Kapopoulos, Sophia Lazaretou Source: Journal of Peace Research, Vol. 33, No. 2 (May, 1996), pp. 229239 Published by: Sage Publications, Ltd. Stable URL: http://www.jstor.org/stable/425439 Accessed: 29/05/2009 12:22
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, noncommercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=sageltd. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a notforprofit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Ltd. is collaborating with JSTOR to digitize, preserve and extend access to Journal of Peace Research.
http://www.jstor.org
1. Introduction debate.'1 But, since the purpose of this This article uses three models to test the article is the statistical study of the military widely held view that there is a systematic behaviour of Greece and Turkey in relation arms race between Greece and Turkey. The to each other, we shall refrain from entering modelling is based on the work of McGuire the debate. Instead, we adopt a broad and (1977), Desai & Blake (1981) and a vector not too controversial definition of the arms autoregression specification. Particular race as a 'dynamic process of interaction beattention is paid to appropriate diagnostic tween countries in their acquisitions of tests, the longrun values and Granger weapons' (Intriligator & Brito, 1990, p. 59). An appropriate measure of the arms race causality. The question of what actually constitutes phenomenon has also been the subject of an arms race between rival states and the debate in the international relations literarelated issue of how militaristic such states ture. Some researchers advocate the use of are have been the subjects of much scholarly narrow measures of military capability, such as the number of specific types of warheads, while others use broader measures, such as * We are grateful to Ron Smith, two anonymous refer the TNT equivalent of the weapons stock ees and the Editor of JPR for helpful comments and (Wiberg, 1990a, b deals with some of these suggestions. We also thank Evamaria LooseWeintraub and Christer Berggren at the Stockholm International issues). In this article, we use an economic Peace Research Institute for providing us with military measure, namely the share of military expenditure data. All errors are ours. Any views in this expenditure in GDP, thus circumventing article are those of the authors and they should in no about the quality and/ part be attributed to the Central Bank of Cyprus or technical arguments or quantity of the armaments stock held by Alpha Credit Bank. This article may be identified the two countries under consideration. The under JEL classification H56, C32.
230
advantage of such an economic measure is that it allows, irrespective of the weapons systems, for a broad measure of belligerence. Furthermore, 'shares have the advantage that they take account of incomes, are insensitive to inflation and exchange rates, and can be used as a signal or a measure of commitment to defence' (Smith, 1989, p.
352).2
The rest of the article is organized as follows. Some stylized facts in the form of a historical description of the data are presented in section 2. The models to be tested are set out in section 3 and several other approaches are discussed. The empirical results are presented in section 4. Finally, Section 5 concludes the article with a brief review of the main findings and a discussion of the avenues for further research. 2. GreekTurkish Relations in Historical Perspective Figure 1 plots Greek and Turkish military expenditure as a percentage of GDP for the period 196090. After an increase in 1960 and 1961, the Turkish share of military expenditure (T) followed a downward trend up to and including 1974. The trend for Greece was also downward, but for a shorter period. Defence spending behaviour largely changed in the post1974 period (i.e. the period following Turkey's invasion of Cyprus). Both countries experienced sharp increases in their shares. As can be seen from Table 1, the increase was sharper for
Figure 1. MilitaryExpenditureas a Share of GDP,
196090
U.UB ; ;
0.07
??.
s4;
0.06
.
.... .................
0.05
.
~

~
W
rl\
jl
r
.
VE ..f
0.04* 
t ;,
.
,I..,. 
