Defence Spending and Economic Growth: A Causal Analysis for Greece and Turkey Paul Dunne, Eftychia Nikolaidou and Dimitrios Vougas Middlesex University Business School The Burroughs London, NW4 4BT United Kingdom Tel 0044(0)181 362 6834 Email E.Nikolaidou@mdx.ac.uk ABSTRACT There are a number of studies which consider the relation between military spending and economic growth using Granger causality techniques rather than a well-defined economic model. Some have used samples of groups of countries, finding no consistent results. Others have focused on case studies of individual countries, which has the advantage of the researchers bringing to bear much more data than the cross country samples and a greater knowledge of the structure of the economy and the budget. This paper adds to the literature by providing an analysis of two countries Greece and Turkey, which are particularly interesting case studies given their high military burdens, the bad relations between the two and the resulting arms race in the area. In addition to analysing the data using standard "pre-cointegration" Granger causality techniques, this paper employs modern vector autoregressive (VAR) methodology that utilises cointegration via Granger's representation theorem. The standard Granger causality tests suggest a positive effect of changing military burden on growth for Greece, but this is not sustained when the cointegration between output and military burden is taken into account. The only evidence of significant Granger causality is a negative impact of military burden on growth in Turkey. Paper presented at the ERC/METU International Conference on Economics, 9 th - 12 th September, 1998, Ankara, Turkey. We are grateful to the participants for comments.
1. Introduction Although the general trend in the post Cold War era shows reduced military expenditure worldwide, there are still some countries that continue to spend a huge amount on defence each year mainly as a result of security issues. Greece and Turkey (both members of the NATO alliance) are examples of such countries. Their military burdens remain the highest in NATO - 5.74% of GDP for Greece and 4.42% for Turkey compared to NATO s 3.5%, for the last decade. This makes understanding the economic effects of military expenditure an important concern for both of the countries. It also provides a valuable case study to add to the growing literature on the defence-growth relationship, which has still to develop any consensus. This paper empirically investigates the hypothesis of a causal relationship between defence spending and economic growth in Greece and Turkey over the period 1960-1996. Instead of relying on the implicit assumption that defence is causally prior to economic growth as most of the previous studies have done, this paper attempts to systematically analyse the presence and direction of the causal relationship between defence and growth in the two countries. Special attention is paid to the integration properties of the series and, in addition to analysing the data using standard pre-cointegration Granger causality techniques, this paper employs modern vector autoregression methodology (VAR). The VAR specification has become increasingly popular in the applied econometrics literature in recent years, its main advantage being that such models are dynamic specifications, free of economic assumptions imposed a priori. Thus, they allow for the testing of causal linkages without the need to first construct arguments and develop hypotheses justifying those linkages (Georgiou et al., 1996). The rest of this paper is organised as follows: Section 2 provides a brief overview of previous work and theoretical perspectives. Section 3, then gives a brief background analysis of the Greek and Turkish economies and of their military expenditures while Section 4 presents the methodology to be employed as well as the empirical analysis. Finally section 5 discusses the results. 2. Military Spending and Economic Growth The investigation of the defence-growth relationship was initiated by Benoit(1973, 1978) who found a positive effect of defence spending on economic growth. This result, which he found surprising, provoked much criticism and a lot of interest among researchers. This led to a large number of studies, mostly using Neoclassical 1 or Keynesian 2 models to provide consistent formal models. While the Neoclassical models concentrated on the supply-side (modernisation, 1 Examples of Neoclassical studies include: Biswas and Ram (1986), Alexander (1990), Mintz and Huang (1990), Mintz (1991), Mintz and Stevenson (1995), Sezgin (1996), Murdoch, Pi and Sandler (1997). 2 Examples of Keynesian studies include: Smith (1980), Faini et al (1980), Deger (1981), Lim (1983), Faini, Annez and Taylor (1984), Antonakis and Karavidas (1990a,b), Kollias (1994), Chletsos and Kollias (1995). 2
