The Military Expenditures and Economic Growth Nexus: Panel Bootstrap Granger Causality Evidence from NATO Countries

Acknowledgment: The authors are deeply grateful to the editor and the two anonymous reviewers for their valuable time, constructive comments, and useful suggestions, which helped us to revise and improve this manuscript significantly. The first version of this report was presented at the 4th International Conferences on Social Issues and Economic Studies organized by the RSEP (Review of Socio Economic Perspectives) (June 29 to 30, 2017, Prague, Czechia).


Introduction
The military expenditures of a country, which are related to its political independence and deterrence, are an important factor in the capital budget because of their economic effects in the public and private sectors. Therefore, such expenditures have long been a subject of interest for both researchers and policy makers.
In addition, whereas understanding how military expenditures affect economic growth has been a main focus in the defense literature, the contribution of the previous literature on the relationship between the variables is to quantify the symmetric or asymmetric causality. In this sense, although many studies have aimed to examine the symmetric causality between these two variables, fewer works have explored the asymmetric causality between them. For example, previous analyses of this relationship have focused particularly on the symmetric causality (see, for example, Alper Ozun and Erman Erbaykal 2014; Jakub Odehnal and Jiri Neubauer, 2016); in contrast Abdulnasser Hatemi-J et al. (2018) investigated the asymmetric causality in six countries. The existing empirical literature, which includes intensive investigations of the relationship between military expenditures and economic growth, does not provide a consensus on the direction and characteristic of this relationship, especially in NATO member countries. Therefore, the present study analyzes the causal relationship between military expenditures and economic growth by applying both panel symmetric and asymmetric causality testing on data from 27 NATO member states.
This research generally focuses on two important contributions to the literature. The first is the investigation of the causality relationship between military expenditures and economic growth in 27 NATO members in terms of theoretical hypotheses, including Keynesian growth, conservative, feedback, and neutrality hypotheses. In the existing literature, studies on the causality relations between military expenditures and economic growth in NATO countries are limited. From the perspective of economic theory, this work presents findings that are consistent with four theoretical hypotheses. In addition, to the best of our knowledge, this is the first study explaining the relationship between the abovementioned variables in NATO members within the asymmetric causality approach. The second contribution is the application of more robust empirical analyses to obtain consistent findings. In this regard, the relationship between military expenditures and economic growth in NATO members is examined by using panel heterogeneity, crosssectional dependency, and symmetric causality approaches. Asymmetric causality analysis is also included in the models to obtain more robust results because symmetric causality analyses may not yield findings consistent with those in the empirical literature. Thus, if significant causality relationships are found in both symmetric and asymmetric analyses, the findings would provide more robust contributions to the defense literature. Therefore, this work may also provide an important empirical contribution to the literature on military expenditures and economic growth by showing the difference between symmetric and asymmetric causality in NATO countries.
The remainder of this report is organized as follows: Section 2 discusses the theoretical and empirical framework of the relationship between military expenditures and economic growth, Section 3 explains the data and econometric methodology applied, Section 4 presents the findings, and Section 5 provides the conclusions.

Theoretical and Empirical Literature
The military expenditure-economic growth nexus is one of the most important theoretical and empirical relationships analyzed in the defense economics literature. However, the results of theoretical and empirical analyses have apparently supported various explanations for controversial findings. Previous studies on the relationship between military expenditures and economic growth have focused mainly on two controversial areas in the impact of military expenditures on economic growth in terms of three different theoretical models of schools of economic thought, and the direction of causality between the two variables (for more details, see A. Henry Dakurah, Stephen P. Devies, and Rajan K. Sampath 2001: 652).