:
0.03
60 62 64 66 ' 68 70 72 74 76 78 80 82 84 86 88 90
,,
. j
j,
a,
j . ,.,,
,
i , I
, , j
Greece. More importantly, in the post1980 period, the Greek share (G) stabilized at higher levels. Indeed, it averaged 6.4% in the 1980s compared with 5.7% in the 1970s, while the standard deviation fell from 1.9% in the 1970s to 0.5% in the 1980s. Moreover, the average rate of growth of real defence spending declined from a high of 6.4% to a low of 1.7% during the same period. What is more interesting is that the crosscorrelation coefficient for the 1970s has a positive large value and is strongly significant. By contrast, in the 1960s and 1980s the coefficient has a negative sign but does not differ significantly from zero. This may be an indication of the existence of an arms race between the two countries, mainly in the 1970s. The period from 1960 to 1970 was a potentially volatile one for GreekTurkish relations. In 1963, three years after gaining independence from Britain, Cyprus experienced its first bout of intercommunal strife. This was followed by a second round in 1964 as a result of which Greece sent troops to the island in 1965 on the pretext of maintaining peace between the Greek and Turkish speaking communities. In April 1967, a rightwing military junta toppled the civilian government in Greece and in November of the same year intercommunal strife flared in Cyprus for the third time since its independence. Turkey threatened to intervene militarily and, as a result, Greece agreed to withdraw its troops.3 The period from 1970 to 1980 was eventful in terms of political and military developments. The five most important events were as follows: (1) the dispute between Greece and Turkey in March 1974 over oil exploration in the Aegean, which almost resulted in war in 1976; (2) the Athens inspired military coup in Cyprus in July 1974, resulting in the overthrow of President Archbishop Makarios; (3) Turkey's invasion of Cyprus during the same month, which was prompted by the coup; (4) following Turkey's invasion, the fall of the Greek and GreekCypriot juntas, and finally (5) the military coup in Turkey in 1980.4 These events would lead one to expect an increase in military expenditure. However,
Modelling GreekTurkish Rivalry Table I. Key Statistics of Defence Spending: Greece and Turkey, 196090 196070 Defence Spending As percentage of GDP Average rate of growth (real) Greece Turkey Greece Turkey L 4.20 4.77 6.86 3.93 u 0.53 0.19 0.38 7.98 0.19 3.79 4.99 18.32 4.83 6.36 0.82 17.05 0.89* 6.55 10.74 p p. 5.71 197180 r 1.19 0.85* 4.57 1.67 0.42 9.92 p u. 6.44 198190 r 0.51
231
p 0.21
0.05
J, is the mean value of the variables in question; o is the standard deviation (computed as percent per year) and p is the contemporaneous crosscorrelation coefficient. The GDP deflator for both nations was used to make defence spending 'real' with 1985 as a base year. Sources: SIPRI, Yearbook, various years and OECD, National Accounts. * Statistically significant at the 5% confidence levels.
the timing of these increases does not conform entirely to expectations (see Figure 1). For example, although Turkey invaded Cyprus in 1974, defence spending during the twoyear period prior to 1974 declined. Surprisingly, the Turkish share in 1974 was the lowest during the period 196074. We would also expect the Greek share to have increased sharply following the events of 1974. As can be seen from the Figure, the increase actually began in 1973. What is interesting about the period following the 1974 invasion of Cyprus is that although the pattern of Turkish and Greek shares is similar, there appears to be a ratchet effect in the case of Greece. Whereas the Turkish share post1975 declined to pre1974 levels, the Greek share remained at over two percentage points above the levels recorded during the pre1974 period. In the 1980s, the two countries were as much preoccupied with other facets of their external relations as with each other, though the potential for aggression between the two remained as great as ever. The year 1985 saw the restoration of a civilian government in Turkey and a commitment by the authorities to embark on a modernization programme with a view to eventual membership of the European Community. A similar commitment was undertaken by Greece which joined the European Community in 1981. Although Cyprus remained a thorn in GreekTurkish relations, it was the Aegean, again, which almost brought the
two countries into open conflict. The dispute over oil exploration flared in March 1987 resulting in a state of military alert in both countries. January 1988 saw the first Davos meeting between the then Greek and Turkish Prime Ministers. The end of the 1980s was also the beginning of a more cordial relationship between Turkey and the Soviet Union, which had previously been one of suspicion and at times hostile. We would thus have expected defence spending as a percentage of GDP for both countries to have, at most, declined or at the very least remained stable during the 1980s. Following the military coup in Turkey in 1980, the Turkish share increased from 4.3% in 1980 to 5.2% in 1982, but thereafter displayed a downward trend, reaching a low of 3.8% in 1988 and 3.9% in 1990. In contrast, the share of Greek military expenditure remained well above 6%, reaching a peak of 7.1% in 1984. Although it then began to decline, it still remained two percentage points above the Turkish share. In summary, the historical rivalry between Greece and Turkey over the past three decades includes three issues of friction (Clogg, 1991; Constas, 1991): (1) the conflicting claims of sovereignty over the continental shelf, sea and air regions of the Aegean, (2) the geostrategic role of Cyprus, and (3) the existence of Greek and Turkish minorities with ethnological and religious differences. The stylized facts reveal that the period 196090 was one of volatility in Greek and Turkish military expenditures.
232
Although they showed similar trends, the turning points do not appear to follow each other. Therefore, it is not clear whether there was a systematic arms race between the two countries, even though political events would lead us to have expected one. In as much as there is any evidence of a race, it seems that Greece was more sensitive to Turkey's actions than vice versa. As Tartter (1988) states: 'Although Greece tended to view Turkey as the principal threat to its security, Turkish troop dispositions were not oriented toward possible conflict with Greece. ... Its (Turkey's) quarrels
always borne out by actual behaviour, namely the proportion of resources allocated to armaments and their maintenance. For this reason, we opt for a descriptive model.5 Within the descriptive methodology there are several alternatives. Two important models are those of McGuire (1977) and Desai & Blake (1981), both of which are based on the wellknown Richardson theSiS.6
with Greece had little direct impact on the overall Turkish military effort except for the financial strain of supporting Turkey's involvement in Cyprus' (p. 317).
