This paper proposes a method for assessing the cost of friction between North Atlantic Treaty Organisation (NATO) allies and highlights internal threats. This is applied to the Greek–Turkish conflict within the NATO context and concerns the functioning of defence expenditure in Greece, modified in such a way as to focus on the causes of friction between these allies. The analysis concentrates mainly on the issue of internal threats to the long-run equilibrium of NATO. The ARDL methodology used modifies the typical error correction model by introducing a mechanism that accelerates the process that leads back to the long-run equilibrium. Along with assessing the cost to an ally in relation to an internal threat, the method proposed allows the time required for the long-run equilibrium of NATO to be restored. The paper concludes that dynamic incidents of friction between allies expressed as an internal threat disturb NATO’s static equilibrium, destabilise an individual ally’s defence policy and contribute to cost being incurred.

The recent literature on North Atlantic Treaty Organisation (NATO) cohesion is dominated by contributions which try to tackle the role of the alliance following the end of the Cold War. There is a wide selection of ideas, suggestions, and propositions to consider in this direction (e.g. ^{1} between its members and threaten its cohesion.

There is, however, an additional case of in-NATO friction, far older than any of the above, the one between Greece and Turkey. Most sources consider the beginning of this conflict to be in 1974 with the Cyprus crisis, although its roots date back even earlier to the start of the 1950s. This issue has developed into a popular topic in the literature as it has grown to represent a typical arms race case (

Given this rather unorthodox background in which two NATO allies have been entangled in an arms race, one cannot help wondering about its impact on the cohesion of NATO. This paper hopes to add to the academic literature by proposing several modifications to a typical defence spending function aimed at reflecting cases in which a NATO member faces a so-called “internal threat”. This is a concept involving the hostile attitude of a member of the alliance to another and is considered as being inversely proportional to the strength of its intra-cohesion. (

Most of the papers on the issue of the defence expenditure function and its determinants use conventional models for a time series or panel analysis employing three main variable categories: economics and production, technology, and geopolitical and security. Several early, well-established contributions include Hewitt (

The case of Greek–Turkish friction and the arms race between the two sides occupies a leading position in the literature (e.g.,

This paper then aims to tackle modifications that a typical demand for defence spending must undertake when there is a threat from within, namely when the identity of an ally coincides with that of an adversary. In such cases, the cohesion of the alliance becomes a leading issue with Lute and Burns (2019, p. 4) who trace its roots to the extent to which all members adhere to the alliance’s values. By contrast, Binnendijk and Priebe (2019, p. 1) are more straightforward when posing the question of cohesion to all NATO members in the event of a Russian attack on a member state. Finally,

Quoting Tardy (2018, pp. 1-18), “one of NATO’s characteristics is that its ^{2}

The difficulty in agreeing on a common external threat (

Focusing on this Greece–Turkey anecdote, which according to the author “demonstrates that the dichotomy between “enemy” and “ally” may be misleading” therefore requires a brief explanatory background. Both countries have been NATO members since 1952. Since then, there have been numerous occasions on which friction between them on a wide selection of issues (territorial waters, national airspace, flight information region [FIR] for the control of flight activity, and the continental shelf), escalated to the brink of war in 1974, 1987 and 1996.^{3} Given this background, one could not really wonder why both countries, despite being NATO members, have been entangled in an arms race (^{4} The reasoning behind this specific choice is based on the fact that the number of daily engagement incidents between aircraft of the two sides constitutes the most aggressive form of challenge given the data series available and is thus a selection much closer to the concept of threat compared to incidents of simple ICAO or FIR violations. The specific variable also assumes a clearly short-term character, as explained earlier on in this paper, and cannot be used in a long-run context, in which one runs the risk of deriving coefficients that may be difficult to interpret.