In this regard, the first controversial topic in the defense literature is whether the defense expenditure can either promote or hinder economic growth. From the theoretical perspective, there are different channels by which military expenditures can have various impacts on economic growth based on different schools of economic thought. For example, the Keynesian aggregate demand approach holds that military expenditures lead to an increase in economic growth. Under this approach, increasing military expenditures stimulate economic growth by increasing employment, capital stock, and profits, thus leading to greater investments in a country (Pei-Fen Chen, Chien-Chiang Lee, and Yi-Bin Chiu 2014: 476). Therefore, there is a Keynesian multiplier mechanism in which military expenditures positively affect economic growth. For example, an increase in capital stock utilization through higher military expenditures results in increased profit rates and may therefore lead to higher investment levels, which may then generate higher economic growth rates with a short-run multiplier coefficient (Christos Kollias, George Manolas, and Suzanna-Maria Paleologou 2004: 556). In contrast, the classical approach has been used to examine the negative effects of military expenditures on economic growth; this has provided support to the defense literature by including the crowding -out effect of military expenditures. Under this approach, the negative effect of defense expenditures in the long-run will depend on two different financial management choices. First, if governments prefer to decrease other important public investments, such as education, public health, and infrastructure, in favor of increasing military expenditures, the defense spending may decrease economic growth in the long-run. Second, if governments finance military expenditures through borrowing or increased taxation, military expenditures may cause a decrease in private investments, known as the crowding-out effect of the public sector (Jonathan Lipow and Camille M. Antinori 1995: 581). The role of military expenditures and arms import in a debt crisis is also an important factor for investment and economic growth (see, for example, Eftychia Nikolaidou 2016). In this regard, borrowing to finance for military expenditures will lower the consumption level in a country and, thus, hamper the economic growth potential (Ourania Dimitraki and Aris Kartsaklas 2018: 716). According to the neoliberal approach, the supply channel means that national defense is a public good that causes opportunity costs. Under this channel, those costs are considered to be associated with certain economic problems, such as inflation, imbalances in the international financial structure, and excessive public debt. On the other hand, if the economic profits are greater than those costs in the supply channel, the military expenditures may contribute to the longrun economic growth (Grzegorz Waszkiewicz 2018: 1-2).
The second controversial topic in the existing empirical literature, which includes intensive investigations of the relationship between military expenditures and economic growth, is the absence of a consensus on the direction and characteristics of such relationship, especially in NATO countries. Although a number of studies have examined the causal relationship between military expenditures and economic growth, there is no general consensus on the direction and magnitude of this relationship. In this context, a subject of intense discussion has been whether or not there is a causal linkage between the variables previously studied in the defense literature. For instance, Dakurah, Davies, and Sampath (2001) analyzed the causal relationship between defense expenditures and economic growth in 62 economies. Their findings showed a unidirectional causality running from defense expenditures to economic growth in 23 countries, and a bidirectional causality existing in 7 countries (see also Nikolaos Dritsakis 2004; Korhan K. Gokmenoglu, Nigar Taspinar, and Mohammadesmaeil Sadeghieh 2015; Melike Bildirici 2016).
The implications of the relationship between military expenditures and economic growth have also been discussed intensively in the empirical literature on the defense-growth nexus. Previous empirical studies have presented different methods of determining the causality between military expenditures and economic growth. Some works have applied time-series techniques for individual countries, whereas several studies have used other types of analysis, such as crosssection, panel data, or spatial data analysis, in different country groups. The empirical research on the causal relationship between military expenditures and economic growth indicates that the findings differ according to the period studied and the econometric method applied in each model, as well as the development level of each country or country group. In this regard, it also seems that the empirical discussions associated with the military expenditures-economic growth nexus have recently continued to grow as such in theoretical discussions (see Table 3 in the Appendix). The general opinion emerging from these discussions is that the theoretical and the empirical literature provide conflicting results and that there is no consensus on the existence and direction of the causality between military expenditures and economic growth. For example, some studies failed to find any significant relationship between military expenditures and economic growth (see A more systematic analysis of the studies in the literature would therefore indicate three main areas of contention, namely, different theoretical models of schools of economic thought, differences in the income levels of countries, and the direction of causality between the variables. First, from the perspective of different theoretical models of schools of economic thought, it may be emphasized that despite the numerous studies carried out in this area, the literature shows that there is a lack of consensus on the positive or negative impact of military expenditures on economic growth. These previous studies have indicated that military expenditures may influence the economy, especially its