Back to Current Issues |
Defense Spending and Economic Growth in Developing Countries: A Causality AnalysisAngson. H. Dakurah* , Stephen P Davies and Rajan. K SampathDepartment of agricultural and Resource Economics Colorado State University Fort Collins, Colorado.*The authors are respectively, a former Graduate Student and Professors at the Department of Agricultural and Resource Economics, Colorado State University, Fort Collins, Colorado. We wish to thank our anonymous referees for their useful comments and suggestions. AbstractThe debate over the impact of military spending on economic growth in the LDCs has continued with no consensus although there is theoretical and empirical evidence that supports both sides of the debate. This study examines the causal relationship between military expenditure and economic growth for 68 developing countries for the period 1975-1995. Granger causality testing procedures are employed for the analysis. The study found some evidence of unidirectional causality from military expenditure to growth and from growth to military expenditure in a number of countries, as well as a feedback relationship in others. However, for a majority of the cases investigated, no evidence of such relationship was found. An examination of the residual plots of these cases revealed evidence of structural change in the data series. The paper concludes that the lack of evidence of causal relationship between the two variables in this study and others before it might be due to violations of the basic assumptions inherent in these testing procedures. These include but are not limited to the presence of structural breaks in the data, the nature and method by which the data was collected and the general lack of equilibrium conditions in the macroeconomic environment in most of these countries. 1. IntroductionBenoit's (1973, 1978) seminal work on the economic impact of military expenditures in LDCs has been the subject of extensive empirical studies^{[1]}. A survey of the large body of empirical evidence in the literature reveals that there appears to be no consensus on either the existence of a relationship between defense spending and economic growth or, if it exists, the nature of such a relationship. In a study of fifty-five developing countries, Chowdhury (1990) found that for fifteen countries, the results indicate that military expenditure caused economic growth; unidirectional causality from economic growth to military expenditure existed in seven; while for three others, there was a feedback relationship between the variables. However, for the remaining thirty countries, no significant relationship between military expenditure and economic growth between the variables was found. Nadir (1993) examined the direction of causality between military expenditure and economic growth in 13 sub-Saharan Africa for the period 1967-1985 using Granger and Hsiao tests. Both test results were similar, and showed that military burden is not determined by economic growth. Most post-Benoit studies have assumed that military expenditure is an exogeneous variable, with the relationship between military expenditure and economic growth going from the former to the latter. The direction of causality between the two variables is important in the sense that, if economic growth causes military burden, but not the reverse then clearly there is no point in analyzing the impact of defense on economic growth (Nadir, 1993). Furthermore, Nadir points out that if the relationship is from defense to economic growth, then econometric models should allow for the effects of the relatively autonomous change in defense spending. This paper presents further evidence on the relationship between military expenditure and economic growth for developing countries in sub-Saharan Africa, Latin America with the Caribbean, and South and East Asia over the period 1975-1995. The period is selected to ensure the inclusion of many countries for which data is available. The rest of the paper is as follows. A brief overview of global trends in military expenditures is presented. This is followed by a brief review of the relevant literature and the methodology. Next, a discussion of the empirical results is presented. The final section of the paper is devoted to sunnnary and conclusions. 