DOES DEFENCE SPENDING IMPEDE ECONOMIC GROWTH? COINTEGRATION AND CAUSALITY ANALYSIS FOR PAKISTAN

Abstract

This study revisits the relationship between defence spending and economic growth via a Keynesian model in Pakistan using the autoregressive distributive lag bounds testing approach to cointegration. Empirical evidence suggests a stable cointegration relationship between defence spending and economic growth. An increase in defence spending reduces the pace of economic growth confirming the validity of Keynesian hypothesis in this case. Current economic growth is positively linked with economic growth of previous periods while a rise in non-military expenditures boosts economic growth. Interest rate is inversely associated with economic growth. Finally, unidirectional causality running from military spending to economic growth is found.

Keywords:

• Defence spending
• Economic growth
• Cointegration
• Causality
• Pakistan

JEL Codes:

• C12
• O16

1. Introduction

The study revisits the impact of defence spending on economic growth using an augmented Keynesian model in case of Pakistan. Existing literature highlights two main channels through which defence spending affects economic growth. On the one hand, an increase in military spending increases aggregate demand by stimulating output, employment and hence economic growth. Additionally, improvements in the quality of human capital due to military spending through education and technological training seems to have positive spill-over effects and so do increases in R&D spending. On the other hand, an increase in military spending may bring about a shift of resources away from the private sector resulting in reduced private spending. This seems to crowd-out both public and private sector investments which retards economic growth (Sandler and Hartley, 1995).

The seminal work of Benoit (1973, 1978) has spurred an increasing attention on the economic impact of defence spending with the relevant literature steadily expanding ever since (inter alia: Chowdhury, 1991; Dakurah et al., 2001; Sezgin, 2001; Atesoglu, 2002; Dunne et al., 2002; Abu-Bader and Abu-Qarm, 2003; Cuaresma and Reitschuler, 2004; Halicioglu, 2004; Bas, 2005; Yildirim et al., 2005; Kollias et al., 2007; Lee and Chen, 2007; Özsoy, 2008; Wijeweera and Webb, 2009). Broadly speaking, the findings are mixed and inconclusive with results depending on the country or sample of countries, the time period or the methodology used (for a survey of the findings and the issues see Dunne et al., 2005). In a number of cases, findings suggest that defence expenditures stimulate economic growth through the demand stimulation channel (inter alia: Fredericksen and Looney, 1982; Stewart, 1991; Ward et al., 1991; Dunne et al., 2001; Atesoglu, 2002, 2009; Yildirim et al., 2005). Military expenditures may also lead to improvements in infrastructure while a similar positive effect is also postulated when it comes to the labour force and the technical skills acquired during military training that can then be useful endowments for the civilian life of soldiers once decommissioned (MacNair et al., 1995). On the other hand, however, there has been ample evidence generated by a plethora of studies pointing to a negative impact on growth using both cross-section as well as time series data-sets in the empirical investigation of the issue at hand (inter alia: Deger and Smith, 1983; Lim, 1983; Faini et al., 1984; Fredericksen and Looney, 1983; Birdi and Dunne, 2002; Karagol and Palaz, 2004; Karagol, 2006; Mylonidis, 2008; Smith and Tuttle, 2008; Pieroni, 2009; Abu-Qarn, 2010).

In case of Pakistan, Tahir (1995) scrutinised the direction of causal relationship between military spending and economic growth for Pakistan and India using the vector error-correction model (VECM) granger causality in bivariate system. The results showed that military spending and economic growth Granger caused each other in both the countries and same inference was drawn by Khilji and Mahmood (1997). Moreover, Khilji and Mahmood (1995) found negative impact of defence spending on economic growth using three-equation model. Khan (2004) investigated the Military Keynesianism Hypothesis (MKH) and reported bidirectional causality between both the variables. Furthermore, he concluded that the MKH does not hold true in case of Pakistan. Nevertheless, the findings by Khan (2004) may be biased due to occurrence of structural break in time series data as East Pakistan got independence and Bangladesh came into being in 1971.

