The relationship between GDP and military spending; case of Serbia

Use current theoretical and empiricalliterature to provide a comprehensive research

project

Abstract

This paper aims the examination of the relationship between GDP

and military expenses in Serbia. We initially review the

empirical and theoretical contradictions continuing with our test

VAR, unit root, grander causality tests, conintegration to test

such relationship between GDP and military spending. Results

present a weak relationship in Serbia.

3

Table of Contents

Introduction..................................................4Literature Review.............................................5 Methodology..................................................7Unit Root.....................................................7

VAR Lag Order Selection Criteria……………..…….…………………………………15

VAR.........................................................16 Test for Grander causality................................19Vector Autoregression Estimates..............................27Conclusion...................................................30

References………………………………………………………………………………….31

INTRODUCTION

Defense industry positions Serbia as far leading country of the

region providing large range of military weapons. In sense of

economic perspective Serbia appears as main exporting country. As

well, it becomes possible due to the fact that Serbia is direct

inheritor of partially customers from former Socialist

Yugoslavia. As result of this, its arms productions are used by

NATO Missions around the world as presented in KIPRED report

2014.

Giorgio et al (2014) believe that military division giving an

assortment for public infrastructures (e. G. , dams,

correspondence networks, roads, airports, highways, Also other

transportation networks) might help a nation to expand the

physical capital, other than enhances human capital through

education, nutrition, medical care, and training. Further, it

will provide R&D experiences also military investing might bring

a sure impact for development through complementarity impacts on

the private production inputs.

5

In the other hand, Lindauer (2013) asked that if it makes much

more sense to assess the actual threats the country faces, to

determine the optimal means of meeting or deterring these threats

with a sufficient degree of confidence, and then to add up the

costs of obtaining the stipulated means? Why can’t the Department

of Defense today defend the country for a smaller annual amount

than it needed to defend the country during the “Cold War”. From

a neoclassical perspective, the effect of government spending

mainly depends on the productivity of the sector compared to the

civilian one and on the amount of resources allocated to it

through taxes (Giorgio et al, 2014).

In addition to this, Lindauer, (2013) asked that if it makes more

sense to evaluate the current threats the country faces on

determine the ideal method for meeting alternately determine

these threats with a enough degree of confidence, in order to

include the costs that are required to keep stimulated this

sector? The reason behind it is that, the countries spend huger

amount of money for the security than they did during the cold

war. From a neoclassical point of view, the impact of government

expenses mainly is depended on the productivity of the country

(Giorgio et al, 2014). Thus, by having a better view on the real

need for this sector we can re-destiny in civilian sector. GDP

measures the value of all goods and services produced during one

year, while military spending means the whole the investment that

state provide for the military sector (Pan et al, 2014).

LITERATURE REVIEW

World military expenditure in 2010 reached $1630 billion,

representing 2.6 per cent of global gross domestic product (GDP)

or $236 for each person. Spending was 1.3 per cent higher in real

terms than in 2009 and 50 per cent higher than in 2001. Most of

the countries have increased military spending rapidly in recent

years, and all are engaging in major military modernization

programmes, not all of which are clearly linked to a perceived

military threat or clear military mission (Perlo-Freeman et al,

2012).

There are at least four views regarding the nature of the

relationship between defense spending and economic growth. The

first is the growth hypothesis, which implies one-way Granger

causality from military spending to economic growth, where

military spending can either promote or retard economic growth.

If there is a unidirectional Granger causality from military

spending to economic growth, with increases in military spending

leading to increased economic growth, then the effect of defense

spending on economic growth is positive, as defense spending may

stimulate economic growth through Keynesian-type aggregate demand

effects (Pan et al, 2014).