1.1 Trends in Military Expenditures: An OverviewWorld military spending of US &864 billion in 1995 was down 34% from an all time high of US $1.36 trillion in 1987 and marked the eighth consecutive drop (ACDA, 1996)'. As can be seen from table 1, the reduction in world military spending is derived from decreases in the developed country group, mainly from Europe, North America, and East Asia. The developed country-group's spending fell by 8% annually from 1991 to 1995. Table 1: World Military Expenditures: Shares and Growth, 1985-1995 In contrast, military spending for the developing country-group, mainly Latin America, Africa, and South Asia, experienced a sizable increase in 1995. Spending by this group has fluctuated repeatedly and peaked earlier in 1982-86 (ACDA, 1997). It declined relatively slowly compared to developed countries, at under 2% over the decade examined here. The trends in military expenditures are becoming more diverse. As shown in Table 1, real spending rates in 1991-95 for seven regions, including the largest, were declining, but five regions had rising trends. The five largest regions in military spending included North America, Western Europe, East Asia and the Middle East, which accounted for 93% of world total in 1995. Trends in Eastern Europe and the Middle East were sharply declining, in North America and Western Europe, moderately declining, and in East Asia, moderately rising. While all five major spending regions except East Asia had failing trends in the second half of the decade, the remaining smaller regions exhibited a more varied picture for that period. Two had falling trends-North Africa and sub-Saharan Africa - while Latin America and South Asia had rising trends. The discussion above can be put in a better perspective by use of other economic indicators, such as population, gross national product (GNP), central government expenditures and total trade. The use of these indicators makes it possible to put the military measures in a socioeconomic context, both within one country and compared across countries. Tables 2 and 3 present a selected number of such indicators for 1995, averaged for the major groupings and regions to permit comparisons across the world. Table 2: The Burden Ratio: Percentage Military Expenditure to GNP. One of the most often-used indicators is the military burden, defined as the ratio of military expenditure to GNP (ME/GNP). In Table 2, the average ME/GNP ratio for the world fell 2.8% in 1995, continuing a trend of consecutive reductions since 1986 (ACDA, 1996). The ratio for the developed group is similar to that of the world's ratio. The developing country ME/GNP ratio, on the other hand, was higher than for developed countries at the beginning of the decade, but it dropped faster until 1990, when it reacted to the Middle East crisis. Again, it dropped faster until 1994-95, when it levelled off at 2.8%. Overall, the ME/GNP ratio declined in all regions over the decade. The Middle East's ratio of 7.9% was the highest of any region in 1995, as six of the eleven countries in the world with estimated ME/GNP ratios IO% or over were in the Middle East. Table 3 cross-classifies all countries in 1995 according to both burden ratio and income level, as measured by GNP per capita. The widespread scatter of countries throughout the matrix suggests that relative income level or stage 'of development is not a critical determinant of burden ratio. Thus, involvement in civil war or external war, military threats by neighbours, or overemphasis on military power are relevant determinants, in addition to the fundamental influence of the absolute income or GNP level. This last point is an important one, especially when it comes to sample selection. It would be ideal to select a sample based primarily on these critical factors for a study of this nature. However, anyone familiar with military expenditure data will admit that this is a daunting task. Cognizance of this study seeks a modest objective of comparing the nature of causality of military burden and economic growth in a selected number of countries for which data is available. Table 3: Relative Burden of Military Expenditures: 1995 2. Literature ReviewThe literature examining the impact of military expenditures on economic growth is extensive, although that on the direction of causation is limited. Benoit's controversial finding of a positive relationship of defense spending to economic growth for 44 LDCs sparked a number of papers. A comprehensive summary of these studies can be found in, among others, Biswas and Ram (1986); Joerding (1986); LaCivita and Frederiksen(1991); Chowdhury (1991); Frederiksen and Looney (1985, 1993); Landau (1994); Alexander (1995); Kusi (1994); and Roux (1996). Kollias (1997) examines the issue of causality between military expenditure and economic growth for Turkey for the period 1954-93. A Granger causality test is applied and his empirical results reveal a lack of causal relationship between military expenditures, expressed as a percentage of GNP, and growth rates. Nadir (1993) examines the direction of causality between economic growth and military burden for 13 sub-Saharan Africa countries using Granger and Hsiao tests for the period 1967-1985. The results for the two tests are similar and show that military burden is not determined by economic growth. Joerding (1986) used Granger causality to check the assumed exogeneity of military spending relative to economic growth. Yearly observations for 57 LDCs were pooled for the period 1962-1977. His results showed that military spending is not a strongly