Hoping to contribute to the ongoing discourse the issue is addressed by this study in case of Pakistan making a twofold contribution. Firstly, the study reinvestigates the impact of military spending on economic growth both for the long run as well as the short run using a Keynesian model over the period of 1972–2009. Secondly, an autoregressive distributive lag (ARDL) bounds testing approach is applied to examine cointegration among the variables. This methodology has not to the best of our knowledge been used before in the case of Pakistan. Thirdly, the study uses the Ng and Perron (2001) unit root test, which provides reliable and consistent results as compared to the other traditional unit root tests such as Augmented Dickey Fuller (ADF), Phillips and Perron (P-P) and Dickey Fuller Generalized Least Square (DF GLS). Finally, the VECM Ganger causality is employed to detect the direction of causality between defence spending and economic growth. The rest of study is organised as following: Section 2 contains the modelling framework and the data presentation; Section 3 explains the estimation strategy. The empirical evidence on the relationship between defence spending and economic growth is discussed in Section 4, and the conclusions and policy implications are to be found in final section.

2 MODELLING FRAMEWORK AND DATA

As noted by Dunne et al. (2005), the Feder-Ram model has been used extensively to examine the impact of military spending on the economy (Feder, 1982; Ram, 1986, 1995). Such studies include Biswas and Ram (1986), Ward et al. (1991), Yildirim et al. (2005) that have examined the issue at hand for both developed and developing countries. Following Romer (2000) and Taylor (2000), Atesoglu (2002) used his own derived macroeconomic model by replacing Investment-Savings, Liquidity for Money (IS-LM) and Aggregate Demand, Aggregate Supply (AD-AS) models to examine the association between military spending and economic growth ¹ as given below:

(1)

where Y is real GDP or aggregate output, C is consumption in real terms, I is real investment, GE is real government expenditures on non-military sectors and X indicates real net exports of an economy or balance of trade. ME represents real military spending of an economy. ² These variables are termed as exogenous variables and written as follows:

(2)

(3)

(4)

(5)

where real taxes and real interest rate are denoted by T_t and R_t , respectively. The present study follows the Halicioglu (2004) approach and real interest rate is considered as an exogenous variable ³ . Following the above discussion, the empirical equation is modelled as follows ⁴ :

(6)

where GDP is real GDP proxy for economic growth, ME denotes real military expenditures, GE shows real government non-military expenditures while IR represents real interest rate. The linear specification of model has been converted into log-linear specification, since log-linear specification provides more appropriate and efficient results as compared to simple linear functional form of model (see for details Shahbaz, 2010). The data on real GDP, real military and real government non-military expenditure have been obtained from Government of Pakistan (GOP) (2008–2009). ⁵ The statistical bulletin of State Bank of Pakistan (SBP) ⁶ (2008–2009) is used to collect the data on real interest rate. The study covers the period from 1972 to 2008.

3 ESTIMATION STRATEGY

The recently developed Ng and Perron (2001) unit root test has been utilized to investigate the order of integration of the variables in this study. This test is particularly suitable for small sample data-sets. Ng and Perron unit root test contains four unit root tests including Phillips and Perron (1988) Z _a and Z_t , Bhargava (1986) R ₁ and Elliot-Rothenberg-Stock (ERS) optimal point tests developed by Elliot et al. (1996). These tests are based on GLS de-trend data . First, let us define

The four statistics are listed below.

and

where if {1} and if ={1,t}

Following this, we have employed the autoregressive distributed lag model or the ARDL bounds testing approach to cointegration developed by Pesaran et al. (2001) as the most appropriate specification to carry out cointegration analysis among the economic growth, defence spending, government non-military expenditures and real interest rate. The bounds testing approach to cointegration has numerous advantages over traditional techniques of cointegration. The main merit lies in the fact that it can be applied irrespective of whether the variables are integrated of order I(0) or integrated of order I(1). Fortunately, the ARDL bounds approach to cointegration is free of any problem faced by more traditional techniques employed in the economic literature. Another merit is that, it has better properties for small sample data-sets. Moreover, a dynamic error-correction model (ECM) can be derived from the ARDL model through a simple linear transformation (Banerjee and Newman, 1993). The ECM integrates the short-run dynamics with the long-run equilibrium without losing information about long run. The ARDL bounds testing approach to cointegration involves estimating the unrestricted error-correction method of the ARDL model as follows:

(7)

(8)

(9)

(10)

where , , , and , , , are the drift components and time trends, and is assumed to be white noise error processes. The optimal lag structure of the first differenced regression is selected by Akaike information criteria (AIC) to ensure that serial correlation does not exist. Pesaran et al. (2001) tabulated two critical bounds (upper and lower critical bounds) to take the decision on the existence or not of a long-run relationship among the variables. The null hypotheses of no cointegration in Equations (7)–(10) are , , and against alternate hypotheses of cointegration which is , , and . The next step is to compare the calculated F-statistic with the lower critical bound (LCB) and the upper critical bound (UCB) tabulated by Pesaran et al. (2001). There is cointegration among the variables if the calculated value of F-statistic is more than the UCB. If the LCB is more than the computed F-statistic then the hypothesis of no cointegration may be accepted. Finally, if the calculated F-statistic is between lower and upper critical bounds then decision about cointegration is inconclusive.