Assume two nations (or groups of nations), X and Y, whose defence expenditures are represented by x and y, respectively. We can write:
(1) (2)
&2
+ a3 + f3 coef
ficients, and a3 and P3 are 'grievance' (or 3. Modelling GreekTurkish Rivalry In modelling the rivalry between Greece sociopolitical terms). Equations (1) and (2) and Turkey, we are faced with several can be used to analyse the consequences of choices. We can adopt a simple Richardson various scenarios. For example, if a1 and 13i specification where it is assumed that there predominate, then the outcome is an arms is a linear relationship between Greek and race which may result in war. If a1 and 13i Turkish military expenditure. Alternatively, were the only terms, then both x and y we can depart from the Richardson would tend to the same infinity which, if approach. Intriligator & Brito (1990) positive, may be interpreted as war. In identify three other types of model. These order for (1) and (2) to be stable, (a2f32) > are the stock adjustment model, the Brito (a431) assuming that the 'grievance' terms a3 and 133are positive. If (a212) < (a1131), model and the differential game model. These approaches are not mutually ex there could be an equilibrium provided that clusive. For example, the differential game the 'grievance' terms were negative. In this model 'is a variant of the Brito model that case, the system would be unstable. There uses theoretic methods to solve a dynamic would be a downward spiral of military optimization problem' (Intriligator & Brito, expenditure toward zero levels for slight 1990, p. 64). Thus, the four approaches can movements below the equilibrium and an be reduced to two broad methodologies: upward spiral of military expenditure for 'descriptive' (Richardsonian specifications) slight movements above the equilibrium. and 'normative' (Britotype specifications). Richardson concluded that increasing straThe advantage of normative models, from tegic budgets would tend to increase the the neoclassical economist's point of view, is probability of war. Equations (1) and (2) are the startingthat they allow the researcher to incorporate the assumption of rationality within the con point for McGuire's seminal quantitative text of constrained optimization. However, study of the arms race between the USA it is not always clear what a country's objec and the Soviet Union. Letting A and R reptive function is with regard to its military resent the strategic strength of the USA and outlay programme, especially in the context USSR, respectively, and assuming that straof rivalry. As we saw in the previous tegic decisions respond with a lag, McGuire section, what we expect of two rivals, given writes (1) and (2) as: certain political and military events, is not (3) A,= a + f3A,_1 + y(R,1 + R,2)
where it is hypothesized, as did Richardson, that a, 13,y, 6, E and < are all positive constants. Using regression estimation techniques, McGuire concludes that (3) and (4) are locally stable. The dissatisfaction with McGuire's conclusion of a stable equilibrium led Desai & Blake (1981) to estimate an alternative specification based on a partial adjustment process,
(5) A21nA, = ao + a1A2lnR, + + a3(lnAlnR),_2 a2(AlnAzlnR),_1
difference equations that allow for lagged responses.7 An increasingly popular specification in the applied econometrics literature is that of the vector autoregression (VAR) specification. The main advantage of VAR models is that they 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. The VAR specification represented by equations (7) and (8) is in error correction form:8 (7) AlnT, = a0 + l1JlnG,_1+ yllnT,_1
+ a1ln7T_I + 61+nGt_1 + u1,
(6)
P2(AlnAAlnR)_1 +? 33(lnAlnR)_2 (8) AlnG, = ao + bl4lnG,_1 + cAllnT,_1 + allnG,i + 8lnT,_1 + u2 where A and R represent US and Soviet where T is Turkey's share of military expenwarheads, respectively. Equations (5) and (6) are based on the diture in GDP, G is Greece's share of militheory of a feedback control mechanism tary expenditure in GDP and u1, and u2tare applied to stabilization policy. The level white noise disturbance terms (Mills, 1991, terms in equations (5) and (6) represent the ch. 5). The A terms are the shortrun re'integral' control, which assumes that each sponse terms, whereas the level terms repof the two superpowers has a desired level resent the longrun response. of weapons stock in relation to each others', As part of the specification search to find and each country acts so as to correct for an appropriate functional form, two other any discrepancy between desired and actual models will also be estimated. The first stock. The terms AlnA and AlnR represent model is the unrestricted version of the 'proportional' control component, since McGuire's specification and is shown by it is assumed that each of the two countries equations (9) and (10), will be concerned with current changes to (9) Tt = a0 + OIT1 + 02G,_1 + a3G2 each other's weapons relative to their own. + U' Assuming also that each country is con+ cerned about the other's rate of weapons (10) G, = P) + 3Gt_i ?32>1 + 13Tt2 + U'2r acquisition, the A2 terms represent 'derivative' control. The two restrictions giving rise to McGuire's Several features of Richardson's original specification are a2 = a3 and 132= /33. model, and by implication later derivatives The second model is the unrestricted based on his work, have been the subject of version of Desai & Blake's specification and critical scrutiny (see Allison, 1974; Intriliga is represented by equations (11) and (12): tor, 1975; Rattinger, 1975; Wibert, 1990a). (11) lnT, = + a1lnG, + a2lnG,_1+ a3lnG,2 ao However, there can be little disagreement + a4lnT/_ + a5lnT,2 + u"1, about the seminal nature of Richardson's (12) lnG, = b0 + b1lnT + b2lnT_i + b3lnT/_2 work and the lasting influence it has had on + b4lnG,_i + bSInG,2 + U'2t even the most sophisticated models of the arms race. It is within the context of a where the restrictions in equation (11) are a2 Richardsontype framework that our model = (2a1a2), a3 = (a1 + a2 a3), a4 = (aT2 ling of GreekTurkish relations is under + 2) and a5 = (a3a21). In equation (12) taken. the restrictions are b2 = (2X31:2), b3 = Assume two countries, Greece and (X3i + 13233), b4 = (132 + 2) and b5 =(X Turkey, which are rivals. We need two 12  1).