exogenous variable and concludes that a number of previous studies, which implicitly assumed military spending to be an exogenous determinant of economic growth, are seriously flawed. LaCivita and Frederiksen (1991) re-examined the direction of causality between military expenditure and economic growth in a sample of 21 LDCs using Hsiao causality tests. Their results show that both the pooled sample and the majority of individual countries exhibited a feedback relationship between defense spending and growth. Chowdhury (1991) also examined the causal relationship between military expenditure and economic growth in 55 LDCs using a Granger causality test. The period covered by the study varied for the individual countries ranging over the period 1961-1987. His results showed a lack of consistent causal relationship between military expenditure and growth across different countries. 3. Model Specification and DataTheoretically, there is no clear-cut prediction of the direction of causation between these military expenditures and economic growth. Military spending can affect economic growth through a number of channels. On one hand, defense spending may retard growth through what is generally referred to as an investment "crowding-out" effect, which is a displacement of an equal amount of civilian resource use (in areas such as health, education and infrastructure). On the other hand, however, defense spending may stimulate growth through Keynesian-type, aggregate demand effects. An increase in demand generated by higher military spending leads to increased utilization of capital stock and higher employment. Increased capital stock may lead to higher profits, which may in turn lead to a higher investment, generating short-run multiplier effects. In addition, growth may be stimulated through spin-off effects such as the creation of socioeconomic structures conducive to growth (Deger, 1986). Although military expenditure may affect growth through these mechanisms, Joerding (1986) contends that economic growth may be causally prior to defense spending. For instance, a country with a high growth rates may wish to strengthen itself against foreign or domestic threats by increased defense spending. However, it is equally plausible that countries with high growth rates may divert resources from defense into other productive uses (Kollias (1997). The foregoing discussion gives rise to four possible outcomes regarding the causal relationship between economic growth and military expenditures: a unidirectional causality from military expenditure to economic growth or vice versa; bi-directional causality between the two variables; and finally, a lack of any causal relationship. Moreover, there are both positive and negative net effects, so that there may be unidirectional causality from military expenditures to economic growth that is on balance either positive or negative. Several policy implications can be derived from understanding directions and magnitude of causality between military expenditures and economic growth. The easiest inferences are made when military expenditures precede economic growth. A positive causal relationship suggests that aggregate demand effects are dominant, while crowding out is the main effect when there is negative causality. If the direction runs from economic growth to military expenditures and is positive, countries are trying to protect growth, or are at a stage of development, where defense spending is seen as a positive social good. If it is negative, there may be some scale economies reached. This paper provides statistical tests for the relationship between two time series of data for military expenditures, {Mexp_{t}}, and economic growth, {GNP_{t}}, using Granger causality testing procedures (Granger (1969)). Data from the United States Arms Control and Disarmament Agency (ACDA) supplemented by other sources are used, and cover the period 1975-1995. In carrying out the Granger causality test, the question of whether GNP causes Mexp is to determine how much of the current value of Mexp can be explained by past values of itself and to see if adding lagged values of GNP improves the explanation of Mexp. Mexp is said to be Granger-caused by GNP when GNP helps in the prediction of Mexp, or alternatively, if the coefficients of lagged GNP are statistically significant. The assumptions of the classical regression model necessitate that both series {GNP_{t}} and {Mexp_{t}} be stationary and that errors have a zero mean and finite variance. In the presence of nonstationary variables, there might be what Granger and Newbold (1974) called a spurious regression. For a simple bivariate model of the type used in this analysis, the pattern of causality can be identified by estimating regressions on Mexp and GNP, using the current and past values of Mexp and GNP as regressors and by testing the appropriate hypotheses. Thus, for this study, the model involves estimating the following