To establish the goodness of fit of the ARDL model, the diagnostic and the stability tests have also been conducted. The diagnostic test examines the serial correlation, functional form, normality of error term and heteroscedasticity associated with the model. The stability test is checked by applying the cumulative sum of recursive residuals (CUSUM) and the cumulative sum of squares of recursive residuals (CUSUM_SQ).

The next step is to detect the direction of causal relationship between economic growth, military spending, government non-military expenditures and real interest by applying standard Granger causality test augmented with a lagged error-correction term. The Granger representation theorem suggests that there will be Granger causality in at least from one direction if there exists cointegration relationship among the variables provided that the variables are integrated of order one or I(1). Engle and Granger (1987) cautioned that if the Granger causality test is conducted at first difference through vector auto regression (VAR) method then it will be misleading in the presence of cointegration. Therefore, the inclusion of an additional variable to the VAR method such as the error-correction term would help us to capture the long-run relationship. To this end, error-correction term is involved in the augmented version of Granger causality test and it is formulated in a bivariate pth order of the VECM which is as follows:

(11)

(12)

(13)

(14)

where is the difference operator; is the lagged error-correction term derived from the long-run cointegrating relationship; and and are serially independent random errors with mean zero and finite covariance matrix. The presence of a significant relationship in first differences of the variables provides evidence on the direction of the short-run causation while a significant t-statistic pertaining to the ECM proposes the presence of significant long-run causation. However, it should be kept in mind that the results of the statistical testing can only be interpreted in a predictive rather than in the deterministic sense. In other words, the causality has to be interpreted in the Granger sense.

4 EMPIRICAL ESTIMATION

The main objective of this paper is to re-investigate the impact of military expenditures on economic growth and the direction of causality between military spending and economic growth in case of Pakistan. A number of cointegration approaches such as Engle and Granger (1987), Johansen (1991, 1992) and Johansen and Juselius (1990), Stock and Watson (1993) and, Phillips and Moon (2001) are available to examine cointegration between the variables. ⁷ The prerequisite of these tests for cointegration is that all variables in the model must have same order of integration. ⁸ The ARDL bounds testing approach to cointegration is more advanced and flexible as compared to other traditional cointegration approaches. The ARDL model can be irrespective of the order of integration of the variables involved. However, it is pointed out by Ouattara (2004) that there is a need to know the order of integration of the variables used in the estimations. The main assumption of the ARDL model is that variables are integrated at I(1) or I(0) and no variable should be stationary beyond that integrating orders. İf any variable is integrated at I(2) then the whole computation of F-statistic for cointegration becomes invalid. Therefore, in order to apply the ARDL bounds testing approach to cointegration, it is necessary to have information about the order of integration of the variables. To this effect, as noted above the Ng and Perron (2001) unit root test was used given since it more powerful and reliable for small data-set as compared to other traditional unit root tests and produces consistent results. The results of unit root test are reported in Table I.

Table I Unit Root Estimation

Variables	MZa	MZ t	MSB	MPT
Ng and Perron at level with intercept and trend
ln GDP t	−4.20307	−1.3811	0.3285	20.9802
ln ME t	−3.2154	−1.2052	0.3748	26.9661
ln GE t	−5.8527	−1.7090	0.2920	15.5669
ln IR t	−8.7644	−2.0398	0.2327	10.5843
Ng and Perron at first difference with intercept and trend
Δln GDP t	−24.1666^∗	−3.4679	0.1435	3.8189
Δln ME t	−30.8187 ^∗	−3.8964	0.1264	3.1208
Δln GE t	−17.1881^∗∗∗	−2.9202	0.1699	5.3696
ln IR t	−27.9289^∗	−3.7181	0.1331	3.3713
Note: ^∗, ^∗∗ and ^∗∗∗ denote significance at the 1, 5 and 10% levels, respectively.