Given our discussion in Sections 3.1 and 3.2, the variable representing threat in the Greek defence expenditure function is considered to refer to the internal dimension of the concept, as it has led to an arms race between Greece and Turkey, i.e., an internal threat (^{5} This means that our specific alliance model needs to be modified to acquire a dynamic dimension. We need therefore to adjust the demand for the defence expenditure function of Greece, as this is derived within the guidelines of the standard theory (

Representing the concept of threat as used in a defence spending function requires caution. For example, the aggregate defence expenditure of a potential adversary, including funds channelled to equipment, personnel and infrastructure may be a reasonable approximation for a possible threat in the future. However, it does not entail the concept of an immediate and direct form of claim against the targeted nation’s sovereignty, or even an attack against it. To put it differently; threat might be regarded as a damoclean sword for the country in focus but is by no means an action exercising threat. The typical defence spending equation, by design, focuses on potential threat and cannot deal with situations where the adversary utilises its resources to challenge the sovereignty rights of the other country. Daily violations of Greek airspace by the Turkish Airforce is a case in point. In this paper, we show how the typical defence spending model can be modified to accommodate such phenomena.

During the last twenty-five years or so, demand for defence models have been based on the so-called vector error correction methodology (e.g. ^{6} It also indicates the loss of efficiency suffered by Greece as a NATO member while frictions against Turkey, an alliance partner, continue.

The SPILL variable is introduced in the analysis to represent the benefits enjoyed by an alliance member because of participating in it (^{7}

It appears, therefore, that the Washington Treaty does not provide for cases of an “internal threat”^{8} and, consequently, one cannot consider the possibility of alliance benefits of the type stated in Article 5.

In terms of a modification to the standard theory, we use defence spending on equipment in the case of Greece as the dependent variable of our demand for defence expenditure function. By contrast, in the case of Turkey, we focus on total defence spending. The choice regarding the Greek case is founded on the grounds that the negative population rates of growth for Greece have led the country to put an emphasis on technology rather than personnel to face the demands of its defence doctrine. Thus, in terms of an indifference curve analysis between manpower (human resources) and technology (property resources), facing the problem of scarcity in the case of the former raises personnel costs and leads to an emphasis on technology. The aim of this choice is to maintain the initial defence output at the original indifference curve (

The corresponding picture in Turkey is quite different. Its population of more than 80 million grows at substantial rates every year. In addition, the country’s defence industrial base (DIB) is thriving and supports more than half of the country’s defence equipment requirements and causes considerable spin-offs and economies of scale. We argue, therefore, that the focus for Greece must be the overall nexus of both human and property resources of Turkey as these interact to represent the threat introduced in the Greek defence spending function.^{9}

A further point worth noting is that we shall use defence spending as shares of gross domestic product (GDP) rather than levels. This option aims to avoid conversion problems between different currencies (Euros or Turkish liras) and conversion between current and constant figures in the absence of reliable price deflators. It also intends to consider the vast difference of the GDP levels between Greece and Turkey (

Since the aim of the paper is to modify the standard demand for defence expenditure function (^{10}:

S = GREQ – GREQT

where GREQ stands for the GDP share of Greek defence spending on equipment and GREQT the corresponding targeted national defence spending which is a function of

GREQT = b_{0} + b_{1}*TURDEF

with TURDEF representing the GDP share of Turkish total defence spending.

Therefore, following the welfare maximisation procedure involving a security function and a budget constraint as indicated by Smith (

GREQ = f (Y, p, TURDEF, SPILL)

where: Y is the Greek GDP, p is the price deflator for defence equipment procured and SPILL enters the specification as a country-specific variable representing NATO’s total defence spending as a GDP share and is added to the account for the environment.

This basic formulation needs to undergo a certain amount of fine tuning to suit the particularities of the Greek case. First, we drop prices due to the absence of a uniform price deflator (^{11} Thus, our basic defence equipment procurement equation will be:

GREQ = f (TURDEF, SPILL) (1)

Intending to stress the importance of distinguishing between property and human resources, we break down the TURDEF into its components as follows: The first is introduced via the trend of the increasing population rate of Turkey D(LTRPOP), whereas the property resources are represented by the GDP share of Turkish equipment purchases (TREQ).