7

The graphical analysis above indicates that as their economies

grew during the non-war years before World War I, the five great

powers in our sample did generally attempt to match—and only

occasionally to outmatch—that growth by increasing their military

expenditures. Graphical and statistical analysis indicate that

the relationship between military expenditures, economic output,

and economic output growth varies over time and across countries

(Castillo, Jasen, et al, 2001). This become obvious when we talk

for Iran country whose military spending in 2006 reached 52% from

16% in 1993, while they spend 15% for the education.

Concerning the impact of the GDP share of total public spending

and the share of military spending on economic growth, the

estimates confirm the nonlinear relationship that emerges from

the theoretical model. In addition to this, Pantelis and

Tzouvelekas (2011) investigated this relation and came to

conclusion that the impact military spending on economic growth

are in nonlinear relationship that emerges from the theoretical

model indicating that military spending in developing countries

is not impacting the economic growth in Bulgaria, Cote d’Ivoire,

Dominican Republic, El Salvador, Guinea, Madagascar, Morocco,

Nicaragua, Pakistan, Papua, and (Pantelis and Tzouvelekas, 2011).

From the above literature review we can conclude that we have two

main categories of thoughts related to the relationship between

military spending and GDP, also it is easy to conclude that it

varies from one to another countries and from one period of time

to another.

METHODOLOGY

In order to study the relationship between the GDP and military

spending, we have extracted available historical data for the

period 1996 to 2013 from World Bank data source. We have employed

the Unit root test, VAR, grander causality test.

Test applied to this paper

Based on the visual

inspection, it leads

to not a co-movement.

Unitroot

This test aims to

examine if the variables are stationary or are not stationary. We

-16

-12

-8

-4

0

4

8

12

1996 1998 2000 2002 2004 2006 2008 2010 2012

gdp% M IL

9

have stationary variables if value of its distributions stands

the constant as time passes by. It should have a constant mean, a

constant auto-covariance structure (how y is associated with its

prior values), variance of lags. In cases we apply regression

methods to non-stationary data, the ending result is unreliable

(Brooks, 2008)

GDP-Gross domestic Product Unit root test

Null Hypothesis: GDP has a unit rootExogenous: Constant, Linear TrendLag Length: 3 (Automatic - based on SIC,

maxlag=3)

t-

Statisti

c

Prob

.*

Augmented Dickey-Fuller test

statistic

-

4.169097

0.029

9Test critical

values:

1%

level

-

4.8864265%

level

-

3.828975

10%

level

-

3.362984

*MacKinnon (1996) one-sided p-values.Warning: Probabilities and critical values

calculated for 20 observations and may not be accurate for a sample

size of 13

Augmented Dickey-Fuller Test EquationDependent Variable: D(GDP)Method: Least SquaresDate: 05/06/15 Time: 10:32Sample (adjusted): 2000 2012Included observations: 13 after

adjustments

Variable

Coeffic

ient

Std.

Error

t-

Statisti

c Prob.

GDP(-1)

-

1.46519

9 0.351443

-

4.169097 0.0042

D(GDP(-1))

0.48366

7 0.247328 1.955575 0.0914D(GDP(-2)) 0.65924 0.241629 2.728336 0.0294

11

5

D(GDP(-3))

0.54945

6 0.223063 2.463234 0.0433

C

7.44229

3 1.857138 4.007399 0.0051

@TREND("1996")

-

0.29191

8 0.076983

-

3.791981 0.0068

R-squared

0.72426

2

Mean

dependent var

-

0.1469

83Adjusted R-

squared

0.52730

7

S.D.

dependent var

0.5427

93S.E. of

regression

0.37318

5

Akaike info

criterion

1.1705

51Sum squared

resid

0.97486

7

Schwarz

criterion

1.4312

96

Log likelihood

-

1.60857

9

Hannan-Quinn

criter.

1.1169

56

F-statistic

3.67729

0

Durbin-

Watson stat

1.6764

43Prob(F-

statistic)

0.05956

8

H0: GDP has a unit root

H1: GDP has not a unit root

If prob >α thus we reject H0, a=0.01 Prob=0.0299

If prob<a thus, we accept H0

Given that P-value is higher than α level of significance Ho is

not rejected, meaning that the series is not stationary. Thus,

first differences must be taken (Brooks. 2008).