The empirical evidence shows that all the variables have a unit root problem at levels. At first differences, GDP, ME, GE and IR are found to be stationary. This shows that the variables in the model have a unique order of integration. In such circumstances, we can apply the ARDL bounds testing approach to cointegration to examine the long-run relationship between the variables. Before proceeding with the two steps the ARDL procedure, it is necessary to select the appropriate lag length of the variables. The AIC were used. The computation of the F-statistic seems to be sensitive to the lag order of the variables in the model (see Feridun and Shahbaz, 2010). The VAR results show that lag order 2 is appropriate. ⁹ The number of total regressions generated by the ARDL methodology is (4 + 1)² = 25 in estimated Equation (7). Table II shows the results of the F-statistic for cointegration. The empirical evidence indicates two cointegration vectors when ME and GE are dependent variables i.e. and For both equations calculated F-statistics are 5.278 and 5.191 and greater than UCB (5.039) at 10% level of significance.

Table II Cointegration Test: ARDL Bounds Test

Estimated model	F-statistics			Lag
F GDP t (GDP t /ME t , GE t , IR t )	3.165			2
F ME t (ME t /GDP t , GE t , IR t )	5.278^∗∗∗			2
F GE t (GE t /GDP t , ME t , IR t )	5.191^∗∗∗			2
F IR t (IR t /GDP t , ME t , GE t )	2.113			2
Estimated model	R ^2	Adjusted-R ^2	F-statistics	Durban–Watson
F GDP t (GDP t /ME t , GE t , IR t )	0.79820	0.55317	3.257^∗∗	2.283
F ME t (ME t /GDP t , GE t , IR t )	0.71432	0.36742	2.059^∗∗∗	2.376
F GE t (GE t /GDP t , ME t , IR t )	0.80145	0.56035	3.324^∗∗	1.915
F IR t (IR t /GDP t , ME t , GE t )	0.65814	0.27070	1.698	2.149
Note: ^∗∗∗denotes significance at the 10% level comparing with critical bounds generated by Turner (2006). b and c also indicate significant at 5% and 10% level of significance.

The existence of cointegrating vectors confirms the long-run relationship between GDP, ME, GE and IR i.e. real GDP, real military spending, government non-military expenditures and interest rate over the period of 1972–2008. The existence of long-run relationship among the variables helps us to find out partial effects of military spending, government non-military expenditures and real interest rate on economic growth in case of Pakistan. Empirical evidence reported in Table III indicates that current economic growth is positively affected by economic growth in previous period. It is concluded that a 1% increase in economic growth in the current period will raise economic growth by 0.8895% in future. The relationship between defence spending and economic growth is negative and significant at 5%. It implies that a 1% increase in defence spending will decline economic growth by 0.4515%. At this point, we can compare our results with Khilji and Mahmood (1997) who reported inverse impact of military spending on economic growth.

Table III Long-Run Elasticities

Dependent variable = ln GDP t
Variable	Coefficient	SE	T-statistic
Constant	0.7188	0.4765	1.5084
ln GDP t	0.8895	0.0270	32.9158^∗
ln ME t	−0.4515	0.1912	−2.3607^∗∗
ln GE t	0.1298	0.0220	5.8988^∗
ln IR t	−0.0418	0.0114	−3.3677^∗
R ^2= 0.9888
Adjusted R ^2= 0.9873
SE of regression = 0.0311
Akaike info criterion = −3.9688
Schwarz criterion = −3.7466
Log likelihood = 74.4542
F-statistic = 662.8248
Prob(F-statistic) = 0.0000
Durbin–Watson = 1.4876
Diagnostic tests		Statistics
J–B normality test		2.0692
Breusch–Godfrey LM test		[1] 1.8188, [2] 2.0671
ARCH LM test		[1] 0.1932, [2] 0.3322
White heteroscedasticity		0.9561
Ramsey RESET		2.9494^∗∗∗
CUSUM		Stable^∗∗
CUSUMsq		Stable^∗∗
Note: ^∗ and ^∗∗ denote significance at the 1 and 5% levels, respectively.