positive externalities from infrastructure, technological spin-offs), the Keynesian ones concentrated on the demand-side (crowding-out of investment, exports, education, health). Thus, Neoclassical models tended to find a positive effect of defence on growth, while Keynesian models found a negative one. To overcome the problem of concentrating on the demand or supply-side only, models were developed with a Keynesian aggregate demand function and a supply-side, in the form of a growth equation derived from an aggregate production function. Developed by Smith and Smith (1980), Deger and Smith (1983), Deger (1986), Roux (1996), Antonakis (1997) and others, these models hypothesised possible positive direct effects of defence on growth through Keynesian demand stimulation and other spin-off effects, and negative indirect effects through reductions in savings or investment. Although they provide a more complete picture of the defence-growth relationship by accounting for the interrelationships between the variables, they have been criticised for not being as strongly based on theory and thus, relying on more ad-hoc justifications. Instead of assuming exogeneity or endogeneity of the defence variable in the growth equation, the direction of Granger causality across defence and growth has also been investigated. Joerding (1986) using Granger causality tests, investigated the direction of causality between defence spending and growth for 57 LDCs over the period 1962-1977. His findings suggested that causality runs from growth to defence, and there was little evidence of causality from defence to growth. On the other hand, Chowdhury (1991) did not find any causality between defence spending and growth for most of a group of 55 LDCs and Kusi (1994) ended up to similar conclusions for 77 LDCs. Other studies have focused on case studies of individual countries. Chen (1993) found no causal link for China, Hasan (1994) finding a positive effect of military spending on growth when reworking Chen s data with Vector autoregression (VAR) models, and Kollias (1997) finding no causal ordering in a study of Turkey. Dunne and Vougas (1998) found that military burden has a negative impact on growth in South Africa, when analysed within a VAR framework. 3
3. Greece and Turkey: Economy and Defence Spending 3.1. Greece Until the late 1950s Greece was an underdeveloped country, with a low productivity agriculture sector and a very weak industrial sector; a situation partly attributable to the Greek Civil War 1944-1949. The US and the army had become important forces in Greek politics and Greece bcame part of Western organisations such as OEEC, the Council of Europe and, in 1952, NATO. Greece s security concerns were the threat from the Warsaw Pact countries and from Turkey and by joining NATO, Greece secured its northern borders but not its eastern ones. Greece and Turkey remained in the contradictory position of state to state adversaries but NATO allies (Sezer, 1991). After 1955 relations with America and Britain became troubled, partly because of resentment over US influence in Greece, but also because of the Cyprus problem (Veremis,1982). Cyprus was a British colony with a population that was 80% Greek and 20% Turkish. The Greek population of the island wanted self-determination and enosis (union) with Greece. Naturally, the Athens government felt sympathy for the Greek-Cypriots, but this provoked tensions with its NATO allies Britain and Turkey. In 1959 Cyprus gained independence from Britain but without enosis. Greek-Turkish tensions flared up in 1964, resulting in UN forces being sent to the island. In the 1960s the Greek economy underwent considerable structural change. 1962 saw the contribution of the industrial sector to national output become greater than that of agriculture for the first time (Kollias, 1996). Bet dent of the Cypriot Republic. In the 1970s the impressive growth rates of the previous decades declined as the structural weaknesses of the Greek economy became apparent. Despite the fact that the annual average growth rate fell to 4.7% it was still well above the average of EC countries, but inflation increased to 13.7% and military burden was increased to an average of 5.75% of GDP. In the early 1970s government controlled defence industries were established because of weapon embargoes during the seven year military government and because Greece wanted some independence in weapon procurement due to the increasing tensions with Turkey (Avramides, 1996). By the mid-1970s the communist threat had all but disappeared, but the Turkish invasion of Cyprus in 1974 led to a huge increase in military spending. It also saw the establishment of democracy in Greece. 4
Figure 1: Real Growth rate of Greek and Turkish GDP 20 Real growth of Greek GDP Real growth of Turkish GDP 15 10 5 0-5 -10-15 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 Growth rate (%) Years In the 1980s the Greek economy deteriorated. The average annual growth rate was 1.6%, compared to 2.3% for Europe as a whole, while inflation increased to 18.4%. Despite these persistent economic problems military expenditures were kept high, with 6.52% of GDP allocated to defence on average. In 1981, Greece became a full member of the European Community (EC) and in 1985 it officially declared a defence doctrine according to which Turkey was identified as the principal threat to its security. During the last decade there has been concern over the events in the Balkans as Yugoslavia began to break up. Greece was particularly upset by the creation of a state called Macedonia (as Macedonia is the name of the northern part of Greece) and the treatment of the Greek minority in Albania. Initially these events seemed to require additional security concerns for Greece, but since none of these countries possesses large military establishments Greek defence policy and military planning was not affected. 5