growth, through different channels. In this regard, there are generally two theses regarding the impact of military expenditures on economic growth: the demand side and the supply side. The supply side mostly, but not entirely, focuses on the negative effect, whereas the demand side argues a positive effect of this relationship (Michael P. Gerace 2002: 2). For example, from the positive effect or demand perspective, military expenditures are theorized to be capable of influencing economic growth through the expansion of total demand, generally called the Keynesian effect. This view holds that an increase in total demand leads to increased utilization of unproductive capital, a lower underemployment level, higher profits, and, therefore, higher economic growth (Suleiman Abu-Bader and Aamer Abu-Qarn 2003: 571). On the other hand, from the negative effect or supply perspective, the theoretical statements assert that military expenditure is a factor that inhibits economic growth. According to this negative perspective, military expenditures can slow down economic growth through some channels, such as the crowding out effect in investments, inflationary pressures, and decreases in publicly available funds for use in productive areas (Kollias, Manolas, and Paleologou 2004: 556). In addition to the abovementioned theoretical relationships, many empirical studies have also shown different results regarding the direction of the effect of those relationships. For instance, Adne Cappelen, Nills Peter Gleditsch, and Olav Bjerkholt (1984), Alex Mintz and Randolph T. Stevenson (1995), and Yildirim, Sezgin, and Ocal (2005)  Second, in light of differences in the income levels of countries, many studies have analyzed whether military expenditures affect economic growth in less developed countries (see, for example, Saadet Deger and Ron Smith 1983; Basudeb Biswas and Rati Ram 1986), developing countries (Na Hou and Bo Chen 2013), and advanced countries (Aynur Alptekin and Paul Levine 2012). More specifically, the theoretical arguments presented in the literature vary across country groups classified by income levels. For example, in examining the relationship between military expenditures and economic growth in less developed countries, it is important to recognize the existence of four channels: resource allocation and mobilization, production organization, sociopolitical structure, and external relations (Deger and Smith 1983: 337). On the other hand, economists assume that, compared with less developed countries, defense expenditures decrease economic growth in developed countries because these expenses use up current resources available for investment (Benoit 1973: 271). In addition, some economists believe that military expenditures increase economic growth in less developed countries. Therefore, poor countries prefer to cut back on expenditures providing high-growth development in favor of larger defense expenditures, whereas richer countries tend to increase development expenditures while maintaining a constant level of military expenditures. Thus, according to this perspective, a negative relationship between military expenditures and economic growth can be expected in poorer countries, and a positive relationship in richer countries (Peter C. Frederiksen and Robert E. Looney 1983: 633).
Finally, from the theoretical and methodological perspective, the literature provides four approaches in considering the direction of the causal relationship between military expenditures and economic growth. The first approach is the growth hypothesis, which focuses on the impact of military expenditures on economic growth. According to this hypothesis, military expenditures positively affect economic growth by increasing the total demand; this is called the Keynesian effect. The literature can be said to support the existence of unidirectional causality from military expenditures to economic growth (see, for example, John Paul Dunne, Eftychia Nikolaidou, and Dimitrios Vougas 2001; Erdal Karagol and Serap Palaz 2004;Chang, Huang, and Yang 2011). The second approach is the conservation hypothesis. In this regard, the conservative argument of Wayne Joerding (1986) claims that, theoretically, an economically growing country may strengthen itself against internal and external threats by increasing its military expenditures. Thus, the literature holds that there is unidirectional causality from economic growth to military expenditures (see, for example, Dritsakis 2004). The third approach is the feedback hypothesis, which asserts the existence of bidirectional causality between military expenditures and economic growth, as indicated in the literature (see, for example, Tsanyao Chang et al. 2001). The last approach is the neutrality hypothesis, which is supported by literature reports indicating the absence of a causal link between military expenditures and economic growth (see, for example, Christos Kollias 1997; Kollias, Manolas, and Paleologou 2004; Chang, Huang, and Yang 2011). However, the differences in the results are mainly caused by the various methods used in the studies.

Data and Econometric Methodology
The present study analyzes the causal relationship between military expenditures and economic growth by using data from 27 NATO member states. The variables used in this study include the per capita real military expenditure (MilEx) and the per capita real gross domestic product (Y), both of which are measured in constant 2010 US dollars 1 . Y is obtained from the World Development Indicators (WDI, 2019), whereas MilEx is from the Stockholm International Peace Research Institute (SIPRI, 2020) 2 . Due the data on military expenditures are not available for the common period, the 27 NATO countries are divided into two panels and periods for the panel data analyses. Panel A includes data for the period between 1960 and 2019 for 15 NATO member countries: Belgium, Canada, Denmark, France, Germany, Greece, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Turkey, the United Kingdom, and the United States. Panel B considers data for the period 1996 to 2019 for 12 NATO countries: Albania, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, and Slovenia.