234
George M. Georgiou et al. Table II. OLS Estimates and Diagnostic Tests for the VAR Model Explanatory Variables Intercept AlnG, l AlnT, 1 InT, , lnGC, SER R2 Diagnostic Tests SC (F ^3) FF (F1 23) N (X2i) H (Fi27) Chow test (Fs,10) Predictive Failure (F3,21) Dependent Variables
4. Empirical Results The three models were estimated using OLS for the period 196090. Table II presents the results for the VAR model. In the light of our discussion in Section 2, the results seem surprising. Although there is some evidence of a relationship between Turkish and Greek military expenditures, it is Turkey which appears to be influenced by Greece rather than vice versa. Specifically, the rate of change of Turkey's share of military expenditure is positively related to the rate of change of the lagged Greek share and inversely related to its own lagged level, whereas none of the explanatory variables for the Greek equation are significant. Both equations are structurally stable and pass the predictive failure test.9 However, the misspecification tests indicate that they are poorly specified and Turkey's equation suffers from heteroskedasticity (Spanos, 1986). Furthermore, both equations produce a low R2. The unrestricted McGuire model provides a more acceptable representation. Table III presents the OLS estimates. Apart from the FF test for Turkey's specification, which just fails at the 5% level, the misspecification tests indicate normally distributed white noise errors. The two equations are structurally stable and predict adequately,'0 and although there is some deterioration in the standard errors they produce a higher R2. Whereas all the explanatory variables are significant in the specification for Turkey, including the two Greek variables (G,1 and G,_2), only the lagged value of Greek military expenditure is significant in the equation for Greece. The negative signs of G_2 and T,2, which indicate an inverse relationship between Turkish and Greek military outlays, are not what we would expect. Table IV reports the results for the unrestricted version of Desai and Blake's model. Apart from evidence of some serial correlation in Turkey's equation, both equations produce acceptable representations in terms of the diagnostic tests.11 However, in the equation for Greece only the one period lagged dependent variable is statistically significant. The rest of the variables do not
AlnT
0.806 (3.26) 0.355 (2.46) 0.286 (1.68) 0.587 (3.55) 0.051 (0.78) 0.087 0.470 2.531 6.376 0.636 10.136 1.419 2.141
ZUnG
0.292 (0.87) 0.238 (1.21) 0.062 (0.27) 0.046 (0.20) 0.127 (1.43) 0.119 0.126 0.058 10.063 0.470 3.630 2.238 0.355
Figures in parentheses are tstatistics which have been rounded to two decimal places. SER is the standard error of the regression; SC is the F version of a Lagrange Multiplier test for firstorder serial correlation; FF is Ramsey's RESET test for functional form, which involves regressing the residuals on the square of the fitted values; N is a skewnesskurtosis test for normality and H is the F version of a test for heteroskedasticity, which involves regressing the squared residuals on squared fitted values. The Chow test tests for the stability of the regression coefficients and the Predictive Failure test is Chow's second test of adequacy of predictions.