There was some slight economic improvement in the early 1990s, as Greece strove to achieve the required criteria for joining EMU. The years 1991-97 saw GDP grow at 1.9%, while military burden fell to 5.5% of GDP because of the tight macroeconomic policies. Inflation was also brought down to an annual average of 11.7% for the same period. The Greek economy remains weak, however, performing well below the EU s average. On top of this, the conflict with Turkey 3 remains unsolved - there are still disagreements 4 over Cyprus, over the continental self of the Aegean Sea and over the control of the airspace above it, and the mlitary burden remains high. Figure 2: MILEX in 1990 mn $ in Greece and Turkey TURME mn$ 1990 GRME mn$ 1990 mn $ 7000 6000 5000 4000 3000 2000 1000 0 1960 1963 1966 1969 1972 1975 1978 Years 1981 1984 1987 1990 1993 1996 3.2. Turkey Since the establishment of a Republic in 1923, Turkey has followed a policy of industrialisation within a closed economy. An inward-oriented policy (import substitution) supported by a high degree of protectionism and high government intervention. The 1960s saw a complicated domestic and external economic situation: high inflation and unemployment, a substantial gap in foreign accounts and social unrest. The military coup of May 1960 against the Democratic Menderes regime symbolised the particular role of the army as the guardians of Kemalists principles, of guided democracy, and of political, social and economic stability (Hershlag, 3 For a comprehensive view of the Greek-Turkish relations, see Constas, D. (1991), Duke (1989).. 4 From the Greek perspective, Turkey is characterised by imperialism and aims to change the status quo which was established by the treaties of Lausanne (1923), Montreux (1936) and Paris (1947).. The 1974 Turkish invasion of Cyprus and the up to date occupation of 40% of the island by Turkish troops is a clear proof of Turkey s ambitions and strategic aims. 6
1988). This social and political unrest and the economic difficulties brought about another military coup in 1970, but the oil crisis in 1973 and the stagflation and unemployment in the industrial countries had a damaging effect on the Turkish economy. In 1974 the Turkish invasion of Cyprus was accompanied by a dramatic increase in military expenditure, which was followed by a decline in GDP growth and an economic crisis in the late 1970s (See figure 1). This was followed by a political crisis and a third military coup took place in 1980. The new government moved to a more outward looking economic strategy which brought dramatic improvements in the Turkish economy. 1981 saw a clear upward trend, with GDP growth of 4.2%, an increase of 1.9% in GDP per capita and a sharp reduction in inflation. Despite the high economic growth during the last two decades, Turkey is still facing serious economic problems such as high inflation (66.1% in 1993) and unemployment (15% in 1993) (Sezgin, 1996). Turkey also faces a number of security concerns and continues to be a big defence spender. Apart from Greece, other external threats for Turkey are Iran, Iraq and Syria. Internally, Turkey is in conflict with Kurdish separists (since 1989) in the South-East of the country, to the extent that the army has been engaged in open warfare (Kollias, 1997). Figure 3: Military Burden for Greece, Turkey and Total NATO Greece Turkey Total NATO 8 7 6 5 % 4 3 2 1 0 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 years 7
4. Methodology and Empirical Analysis The approach adopted in this paper is to analyse the statistical causality of military burden and growth within a VAR framework, starting off by investigating the integration of the two series. If the two series are integrated of order one [I(1)], Granger causality must exist in at least one direction, at least in the I(0) variables (Engle and Granger, 1987). The Granger representation theorem demonstrates how to model cointegrated I(1) series in the form of a VAR model. The VAR can be constructed in terms of the levels of the data or in terms of their first differences with the addition of an error correction term (ECT) to capture the short run dynamics. These features are used as pre-test strategy to establish whether causality exists prior to identifying the direction via standard Granger-type tests. Specifically, the empirical analysis will rely on the following steps: The first step prior to applying Granger causality tests is to establish the integration properties of our series by using the Dickey-Fuller tests. If both series are found to be I(1), then there might exist a long-run relationship between them (that is, they may be cointegrated). The second step involves testing for cointegration via Johansen s maximum likelihood approach. If cointegration exists then either unidirectional or bidirectional Granger causality must exist in at least the I(0) variables. The third step involves the construction of standard Granger causality tests augmented with an appropriate error correction term derived from the long-run cointegrating relationship. Granger tests presume the use of stationary data so, they must be applied on I(0) series in order to derive valid inferences. Finally, we must allow for some dummy variables to capture significant changes due to specific incidents in each country. 4.1. Testing for unit roots The first step in order to establish the integration properties of the Greek and Turkish time series is to apply the Dickey-Fuller (1979) unit root tests. According to these, for a time series, X t, DF test is based on an autoregression that includes either only an intercept or both an intercept and a linear trend. For the first case, where only an intercept is included, we have the following autoregression: xt = β0 + α1 xt-1 + k i= 1 α i xt-i + εt (1) and for the second case (both an intercept and a linear trend is included), we have: 8