GREQ = f [ TREQ, D(LTRPOP), SPILL] (1’)

According to theory, equations (1) or, alternatively (1’) represent the long run equilibrium. Estimating these equations, however, requires some caution. If the variables involved are non-stationary/random walks.^{12} As Granger and Newbold (

Writing equations (1) and (1’) in estimation form after the fine tuning mentioned above and presenting the variables in log form yields:

LGREQ_{t} = b_{0}+ b_{1}*LTURDEF_{t} + b_{2}*LSPILL_{t} + u_{t}, (1a)

LGREQ_{t} = b_{0}+ b_{1}*LTREQ + b_{2}*D(LTRPOP) + b_{3}*LSPILL_{t} + u_{t}, (1’a)

where u_{t}, the error term represents deviations from the long run equilibrium. For reasons that will become obvious below, the two equations can be written as

u_{t} = LGREQ_{t} - b_{0}- b_{1}*LTURDEF_{t} - b_{2}*LSPILL_{t}, (2)

u_{t} = LGREQ_{t} - b_{0} - b_{1}*LTREQ - b_{2}*D(LTRPOP) - b_{3}*LSPILL_{t}. (2’)

Following standard practices, our first step is to examine the stationarity properties of the variables. Inspection of the graphs of the variables, see ^{13} ^{14}

Variable Historical Trends

Typical ADF Tests for the Variables Used

Levels | Differences | |
---|---|---|

LGREQ | 0.63 | 0.00 |

LTURDEF | 0.71 | 0.00 |

LSPILL | 0.60 | 0.00 |

LTREQ | 0.08 | 0.00 |

D(LTRPOP) | 0.0 | 0.00 |

LDOGS | 0.08 | 0.00 |

It turns out that the hypothesis that the two models are cointegrated cannot be rejected with any reasonable level of confidence; therefore, we can proceed with an estimation.

As we have discussed, cointegration implies the existence of an error correction mechanism. The empirical model (2_and (2’) can be written as:

The equations above are the typical representations of an error correction model. Changes of the dependent variable depend on the history of all variables^{15} of the long run model, deviations from equilibrium and, possibly on variable(s) z which might impact upon the adjustment process, in our case fighters’ engagements. Note the critical role of the term α which must be negative if an equilibrium is to exist. So, if we have a positive deviation, a negative force is exercised upon the dependent variable to restore equilibrium, and if we have a negative deviation, a positive force is exercised.^{16} e is the residual.

The dataset used in this study contains the variables in

The Dataset

Code | Data Series | Source |
---|---|---|

GREQ | Greece: Expenditure on Defence Equipment / GDP | NATO and SIPRI |

SPILL | NATO Defence Expenditure / GDP | NATO and SIPRI |

Y | Rate of change of Greek GDP | ELSTAT |

TURDEF | Turkey: Total Defence Expenditure / GDP | NATO and SIPRI |

TREQ | Turkey: Expenditure on Defence Equipment / GDP | NATO and SIPRI |

D (LTRPOP) | Turkey: Population Growth Rate | UN STATISTICS |

DOGS | Greek / Turkish Fighters Engagements | Hellenic Air Force General Staff |

The estimates^{17} of our long-run defence spending function for Greece (1a), after adjusting it for the drop in prices and the GDP, are shown in

Greece: The Long - Run Defence Expenditure Function

Variable | Coefficient | Std. Error | t-Statistic | Prob. |

LTURDEF | 0.980636 | 0.258263 | 3.797042 | 0.0007 |

LSPILL | 1.401085 | 0.239635 | 5.846745 | 0.0000 |

C | -3.974160 | 0.323410 | -12.28832 | 0.0000 |

u = LGREQ - (0.9806*LTURDEF + 1.4011*LSPILL -3.9742 ) | ||||

LDOGS(-2) | 0.052101 | 0.013027 | 3.999494 | 0.0004 |

α | -0.818133 | 0.107117 | -7.637769 | 0.0000 |

Greece: The Modified Long - Run Defence Expenditure Function

Dep. Variable: LGREQ)

Variable | Coefficient | Std. Error | t-Statistic | Prob. |
---|---|---|---|---|

LTREQ | 0.577979 | 0.167435 | 3.451960 | 0.0018 |

LSPILL | 1.09101 | 0.286677 | 5.264123 | 0.0000 |

D(LTRPOP) | 71.15834 | 22.14052 | 3.213942 | 0.0033 |

C | -4.082036 | 0.374659 | -10.89534 | 0.0000 |

u = LGREQ - (0.5780*LTREQ + 1.5091*LSPILL + 71.1583*D(LTRPOP) -4.0820) | ||||

LDOGS (-2) | 0.081285 | 0.013554 | 5.996990 | 0.0000 |

α | -0.79482 | 0.096704 | -8.220775 | 0.0000 |

The corresponding estimates of the short run/adjustment parameters are shown in Tables A1 and A2, respectively, in Appendix A.