T-stat -4.169097is lower than t- critical of -3.828975. Therefore

we reject the H0, at 0.05 level of significance.

T-stat -4.169097is higher than t- critical of -4.886426.

Therefore we accept the H0, at 0.01 level of significance.

Null Hypothesis: D(DGDP) has a unit

rootExogenous: ConstantLag Length: 0 (Automatic - based on SIC,

maxlag=3)

t-

Statisti

c

Prob

.*


statistic

-

7.406914

0.000

0Test critical1% -

13

values: level 4.0044255%

level

-

3.09889610%

level

-

2.690439



size of 14

Augmented Dickey-Fuller Test EquationDependent Variable: D(DGDP,2)Method: Least SquaresDate: 05/06/15 Time: 12:52Sample (adjusted): 1999 2012Included observations: 14 after

adjustments

Variable

Coeffic

ient

Std.

Error

t-

Statisti

c Prob.

D(DGDP(-1))

-

1.62554

7 0.219464

-

7.406914 0.0000C 0.00138 0.181518 0.007633 0.9940

5

R-squared

0.82052

7

Mean

dependent var

0.0434

04Adjusted R-

squared

0.80557

1

S.D.

dependent var

1.5395

42S.E. of

regression

0.67884

8

Akaike info

criterion

2.1947

23Sum squared

resid

5.53000

7

Schwarz

criterion

2.2860

17

Log likelihood

-

13.3630

6

Hannan-Quinn

criter.

2.1862

72

F-statistic

54.8623

7

Durbin-

Watson stat

2.4103

19Prob(F-

statistic)

0.00000

8

15

Military spending Unit root test

Null Hypothesis: DM has a unit rootExogenous: ConstantLag Length: 0 (Automatic - based on SIC,

maxlag=3)

t-

Statisti

c

Prob

.*


statistic

-

5.892940

0.000

3Test critical

values:

1%

level

-

3.9591485%

level

-

3.08100210%

level

-

2.681330



size of 15

Augmented Dickey-Fuller Test EquationDependent Variable: D(DM)Method: Least SquaresDate: 05/06/15 Time: 12:32Sample (adjusted): 1998 2012Included observations: 15 after

adjustments

Variable

Coeffic

ient

Std.

Error

t-

Statisti

c Prob.

DM(-1)

-

1.45196

0 0.246390

-

5.892940 0.0001

C

-

0.23577

2 1.774839

-

0.132841 0.8964

R-squared

0.72761

6

Mean

dependent var

0.5579

76Adjusted R-

squared

0.70666

3

S.D.

dependent var

12.655

15S.E. of

regression

6.85409

9

Akaike info

criterion

6.8111

37Sum squared

resid

610.722

8

Schwarz

criterion

6.9055

43Log likelihood- Hannan-Quinn 6.8101

17

49.0835

3 criter. 31

F-statistic

34.7267

4

Durbin-

Watson stat

2.0380

31Prob(F-

statistic)

0.00005

3

H0: DM evidenced a unit root

H1: DM has not a unit root

a = 0.01; 0.05

Prob= 0.0003

P-value is lower than a, thus, we reject Ho, meaning that the

series is stationary. But we identify that the constant is not

statistically significant, thus we take it (Brooks, 2008).

19

VAR Lag Order Selection Criteria

VAR Lag Order Selection

Criteria

Endogenous variables:

DGDP DMExogenous variables: CDate: 05/06/15 Time:

10:30Sample: 1996

2013

Included observations:

15

Lag LogL LR FPE AIC SC HQ

0 -61.93243NA* 17.26991 8.524324

8.61873

1* 8.523318

1 -57.46701 7.144675

16.4026

6*

8.46226

8* 8.745488

8.45925

1*

* indicates lag order selected

by the criterion LR: sequential modified LR test statistic

(each test at 5% level) FPE: Final prediction

error AIC: Akaike information

criterion SC: Schwarz information

criterion HQ: Hannan-Quinn information

criterion

Regarding the table, we conclude that only one lag will be used

in the tests we will run.

