FDI, defense spending, and economic prosperity

2.1.1 Data description

Level data are normalized using current US dollars. SIPRI is the main source for military expenditures data while the World Bank economic indicators (WDI) provide data on all economic variables, including FDI inflows. In our robustness check analysis, we utilize WDI natural resources data, while conflict data are obtained from the Uppsala Conflict Data Program (UCDP).10 The International Country Risk Guide (ICRG) variables are retrieved from the subscription-based Political Risk Services (PRS) group,11 which is an online service providing data on foreign investment and country-specific political and economic factors to international investors.

By this transformation, negative and zero values are retained. We also measure FDI inflows in dollars (not in millions), hence adding 1 in the previous equation is equivalent to adding one dollar to gross inflows.

All other macrovariables in the panel are used in their natural logarithmic forms, ICRG risks and education are used in levels. Table 1-A shows the descriptive statistics; namely, mean value, standard deviation, and number of observations of all variables in the panel. As stated by SIPRI,12 military spending is a measure of all costs incurred as a result of current military forces, consisting of expenditure on personnel, operations, and maintenance. SIPRI also highlights the general lack of transparency and accountability regarding military matters in the MENA region.13 Between data availability for some countries and uncertainty due to the lack of detail in public documents (high level of secrecy), military spending data for the region are to be used with a precaution. This issue will not, however, limit researchers from using the available data for economic analysis.

To analyze the relationship between military expenditures and FDI, we start by plotting the average of country-year observations on these two variables for the period 1984 to 2020. Plots in Figure 3 depict the average FDI inflows and the average military expenditures (in log forms) for all countries (averages by country). A closer look at the scatterplot in Figure 3 shows a strong divergence among countries between the average military spending and the attractiveness of foreign capital. Israel is the frontrunner when both military spending and FDI attractiveness are considered. Saudi Arabia is by far the largest military spender in the region and was the world's fifth largest spender in 2020. FDI inflows to Saudi Arabia have, however, followed a downward trend over the last decade, due mainly to political factors and lower oil prices. Iraq, Libya, and Yemen are the only three other countries in the region attracting similar level or less FDI than Saudi Arabia. Overall, the scatterplot in Figure 5a suggests a slightly positive relationship between the two variables. This association is also highlighted in the correlation matrix shown in Table 1-B where FDI inflows is significantly and positively correlated with military spending (in levels). The significant Pearson correlation coefficient value of 0.497 confirms what was apparent from Figure 5a. Table 1-B also reports the correlation coefficients between FDI and military expenditures per capita ($${\bm{lMilexCap}}$$) and military burden14 ($${\bm{MilexGDP}}$$), showing a significant and positive coefficient for $${\bm{lMilexCap}}$$ and a negative one for $${\bm{MilexGDP}}.$$ A detailed description of the data is provided in the Online Appendix.

2.1.2 Panel unit root tests

In the recent literature of dynamic panel data, unit root tests have become commonplace. With data sets including time period, which is fairly long (37 years), it is very likely that some of the macroeconomic variables will follow a unit root process (Nelson & Plosser, 1982). The first generation of these tests includes Harris and Tzavalis (1999), Maddala and Wu (1999), Choi (2001), and Im et al. (2003). These tests relay on the assumption that the individual time series in the panel are cross-sectionally independently distributed. A second generation of these tests is developed around the concerns of the existence of cross-sectional dependencies. Various tests have been proposed including the works of Choi (2002), Ploberger and Phillips (2002), Chang (2002), Pesaran (2003), Phillips and Sul (2003), Moon and Perron (2004), and Bai and Ng (2004).

Nonstationarity is a very important issue, which has been taken into consideration, in this work, by applying selected first-generation tests: Harris and Tzavalis (1999), Maddala and Wu (1999), and Im et al. (2003). Second-generation unit root tests: Pesaran (2007) and Bai and Ng (2004) are also applied. Panel unit root results are shown in Tables 2 and 3. Results report that the series are a mix of I(0) and I(1) and that no series is I(2), allowing us to employ the panel autoregressive distributed lag (ARDL) approach in the estimation stage (Pesaran, 1997; Pesaran et al., 1999).