Before discussing our estimates, it is important to establish their reliability, i.e. in the statistical jargon whether they are unbiased, consistent, and efficient. ARDL estimates are unbiased, consistent, and efficient, provided that the residuals have constant variance. To establish that the variance is constant and the residuals normally distributed, we use the following tests:

Correlogram: to establish that the residuals are not correlated over time.

Correlogram of squared residuals: to establish that the variance of the residuals is not changing over time.^{18}

The Breusch–Pagan–Godfrey test: to establish that the variance is constant.

The Jarque–Bera test: establish that the residuals are normally distributed.

Detailed results of these tests can be found in Appendix A. Clearly, the hypothesis of constant variance, normally distributed residuals cannot be rejected at any reasonable level of confidence. Therefore, we are certain that our estimates are unbiased, consistent, and efficient.

The results presented in the long-run versions of the equations (Tables 3 and 4 above) show that all determinants are significant in explaining the behaviour of the demand for defence equipment and bear the expected sign. More specifically, commenting on each determinant separately, we start with the behaviour of equipment procurement to changes in LTURDEF (a variable including both the property and human resources of Turkey). The derived coefficient is close to unity, bears a positive sign and must be interpreted as corresponding to that derived for the variable representing the threat in conventional estimates (e.g. ^{19} In fact, this estimate indicates that the Greek defence doctrine is much more concerned with the long-run developments on the Turkish human resources side rather than the equipment purchased by the ally-adversary. This is to be expected if one compares the impressive rates of Turkish population growth to those of Greece, the corresponding figures of which have started to become negative over the last few years.

As regards the importance of the NATO alliance to the Greek defence equipment procurement, as represented by SPILL, the derived coefficient is significant and positive indicating the absence of any form of free-riding policies.^{20} In fact, the derived coefficient is in line with the pronounced commitment of Greece to her alliance obligations with the elastic coefficient being higher than most sources in the literature on this issue (between 0.3 and 1.3 in e.g.

Finally, we turn our attention to the coefficient of DOGS and α, the coefficient of adjustment. The coefficient of DOGS is positive, indicating the tight short-run margins allowed to the Greek side to increase its defence equipment. The speed of adjustment coefficient, α, (about -0.8) in Tables 3 and 4 is highly significant and points to an adjustment period of slightly more than one year.^{21}

Concluding this discussion on the results derived, we need to point out that the reliability of these estimates is further supported by the fact that our method of estimation captures the actual Greek defence spending behaviour without resorting to the help of any dummy variables to interpret the effects of major political or geostrategic events. Attention is drawn, however, to cases in which our estimates may be compared to those encountered in the literature thus far due to the different approach used herewith.

One of the main issues of the paper is to highlight the measure of alliance sub-optimality, approximated by the number of daily engagements between HAF and THK fighters when the latter attempt to violate Hellenic air space as this is seen from the Greek side (Figures 2 and 3). As pointed out in Section 3, this is clearly an adjustment reaction and any additional procurement (over and above the purchases provided in the 5-year procurement programmes – EMPAE) following such attempts indicates the extra cost to the adjustment/corrective process to restore long–run equilibrium in our alliance model.^{22} Thus, we argue that the role of this specific variable in the analysis is to approximate the degree of sub optimality and, consequently, the loss of efficiency suffered by Greece while incidents of friction with Turkey, an alliance member, continue.

^{23}

More specifically, following the Greek–Turkish 1985 crisis and a local maximum of the number of engagements, Greece purchased two batches of Mirages and F-16s a couple of years later. About two years after the period 1994–1999 and the prolonged friction between the two sides, Greece placed an order for an additional batch of F-16s followed by an order for Mirages in the year 2000. With the turn of the century, the number of engagements increased dramatically assuming a global maximum in 2003, followed by the last F-16 batch purchase in 2006. Based on this behaviour, therefore, one can safely argue that the engagements accelerated defence spending.