After the data collection, the Granger causal linkage between military expenditures and economic growth is examined by applying symmetric and asymmetric bootstrap Granger causality analysis. In the first point, we use the symmetric bootstrap panel causality test developed by Konya (2006) for the relationship between military expenditures and economic growth. This test has some advantages. First, the level values of the variables show the initial values even if the variables are not stationary because the model does not require preliminary testing, such as unit root and cointegration tests. Second, this test considers the correlation between countries because it allows cross-sectional dependence. Third, the approach enables a comparison of the relationship between two variables across countries, providing separate results for each country. The bootstrap panel Granger causality test is carried out through a two-step process. In the first step, the cross-sectional dependence and panel heterogeneity are tested. In the second step, the causality between the variables is estimated with seemingly unrelated regression (SUR) system estimation. In this regard, the asymmetric approach proposed by Abdulnasser Hatemi-J (2012) is applied to the bootstrap Granger causality test developed by Kònya (2006) to investigate the causal relationship between military expenditures and economic growth. The following sections first discuss how the asymmetric approach obtains positive and negative series from the original forms of the variables and then explain the essential processes for the application of the bootstrap Granger causality test developed by Kònya (2006).

Asymmetry
According to the theoretical literature, economic agents in the cycles of economic growth and recession are generally accepted to favor different strategies, and this behavior is particularly true in the real and monetary sectors. Policymakers and domestic institutions face difficulties in these times because they are under pressure to determine how to respond to economic growth or decline. Because the asymmetry approach is incapable of providing full details on the situation of economic growth or recession, these difficulties are neglected in the calculation of symmetric estimation methods when the relationship between economic variables and symmetric estimation methods are evaluated. On the other hand, this approach, which presents a non-linear relationship between two variables, is believed to provide a more complete picture of the causality relationship between the variables as a more general tool because it provides more insights on the situation under both economic growth and recession. Thus, in terms of model specification, the asymmetric approach is said to be more successful than the symmetric approach. Furthermore, bootstrap simulations are more effective in generating critical values because economic series are typically non-normal with time-varying volatility. Otherwise, in such situations, asymptotic critical values will not be accurate. In this sense, Konya's (2006) panel Granger causality test establishes bootstrap critical values and provides the cross-sectional dependency and heterogeneity. In the present work, the asymmetric approach proposed by Hatemi-J (2012) is adapted to the symmetric panel Granger causality approach developed by Konya (2006). If both symmetric and asymmetric approaches indicate the causal relationship between military expenditures and economic growth, then the variables are strongly causal. In certain situations, even if the symmetric causality is insignificant, the asymmetric causality findings may indicate a significant causal relationship between the economic variables. In this case, ignoring the asymmetric relationship in the analysis could lead to a tendency to neglect the possibility of a causality relationship.
The fundamental asymmetric approach of Hatemi-J (2012) mainly derives positive and negative cumulative sum series from the original forms of the series and then explores the causal relationship between these cumulative sums. (Veli Yılancı and Mücahit Aydın, 2017: 12). To explain this approach, first, suppose that two series, such as and , follow the random walk processes in Eqs. (1) and (2), where (1) ,0 and ,0 are the initial values for the series in the equations, and ~(0, 2 ) and ~(0, 2 ) are disturbance terms with a white noise distribution characteristic. In this regard, the empirical representation of the positive and negative shocks of the series is defined as: + = max( , 0), − = min( , 0) (3) + = max( , 0), − = min( , 0) Due to the disturbance terms identified in the form of = + + − and = + + − , Eqs. (1) and (2) can be written as Eqs. (5) and (6): With the help of Eqs. (5) and (6), the positive and negative shocks in each variable can be expressed as Eqs. (7) and (8):

Testing Cross-Sectional Dependence
The first step in the Kònya bootstrap Granger causality test is to examine the cross-sectional dependency for the series.
In the application of cross-sectional dependence, it is often useful to test whether there is a connection between crosssectional units and whether these units are equally affected by shocks to related series (Yusuf Ekrem Akbas, Mehmet Senturk, and Canan Sancar 2013: 796). Given the expansion of globalization and the increased interdependence among countries today, any analyses that do not consider cross-sectional dependence would provide biased and inconsistent results (M. Hashem Pesaran 2004). In the case of cross-sectional dependence, first generation tests lead to spurious results because of size distortion. Trevor S. Breusch and Adrian R. Pagan (1980) proposed the Lagrange multiplier (LM) test in Eq. (9) to examine cross-sectional dependence.