Our empirical evidence are in line with the findings by studies such as Atesoglu (2002) for the USA, Karagol and Palaz (2004) for Turkey, Smith and Tuttle (2008) for the USA and Keller et al. (2009) for OECD countries who found inverse relationship between defence spending and economic growth. The impact of non-military spending on economic growth is positive and it is statistically significant at 1% level of significance. It is found that a 5% increase in non-military expenditures raises economic growth by 0.649%. The findings are consistent with existing defence economics literature including Atesoglu (2002) for the USA, Halicioglu (2004) for Turkey, Yildirim et al. (2005) for Middle Eastern countries, Tang (2008) for Malaysia and Wijeweera and Webb (2009) for Sri Lanka. Finally, real interest rate is inversely correlated with economic growth. It is documented that a 1% increase in real interest rate is linked with 0.0418% decline in economic growth. These findings are consistent with the empirical results of Atesoglu (2002), Halicioglu (2004) and Wijeweera and Webb (2009).

The lower portion of Table III reflects that long-run model passes all diagnostic tests against serial correlation, autoregressive conditional heteroscedasticity, non-normality of residual term, white heteroscedasticity and misspecification of model. The long-run estimates are stable because diagrams of CUSUM and CUSUM_SQ are lying between critical bounds. To examine the short-run impact of independent variables including lagged error term ECM version of Ordinary Least Square (OLS) is used. The results of short-run model are reported in Table IV. The coefficient of lagged error term or indicates the speed of adjustment from short span of time towards long-run equilibrium path is significantly negative. It is suggested by Bannerjee et al. (1998) that significance of lagged error term further validates the established long-run relationship among the variables. Our empirical exercise indicates that coefficient of is −0.6057 and significant at 5%. It implies a 60.57% of disequilibrium from the current year’s shock seems to converge back to the long-run equilibrium in the next year.

Table IV Short-Run Elasticities

Dependent variable = Δln GDP t )
Variable	Coefficient	SE	T-statistic
Constant	−0.0010	0.0058	−0.1815
Δln GDP t −1	0.7954	0.1454	5.4671^∗
Δln ME t	−0.2147	0.7151	−0.3002
Δln GE t	0.0550	0.0256	2.1512^∗∗
Δln IR t	−0.0406	0.0146	−2.7739^∗
ECM t −1	−0.6057	0.2333	−2.5959^∗∗
R ^2 = 0.5358
Adjusted R ^2 = 0.4529
SE of regression = 0.0275
Akaike info criterion = −4.1840
Schwarz criterion = −3.9146
Log likelihood = 77.1285
F-statistic = 6.4639
Prob(F-statistic) = 0.0004
Durbin–Watson = 2.1083
Diagnostic tests		Statistics
J–B normality test		0.9678
Breusch–Godfrey LM test		[1] 0.8106, [2] 1.7890
ARCH LM test		[1] 0.9308, [2] 0.4728
White heteroscedasticity		1.1003
Ramsey RESET		1.6688
CUSUM		Stable^∗∗
CUSUMsq		Stable^∗∗
Note: ^∗ and ^∗∗ show significant at 1% and 5% levels respectively.

In the short run, economic growth is affected positively by 0.7954% in future by 1% rise in economic growth in the current period. The relationship between defence spending and economic growth is negative but it is insignificant. It is documented that a 1% increase in military spending will lower economic growth by 0.2147% in short span of time, but it is statistically insignificant. There is a positive association between government non-military expenditures and economic growth. The results indicate that a 1% rise in non-military spending shows very minimal effect on economic growth i.e. 0.0550%. Finally, the link between real interest rate and economic growth is negative, and significant at the 1% significance level. The coefficients of long run and short run for real interest rate are more or less the same.

For the short-run model, diagnostic tests also indicate that there is no evidence of serial correlation and the error term is normally distributed. There is no evidence of autoregressive conditional heteroscedasticity and white heteroscedasticity. Finally, the short-run model is well specified as confirmed by Ramsey RESET test. The stability of the long-run and short-run estimates is checked by applying the cumulative sum (CUSUM) and the cumulative sum of squares (CUSUM_SQ) tests. The results of CUSUM and CUSUM_SQ reveal that both short-run and long-run estimates are stable and reliable Figures 1–4.

4.1 The VECM and Direction of Causality between Defence Spending and Economic Growth

The presence of cointegrating among the variables leads us to perform the Granger causality test to provide a clearer picture for policy-makers to formulate defence and economic policies by understanding the direction of causality between defence spending and economic growth. It is reported that variables are cointegrated for long-run relationship and this leads us to apply the VECM framework to detect direction of causality between the variables both for short and long runs. The results of Granger causality test are reported in Table V.