equations: (1) (2) The coefficients a_{i} and b_{j} describe the effects of k current and past values of GNP and Mexp on Mexp, while f_{i} and g_{j} and describe the effects of k current and past values of GNP and Mexp on GNP_{t}. The e_{1t} and e_{2t}, are assumed to be uncorrelated. Uni-directional causality from Mexp to GNP is indicated if the estimated coefficients of lagged GNP in equation (1) are statistically different from zero, i.e. (Sa_{i… }0). If unidirectional causality is from GNP to Mexp, then the coefficients of lagged Mexp in equation (2) are statistically different from zero as a group, i.e. Sg_{j… }0. If both Sa_{i… }0 and Sg_{j… }0, then there is bi-directional causality and both variables are related to current and/or past values of the other variable. The computations of the relevant Wald test statistic presupposes that variables entering the vector autoregression (VAR) specification are covariance stationary. If the variables in question are non-stationary, the form of non-stationarity should be determined prior to the implementation of the Granger causality test so that the danger of drawing misleading inferences can be avoided (Nadir 1993). 3.1 Testing for StationarityFollowing Murthy (1993), a series X, is said to be integrated of order d, denoted by X_{t- }I(d), if it becomes stationary after differencing d times. Thus, X_{t} contains d unit roots. A series that is I(O) is said to be stationary. To determine whether a series is I(l) against the alternate I(O), unit root tests developed by Dickey and Fuller (198 1) are used. The most general form of the Dickey-Fuller test, the Augmented Dickey-Fuller (ADF) test, estimates the auxiliary regression below by OLS: (3) where D is the first difference operator, t is a linear trend, and e_{t} is a normally distributed term with zero mean and constant variance. The lagged DMexp_{t-i} are included to account for possible auto correlation in e_{t}. The ADF tests can be done with or without the constant (a_{0}) and time trends (a_{l}t) above. Addition of the constant term in effect amounts to conducting the test on the residual variation conditioned on a deterministic trend. A quadratic deterministic time trend is assumed when both the constant and trend term are added. If the a_{0} and time trends a_{l}t are not included, and there is no autocorrelation, so that the c_{i} are zero, the test is on a pure random walk model. If the constant is included, the test is for a random walk with drift (Anwar et. al 1996). The parameter of interest in all these regressions is a_{2}, because if a_{2}=0, then {Mexp_{t}} contains a unit root. The test involves estimating one or more variations of equation (3) using ordinary least squares (OLS) to obtain the estimated _{2} and associated standard error. Comparing the resulting t-statistic with Mackinnon critical values permits the researcher to accept or reject the null hypothesis that a_{2}=0. If the calculated t-ratio is less than the critical t-value, then the null hypothesis of a unit root (non-stationarity) is rejected and {Mexp_{t}} is integrated of order zero, i.e. I(O). If it is found that an individual time series is integrated of order I(l), then that series is non-stationary, and contains a stochastic trend. 3.2 Testing for CointegrationThe next step is to test whether stochastic trends (or unit roots) in the series are cointegrated to avoid running spurious regressions. Engle and Granger (1987) point out that a linear combination of two or more non-smionary series may be stationary, or I(O), and the non-stationary time series are said to be cointegrated. Testing the two series for cointegration assesses whether the individual stochastic trends are related to each other. Once the order of integration is determined by Dickey-Fuller tests, the next step is to examine whether the series are cointegratecl or not, and if they are, to identify the co-integrating (long-run equilibrium) relationships. The stationary linear combination is called the cointegrating equation and mav be interpreted as a long-run equilibrium relationship between the variables. Intuitively, if the series {GNP_{t}} and {Mexp_{t}} were both I(d), we could run a regression of the form: (4) If the residual e_{t} from the regression is I(O) then, GNP and Mexp are said to be cointe- grated and have a long-term relationship. Equation (4) is the cointegrating equation; the cointegrating vector would be (I-b), formed with the coefficients when the first term on the right is taken to the left hand side. Johansen (1991) developed a modified vector error correction (VEC) model that has cointegrating restrictions built into the specification, which was designed for use with cointegrated, non-stationary series. The VEC specification restricts the long-run behavior of the endogenous variables to converge to their cointegration relationships while allowing a wide range of short-run dynamics. The cointegrating variable is known as the error correction term (ECT,), and is equivalent to the e_{t} given in equation (4) above. The deviation from long-run equilibrium is corrected gradually through a series of partial short-run adjustments (Enders, 1995). In long-run equilibrium, the error correction term is zero. However, if Mexp and GNP deviate from long-run equilibrium during some past periods, the error correction term is nonzero and each variable adjusts partially each subsequent period restore equilibrium. The coefficients (l_{1} and l_{2}) of the error correction term in equations (5) and (6) below measure the speed of adjustment. 