VAR-Vector Autoregressive model

Vector Autoregressive model is a system of equations (Brooks,

2008). The number of variables presented as endogenous is same

21

with the number of equations. VAR analyses are developed upon

supposition that all components within VAR are stationary and it

aims to test the inter-dependence among time series (Maddala and

Lahiri, 2009).

Vector Autoregression Estimates Date: 05/06/15 Time: 14:33 Sample (adjusted): 1999 2012 Included observations: 14 afteradjustments Standard errors in ( ) & t-statistics in [ ]

DGDP DM

DGDP(-1) 0.272607 5.219292 (0.18932)

(2.69864)

[ 1.43990]

[ 1.93404]

DM(-1) -0.006558-0.838832 (0.01366)

(0.19470)

[-0.48012]

[-4.30829]

C -0.073623 1.352787 (0.08714)

(1.24213)

[-0.84487]

[ 1.08909]

R-squared 0.779058 0.674836 Adj. R-squared 0.680862 0.530319 Sum sq. resids 0.865442 175.8401 S.E. equation 0.310097 4.420157 F-statistic 7.933674 4.669588 Log likelihood -0.380136-37.57876 Akaike AIC 0.768591 6.082680 Schwarz SC 0.996826 6.310915 Mean dependent -0.101195 1.053328 S.D. dependent 0.548919 6.449649

Determinant residcovariance (dof adj.) 1.867074 Determinant residcovariance 0.771597 Log likelihood -37.91523 Akaike informationcriterion 6.845033 Schwarz criterion 7.301502

VAR Model:

Left hand side Right hand side

PRF: DGDPi = α0 + α1GDPi-1 + α2DM i-1+ U1,i

PRF: DMi=β0+ β1GDP i-1+ β2DM i-1 + U2,i

SRF: ^DGDPi= -0.073623+ 0.272607GDPi-1-0.006558DM i-1

SRF: ^DMi= 1.352787+ 5.219292GDP i-1- 0.838832DM i-1

The null hypotheses:

First VAR model

H0: α2=0

H1: α2≠ 0

23

Second VAR model

H0: β1=0

H1: β1≠0

We will continue by comparing t tc

VAR (1) t and tc :

t=β−bose

(β )

= -0.48012; tc=t a/2; T-m= t 0,025; 15-2≈1.77) and 2.65 at

0.01 level of significance, tc>t thus H0 is rejected meaning that

DGDP has an impact on DM (Brooks, 2008).

If DGDP increases by 1 % DM increases by 0.48012 % and if DGDP

decreases by 1 % DM decreases by 0.48012 %.

VAR (2) t and tc are :

t=β−bose

(β )

=1.93404; tc=t a/2;T-m= t 0,025;15-2≈2.9467) and 2.65 at

0.01level of significance, tc>t at 0.005 and 0.01 level of

significance, thus we reject H0 meaning that DM impact DGDP.

Test for grander causality

Pervaise Grander causality

Causality test aims to answer simple questions of the kind, ‘Does

change in y1 influences changes in y2?’ The logic followed is

that if y1 causes y2, (Brooks, 2008)

Pairwise Granger Causality TestsDate: 05/05/15 Time: 19:23Sample: 1996 2013Lags: 1

Null Hypothesis: Obs

F-

Statis

tic Prob.

DM does not Granger Cause

DGDP 15

0.005

87 0.9402

DGDP does not Granger Cause DM

2.445

16 0.1439

1ST hypotheses

Ho: DM does not Granger Cause DGDP

H1: DM does Granger Cause GDP

2nd hypotheses

H0: DGDP does not Granger Cause DM

25

H1: DGDP does Granger Cause DM

We compare Prob and α at any level of significance.