In the LM test, the null hypothesis that there is no cross-sectional dependence is = ( , ) = 0 for all t and ≠ , whereas the alternative hypothesis that there is cross-sectional dependence for at least one pair is 1 = ( , ) ≠ 0 for at least one pair of ≠ . ̂ indicates a pair-wise Pearson correlation coefficient among residuals acquired through OLS estimation for each i. The LM statistic is used to test for cross-sectional dependence when → ∞ and N is constant, that is, when T > N. In addition, under these conditions, the statistics has an asymptotic chi-squared distribution with N(N-1)/2 degrees of freedom. Thus, the power of the LM statistic decreases when N increases. To solve this problem, Pesaran (2004) suggested two different cross-sectional dependence test methods with an asymptotic standard normal distribution. According to Pesaran, these methods are under → ∞ and → ∞, if T>N, the CDLM test is used; if N>T, the CD test is used. Under the null hypothesis that there is no cross-sectional dependence, the empirical representations of the CDLM2 and CD tests are modeled, respectively, as follows: However, in the stationary dynamic panel data models, the CDLM2 and CD tests do not have a tendency to reject the null hypothesis if the group means are zero but the individual means are different from zero. To solve this problem, M. Hashem Pesaran, Aman Ullah, and Takashi Yamagata (2008) suggested a bias-adjusted modified LM statistic that uses the mean and variance of the LM statistic. The bias-adjusted LM equation is expressed as: In Eq. (12), the terms and 2 , respectively, are the mean and variance of ( − )̂2 derived by Pesaran, Ullah, and Yamagata (2008).
has an asymptotically standard normal distribution when → ∞ and → ∞ under the null hypothesis that there is no cross-sectional dependence.

Testing Slope Homogeneity
The second step in the Kònya bootstrap Granger causality test is to determine whether the slope coefficients are homogenous. The traditional method used to determine whether there is evidence of homogeneity of slope coefficients is the standard F test. However, the F test is acceptable only when the number of cross sections (N) is small or the time period is large and when the explanatory variables are strictly exogenous or the error variances are homoscedastic. Paravastu A. Swamy (1970) developed a test to minimize the lack of homoscedasticity assumption in the F test. However, this test, similarly to the F test, is valid only under T>N. M. Hashem Pesaran and Takashi Yamagata (2008) recently proposed a standardized version of the Swamy's test, called the ∆ test, which is valid for larger panels. The first step in this testis to calculate the revised version of Swamy statistic. The test is formulated as defined in Eq. (13): In Eq. (13), ̂ denotes the pooled OLS estimator, ̂ is the weighted fixed effect pooled estimator, is an identity matrix, and ̃2is the estimator of 2 . Standardized dispersion statistics are later developed in Eq. (14) as: On the other hand, the ∆ test has an asymptotically normal distribution. We test the null hypothesis, 0 : = for all i, which states that the slope coefficient is homogeneous, against the alternative hypothesis, 1 : = for i ≠ j, which holds that the slope coefficient is heterogeneous under ( , ) → ∞, provided that √ ⁄ → ∞. The biasadjusted ∆ test,which is used for small samples and indicates a normal distributive property in the disturbance term, is defined in Eq. (15) as: In Eq. (15), (̃) = is the mean, and (̃) = 2 ( − − 1)/( + 1) is the variance.

Bootstrap Panel Granger Causality
When the existence of cross-sectional dependence and heterogeneity across the NATO countries has been determined, a bootstrap panel Granger causality test that should account for these dynamics is applied. The bootstrap panel causality test proposed by Kónya (2006) can account for both cross-sectional dependence and country-specific heterogeneity. This approach is based on the SUR estimation of the set of equations and on Wald tests with individual country-specific bootstrap critical values. The variables in the system do not need to be stationary when the critical values for the cross sections are derived from the bootstrap. Therefore, the level values of the variables can be used directly without applying preprocessing tests, such as unit root and cointegration tests, in the system estimation. The bootstrap Granger causality approach has two sets of equations: , including + and − for asymmetric causality; and , including + and − for asymmetric causality. The two sets of equations to be estimated are expressed in Eqs. (16) and (17) N and t denote the number of countries for two panels (A and B) and the time period, respectively. More specifically, N is 13 for Panel A and 14 for Panel B; t is 55 for Panel A and 20 for Panel B. , , and are common factors, whereas and l refer to the disturbance and lag length, respectively. Changing of lag lengths in the equation systems is allowed; however, it is assumed that the individual lag length does not change according to the cross sections. The maximum lag length considered in this work is 4, which was determined based on the Akaike (AIC) and the Schwarz (SBC) information criterion for systems. The general functional forms for these information criteria may be written as in Eqs. (18) and (19): Here, W, N, q, and T are, respectively, the residuals covariance matrix, number of equations, number of coefficients per equation, and sample size.
Each equation of the system analysis has various predetermined variables and disturbance terms. The disturbance terms in the equations are assumed to be closely interrelated with each other, that is, to have cross-sectional dependence. Hence, each set of equations is a SUR system. On the other hand, the Wald test is used to analyze causality. Four types of Granger causality may occur in the system estimation (Kònya 2006: 981).