x t = β0 + β1 t + α1 x t-1 + k i= 1 α i x t-i + εt (2) In both cases the null hypothesis of a unit root is the same (H 0 : α 1 = 0) but the critical values differ. In order to select the maximum lag order 5 criteria which usually give quite contradictory results., k, we have to consult some information It is important to establish integration properties of involved time series. There are cases where we care about the presence of a unit root because we want to establish the nature and presence of trend in a time series (see Newbold and Vougas (1996)). However, in most cases, we need to detect the presence of a unit root to justify further cointegration analysis. Table 1 gives the DF tests for unit roots when only an intercept is included, for the following level series: the logarithm of Greek GDP (Y t ), Greek military burden (SM t ), the logarithm of Turkish GDP (TY t ) and Turkish military burden (TSMt). But before we conclude about the integration properties of our series we have to examine also DF tests with an intercept and a linear trend. These results are presented in Table 2. Tables 1 and 2 show that, with Dickey-Fuller tests of various lag orders, the logarithm of Greek real GDP is integrated of order one, and the same is true for the Greek military burden (the share of military expenditure in GDP). For Greece, there is persistence of military burden to either low levels, when the burden is low, or to persistently higher spending levels, when burden is high. Indeed, one observes that a considerably higher proportion of Greece s GDP is devoted to defence spending after 1974 (after the Turkish invasion of Cyprus) than before. (See Figure 3). In contrast, the results for Turkey show no evidence against the presence of a unit root in real GDP. However, there is evidence in favour of trend stationarity (non-integration) of Turkish military burden. For example, the Dickey-Fuller test based on the model AIC selects, that is an auxiliary autoregression with a first order lagged difference, rejects the unit root null at the 5 % significance level. Furthermore, this is only marginally not supported by the selection of SBC criterion, which is an autoregression with no lagged differences. Inspection of a time series plot of Turkish military burden in Figure 3 indicates, a rather non-integrated behaviour of Turkish military burden. In view of this, we are very reluctant to classify Turkish military burden as integrated or non-integrated with no further study. 5 Lagged differences of the series are added to whiten the error of the autoregressions 9
Instead, we estimate by exact maximum likelihood all ARMA(p,q) models, p + q 2, with a fitted intercept, in the absence of any obvious trend. Not surprisingly, the model selected by both AIC and SBC criteria is a MA(1) model with intercept: TSM = 0.047122 + E + 0.65648*E(-1) s = 0.0047571 (37.7803) (6.3666) with asymptotic t-ratios in parentheses. This means an autoregressive unit root is out of the question. Furthermore, we observe that Turkey may have maintained a higher than average military burden in the years following the Cyprus invasion. This is captured by a variable D7578, that takes value 1 in 1975, 1976, 1977, and 1978 and zero otherwise. Again, we fitting by exact maximum likelihood all ARMA(p,q) models, p + q 2, with a fitted intercept and D7578, leads to the selection of an MA(1) structure by both AIC and SBC 6. The model is: TSM = 0.045568 + 0.014871*D7578 + E + 0.56531*E(-1) s = 0.0032236 (55.6413) (6.9208) (5.1529) In view of the above specification search, we conclude that Turkish military burden is not integrated. Hence there is no possibility of cointegration between the logarithm of real GDP and military burden of Turkey, and further causal analysis from output to military burden and vice versa should rely on a VAR which includes real GDP growth (first difference of the logarithm of real GDP) and military burden, expressed as a fraction of GDP. To justify the conclusion that the rest of the series (Greek GDP, Greek defence burden and Turkish GDP) are I(1), we also present the DF tests for the differenced series in table 3. If the differenced series are proved to be stationary, then by induction, the level series are I(1). If on the other hand, the differenced series are still non-stationary, but the second differenced series are stationary that means that the levels series are I(2) and so on. For our first differenced series the hypothesis of a unit root is rejected at 95% level of significance. So, we can say with confidence that the levels series are I(1) apart from the Turkish military burden, which as shown 6 It should be noted that the AR part of the other models (not reported) were not significant most of the time. 10
above is considered to be I(0). 11