IF prob < α -reject H0

1 st hypotheses

Level of significance 0.01 and 0.05; Prob; 0.9402

0.9402>0.01 (level of significance); 0.9402>0.05 (level of

significance), we accept null hypothesis meaning that DM does not

Granger Cause DGDP at 0.01 and 0.05 level of significance.

2 nd hypotheses

Level of significance 0.01 and 0.05; Prob: 0.1439

0.1439>0.01; 0.1439>0.05 thus we ACCEPT H0 thus, DGDP doesn’t

Granger Cause DM at 0.01 and 0.05 level of significance.

Test for cointegration

A group of variables is defined as cointegrated if an in lines

arrangement of them is stationary. Several times series are non-

stationary. Over times, there appear influences on the series

which implies that the two series are bound by some relationship

in the long run. A cointegration can be seen as a long-term or

equilibrium phenomenon (Brooks, 2008)

-16

-12

-8

-4

0

4

8

12

1996 1998 2000 2002 2004 2006 2008 2010 2012

gdp% M IL

Null Hypothesis: U has a unit root-not

stationaryExogenous: ConstantLag Length: 0 (Automatic - based on SIC,

maxlag=3)

t-

Statisti

c

Prob

.*


statistic

-

3.868924

0.011

0Test critical

values:

1%

level

-

3.9203505%

level

-

3.065585

27

10%

level

-

2.673459



size of 16

Augmented Dickey-Fuller Test EquationDependent Variable: D(U)Method: Least SquaresDate: 05/05/15 Time: 18:30Sample (adjusted): 1997 2012Included observations: 16 after

adjustments

Variable

Coeffic

ient

Std.

Error

t-

Statisti

c Prob.

U(-1)

-

1.01715

6 0.262904

-

3.868924 0.0017

C

-

0.22131

1 1.298714

-

0.170408 0.8671

R-squared

0.51671

8

Mean

dependent var

-

0.1694

81Adjusted R-

squared

0.48219

7

S.D.

dependent var

7.2188

57S.E. of

regression

5.19458

1

Akaike info

criterion

6.2495

77Sum squared

resid

377.771

4

Schwarz

criterion

6.3461

51

Log likelihood

-

47.9966

2

Hannan-Quinn

criter.

6.2545

23

F-statistic

14.9685

7

Durbin-

Watson stat

1.9675

68Prob(F-

statistic)

0.00170

3

Null Hypothesis: U has a unit rootExogenous: Constant, Linear TrendLag Length: 0 (Automatic - based on SIC,

maxlag=3)

t-

Statisti

c

Prob

.*

Augmented Dickey-Fuller test- 0.044

29

statistic 3.800066 7Test critical

values:

1%

level

-

4.6678835%

level

-

3.73320010%

level

-

3.310349



size of 16

Augmented Dickey-Fuller Test EquationDependent Variable: D(U)Method: Least SquaresDate: 05/05/15 Time: 18:31Sample (adjusted): 1997 2012Included observations: 16 after

adjustments

Variable

Coeffic

ient

Std.

Error

t-

Statisti

c Prob.

U(-1) -

1.02526

0.269801 - 0.0022

3 3.800066

C

-

1.63304

6 2.795324

-

0.584206 0.5691

@TREND(1996)

0.16603

8 0.289107 0.574314 0.5756

R-squared

0.52867

6

Mean

dependent var

-

0.1694

81Adjusted R-

squared

0.45616

5

S.D.

dependent var

7.2188

57S.E. of

regression

5.32355

9

Akaike info

criterion

6.3495

22Sum squared

resid

368.423

7

Schwarz

criterion

6.4943

82

Log likelihood

-

47.7961

8

Hannan-Quinn

criter.