differ significantly from zero. In the equation for Turkey, lnT,_i and lnG,_2 are significant at the 5% level, whereas lnG,_1 is significant at the 10% level. The positive sign of lnGt_i is consistent with the hypothesis of an arms race, while the negative sign of lnGt_2contradicts such a hypothesis. Equations (7) and (8) are the reduced form of a more general model written as:
(13) (14) + + + AlnTt = cao) 30AlnG, 4llnG,.i ylAlnT,_i + allnT,_ + 4 lnG,_1 + ul1 = zAlInGC ao + boAlnTf + b1AllnG_1 + clAlnT_1 + dlnT,_1 + allnG,_1 + u2
Modelling GreekTurkish Rivalry Table III. OLS Estimates and Diagnostic Tests for the Unrestricted McGuire Model Explanatory Variables Intercept G,
l
235
Table IV. OLS Estimates and Diagnostic Tests for the Unrestricted Desai & Blake Model Explanatory Variables Intercept InT
lnT,_l
Dependent Variables T G / G 1.913 (2.91) 0.422 (3.22) 0.540 (3.79) 0.391 (3.04) 0.416 0.537 0.159 4.960 2.945 0.150 2.222 1.040 1.069 (0.96) 0.883 (9.53) 0.053 (0.20)
Dependent Variables InT 0.816 (3.18) 0.699 (4.00) 0.288 (1.65) 0.032 (0.21) 0.441 (1.96) 0.362 (2.39) 0.089 0.538 3.048 0.837 1.115 0.911 1.269 2.023 Inc 0.340 (0.82) 0.059 (0.21) 0.058 (0.19) 0.079 ( 0.32) 1.135 (4.87) 0.259 (1.15) 0.122 0.826 0.220 1.189 0.523 0.917 2.051 0.325
T/_i G,2 Tr_2 SER R2 Diagnostic Tests SC (F1,24) FF(FI,24) N (X2i) H (F1,27) Chow test (F4,21) Predictive Failure (F322)
InT,_2 0.131 (0.51) 0.633 0.799 0.366 1.965 1.811 0.321 1.866 0.433 InG
lnG,_l
InGC, SER R2 Diagnostic Tests SC (F1.4) FF(F1,24) N (X2i) H (F1,27) Chow test (F4,21) Predictive Failure (F3.22)
(12). The problem with (11), (12), (13) and (14) is that they are not identified. For example, AlnG, is endogenous in (13) and AlnT, is endogenous in (14), but there are no exogenous variables excluded from one but not the other. 12 One way of overcoming this problem is to use the following simple Richardson model:
(15) (16) AlnTt = a1 j31A1nG, + yllnT,_1 AlnG, = a2 + 2ATlnT +? y2lnG,_l
The model is identified because G,_1 does not appear in (15) and T,_1does not appear in (16). The empirical results for the identified model (OLS and 2SLS estimates) are presented in Table V.13 The shortrun response variables are insignificant for both equations. In the equation for Turkey, the variable lnT,_1 is significant but has a negative sign, whereas the lagged dependent variable in the equation for Greece does not significantly differ from zero and has a negative sign too. When the model is reestimated by the 2SLS method, the lagged dependent variables in both equations have the correct sign but they are not significant.
Moreover, the equation for Turkey fails the tests for functional form and normality, while the equation for Greece satisfies the diagnostics. Both equations, however, produce a very low R2. Summarizing, we note that the OLS estimates for all models do not provide strong evidence of a systematic rivalry between Greece and Turkey. Although some of the results are statistically acceptable, they are, in the case of the McGuire and Desai & Blake models, inconsistent with a priori expectations, especially with regard to the signs of the explanatory variables. Of the three models, McGuire's specification performs best but only for Turkey. Even here caution is needed, since the negative sign of Gt2 has no theoretical explanation and the equation fails the test for functional form, though only just. The poor results are further reinforced by the longrun relationships shown in Table VI. The low values of the longrun elastici
236
George M. Georgiou et al. Table VI. Longrun Relationships T Vector Autoregression Model Unrestricted McGuire Model Unrestricted Desai & Blake Model T G T G T C 0.087 0.672 0.071 0.646 0.081 G 0.360
Table V. OLS and 2SLS Estimates and Diagnostic Tests for the Identified Richardson Model Explanatory Variables Intercept Dependent Variables AlnT 0.618 (2.44) AlnG 0.190 (1.42) (1.04)
2.054
0.317
AlnG
AlnT
InG, l
The values for the vector autoregression and Desai & Blake models are calculated using the coefficients of variables expressed in logs; hence the longrun values are longrun elasticities.