FIGURE 1 Plot of cumulative sum of recursive residuals (Long run model) Note: The straight lines represent critical bounds at 5% significance level

FIGURE 2 Plot of cumulative sum of squares of recursive residuals (Long run model) Note: The straight lines represent critical bounds at 5% significance level

FIGURE 3 Plot of cumulative sum of recursive residuals (Short run model) Note: The straight lines represent critical bounds at 5% significance level

FIGURE 4 Plot of cumulative sum of squares of recursive residuals (Short run model) Note: The straight lines represent critical bounds at 5% significance level

Table V The Results of Granger Causality

Dependent variable	Type of Granger causality
	Short run				Long-run	Joint (short- and long-runs)
	Δln GDP t	Δln ME t	Δln GE t	Δln IR t	ECT t −1	Δln GDP t , ECT t −1	Δln ME t , ECT t −1	Δln GE t , ECT t −1	Δln IR t , ECT t −1
	F-statistics [p-values]				[t-statistics]	F-statistics [p-values]
Δln GDP t	–	2.7751^∗∗∗ [0.0816]	0.5497 [0.5839]	3.5256^∗∗ [0.0448]	–0.04757^∗∗ [−2.0951]	–	5.0604^∗∗∗ [0.0071]	2.2172 [0.1110]	3.2058^∗∗∗ [0.0403]
Δln GDP t	1.7899 [0.1877]	–	0.8173 0.4530]	0.5634 [0.5763]	−0.0543 [−1.0155]	2.1834 [0.1151]	–	1.1099 [0.3637]	0.8882 [0.4607]
Δln GE t	0.0767 [0.9263]	3.0751^∗∗∗ [0.640]	–	5.3668^∗∗ [0.0115]	−0.3752^∗ [−2.7935]	2.7262^∗∗∗ [0.0654]	7.6500^∗ [0.0000]	–	6.2417^∗ [0.0026]
Δln IR t	3.2052^∗∗ [0.0577]	0.7267 [0.4934]	0.0540 [0.9447]	–	−0.4838^∗∗∗ [−1.8536]	4.6518^∗∗ [0.0102]	3.3255^∗∗ [0.0358]	4.5179^∗∗ [0.0107]	–
Note: The asterisks ^∗∗∗, ^∗∗ and ^∗ denote the significant at the 1, 5 and 10% levels, respectively.

The causality relation can be divided into short and long-run causation as variables are cointegrated. The long-run causality is indicated by the significance of coefficient of the one-period lagged error-correction term in Equations (11)–(14) using t-test. The short-run causality can be detected by the joint significance F test of the lagged explanatory variables in the equation. Our empirical results suggest that the is having negative sign and statistically significance in all the VECM equations except in Equation (12). The results show unidirectional causal relationship running from military spending to economic growth in short run as well in long run over the period of 1972–2008. It is concluded on the basis of our empirical exercise that rise in defence expenditures will inversely Granger cause economic growth. These findings are contradictory to those of Tahir (1995), Khilji and Mahmood (1997) and Khan (2004) for Pakistan who reported bidirectional causality between the variables which may be biased and inconsistent due to different data span used in the studies. However, our empirical evidence is consistent with the existing defence economic literature such as Abu-Bader and Abu-Qarm (2003) for Egypt, Israel and Syria; Karagol and Palaz (2004) and Özsoy (2008) for Turkey; Tang (2008) for Malaysia and Smith and Tuttle (2008) for the USA.

Bidirectional causal relationship is found between economic growth and government non-military spending in long run. Our findings corroborate with the view by Abu-Bader and Abu-Qarm (2003) who reported that a rise in government non-military spending will stimulate the pace of economic growth and in turn, government allocates more resources to productive and efficient ventures to sustain the rate of economic growth. There is also bidirectional causal relation between economic growth and interest rate. It can be inferred on the basis of our findings that a rise in interest rate will Granger cause economic growth inversely through investment-declining effect while economic growth inversely Granger causes interest rate through real money balances enhancing effect. The unidirectional causality is reported to be running from interest rate and defence spending to government non-military expenditures in short and long runs. This shows that an increase in interest rate will increase the rate of inflation that makes government non-military spending less efficient and expensive. The causality running from defence spending to government non-defence spending lends the support for popular perception that a rise in defence expenditures is generally accompanied with the decline in development expenditures. Finally, there is unidirectional causality is found running from defence spending and government non-military expenditures to interest rate in long run. Overall, our results report that military spending Granger causes economic growth which indicates that high military spending is retarding economic growth both for short and long runs.