Table 1 The Dickey-Fuller regressions with an intercept 33 observations used in the estimation of all ADF regressions (1964-1996) Unit root tests for variable Yt DF -6.1440 78.5373 76.5373 75.0408 76.0338 ADF(1) -4.2924 78.5425 75.5425 73.2978 74.7872 ADF(2) -4.1798 78.8932 74.8932 71.9002 73.8862 ADF(3) -3.3270 79.0174 74.0174 70.2761 72.7585 95% critical value for the ADF statistic = -2.9528 Unit root tests for variable SM t DF -1.8475 128.6736 126.6736 125.1771 126.1701 ADF(1) -1.8618 128.7559 125.7559 123.5112 125.0007 ADF(2) -1.6754 128.9879 124.9879 121.9949 123.9809 ADF(3) -1.8951 129.7788 124.7785 121.0372 123.5197 95% critical value for the ADF statistic= -2.9528 Unit root tests for variable TY t DF -.82211 34.0512 32.0512 30.5547 31.5477 ADF(1) -.85809 34.3046 31.3046 29.0599 30.5494 ADF(2) -.83242 34.5668 30.5668 27.5738 29.5598 ADF(3) -.83619 34.7685 29.7685 26.0272 28.5097 95% critical value for the ADF statistic= -2.9528 Unit root tests for variable TSM t DF -2.6296-22.5505-24.5505-26.0470-25.0540 ADF(1) -3.0512-21.4131-24.4131-26.6579-25.1684 ADF(2) -2.7523-21.3450-25.3450-28.3380-26.3521 ADF(3) -2.0644-21.2251-26.2251-29.9664-27.4839 95% critical value for the ADF statistic= -2.9528 LL=Maximised log-likelihood, AIC= Akaine Information Criterion, SBC= Schwarz Bayesian Criterion, HQC= Hannan-Quinn Criterion
Table 2 The Dickey-Fuller regressions with an intercept and a linear trend 33 observations used in the estimation of all ADF regressions (1964-1996) Unit root tests for variable Yt DF -1.7747 78.5540 75.5540 73.3092 74.7987 ADF(1) -1.7486 78.5645 74.5645 71.5715 73.5575 ADF(2) -1.6676 79.0296 74.0296 70.2883 72.7708 ADF(3) -1.6397 79.0882 73.0882 68.5987 71.5776 95% critical value for the ADF statistic= -3.5514 Unit root tests for variable SM t DF -1.5007 128.6762 125.6762 123.4315 124.9209 ADF(1) -1.5376 128.7817 124.7817 121.7887 123.7747 ADF(2) -1.1960 128.9879 123.9879 120.2467 122.7291 ADF(3) -1.5541 129.8568 123.8568 119.3673 122.3462 95% critical value for the ADF statistic= -3.5514 Unit root tests for variable TY t DF -2.1024 35.9880 32.9880 30.7432 32.2327 ADF(1) -2.4973 37.1386 33.1386 30.1456 32.1315 ADF(2) -2.3354 37.1485 32.1485 28.4073 30.8897 ADF(3) -2.2122 37.1685 31.1685 26.6790 29.6579 95% critical value for the ADF statistic= -3.5514 Unit root tests for variable TSM t DF -2.6892-22.2474-25.2474-27.4921-26.0027 ADF(1) -3.1109-21.0593-25.0593-28.0524-26.0664 ADF(2) -2.8312-20.9583-25.9583-29.6995-27.2171 ADF(3) -2.1456-20.8462-26.8462-31.3357-28.3568 95% critical value for the ADF statistic= -3.5514 LL=Maximised log-likelihood, AIC= Akaine Information Criterion, SBC= Schwarz Bayesian Criterion, HQC= Hannan-Quinn Criterion
Table 3 The Dickey-Fuller regressions with an intercept 32 observations used in the estimation of all ADF regressions (1965-1996) Unit root tests for variable D Yt DF -3.3006 68.1697 66.1697 64.7040 65.6838 ADF(1) -2.3934 68.8647 65.8647 63.6661 65.1359 ADF(2) -1.4795 70.8430 66.8430 63.9115 65.8713 ADF(3) -1.2713 70.9933 65.9933 62.3290 64.7787 95% critical value for the ADF statistic = -2.9558 Unit root tests for variable DSMt DF -5.4916 122.9183 120.9183 119.4526 120.4325 ADF(1) -4.6316 123.4909 120.4909 118.2923 119.7621 ADF(2) -3.1333 123.6937 119.6937 116.7622 118.7220 ADF(3) -3.4694 124.8042 119.8042 116.1399 118.5896 95% critical value for the ADF statistic= -2.9558 Unit root tests for variable DTYt DF -4.8818 32.3871 30.3871 28.9213 29.9012 ADF(1) -4.0733 32.6506 29.6506 27.4520 28.9218 ADF(2) -3.5563 32.8279 28.8279 25.8964 27.8562 ADF(3) -3.5980 33.5746 28.5746 24.9102 27.3599 95% critical value for the ADF statistic= -2.9558 Unit root tests for variable DTSMt DF -5.4241-25.5554-27.5554-29.0211-28.0412 ADF(1) -4.6479-24.8992-27.8992-30.0978-28.6280 ADF(2) -4.7720-23.2893-27.2893-30.2208-28.2610 ADF(3) -4.2683-22.5880-27.5880-31.2523-28.8026 95% critical value for the ADF statistic= -2.9558 LL=Maximised log-likelihood, AIC= Akaine Information Criterion, SBC= Schwarz Bayesian Criterion, HQC= Hannan-Quinn Criterion 14
4.2. Testing for Cointegration Now that we have established the integration properties of the series, we can examine whether there is a long-run relationship between the Greek series, which are integrated of the same order [I(1)] as we showed in the previous section. In other words we can test for the existence of cointegration between the logarithm of real Greek GDP (Y) and the Greek military burden. If they are cointegrated, this will have important consequences when examining short-run Granger causality between them. We cannot do the same for the Turkish series as they are not integrated of the same order, Turkish GDP is I(1) but Turkish military burden is I(0). We test for cointegration using Johansen s (1988) cointegration method. We estimate a VAR(1) model with restricted intercept, as the two series are quite different in nature (See Figures 1 and 3). Table 5 gives the results of two tests for cointegration - the Likelihood Ratio test based on the maximum Eigenvalues of the stochastic matrix and one based on the Trace of the stochastic matrix. The null hypothesis of no cointegration (H0: r=0) is rejected by both tests in favour of the alternative (r=1), indicating that there is one cointegrating vector in the Greek series. Results from a VAR(2) model point to the same conclusions. Table 4 Cointegration with restricted intercepts and no trends in the VAR for Greece 36 observations