1
larly well suited to investigating causal linkages, we take as our startingpoint the model represented by equations (7) and (8). Table VII presents the Fvalues. It is found that causality runs one way only, from AlnC to AlnT but not from AlnT to AlnG. Evidently, these results do not support the basic hypothesis of an arms race process according to which current defence expenditure decisions are based on the opponent's past behaviour. 5. Concluding Remarks The purpose of this article has been to test for the existence of a systematic rivalry between Greece and Turkey. The modelling is based on the work of McGuire (1977), Desai & Blake (1981) and a vector autoregression specification. Such rivalry is well documented by political commentators and historians. However, the statistical evidence
Table VII. Tests of Granger Causality Fvalues AnT Null hypothesis AlnT is not Granger caused by AlnG Null hypothesis AlnG is not Granger caused by AlnT 3.912* (2,24) 0.040 (2,24) AlnG
0.105
(1.03) InT, l 0.406 (2.49) 0.672 (1.17) 0.103 [0.149] 0.192 [0.079] 1.438 4.993 11.090 1.121 0.348 1.285
SER R2 Diagnostic Tests SC (F,2() FF (F1,()) N (X2) H (F1,2X) Chow test (F3,24) Predictive Failure (F3.24)
0.116 [0.113] 0.068 [0.084] 1.351 0.141 0.942 0.002 1.040 0.422
See notes at the foot of Table II. 2SLS estimates are presented in square brackets. As instruments, we choose a constant and the shares of military spending for Greece and Turkey in the current period and with two lags.
ties for the VAR and Desai & Blake models indicate that the two countries are insensitive to each other's military outlays. Furthermore, the negative sign derived from the equations for Greece indicates that in the longrun the response of Greek procurement to changes in Turkey's outlays is likely to be an inverse one. Discussion so far has centred on the significance and sign of the OLS estimates together with consideration of the concomitant diagnostic tests. These estimates, however, do not indicate the direction of the causality. Since the VAR model is particu
Ftest for the addition of (i) InG, , and AlnG,_1; (ii) lnT,7_ and AlnT,_,. Degrees of freedom are presented in parentheses below the Fvalues. * Statistically significant at the 5% confidence level.
237
provides little corroboration of the widely held view that there is an arms race between Greece and Turkey.14 Such results should not be entirely surprising. First, we have assumed that rivalry is synonymous with an arms race. Yet, rivalry can take many other forms, such as periodic exchanges of bellicose rhetoric; economic, political and diplomatic manoeuvring; lobbying within existing alliances; political, historical and cultural propaganda. One or all of these may be present without a concomitant commitment of financial resources to military outlays. Second, Greece and Turkey are both members of NATO which, prior to the collapse of the Warsaw Pact, was primarily concerned with the containment of communism. Hence, despite deeprooted historical animosity, Greece and Turkey tended to view their East European and Soviet neighbours as, ultimately, their common threat.'5 This would have been reflected in their military procurement decisions. Specifically, the participation of both countries in the NATO Alliance may cause a positive crosscorrelation of their shares of GDP that are spent for defence, which is not attributed to the existence of an arms race between these two countries. If both countries exhibit the same behaviour within the Alliance, either both freeride or both follow, the time pattern of Greek and Turkish shares will be similar.16 Third, as Anderton (1989) argues, the share of military expenditure to GDP does not adequately depict the military capabilities of rival states. One way to probe the generality of the empirical findings is to use indices of military capabilities that can capture both the quantitative and qualitative dimensions of arms races. Furthermore, the arms race models provide an analytical approach by which to derive a demand function for the defence spending of a state visavis the rival state. Given that the statistical evidence provides little support on the widely held view that there is an arms race between Greece and Turkey, there are possibly other explanations for their military expenditures. Therefore, on the empirical side, the most important issue for future research of the
determinants of military expenditures in both countries would be the consideration of models that incorporate the impact of various strategic, political and economic factors that may be important determinants of defence spending. NOTES
1. On the issue of arms races, see Gleditsch (1990) and references therein. On the more general question of militarism, see Geyer (1990), Klare (1980), Regehr (1980), Skjelsbtek (1980), Thee (1980). 2. Smith (1989) actually tests for the statistical significance of the share of military expenditure in output and finds it to be the appropriate defence variable. The use of military expenditure is not confined to Smith only. See, for example, Edelstein (1990) and the references cited therein. 3. For further discussion of the role of Greece and Turkey in Cyprus during this period, see Hitchens (1984). 4. For further details of some of these events, see Hitchens (1984) and Seeley (1984). 5. An alternative to the normative/descriptive dichotomy is discussed by Wiberg (1990a) and Smoker (1990). 6. See Intriligator & Brito (1990), Rapoport (1957), Richardson (1939, 1960), Smoker (1990), Wiberg (1990a). For an early statement of the actionreaction thesis, see von Clausewitz (1968). 7. As Wiberg (1990a) has pointed out, Richardson's model, which is basically two differential equations, does not allow for the use of time, and hence decisions. The latter can only be modelled by using difference equations. 