3.3 Granger Causality with Cointegrated variablesCorrelation, even in the long run among cointegrated variables, does not necessarily imply causality. If several series are cointegrated, then a Granger causality test can be constructed by augmenting the earlier construction with an appropriate error correction term (ECT) derived from the cointegrating equation. For example, if the two series are l(l), the Granger causality test for a bivariate regression would be applied after taking their first differences, and equations (1) and (2) would take the following forms: (5) (6) for all possible pairs of GNP and Mexp in the group. The A denotes the first difference operator and Ect_{t-p} is the error correction term, lagged p periods and derived from the cointegrating equation. The reported F-statistics are based on standard tests for the joint hypothesis that: (7) (8) for each equation. The null hypothesis is therefore that GNP does not Granger-cause Mexp in the first equation and that Mexp does not Granger-cause GNP in the second regression. 4. Empirical Results and Policy ImplicationsIn this section, we report the test results for the 62 countries used to examine the direction of causality between Mexp and GNP, which are based on four possible combinations of stationarity outcomes of {Mexp_{t}} and {GNP_{t}}. The first case analyzed is when {Mexp_{t}} and {GNP_{t}} are both stationary and the typical regression model for Granger causality can be used. Secondly, when {Mexp_{t}} and {GNP_{t}}are integrated of different orders, the regression would be spurious, so we do not examine these countries further. Thirdly, when non-stationary {Mexp_{t}} and {GNP_{t}} are integrated of the same order but the residual contains a stochastic trend, they are not co-integrated and the causality regression is spurious because all errors are permanent. Often it is recommended to estimate the equation using first differences, which we did. Fourthly, non-stationary {Mexpd and (GNP,L may be integrated of the same order and be cointegrated, so that they have stationary residuals. Under these circumstances, the approach to causality analysis that accounts for the existence of cointegrating relations should be used. The Dickey and Fuller (1979) equation (3) given above were used to test for the presence of unit roots in our data. Results of the unit root tests for the ADF procedures described above are reported in Tables 4, 5, 6 and 7. Following Ng and Perron (1995), the maximum lag length k required for correcting serial correlation in all ADF-type aux- iliary regressions was selected based on evidence provided by a IO% level of sequential t-ratio test on the significance of the highest order lag in the estimated autoregressions. In this case, the maximum lag period was 2 periods. The two variables, Mexp and GNP, were found to be integrated of different orders in 13 countries (Table 4). In two countries, we found GNP to be a I(2) process, while military expenditure was an I(2) process in one. An examination of the residual plots of the ADF auxiliary regression in these countries revealed structural breaks in the data sets. In the few cases where structural breaks were found close to the beginning or end of the data, observations were dropped and the I(2) series reverted to either I(0) or I(1) processes. In the case of the 13 countries, structural breaks were found in the mid-section of the data, and because of the short series of data, we could not drop observations. As noted by Enders (1995), when there are structural breaks, the various Dickey-Fuller and Phillips-Perron test statistics are biased towards the non-rejection of a unit root. He suggested splitting the sample into two parts and running the tests on each part. The problem with this procedure, given our short data series, is that degrees of freedom are lost, which in turn reduces the power of the test. In part, the lack of relationships found in this study and others before it could be attributed to this problem. The Mexp and GNP series in the remaining 49 countries were either I(0) or I(1) processes. Nigeria and Barbados were exceptions, where both Mexp and GNP turned out to be I(2) processes. Cointegration tests, using the Johansen's (1995) test procedures were then applied to these countries (See Tables 5 and 6). These analyses were done to determine