6.3569

40

F-statistic

7.29094

1

Durbin-

Watson stat

2.0007

90Prob(F-

statistic)

0.00752

6

Ho residuals are not stationary

H1 residuals are stationary

31

If residuals are stationary, residuals are coo integrated, if not

residuals are not stationary.

Given that P-value is lower than a, thus, we reject Ho, meaning

that residuals are stationary and tended.

As a result of this we can conclude that residuals are stationary

and cointegrated (Brooks, 2008).

JOHANSON TEST-Trasor statistic &max eigenvelnes

Date: 05/05/15 Time: 18:41Sample (adjusted): 1998 2012Included observations: 15 after

adjustmentsTrend assumption: No deterministic

trendSeries: GDP_ MILLags interval (in first differences): 1

to 1

Unrestricted Cointegration Rank Test

(Trace)

Hypothesi

zed Trace 0.05No. of

CE(s)

Eigenvalu

e Statistic

Critical

Value

Prob.*

*

None * 0.788841 24.08988 12.32090

0.000

4

At most 1 0.049577 0.762725 4.129906

0.439

9

Trace test indicates 1 cointegrating eqn(s)

at the 0.05 level * denotes rejection of the hypothesis at the

0.05 level **MacKinnon-Haug-Michelis (1999) p-

values

Unrestricted Cointegration Rank Test (Maximum

Eigenvalue)

Hypothesi

zed Max-Eigen 0.05No. of

CE(s)

Eigenvalu

e Statistic

Critical

Value

Prob.*

*

None * 0.788841 23.32716 11.22480

0.000

3

At most 1 0.049577 0.762725 4.129906

0.439

9

33

Max-eigenvalue test indicates 1 cointegrating

eqn(s) at the 0.05 level * denotes rejection of the hypothesis at the

0.05 level **MacKinnon-Haug-Michelis (1999) p-

values

Unrestricted Cointegrating Coefficients

(normalized by b'*S11*b=I):

GDP_ MIL-0.276403 0.188721 0.075576 -0.369465

Unrestricted Adjustment Coefficients

(alpha):

D(GDP_) 4.030562 -0.999891D(MIL) 0.387452 0.040787

1 Cointegrating

Equation(s):

Log

likelihoo

d -46.29677

Normalized cointegrating coefficients

(standard error in parentheses)GDP_ MIL

1.000000 -0.682775 (0.16568

)

Adjustment coefficients (standard error

in parentheses)D(GDP_) -1.114060

(0.39506

)D(MIL) -0.107093

(0.02167

)

1 st line Ho- none cointegrated equation against

H1- not non cointegrated (at least one coint.)

Prob <alfa thus we reject Null Hypothesis and not non

cointegrated (at least one cointegrated).

2 nd line Ho at most one cointegrated .

H1 not at most one coint. Equation(at least two)

Prob >alfa thus we accept Null Hypothesis meaning a most one

cointegrated (Maddala, 2006).

35

Vector Autoregression Estimates

VAR is estimated in order to examine whether there are lead--lag

relationships for the GDP against military spending (Brooks,

2008).

Vector Autoregression Estimates Date: 05/05/15 Time: 19:37 Sample (adjusted): 1999 2012 Included observations: 14 after

adjustments Standard errors in ( ) & t-

statistics in [ ]

DGDP DM

DGDP(-1) 0.353099 0.016273 (0.36753

)

(0.03988

)[ 0.96074

]

[ 0.40804

]

DGDP(-2) 0.279251-0.055591 (0.27449

)

(0.02978

)[ 1.01735[-

] 1.86644]

DM(-1) -4.027351 0.095493 (3.22115

)

(0.34952

)[-

1.25028]

[ 0.27321

]

DM(-2) -3.490675 0.193294 (2.03305

)

(0.22060

)[-

1.71697]

[ 0.87622

]

C 0.148839-0.096684 (0.90196

)

(0.09787

)[ 0.16502

]