8. For a discussion of error correction mechanisms, see Alogoskoufis & Smith (1991), Salmon (1982), Stewart (1991). 9. The subperiod used to calculate the Chow test is, for all three models, 196076 and the forecast period for the predictive failure test 198790. 10. They also satisfied the CUSUM and CUSUMSQ tests, whose plots were obtained using MICROFIT version 3. 11. They also satisfy the CUSUM and CUSUMSQ tests. 12. The OLS method does not give consistent estimates because of the correlation between the residuals and the regressors. However, the presentation of the OLS results in Table IV can be justified by saying something about the direction of the asymptotic bias. In a simple twoequation model with two exogenous variables, it is easily pointed out (Maddala, 1977, ch. 11, pp. 248249) that leastsquares estimation can still be worthwhile, since the introduction of exogenous variables that are not highly intercorrelated can considerably reduce the asymptotic bias. A general linear statistical model may be written as YTXMFMxM+XTXKBK)XM+UTXM = 0. We may
238
George M. Georgiou et al. arbitrarily set the value of one coefficient to be 1, say y,, = 1 (normalization rule). Thus, the system is written as: Yr,+XBi+ U, = o or Yi = Yiyi+XiP i+ Ui = (Y,, Xi) (3i) + ui = zii+ ui where Y, and Xi are those endogenous and exogenous variables that appear in the i equation with nonzero coefficients. The LSestimate of 8, is  = (Z',iZi)1 Z'iYi. As the sample size increases, , does not converge to 8, in probability, since plim $, = i8, + plim (T'Z'iZ7)(T1Z'iU,) t6i. In Desai and Blake's model, we set Zi = (InG,, lnG,_1, lnG,_2, InT,_1, InT,2) and 8i = (al, a2, a3, a4, a5)'. The bias is plim $i8i = SX lzzz1ui; ?O, since z1_2u"'O because Z1 contains the variable inG, that is not independent from the stochastic term u"'. Calculating 'zizi and Eziuh we find that plim 8i$,<0. Table V reports the OLS estimates along with those obtained from the 2SLS method (the estimated parameters and the tvalues are presented in square brackets). In exactly identified models, the OLS estimates are inconsistent. However, their reporting can be justified by the fact that the OLS method is more robust against specification errors than many of the simultaneousequation methods and also that the predictions from equations estimated by OLS often compare favourably with those obtained from equations estimated by the simultaneousequation methods. This result reinforces previous empirical evidence on the nonexistence of interdependence between the two countries under consideration (see Georgiou, 1990 and Stavrinos, 1992). By contrast, Kapopoulos & Lazaretou (1993) estimate a twopart model of demand for Greek military expenditure and find that Turkish military spending is an important factor in Greek defence policymaking. For an example of a model of military expenditure where strategic considerations are an important explanatory variable, see Smith (1989). It has been argued (Kapopoulos & Lazaretou, 1993) that Greece is a freerider on US spending in the longrun, while in the shortrun it acts as a follower. Clogg, Richard, 1991. 'GreekTurkish Relations in the Post1974 Period', pp. 1223 in Constas. Constas, Dimitrios, ed., 1991. The GreekTurkish Conflict in the 1990s. London: Macmillan. Desai, Meghnap & David Blake, 1981. 'Modelling the Ultimate Absurdity: A Comment on a Quantitative Study of the Strategic Arms Race in the Missile Age', Review of Economics and Statistics, vol. 63, no. 4, pp. 629632. Edelstein, Michael, 1990. 'What Price Cold War? Military Spending and Private Investment in the US, 19461979', Cambridge Journal of Economics, vol. 14, no. 4, pp. 421437. Eide, Asbj0rn & Marek Thee, eds, 1980. Problems in Contemporary Militarism. London: Croom Helm. Georgiou, George M., 1990. 'Is There an Arms Race Between Greece and Turkey?: Some Preliminary Econometric Results', Cyprus Journal of Economics, vol. 3, no. 1, pp. 5873. Geyer, Michael E., 1990. 'Militarism and Capitalism in the 20th Century', pp. 247275 in Gleditsch & Nj0lstad. Gleditsch, Nils Petter, 1990. 'Research On Arms Races', pp. 114 in Gleditsch & Nj0lstad. Gleditsch, Nils Petter & Olav Nj0lstad, eds, 1990. Arms Races: Technological and Political Dynamics. London: Sage. Hitchens, Christopher, 1984. Cyprus. London: Quartet. Intriligator, Michael D., 1975. 'Strategic Considerations in the Richardson Model of Arms Races', Journal of Political Economy, vol. 83, no. 2, April, pp. 339353. Intriligator, Michael D. & Dagobert L. Brito, 1990. 'Arms Race Modelling: A Reconsideration', pp. 5877 in Gleditsch & Nj0lstad. Kapopoulos, Panayotis T. & Sophia Lazaretou, 1993. 'Modelling the Demand for Greek Defence Expenditure: An Error Correction Approach', Cyprus Journal of Economics, vol. 6, no. 1, pp. 7386. Klare, Michael T., 1980. 'Militarism: The Issue Today', pp. 3646 in Eide & Thee. Maddala, George S., 1977. Econometrics. New York: McGrawHill. McGuire, Martin, 1977. 'A Quantitative Study of the Strategic Arms Race in the Missile Age', Review of Economics and Statistics, vol. 59, no. 3, pp. 328339. Mills, Terence, 1991. Time Series Techniques for Economists. Cambridge: Cambridge University Press. Rapoport, Anatol, 1957. 'Lewis F. Richardson's Mathematical Theory of War', Journal of Conflict Resolution, vol. 1, no. 3, September, pp. 249299. Rattinger, Hans, 1975. 'Armaments, Detente and Bureaucracy: The Case of the Arms Race in Europe', Journal of Conflict Resolution, vol. 19, no. 4, December, pp. 571595. Regehr, Ernie, 1980. 'What is Militarism?', pp. 127139 in Eide & Thee. Richardson, Leon F., 1960. 'Generalised Foreign Politics', British Journal of Psychology Monographs, Supplement, vol. 23, pp. 691. Richardson, Leon F., 1960. Arms and Insecurity. Pittsburgh, PA: Bonwood.