whether a long-run equilibrium relation exists between the two variables. Among these countries, there was evidence of unidirectional causation from economic growth to military expenditure in 10, unidirectional causation from military expenditure to growth in 13, while there was a feedback relation in 6. On a regional basis, unidirectional causality from growth to military expenditure was found in 4 countries in sub-Saharan Africa, 5 in Latin America and I in South and East Asia. Uni-directional causation from military expenditures to growth was found in 8 sub-Saharan countries, 4 in Latin America, and only 1 in South and East Asia. Following Anwar et. al. (1997), for I(1) variables that are not cointegrated in levels, cointegration between the two variables was achieved after first differencing both. series to I(0), and then causality tests were then applied (Table 6). There was no evidence of a causal relationship for 18 countries even though the two variables were cointegrated (Table 7). That is, although there is evidence of a long-term relationship between variables, neither was found to cause the other in Granger's sense. Table 4: Unit Root and Stationarity Tests for Countries with differing orders of integration, (1975-1995) Table 5a: Unit Root and Cointegration Tests for Countries with same Order of Integration. (1975-1995). Table 5b: Unit Root and Cointegration Tests for Countries with same Order of Integration, (1975-1995) Table 6: ADF and Cointegration Testsfor Countries for which cointegration is achieved after differencing the series, (1975-1995) Table 7: Unit Root and Cointegration Tests for Countries with Cointegration but no Causality, (1975-1995) 4.1. Policy ImplicationsThe results in Tables 7 and 8 shed light on a number of policy implications discussed in the introduction. First of all, the even split in the direction of causality, with 20 countries showing causality from military expenditures to economic growth and 17 the reverse (albeit some bidirectional), suggests that a number of factors are at work that may differ significantly across countries. Table 8: Summary Table of Stationarity Tests and Causality Analysis between GNP and Mexp for 62 LDCs The perhaps surprising prevalence of causality from economic growth to military expenditures suggests that many goverru-nents take the state of their economy into account when making decisions regarding defense expenditures. The net effect of the causal relation, whether positive or negative, is informative as well. (This is analyzed by F tests on whether the sum of terms in the hypothesis tests of equations (7) and (8) are positive or negative, after having concluded that they are different from zero). Of the 23 countries with unidirectional causality, 16 had positive net effects while only 7 had negative effects, regardless of the direction of causality, inferring that defense spending most often raises economic growth or economic growth causes greater defense spending. These net positive effects support the belief that military expenditures and growth are causally related in developing countries most generally through an expansion of aggregate demand. Investment in infrastructure and human capital development in LDC economies operating below full employment thus has positive Benoit-type spillover effects from military expenditures. There is less evidence to suggest that military expenditures in developing countries negatively impact growth. Additionally, positive net effects when causality runs from economic growth to military expenditures imply that many LDCs are still at a stage where military expenditures are constrained by low income and will grow along with the economy. They are not yet in a position to have defense expenditures grow less than proportionally with economic growth. These conclusions must be tempered by the large number of countries that did not exhibit either causality or similar levels of integration in the two variables. At one level, this suggests that, in many countries, military expenditures and economic growth are not closely related. This conclusion is similar to Chowdhury's, with differences in time periods covered and data sources perhaps explaining variations in specific country results between the two studies. Table 9 shows a final perspective on the relationship between Mexp and GNP, focusing on long run relations. The speed of adjustment coefficients, in combination with the sign of the coefficient in the cointegrating vector, describe whether there is likely to be a true economic relation. If the b is negative, then Mexp and GNP are positively related, while the reverse is true when the cointegrating coefficient is positive. Given the sign on b, there should be a prevalence of ls of a certain sign if a variable is equilibrating. For example, our cointegrating equation for Burundi given in Table 5a, normalized on GNP, is: (9) From the perspective of a typical structural equation, or the