All variables, except for RGDP, are

significant, most at the 1% level. Dynamic panel analyses’ findings in the table show that lagged values of military expenditures have a positive effect on their present value. The results show that as inequality (THEIL), the size of armed forces (AF), and number of terror occurrences (TERROR) increase military expenditures as a percentage of government expenditures (MECGE) increase as well (Model 1.1). On the other hand, arm imports as a percentage of total imports (ARMIMP) and real GDP per capita (RGDP) have negative relationships with military expenditure. These relationships are also valid when year dummies are added (Model 1.2). While the explanation for the positive relationship between income inequality, armed

forces, terror, and share of military expenditure in the central government budget is more

DEFENSE SPENDING, NATURAL RESOURCES, AND CONFLICT

straightforward, the impact of the share of arms imports in total imports and real GDP per capita deserves more attention. Before discussing the latter two variables, we note that the negative sign of income inequality in each different model we define in our paper can be used as clear evidence, in line with the previous literature, that our Hypothesis1 that countries with more deteriorated income distribution spend more on military, and vice versa holds. Regarding arms imports to total imports, the ratio can increase either if arms imports increase (more than the increase in total imports) or total imports decrease (more than the decrease in arms imports). Considering its comparative advantage in trade, it could be the case that the country can obtain armaments at relatively lower cost by importing rather than producing them domestically. Therefore, the country spends less on armaments as a share of budget. In our estimation results, ‘arms imports to total imports’ have a negative sign. This is true not just in the baseline regressions but also when the model is controlled for the welfare regimes by using dummy variable for being a ‘social democratic welfare regime’ (Model 2.6). In addition, the coefficient remains negative when the model is controlled for all welfare regime types by taking ‘social democratic welfare regime’ as the base category, and also controlled for all political regimes. Hence, our data support the comparative advantage argument for these countries since they spend less on armaments as a share of budget.9