13.
14.
15. 16.
REFERENCES Allison, Graham T., 1974. 'Questions About the Arms Race: Who's Racing Whom?' pp. 6298 in R. L. Pfaltzgraf, ed., Contrasting Approaches to Strategic Arms Control. Lexington: Heath. Alogoskoufis, George & Ron Smith, 1991. 'On Error Correction Models: Specification, Interpretation, Estimation', Journal of Economic Surveys, vol. 5, no. 1, pp. 97128. Anderton, Charles, H., 1989. 'Arms Race Modelling: Problems and Prospects', Journal of Conflict Resolution, vol. 33, no. 2, June, pp. 346367. Clausewitz, von Carl, 1968. On War. London: Penguin. [First published in Berlin, 1832].
239
metricModelling.Cambridge: Salmon, Marc, 1982. 'ErrorCorrectionMechanisms', Cambridge University EconomicJournal,vol. 92, no. 4, pp. 615629. Press. Seeley, Jo Anne Browning, 1988. 'Governmentand Stavrinos, Vassilis, 1992. 'Defence Expendituresin A Arms Competition: Politics', in P. M. Pitman, ed., Turkey: Country Modellingand Causality Analysis', Greek Economic Review, vol. 14, no. 1, pp. 115Study. Federal Research Division, United States Government. 128. SIPRI (StockholmInternational Peace ResearchInsti Stewart,Jon, 1991. Econometrics. Hemel Hempstead: tute), World Armaments and Disarmaments, SIPRI PhilipAllan. Yearbook(variousyears). London:OxfordUniver Tartter,Jean R., 1988. 'NationalSecurity',in P. M. Pitman, ed., Turkey: A Country Study. Federal sity Press. ResearchDivision,United StatesGovernment. its Skjelsbak, Kjell, 1980.'Militarism, Dimensionsand and Militarisation Corollaries: An Attempt at Conceptual Clarifi Thee, Marek, 1980. 'Militarism in International cation', pp. 77105 in Eide & Thee. Relations',pp. 1535 in Contemporary Eide & Thee. Smith,Ron P., 1989.'Modelsof Military Expenditure', Journal of Applied Econometrics, vol. 4, no. 4, pp. Wiberg,Hakan, 1990a. 'Arms Races, FormalModels and QuantitativeTests', pp. 3157 in Gleditsch& 345359. Smoker, Paul, 1990. 'ArtificialIntelligenceModelsof Nj0lstad. Arms Races', pp. 7886 in Gleditsch& Nj0lstad. Wiberg,Hakan, 1990b. 'ArmsRaces Why Worry?', pp. 352375 in Gleditsch& Nj0lstad. Spanos, Aris, 1986. Statistical Foundations of Econo
GEORGE M. GEORGIOU, b. 1956, Central Bank of Cyprus (International Division), PhD in Economics (Birkbeck College, University of London, 1984), MSc in Economics (Birkbeck College, University of London, 1980), and BA in Social Science (Polytechnic of Central London, 1979). PANAYOTIS T. KAPOPOULOS, b. 1964, Alpha Credit Bank (Economic Research Division), PhD in Economics (Athens University of Economics and Business, AUEB, 1993), MA in Economics (AUEB, 1989) and BA in Economics (AUEB, 1987). SOPHIA LAZARETOU, b. 1964, Lecturer, University of Macedonia, Department of International and European Economic and Political Studies, PhD in Economics (Athens University of Economics and Business, AUEB, 1993), MA in Economics (AUEB, 1989) and BA in Economics (AUEB, 1986).