1. If 2, = 0 for all i when 1, ≠ 0 for each i, there is one-way Granger causality from MilEx (or positive and negative components of MilEx) to Y (or positive and negative components of Y). 2. If 2, ≠ 0 for all i when 1, = 0 for each i, there is one-way Granger causality from Y (or positive and negative components of Y) to X (or positive and negative components of MilEx). 3. If 1, ≠ 0 and 2, ≠ 0 for all i, there is bidirectional Granger causality between X (or positive and negative components of MilEx) and Y (or positive and negative components of Y).
4. If 1, = 0 and 2, = 0 for all i, there is no Granger causality between X (or positive and negative components of MilEx) and Y (or positive and negative components of Y). Table 1 shows the results of the cross-sectional dependence (i.e. CDLM1, CDLM2, CD, and LMadj) and slope homogeneity tests ( . . ∆ and Δ adj ) applied to both panel groups. The results of the cross-sectional dependence test show that the null hypothesis of no cross-sectional dependence in the series is clearly rejected in both panel groups. Only the CD-test statistic for the negative cumulative shock series of Y in Panel A accepts the null hypothesis of no cross-sectional dependence. In contrast, the results of the three other tests of cross-sectional dependence indicate that the null hypothesis of no cross-sectional dependence for the negative cumulative shock series of Y is also clearly rejected. The results of the slope homogeneity test indicate that the null hypothesis claiming homogeneity is clearly rejected. Thus, the assumption of country-specific heterogeneity is valid because the economic relations of individual countries in a panel group may vary. In this regard, the bootstrap panel Granger causality approach can be applied under both cross-sectional dependence and slope heterogeneity.

Conclusion
This research empirically examines the symmetric and asymmetric causal relationships between military expenditures and economic growth in 27 NATO member states by using the bootstrap panel Granger causality analysis. The advantage of this approach is that it allows testing for cross-sectional dependence and country-specific heterogeneity. Because data on military expenditures are not available for all countries for the common period, the 27 NATO countries are divided into two groups (i.e., Panels A and B). Panel A includes data for the period of 1960-2019 for 15 countries: Belgium, Canada, Denmark, France, Germany, Greece, Italy, Luxembourg, the Netherlands, Norway, Portugal, Spain, Turkey, the United Kingdom, and the United States. Panel B considers data for the period 1996-2019 for 12 countries: Albania, Bulgaria, Croatia, Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, and Slovenia. The empirical results indicate that there is both symmetric and asymmetric Granger causality between military expenditures and economic growth, which vary from country to country. Here, the results are presented in terms of four outcomes. First, considering the asymmetric relationships between military expenditures and economic growth, the results suggest more empirical evidence for the causal linkages between military expenditures and economic growth. Second, there is no symmetric and asymmetric causality between military expenditures and economic growth only in Italy and the United States. Third, there is unidirectional and strong symmetric and asymmetric causality from military expenditures to economic growth in Denmark, France, and Netherlands and from economic growth to military expenditures in Latvia. Finally, the results of the causality tests indicate that there is bidirectional causality between military expenditures and economic growth in Belgium, Bulgaria, Canada, Germany, Greece, Latvia, Norway, Poland, Romania, Slovak Republic, Spain, and the United Kingdom. By evaluating together the symmetric and asymmetric causality results, some economic implications are determined. For example, the findings indicate that the neutrality hypothesis is valid in Italy and the United States. This hypothesis implies that military expenditures do not influence economic activities and that economic growth is not a determinant of military expenditures. In addition, our findings support the military expenditure-led growth hypothesis in Albania, Czechia, Denmark, France, Hungary, Lithuania, Luxembourg, Netherlands, and Portugal. Thus, the Keynesian theory is confirmed, which implies an impact of military expenditures as a public expenditure on employment and output levels through an increase in demand in the country. On the other hand, the results of the present study indicate that the conservative hypothesis is valid in Croatia, Estonia, Slovenia, and Turkey. Therefore, there is unidirectional causality from economic growth to military expenditures in these countries. Finally, the findings confirm that the feedback hypothesis is valid in Belgium, Bulgaria, Canada, Germany, Greece, Latvia, Norway, Poland, Romania, Slovak Republic, Spain, and the United Kingdom. Thus, the causal relationship between military expenditures and economic growth differs from one country to another. The reasons why the causality relationship between military expenditures and economic growth gives different results across countries are based on various preferences applied in countries' defense and economic growth policies.