from 1961 to 1996. Order of VAR = 1 List of variables included in the cointegrated vector: Y, SM, Intercept List of eigenvalues in descending order:.79953.15350 0.00 Cointegration LR test based on Maximal Eigenvalue of the stochastic matrix Null Alternative Statistic 95% critical value 90% critical value r = 0 r = 1 57.8560 15.8700 13.8100 r <= 1 r = 2 5.9991 9.1600 7.5300 Cointegration LR test based on Trace of the stochastic matrix Null Alternative Statistic 95% critical value 90% critical value r = 0 r >= 1 63.8551 20.1800 17.8800 r <= 1 r = 2 5.9991 9.1600 7.5300 The existence of one cointegrating vector between the series should be taken into consideration when we examine the short-run causality between the variables. On the basis of such results we can establish that Yt and SMt are causally related, since they are cointegrated, but to find the direction of the causality we have to apply the standard Granger tests augmented by the error 15
correction term 7 (ECT), derived from the long-run cointegrating relationships. The cointegrating regression for Greece is: Y t = 9.34 + 28.82 SM t + ε t (3) (36.64) (6.28) 4.3. Granger Causality tests Bearing these results in mind, the cointegrating series can be modelled as a VAR. The choice of the VAR s order (the lag-length) was made on the basis of minimising Akaine s final prediction error and for both Greece and Turkey a second order VAR model is accepted. For Greece we present three specifications of the VAR models. First, the standard Granger causality test (that is we ignore the existence of cointegration), second the Granger causality test augmented by the error correction term derived from the cointegrating regressions and finally, the above specifications are further augmented with some dummy variables that capture important changes due to specific events in the two countries. For Turkey, the specification that includes the ECT is omitted since no cointegrating vector was found for the Turkish series. The standard Granger causality test (when cointegration is not taken into account) assumes that the information for the prediction of the variables X t and Z t is contained only in the time-series data of these variables. The test involves estimating the following regressions: Xt = k i= 1 α i Zt-i + k β j j= 1 Χ t j + u1t (5) Z t = m λ i i= 1 Z t-i + m δ j j= 1 X t-j + u 2t (6) Equation 5 postulates that current X is related to past values of X itself as well as of Z and equation 6 postulates a similar behaviour for Z. Generally, if Z Granger causes X, then changes in Z should precede changes in X. Therefore, in a regression of X on other variables (including its own past values) if we include past or lagged values of Z and it significantly improves the prediction of X, then we can say that Z Granger causes X. A similar definition if X Granger causes Z. An important feature of the Granger causality tests is that they presume the use of 7 The ECT is nothing more than the lagged value of the estimated residuals from each of the above equations. 16
stationary data. Since both the Greek level variables are I(1), we have to use their first differences which are I(0). But for the Turkish series we use the difference in the logarithm of GDP only since the military burden variable is I(0). So, the VAR(2) for Greece (similarly for Turkey) when cointegration is ignored is: Y = a 0 + a 1 Y(-1) + a 2 Y(-2) + a 3 SM(-1) + a 4 SM(-2) + u (7) SM = β 0 + β 1 SM(-1) + β 2 SM(-2) + β 3 Y(-1) + β 4 Y(-2) + ε (8) The null hypothesis in equation 7 is that SM(-1) and SM(-2) do not Granger cause Y (H 0 : a 3 = a 4 = 0) and in equation 8 that Y(-1) and Y(-2) do not Granger cause SM (H 0: β3 = β4 = 0). The results for Greece and Turkey are presented in tables 8 and 9 respectively. To take into account the long-run relationship that exists between Greek growth and military burden, we have to include the ECT from equation 3 in the Granger causality tests. Following these, the VAR(2) model for Greece is: Y = a 0 + a 1 Y(-1) + a 2 Y(-2) + a 3 SM(-1) + a 4 SM(-2) + ECT(-1) + u (9) SM = β0 + β1 SM(-1) + β2 SM(-2) + β3 Y(-1) + β4 Y(-2) + ECT(-1) + ε (10) In equation 9, the null hypothesis is that SM(-1) and SM(-2) do not Granger cause Y, (H 0 : a 3 = a 4 = 0) and in equation 10 that Y(-1) and Y(-2) do not Granger cause SM (H 0: β 3 = β 4 = 0). Results for the direction of causality between growth and military burden for Greece using the equation augmented by the ECT for Greece are presented in Table 8. Finally, we consider the inclusion of some dummy variables in the VAR models for Greece and Turkey in order to capture some impulse shocks. For Greece we include D74 (taking the value zero till 1974 and 1 thereafter) to capture the effect of the Turkish invasion of Cyprus in 1974 on growth and military burden. From the results in Table 8, it is clear that the invasion had a significant negative effect on Greek growth. For Turkey a dummy was included for the years 1975-1978 (D7578) to capture the effect of the economic crisis that took place in the country. D74 was not found to be significant. 17