To analyze the relationship between military expenditures and FDI, we start by plotting the average of country-year observations on these two variables for the period 1984 to 2020.15

2.2.1 Bivariate analysis

We start our analysis by checking the empirical relationship between FDI inflows ($$lFD{{I}_{it}}$$) and military expenditures ($$lMILE{{X}_{it}}$$) in a bivariate setup. Bivariate analysis is the empirical analysis of two variables to ascertain if a relationship exists between them. Specifically, this strategy can be used to describe the strength and the direction of the relationship between $$lFD{{I}_{it}}$$ and $$lMILE{{X}_{it}}.$$ The aim is to check whether an association between these two variables exists, without controlling for any other factor. Therefore, and rather than moving forward with the inferential statistics, we first use a bivariate analysis to determine if any potential connection exists between the two variables at the observational level.

FDI inflows and military spending vary widely across the MENA region. As shown in Figure 3, countries like Israel, Egypt, Tunisia, and Morocco have traditionally attracted the most significant FDI in the region. On the other hand, Saudi Arabia is one of the largest military spenders globally, often ranking in the top five. A significant portion of its budget is allocated to defense due to regional security concerns (e.g., the war in Yemen, tensions in the region). FDI inflows are typically low in countries like Iraq, Libya, and Yemen, where conflicts have led to political and economic instability, but military spending remains relatively high due to ongoing conflict.

Trends, over the time period of this study (Figure 4), show that both variables increase significantly, with $$FD{{I}_{it}}$$ showing a more uneven path, with a setback in 2008 when it reaches a historic level for the MENA region. All lines in this figure show an overall upward trend, which indicates that FDI inflows have appeared to substantially increase while military spending has also increased, suggesting the possibility of a positive relationship between the two variables.

Going deeper into the relationship between country-level averages and in order to visualize this association without resorting to trend lines, we rely on a local smoothing technique. Smoothing methods attempt to find functional relationships between different measurements. Essentially this means drawing lines through the data nearest to the independent variable, not from the entire set of points. Figure 5a shows a scatterplot of the data (averages), together with a kernel smooth using the local smoothing technique. The smooth is constructed using the bi-weight kernel and a bandwidth of 3. The smooth fit captures the main trend of the data, showing a positive relationship with an $${{R}^2} = 0.37.$$ To further check the possibility of a quadratic relationship between the two variables, the augmented component-plus-residual plot is used to detect any possible nonlinearity in the data. As shown in Figure 5b, the low-smoothed line does not appear to deviate too much from the line of best fit (polynomial pattern). Further investigation using a formal test16 is conducted and we conclude that nonlinearity is not an accurate relationship between FDI inflows and military expenditures, pointing toward a monotone relationship between these variables.

A more sophisticated smoothing procedure is also available via the locally estimated scatterplot smoothing technique (LOWESS). The LOWESS smoother uses a regression model with polynomial terms on a local subset of the data (avoiding sensitivity to extreme data points) to estimate a fitted value for each observation (see Cleveland, 1979, and Mallows, 1986, for more details about these techniques). The positive association between the two variables is again put forward by LOWESS, even though post–2008 financial crisis negatively impacted the flow of FDI to the region. The impact on FDI inflows leads to a curved shape albeit very minimal (See Figure 6). As a result, these probably indicate that there may be a (weak) negative relationship between the variables of interest after 2008.

Finally, and in order to summarize, the association between the two variables classified categorically as to whether observations are above the 90th percentile of FDI or military spending, respectively (defined as high observations). Table 4 displays a breakdown of the data by these two categories (high and low, with the later defined as observations below the 90th percentile) along with the Pearson chi-square test. The aim here is to see if high levels of FDI inflows are associated with high level of military expenditure in the MENA region (results shown in Table 4). It is evident, though, that low FDI is associated with low military spending as 95% (567/599) of “low” FDI inflows are associated with “low” military spending. Only 5% (32/599) of low military expenditure observations are linked to high FDI inflows while 52% (35/67) of high military spending coincide with high FDI observations, thus, providing some evidence that high FDI inflows and high military spending are linked together. The Pearson's $${{\chi }^2}$$ statistic is significant at all confidence levels with a p-value < 0.000, thus providing ample confidence that the above associations are not due to random chance.

To summarize the bivariate analysis, we can conclude that an empirical observed positive relationship exists between the two variables of interest in this study, FDI inflows, and military expenditures. We will, now, proceed with this association between the two variables in the next section, adding more controls and testing the relationship in a multivariate framework.