cointegrating equation in (4), the signs on the constant and Mexp_{t} are reversed because they are on the left side of the equation. The error term thus describes departures from long-run equilibrium and is equivalent to ECT_{t}. From an economic perspective, Mexp is seen to be positively related to GNP when Mexp is taken to the right side of the equation. The question of causality arises in evaluating whether the signs on the ls are consistent with a long-run relationship. If there is a positive value of ECT_{t-p}, then either Mexp is too high or GNP is too low, and the adjustment back to equilibrium in later periods requires that either GNP fall or Mexp rise; in fact, any combination of the two variables moving correctly will return the relation to equilibrium. In error correction equations like (5) and (6), GNP and Mexp should thus respond negatively and positively to positive values of ECT_{t-p}, and the ls should be negative for GNP and positive for Mexp. On the other hand, if Mexp has a positive coefficient in the, cointegrating equation, implying a negative economic relation to GNP, then the adjustment over subsequent periods would have both GNP and Mexp decline to return to equilibrium. Naturally, if there is a negative shock, so that ECT_{t-p} is negative, the movements are reversed but the signs should be the same. Table 9: Summary Speed of Adjustment Analysis The comparison of these signs is given in Table 9. The first two columns contain the results for countries with negative signs for b in the cointegrating equation, and which have a positive economic relation between GNP and Mexp. The appropriate speed of adjustment coefficient (l) for a positive value of ECT_{t-p}, is for GNP to decline to facilitate a return to equilibriums but Mexp, because of the negative sign in the cointegrating equation, should increase to offset the positive error term. The countries with a positive l in the Mexp block and those in the GNP block with a negative l are correct, which is the bulk of the countries having a negative b. When the b is positive, as in the third and fourth columns, both Mexp and GNP should decline to offset disequilibria, and the correct responses are for the ls in both Mexp and GNP to be negative. Again, the bulk of the countries have the correct responses. 5. Summary and ConclusionsThis study examined the causal relationship between military expenditure and economic growth in a sample of 62 developing countries in sub-Saharan Africa, Latin America, South and East Asia. In summary, of the 62 countries examined, 14 had differing orders of integration and could not be examined further. Unidirectional causality from military expenditure to growth was found for 13 countries, from growth to military expenditures for 10 countries and a feedback relation for 6. For 18 countries, no causal relationship was found despite evidence of a long-term relationship between the variables. Finally, for 8 other countries, cointegration was found only after first differencing the two variables. The even split in the direction of causality, with 20 countries showing causality from military expenditures to economic growth and 17 the reverse (albeit some bidirectional), suggests that a number of factors are at work that may differ significantly across countries. (This split in direction of the relationship were seen further in Table 9, which examined the cointegrating equations and speeds of adjustment for countries with I(1) processes. More than half of the countries showed a positive long run relation between Military expenditures and growth). The perhaps surprising prevalence of causality from economic growth to military expenditures suggests that many governments are forced to take the state of their economy into account when making decisions regarding defense expenditures. Moreover, of the 23 countries with unidirectional causality, 16 had positive net effects while only 7 had negative effects, regardless of the direction of causality, inferring that defense spending most often raises economic growth or economic growth causes greater defense spending. These positive net effects when causality runs from economic growth to military expenditures imply that many LDCs are still at a stage where military expenditures are constrained by low income and will grow along with the economy. They are not yet able to have defense expenditures grow less than proportionally with economic growth. With more than half of the countries either exhibiting differing orders of integration or a lack of causality, this study supports findings in Chowdhury's study of 5 I developing countries cited above. Variation in time and data sources might be reasons why differences exist in the conclusions reached with respect to specific countries in the two studies. The lack of strong statistical evidence of a causal relationship between military expenditure and growth in LDCs in this paper and others like Chowdhury, Kollias, and