Table 5 Granger Causality : VAR(2) for Greece (1963-96) Dependent D Y Dependent DSM Eq. 7 Eq. 9 Eq. 11 Eq. 8 Eq. 10 Eq. 12 DY(-1) 0.44 (2.49)** 0.35 (2.21)** -0.02 (0.10) -0.02 (0.54) -0.009 (0.28) -0.04 (0.76) DY(-2) 0.28 (1.68) 0.18 (1.16) -0.08 (0.47) 0.03 (0.79) 0.04 (1.13) 0.02 (0.47) DSM(-1) 2.67 (2.75)*** 1.43 (1.45) 0.88 (0.96) 0.008 (0.04) 0.14 (0.70) 0.11 (0.50) DSM(-2) 0.02 (0.02) -0.68 (0.68) -0.25 (0.27) -0.07 (0.35) 0.005 (0.02) 0.03 (0.16) Constant 0.008 (1.04) 0.02 (2.19)** 0.07 (3.40)*** 0.001 (0.09) 0.001 (0.51) 0.003 (0.58) D74 ------- -0.04 (2.71)** -------- -0.003 (0.81) ECT(-1) -0.06 (2.77)*** -0.06 (3.33)*** 0.006 (1.46) 0.006 (1.36) R 2 0.43 0.55 0.65 0.04 0.11 0.14 SE 0.03 0.03 0.02 0.005 0.005 0.005 DW 1.87 1.92 1.99 2.05 2.10 2.08 Causality Tests LM X 2 (2)=7.09 [.029] X 2 (2)=2.89 [.236] X 2 (2)=1.20 [.549] X 2 (2)=0.75 [.687] X 2 (2)=1.56 [.458] LR X 2 (2)=7.95 X 2 (2)=3.02 X 2 (2)=1.22 X 2 (2)=0.76 X 2 (2)=1.60 [.019] [.221] [.543] [.684] [.450] F F(2,29)=3.8 F(2,28)=1.3 F(2,27)=0.5 F(2,29)=0.3 F(2,28)=0.7 [.034] [.289] [.615] [.723] [.518] t-ratios for regression results in parenthesis and probabilities for causality tests in brackets X 2 (2)=1.09 [.579] X 2 (2)=1.11 [.574] F(2,27)=0.4 [.643] 18
Table 6 Granger Causality: VAR(2) for Turkey (1963-96) Dependent D TY Dependent TSM Eq. 5 Eq. 7 Eq. 6 Eq. 8 DTY(-1) -0.05 (0.29) 0.02 (0.13) 1.20 (1.19) 0.57 (0.69) DTY(-2) -0.25 (1.52) -0.19 (1.21) -0.45 (0.44) -1.03 (1.24) TSM(-1) -0.008 (0.26) 0.02 (0.79) 0.81 (4.46)*** 0.52 (3.19)*** TSM(-2) -0.08 (2.46)** -0.08 (2.71)*** -0.25 (1.29) -0.23 (1.50) Intercept 0.47 (3.47)*** 0.34 (2.42)** 2.03 (2.44)** 3.28 (4.48)*** D7578 ------- -0.11 (2.36)** ------- 1.01 (4.13)*** R 2 0.29 0.41 0.44 0.65 SE 0.08 0.07 0.48 0.38 DW 2.21 2.21 1.99 1.62 Causality Tests LM X 2 (2)=9.21 [.010] X 2 (2)=7.60 [.022] X 2 (2)=1.76 [.414] X 2 (2)=2.29 [.318] LR X 2 (2)=10.74 [.005] X 2 (2)=8.61 [.014] X 2 (2)=1.81 [.404] X 2 (2)=2.37 [.305] F F(2,29)=5.38 [.010] F(2,28)=4.03 [.029] F(2,29)=0.79 [.462] F(2,28)=1.01 [.377] t-ratios for regression results in parenthesis and probabilities for causality tests in brackets The results for Greece in Table 8 show that if we were to take the usual approach to Granger causality testing, in column 1, we would find that military burden has a significant positive effect upon growth, with no significant effect from military burden to growth. Military spending would appear to Granger cause growth. This is, however, the result of a mispecification. It fails to allow for the long run properties of the data and the possible structural break caused by the shock of the Cyprus invasion. When either or both of these are considered the significant positive effect disappears and there is no causality from military burden to growth nor vice versa. For Turkey, there is no cointegration between growth and military burden, but there is evidence of Granger causality from the level of the military burden to the growth of output with the effect being negative, but not the other way around. 19
5. Conclusions This paper has provided an empirical analysis of the relation between defence spending and economic growth in Greece and Turkey over the period 1960-1996. It has systematically analysed the presence and direction of the causal relationship between defence and growth, paying attention to integration properties of the series and using vector autoregression methodology (VAR). In this way it has extended the methodology commonly employed. The result of this is to show that the standard approach can lead to spurious findings of Granger causality. This is the case for Greece, where a positive effect of military burden on growth is found for the standard test, but does not survive the introduction of long run information, nor a dummy to allow for the impact of the Cyprus invasion in 1974. In contrast, there is no cointegrating relationship between military burden and output in Turkey, but a significant negative Granger causal link from military burden to growth is found. For neither country is there any evidence of growth leading to changes in military burden. 20
References Atesoglou, S. & Mueller, M.J. (1990), Defence Spending and Economic Growth, Defence Economics, 2(1), pp.19-28 Athanassiou, E. & Kollias, C. (1998), Economic Aspects of the Greek-Turkish Conflict: A Preliminary Investigation of the Impact of Tension on Foreign Direct Investment, Paper presented at the International Conference on Defence Economics and Security in the Mediterranean and Sub-Saharan Countries, Universidade Technical de Lisboa, Lisbon, Portugal, 5 & 6 June 1998 Avramides, Christos (1997), Alternative Models of Greek Defence Expenditures, Defence and Peace Economics, Vol. 8, pp.145-187 Balfoussias, Athanassios & Stavrinos, Vassilios (1996), The Greek Military Sector and Macroeconomic Effects of Military Spending in Greece, The Peace Dividend, Ch.10, pp.191-213 Chan, Steve (1985), The Impact of Defence Spending on Economic Performance: A Survey of Evidence and Problems, Orbis, 29(3), pp.403-434 Chletsos, Michael & Kollias, Christos (1995), Defence Spending and Growth in Greece 1974-90: Some Preliminary Econometric Results, Applied Economics, 27, pp.883-890 Chowdhury, A. (1991), A Causal Analysis of Defence Spending and Economic Growth, Journal of Conflict Resolution, 35(1), pp.80-97 Clogg, R. (1991), Greek-Turkish relations in the post-1974 period in Constas (ed) Constas, D. (1991), The Greek-Turkish Conflict in the 1990s, London, MacMillan Deger, S. & Sen, S. (1983), Military Expenditures, Spin-off and Economic Development, Journal of Development Economics, 13, pp.68-83 Deger, Saadet (1986), Economic