2.2.2 Multivariate analysis

This section will introduce control variables into our analysis, in order to test the assertion that a positive association exists between the FDI inflows and military spending, as highlighted in the bivariate case.

Our empirical model fills the gaps in the literature. A panel model on all MENA countries tests the relationship between FDI inflows ($$FD{{I}_{it}}$$) and military expenditures ($$MILE{{X}_{it}}$$), controlling for institutional, economic, and political factors ($${{Y}_{it}}$$) over a period of 37 years (1984−2020).

As it is well known in the literature, allowing for the presence of subject-specific unobserved heterogeneity represents one of the key advantages of using panel data. In this context, we deal with this unobserved heterogeneity by using the one-way fixed effects models, or by taking first differences if the second dimension of the panel is a proper time series, that is, dynamic panel. Dynamic panels in which the regressors include the lagged dependent variable are developed around the work of Arellano and Bond (1991), Arellano and Bover (1995), and Blundell and Bond (1998). The resulting generalized method of moments (GMM) approach, and its extension to the “System GMM” context, are, however, estimators designed for situations with, “small $$T$$, large $$N$$” panels, that is, few time periods and many individual units.

The analysis of panel data with both large $$N$$ and $$T$$ becomes increasingly important in the literature. Pesaran and Smith (1995), Pesaran (1997), and Pesaran et al. (1999) present the ARDL in the error correction form allowing both the short-run dynamic, and the long-run equilibrium, relationships between $$FDI$$ and military expenditures to be separately identified. When both $$N$$ and $$T$$ are sufficiently large, there have been two approaches in the literature to analyzing dynamic panels. The fixed versus random effects estimators, on the one hand, and the mean group (MG) versus the pooled mean group (PMG) estimator on the other hand. The dynamic fixed effects (DFE) being another option used for situations in which it is reasonable to assume that short- and long-run parameters are identical for all individuals in the panel. Choosing among these estimation techniques is based on a Hausman test of the null hypothesis that the difference between the $$MG$$ and the $$PMG$$ (or $$DFE$$, assuming a dynamic panel) estimator is zero. The $$MG$$ estimator is selected if the null hypothesis is rejected; otherwise, the more efficient pooled estimator is to be preferred. As shown by Pesaran (1997) and Pesaran et al. (1999), these models can be estimated with the variables being I(0) and/or I(1).

Furthermore, in order to assess simultaneously both in the short- and long-run, we also consider the panel ARDL technique. The ARDL approach estimates the effect of a particular variable on military expenditures separating the short-run from the long-run effects (Bentzen & Engsted, 2001). ARDL is based on Pesaran (1997) and Pesaran et al. (2001). The dynamics in these models is incorporated into the error correction term by using lags of the dependent and independent variables. This incorporation allows for rich dynamics in the sense that the dependent variable adapts to changes in the explanatory variables.

The key parameters are defined as $${{\varphi }_i} = - ( {1 - \sum_{p = 1}^P {{\lambda }_{ip}}} );\;{{\theta }_i} = \sum_{j = 0}^q {{\beta }_{ij}}/( {1 - \sum_k {{\lambda }_{ik}}} );$$ $$\lambda _{ip}^* = - \sum_{m = p + 1}^P {{\lambda }_{im}}, p = 1,2,\ldots,P - 1, {\mathrm{and}}\;\beta _{ij}^* = - \sum_{m = j + 1}^q {{\beta }_{im}}, j = 1,2,\ldots,q - 1.$$ Of interest are the parameter $${{\varphi }_i}$$, which is the error-correcting speed of adjustment term, and the vector $${{\theta }_i}$$, which contains the long-run relationships between the variables. The error-correcting term $${{\varphi }_i}$$ is expected to be significantly negative under the prior assumption that the variables show a return to a long-term equilibrium.18 If the coefficient is rather zero, there is no evidence of long-term relationship.

In our empirical work, we utilize a panel regression method for the 18 MENA countries. The analysis covers the period $$1984 - 2020$$. The length of each time series in the panel is determined by data availability. Our dependent variable is the net FDI inflows into the host country in a given year, measured in current US dollars. We use the World Bank measure of FDI inflows in this analysis. Military expenditure data are produced by the SIPRI and GDP is retrieved from the WDI database. All other control variables are from the PRS group or constructed from the World Bank indicators.