Nadir could be attributed to possible violations of some or more of the assumptions of the testing procedures used here. The existence of structural breaks, the short period nature of the data, and the very nature and procedures in gathering data on military expenditures are very important considerations in arriving at any conclusions about test results. The test results assume that the economies from which the series are generated are in equilibrium. This is often not the case for developing countries. Given that these are the best possible data available to us, care should be taken in making inferences regarding any test results. Research, using longer data period if available and taking into consideration the structural breaks in the data will help resolve this debate. ReferencesAlexander, R. (1995) Defense spending: burden or growth promoting? Defense and Peace Economics 6: 13-25. Anwar M Sarfraz, Davies, S.P. and Sainpath, R.K. (1996) Causality between Govenunent Expenditures and Economic Growth: An Examination using Cointegration Techniques. Public Finance 51(2): 166-184. Benoit, E. (1973) Defense Spending and Economic Growth in Developing Countries, Lexington: Lexington Books. Biswas, B. and Ram, R. (1986), Military Expenditures and Economic Growth in the Less Developed Countries: An Augmented Model and Further Evidence, Economic Development and Cultural Change 34: 361-372 Boulding, K. E (1974) Defense Spending: Burden or Boon? WarlPeace Report, 13. Chatte@ i, M (1992) Regional conflict and Military Spending in the Developing Countries, in Economics of Arms Reduction and the Peace Process (W. Isard and C.H Anderson, Eds.) Amsterdam: North-Holland, pp. 235-248. Chowdhury, A.R (1991) Defense Spending and Economic Growth: A Causal Analysis, Journal of conflict Resolution 35: 80-97. Dickey, D.A. and Fuller W.A. (1979) Distribution of the Estimators for Autoregressive Time Series with Unit Roots, Journal of American Statistical Association 75: 427-431. ________(1981) The Likelihood Ratio Statistics for Autoregressive Time Series with Unit Roots, Econometrica 49: 1057-1072. Enders, Walter (1995), Applied Econometric Time Series. New York: Wiley and Sons, Inc.(lst Edition). Engle, Robert F and Ng, Victor K. (1993) Measuring and Testing the Impact of News Volatility, Journal of Finance, 48: 1022-1082. Engle, Robert and Granger, C (1987) Cointegration and Error Correction Representation, Estimation and Testing, Economitrica 55: 251-276. Frederiksen, P. C and Looney, R. E (1983) Defense Expenditures and Economic Growth in the Developing Countries, Armed Forces and Society 9: 633-645. _______(1985) Another Look at Defense Spending and Development Hypotheses, Defense Analysis 1: 205-210.
_______(1993) Arms Imports and Third World Growth in the 1980s, in Defense Spending and Economic Growth Edited by James, E. Payne and Anandi, P. Sahu. Boulder: Westview Press, pp. 237-254. Johansen, Soren (1995) Likelihood-based Inference in Cointegrated Vector Autoregressive Models, London: Oxford University Press. Joerding, Wayne (1986) Economic Growth and Defense Spending, Journal of Development Economics 21: 35-40. Kollias, Christos (1997) Defense Spending and Growth in Turkey 1954-1993: A Causal Analysis, Journal of Defense and Peace Economics 8: 189-204. Kusi, N. K. (1994) Economic Growth and Defense Spending in Developing Countries, Journal of conflict Resolution 3 8: 152-159. LaCivita and Frederiksen (199 1) Defense: Spending and Economic Growth: An Alternative Approach to the Causality Issue, Journal of Development Economics, 35: 117-126. Landau, Daniel. (1983) Government Expenditure and Economic Growth: A Cross- country Study, Southern Economic Journal 49: 783-791. Madden and Haskhurst (1995) Causal Analysis of Australian Economic Growth and Military Expenditure: A Note, Journal of Defense and Peace Economics 6: 115- 121. Murthy, N.R. Veasudeva (1993) Further Evidence of Wagner's Law for Mexico: An Application of Cointegration Analysis, Public Finance/Finance Publiques, 48: 92-96. Ng, S. and P. Perron (1995) Testing for Unit Roots in Flow Data Sampled at Different Frequencies Economics Letters, 47 (3-4): 237-42. Perron, Pierre and Timothy Vogelsang (1992) Nonstationary and Level Shifts with Application in Purchasing Power Parity, Journal of Business and Economic Statistics 10: 301-20. Phillips, P.C.B. and P. Perron (1988) Testing for a Unit Root in Time Series Regression, Biometrika 75: 335-346 Roux A.(1996) Defense Spending and Economic Growth in South Africa, Journal for Studies in Economics and Econometrics 20 (1): 19-34. Sivard Ruth L (1982) World Military and Social Expenditures, Washington, D.C.: World Priorities, Inc. (7th edition). United States Arms Control and Disarmament Agency, (ACDA), (WMEAT WebPages) Data Tables, 1996 and various years. ^{[1]}See Joerding, 1986; Chowdhury, 1991; LaCivita and Frederiksen, 1991; Alexander, 1995; Madden and Haskhurst, 1995; Chattedi, 1992; and Kusi, 1994. |