Development and Defence Expenditure, Economic Development and Cultural Change, pp.179-196 Dunne, P. & Vougas, D. (1998), Military Spending and Economic Growth: A Causal Analysis for South Africa, Paper presented at the International Conference on Defence Economics and Security in Mediterranean and Sub-Saharan Countries, Universidade Technical of Lisboa, Lisbon, Portugal, 5 & 6 June 1988 Dunne, P. & Nikolaidou, E. (1998), Military Spending and Economic Growth: A Case Study of Greece, 1960-1996, Paper presented at the International Conference on Defence Economics and Security in Mediterranean and Sub-Saharan Countries, Universidade Technical of Lisboa, Lisbon, Portugal, 5 & 6 June 1988 Dunne, Paul (1996), Economic Effects of ME in Developing Countries: A Survey, in Gleditsch et al The Peace Dividend, Elsevier, Ch. 23, pp.439-464 Dunne, Paul & Mohammed, Nadir (1995), Military Spending in Sub-Saharan Africa: Some Evidence for 1967-85, Journal of Peace Research, 32, no. 3, pp.331-343 21
Georgiou, G., Kapopoulos, P. and Lazaretou, S. (1996), Modelling Greek-Turkish Rivalry: An Empirical Investigation of Defence Spending Dynamics, Journal of Peace Research, Vol.33, no.2, pp.229-239 Georgiou, George (1990), Is There an Arms-Race between Greece and Turkey? Some Preliminary Econometric Results, Cyprus Journal of Economics, Vol.3, no.1, pp.58-73 Hartley, Keith & Sandler, Todd (1995), Handbook of Defence Economics, Vol.1, Elsevier Hershlag, Z. Y. (1988), The Contemporery Turkish Economy, Routledge Hess, P. (1989) The Military Burden, Economic Growth and the human suffering index: evidence from the LDCs, Cambridge Journal of Economics, 13 (4), 497-515 Ifestos, P. and Platias, A. (1992), Greek Deterrence Strategy, Papazisis Publications, Athens Joerding, W. (1986) Economic Growth and Defence spending: Granger causality, Journal of Development Economics, 21, 35-40 Kapopoulos, P. & Lazaretou, S. (1993), Modelling the Demand for Greek Defence Expenditure: An Error Correction Approach, Cyprus Journal of Economics, Vol.6, no.1, pp.73-85 Khanna, J.; Sandler, T.; Shimizu, H. (1998), Sharing the Financial Burden for U.N. and NATO Peacekeeping, 1976-1996, Journal of Conflict Resolution, Vol.42, no.2, pp.176-195 Kollias, Christos and Makrydakis, S. (1997) Is there a Greek-Turkish arms-race? Evidence from cointegration and causality tests, Defence and Peace Economics, Vol.8, pp.355-379 Kollias, Christos (1997), Defence Spending and Growth in Turkey 1954-1993: A Causal Analysis, Defence and Peace Economics, 8, pp.189-204 Kollias, Christos (1996), The Greek-Turkish Conflict and Greek Military Expenditure, 1960-92, Journal of Peace Research, Vol.33, no.2, pp.217-228 Larrabee, S. (1992), Instability and Change in the Balkans, Survival, Vol.34, pp.31-49 Lim, D. (1983) Another look at Growth and Defence in Developed countries, Economic Development and Cultural Change, 31 (2), 377-384 Looney, Robert & Frederiksen, P.C. (1986), Defence Expenditures, External Public Debt and Growth in Developing Countries, Journal of Peace Research, Vol.23, no. 4, pp. Macnair, E.; Murdoch, J.; Pi, C.R.; Sandler, T. (1995), Growth and Defence: Pooled Estimates for the NATO Alliance, 1951-1988, Southern Economic Journal, Vol.61, no.3, pp.846-860 Majeski, S. (1985), Expectations and Arms-Races, American Journal of Political Science, Vol.29, pp.217-245 Majeski, S. and Jones, D. (1981), Arms-Race Modelling: Causality Analysis and Model Specification, Journal of Conflict Resolution, Vol.25, pp.259-288 22
Murdoch, J.; Pi, C-R.; Sandler, T. (1997), The Impact of Defence and Non-Defence Public Spending on Growth in Asia and Latin America, Defence and Peace Economics, Vol.8, pp.205-224 Newbold, P. & Vougas, D. Drift in the Relative Price of Primary Commodities: A Case where we care about Unit Roots, Applied Economics, Vol.28, 653-661. Nikolaidou, Eftychia (1998), Military Spending and Economic Growth in Greece, 1950-1996), Paper presented at the International Conference on The Economics of Military Expenditure in Developing and Emerging Economies, Middlesex University Business School, London, U.K., 13 & 14 March 1998 Oxley, L. (1993), Cointegration, Causality and Export-led Growth in Portugal, 1865-1985, Economics Letters, 43, pp. 163-166 Platias, A. (1991), Greece s Strategic Doctrine: In Search of Autonomy and Deterrence, in Constas (ed) Sandler, Todd & Hartley, Keith (1995), The Economics of Defence, Cambridge University Press Scheetz, Thomas (1991), The Macroeconomic Impact of Defence Expenditures: Some Econometric Evidence for Argentina, Chile, Paraguay and Peru, Defence Economics, 3(1), pp.65-81 Sezgin, Selami (1998), An Empirical Analysis of Defence-Growth Relationships with a Demand and Supply Side Multi Equation Model: Time Series Evidence from Turkey, Paper presented at the International Conference on The Economics of Military Expenditure in Developing and Emerging Economies, Middlesex University Business School, London, U.K., 13 & 14 March 1998 Sezgin, Selami (1996), Country Survey X: Defence Spending in Turkey, Defence and Peace Economics, 8, pp.381-409 Smith, Ron (1980), ME and Investment in OECD Countries, 1954-1973, Journal of Comparative Economics, 4, pp.19-32 Summers, R. & Heston, A. (1988), A New Set of International Comparisons of Real Product and Price Levels Estimates for 130 Countries, 1950-1985, Review of Income and Wealth, Vol.34, no.1, pp.1-25 23