FDI inflows are known to be linked to domestic conditions of the host country. A number of institutional, political, and economic factors are to be considered when dealing with the relationship between military spending and FDI. Following Hite-Rubin (2015) and Drezner and Hite-Rubin (2016), we use the Political Risk Services Group's International Country Risk Guide (PRS-ICRG) risk measures for control variables in our empirical work. The PRS-ICRG is a subscription-based service that provides data on foreign investment and country-specific institutional, political, and economic factors.19 According to the PRS-ICRG methodology, an index is created for each of three main categories, which can be defined as political, economic, and financial risk categories. The political risk index is based on 100 points, financial risk index is based on 50 points, and economic risk index is based on 50 points. Likewise, each category is divided into subfactors where each factor has its own weight.20 Six of these factors are used in our data set. In order to assess the impact of these factors, we include control variables in our model estimation in a stepwise fashion.

Military expenditure, $$lMilex$$, is the main explanatory variable in this paper. This variable is constructed using military expenditure by country, in millions of US$ at current prices, converted at the exchange rate for the given year (in log form). In the defense economics literature, the choice of the military spending proxy has a long tradition of empirical and theoretical debate before settling on a specific choice. In the military spending–growth literature, whether the level or share is the appropriate measure, military burden (the share of national income going into military expenditure) is usually chosen (see Brauer, 2002, for an excellent discussion and Smith, 2017, for a recent confirmation). In the demand for military expenditures literature, levels are often used (Smith, 2017). Each approach provides different insights into a country's defense capabilities and priorities. On the one side, military spending as a share of GDP offers insights into the economic impact of military expenditures, as it captures the relative burden of this category of spending on a country's overall budget. On another side, military spending (in levels) primarily measures the financial resources allocated by a government to its defense and military capabilities. It provides insight into the extent to which a country invests in its armed forces, military capabilities, defense infrastructure, and security-related activities. This variable is then considered as a good proxy for a nation's commitment to national security. This paper uses the level of military spending (in log form) as a signal and test of the country's attractiveness of the business environment. International investors may interpret this signal as a sign that the country is committed to maintaining stability and security. In contrast, high military spending at the expense of crucial social and economic investments may, also, raise concerns about the government's priorities and its commitment to creating a business-friendly environment, which might discourage foreign investors.22

We test the relationship among our variables controlling for a number of other factors,23 including: income proxied by log of GDP expressed in current US dollars ($$lGDP$$); log of GDP per capita expressed in current US dollars ($$lGDPPC$$); the ICRG's risk points for Corruption ($$Corrup$$); the ICRG's combined economic risk rating ($$eRisk$$); a country measure of the number of active Bilateral Investment Treaties in a given year ($$BIT$$); as an alternative to $$BIT$$, the degree of economic openness of a given country is used to measure the degree to which nondomestic transactions take place and affect the size and growth of a national economy ($$Open$$). Other control variables accounting for country's financial and political risks are: the ICRG's risk points for gross foreign debt expressed as a percentage of GDP (financial risk, $$FDebt$$); the ICRG's risk points for exchange rate stability ($$ER$$); the ICRG political risk measure of the strength and impartiality of the legal system as long as an assessment of the observance of law in practice ($$LawO$$) and the ICRG's risk points for international liquidity (financial risk, $$IntLiq$$).

The energy consumption is measured in quadrillions British thermal unit (BTU) per capita (in log form, $$lEU$$). Education is measured by the Human Development Index (HDI), mean years of schooling ($$Educ$$). In order to account for country-specific banking-sector vulnerabilities or certain types of recessions, a financial dummy variable is constructed $$( {FinDum} ).$$

In this equation, we control for most institutional, political, financial, and economic factors that, in our view, have an impact on the FDI inflows into a given country in the MENA region from one year to the next. As stated before, dynamic panel regressions often employ any one of the following three estimators: the DFE, the MG, and the PMG estimator. The choice among the three estimators is determined by the joint Hausman test. In this case, PMG is tested against both MG and DFE as reported in the last panel of Table 5. The Hausman test indicates that PMG is consistent and more efficient when compared to MG and DFE. The PMG technique is known for pooling the long-run parameters, restricted to be homogeneous across countries. This technique avoids, at the same time, the inconsistency problem flowing from the heterogeneous short-run dynamics.24 This is particularly useful where there are reasons to accept that the long-run equilibrium relationship between the variables is similar across countries or at least, a subset of them, which is the case for countries in the MENA region.