[-

0.98789]

ECT -1.548273-0.029656 (0.44630

)

(0.04843

)[-

3.46914]

[-

0.61239]

R-squared 0.870161 0.788952 Adj. R-squared 0.789012 0.657046

37

Sum sq. resids 70.21338 0.826689 S.E. equation 2.962545 0.321459 F-statistic 10.72298 5.981201 Log likelihood -31.15251-0.059449 Akaike AIC 5.307501 0.865636 Schwarz SC 5.581383 1.139517 Mean dependent 1.053328-0.101195 S.D. dependent 6.449649 0.548919

Determinant resid

covariance (dof adj.) 0.889708 Determinant resid

covariance 0.290517 Log likelihood -31.07762 Akaike information

criterion 6.153946 Schwarz criterion 6.701709

We can evidence from the table that the coefficients of periods

for both equation are not statistically significant. So, we can

evidence that change in today GDP and M cannot in a long run be

used to describe the variation (Maddala, 2006).

CONCLUSION

In this study we proved contradictions of school of thoughts

related to relationship among GDP and military spending based on

literature and empirical results. With aim to test such

relationship for Serbia, we have employed available data for the

period 1996 to 2013. This data was taken from St. Louis Federal

Reserve Bank. Usually, such relationship is meant as not existing

or at weak level contradicted with the viewpoint that provides

certain arguments in favor of it.

Our results indicate that the GDP becomes statistically at first

difference while the military expenses are stationary at level.

On the other hand, DGDP has an impact on DM as well as DM has

impact on GDP of Serbia. Related to the causality, we noticed

that DM does not Granger Cause DGDP and versa. Residuals are

stationary and cointegrated while the GDP and military spending

are cointegrated at least one lag. And at the end after applying

Vector Autoregresion Estimates we can evidence that change in

today GDP and M cannot in a long run be used to describe the

variation.

39

References

Brooks,C.(2008).Introductory Econometrics for Finance, Ed.

Cambridge University

Press 2nd edition. Cambridge UK.

Castillo, Jasen, et al.. (2001). EMPIRICAL DETERMINATION OF THE

GROWTH–

MILITARY EXPENDITURES RELATIONSHIP. In Chpaper: Military

expenditures and economic growth. CA: RAND ARROYO CENTER SANTA

MONICA. p. 11-48.

Farzanegan, M. (2014). Military spending and economic growth: the

case of

Iran. Defence and Peace Economics, 25(3), p.247-269.

Giorgio, D. et al (2014). Does military spending stimulate

growth? An empirical

investigation in Italy. University Library of Munich, Germany

KIPRED. (2014). Security and Defence Context of the Western Balkans.

Available:

http://www.kipred.org/advCms/documents/18007_Kosovo%20in

%20the%20Security%20Context%20and%20Defense%20Context%20of

%20the%20Western%20Balkans.pdf. Last accessed 01.05.2015.

Lindauer, J. (2013). The General Theories of Inflation, Unemployment, and

Government Deficits. iUniverse.

Maddala, G.S. (2006) Introduction to Econometrics. Third edition.

England: John

Willey & Sons, LTD.

Pan, C. et al. (2014). Military spending and economic growth inthe Middle East

countries: bootstrap panel causality test. Defence and PeaceEconomics,

p1-14.

Pantelis, K and Tzouvelekas, V. (2011). Military Spending And The

Growth‐

Maximizing Allocation Of Public Capital: A Cross‐Country

Empirical Analysis. Economic Inquiry. 49 (4), p.1029-1040.

Perlo-Freeman, S., Cooper, J., & Ismail, O. (2012). 4. Militaryexpenditure.

http://www.kipred.org/advCms/documents/18007_Kosovo%20in%20the%25

http://www.kipred.org/advCms/documents/18007_Kosovo%20in%20the%25

The relationship between GDP and military spending; case of Serbia

Documents