5 CONCLUSIONS AND POLICY IMPLICATIONS

The allocation of military and non-military expenditures for developing economies is one of the major policy issues which can direct the pace of economic growth. Therefore, the issue of military spending growth nexus has been investigated using cross-section and time series data analysis across developed, developing and least-developed economies by the researchers frequently. Various approaches including classical, neoclassical and Keynesian were used to explore the nature of relationship between defence spending and economic growth, and produced mixed results. Using time series data-set and the ARDL bounds testing approach to cointegration, relationship between military spending and economic growth in case of Pakistan has been re-investigated over the period of 1972–2008.

The empirical exercise has confirmed cointegration between economic growth, military spending, government spending and interest rate. Moreover, results have indicated negative effect of military spending on economic growth for Pakistan’s economy. These findings are consistent with the existing literature such as Khilji and Mahmood (1997), Atesoglu (2002), Karagol and Palaz (2004), Smith and Tuttle (2008), Tang (2008) and Keller et al. (2009). The estimated coefficient of government non-military spending is showing positive impact on economic growth supporting the views of Halicioglu (2004), Yildirim et al. (2005) and Wijeweera and Webb (2009). The inverse relationship is also witnessed between real interest rate and economic growth, and is consistent with findings of Halicioglu (2004). Finally, unidirectional causal relationship running from military spending to economic growth has been found.

In the background of our empirical investigation, it can be highlighted that both Pakistan and India are strategically important nuclear states, and their cordial mutual relationship is important for South East Asian region as well as the global economy and peace. Therefore, it is highly appropriated if both the governments initiate bilateral talks to develop mutual confidence and harmony to fight against terrorism and poverty. The population size and population growth rate of both the countries do not permit them to invest such a huge chunk of their annual budgets on their military spending. It is strategically important for them to start dialogue to reach at a consensus for peace and prosperity by reducing their military size and expenditures. This may result in reducing the arms race between Pakistan and India which will shift resources to developmental projects and stimulate the pace of economic growth.

In the context of policy implications for Pakistan, defence expenditures are escalating due to the mutiny and unrest as a consequence of terrorism, violence and carnage. Terrorism is instigated and noticed in tribal areas like Federally Administered Tribal Areas (FATA) and others areas in Pakistan where per capita income appears to be very low. It can be highlighted that the terrorism in these areas may be due to low expenditure on the basic needs of health, education and infrastructure. Therefore, the Government of Pakistan should initiate development projects in the areas with low per capita income, scarcity of resources, penury and abject poverty by reducing military spending. Employment generating activities should be supported and emphasis should be placed on schooling, edification and civilisation. Currently, industries are established in Sind, especially in Karachi and Punjab such as Gujranwala, Sailkot, Faisalabad and Wazirabad. Government of Pakistan should pay special attention to establish industries in less-developed areas like FATA and other tribal areas to increase employment opportunities for the people of that area which will help to enhance their living standards. It will be possible by cutting down the defence spending and shift these resources to production ventures to sustain economic growth, reduce poverty and decline income inequality in the country.

We used Keynesian approach to examine the impact of military spending on economic growth using time series for Pakistan and this approach has its own limitations indicated by Dunne et al. (2005). The study can be extended for future research by including net exports, capital, labour and natural resources and technology (Dunne et al., 2005; Dunne and Uye, 2009) for more efficient and consistent results following endogenous growth model. d’Agostino et al. (2010) pointed out that modified endogenous growth model may provide more help in investigating the effect of military spending on economic growth by including the above mentioned variables.

Notes

¹Atesoglu (2002) extended the macroeconomic model by including military expenditures.

²The formation of the empirical model is totally based on Halicioglu (2004).

³For more details, see Atesoglu (2002) and Halicioglu (2004).

⁴Atesoglu (2002) has used many other additional equations to investigate the impact of military expenditures on aggregate output in the case of the USA with the help of new macroeconomic model, but ignored the effect of real interest on aggregate output (Halicioglu, 2004).

⁵GOP.

⁶SBP.

⁷Engle and Granger’s approach seems to produce less satisfactory when one cointegrating vector is present in multivariate case.

⁸There are several factors that can cause structural changes such as for instance changes in economic policies, financial or economic crisis/es and institutional change.

⁹The VAR lag length selection results are not reported but available upon request from authors.