Annual HAF and THK Engagements Measure Alliance Sub-Optimality: Extra Procurement Acts as an Error-Correction Mechanism Restoring Convergence

The economic crisis of 2010 to 2019 is an interesting period to discuss, as there are no equipment purchases, following the “Troika” guidelines and despite the continuing aggressiveness from Turkey. In fact, there are sources in the literature (

THK Activity in Hellenic Airspace

To the extent that such an approach implies that the continuous pressure exercised by Turkey is intended to bring the Greek economy to its knees, facts and figures prove that it is a rather superficial consideration for the specific case. To begin with, we have already shown, both in this paper (^{24} annual cost of such engagements for the Greek side amounts to about €15 million at most.^{25} This is a rather low price to pay for two-minute readiness annual drills of the Hellenic Air Force, which keep its pilots in top shape. The only explanation left to consider, therefore, concerning the absence of additional defence equipment purchases is the political cost involved in such cases, which discourages governments to purchase additional defence equipment.

Zooming on the Engagements between the HAF and THK during the Crisis

Source: HAF, NATO

Concluding with the section of policy implications, we thought that it would be appropriate to embark on a forecasting exercise given the encouraging results of the recursive regression exercise shown in Figure A1. The values assumed by the explanatory variables have been input as follows: The TREQ figures are based on the provisions of the $150 billion long-term (2000–2025) procurement programme of the Turkish armed forces, while the DTRPOP figures retain the current year growth rate for the forecasted period.^{26} Finally, the SPILL figures assume that the NATO spending figures as a GDP percentage will remain broadly stable for the forecasted period.

Greece: Defence Expenditure on Equipment (GREQ) Projections (% GDP)

Source: NATO, Authors’ Projections

The forecasted defence spending on equipment as a percentage of GDP for Greece speaks for itself (

The aim of this paper has been to assess the extent of friction introduced in an alliance environment and the inevitable cost incurred when two of its members resort to the extreme option of continuous aggressive behaviour, as indicated in Section 3.1. Using the case of Greece versus Turkey, we have modified the typical defence expenditure function of the former to include an accelerator to the error-correction mechanism that assesses the cost due to such friction. This is approximated using the daily engagements of the Hellenic Air Force (HAF) fighters against the corresponding THK when attempting a violation of the Greek airspace as the Greek side considers it. To this end, we have modified the traditional defence spending function for Greece as it appears in the literature, in order to provide a dynamic dimension to its long-run static Nash equilibrium environment. In the context of this modification, we consider the daily engagements and the acquisition cost of the additional defence equipment involved in such cases, to approximate the degree of sub optimality and, consequently the loss of efficiency suffered by the NATO alliance while frictions between Greece and Turkey continue. We believe that this specification and estimation method reflects the actual picture much better than most of the literature sources, a fact supported by the absence of any dummy variables representing political geostrategic events.

In this context, we show that engagements accelerate defence spending, a cost which leads to a waste of resources, threatens NATO cohesion, and prevents it from focusing on common external threats that are clearly evident. Thus, cases in which commitment to NATO is weakened due to a member’s hostile attitude or, even worse, cases in which a member state employs defence equipment manufactured by the country which until now has been considered as an external threat do not exactly contribute to NATO cohesion. Greece’s irregular, but frequent disturbances de-trend the Greek defence policy and represent the adjustment cost suffered by this type of threat. In cases in which such a threat originates from external sources, then the alliance’s members would be expected to contribute to the cost incurred along the lines provided by Article 5 of the NATO Treaty. The problem for Greece is that in her case, the threat is internal with reference to the alliance and, as such, there is no contribution to be expected from its allies, either individually, or as a NATO entity collectively. In this case, therefore, this adjustment cost is borne exclusively by the country which suffers assault, namely Greece.

We are grateful to Professor Keith Hartley, University of York for valuable contributions to earlier drafts. We are also indebted to three unknown referees for their constructive comments. Finally, thanks are due to Nicolas Zonzilos, a colleague in the Research Department of the Bank of Greece for his technical support.

No potential conflict of interest has been reported.

The term “friction” is used along the lines found in the recent literature (e.g. Bazin and Kounertova, 2018, pp. 1-12) pointing to sustained “civil-military frictions on both NATO and national levels,” which tend to develop and represent “internal threats.” The first such incident that comes to mind is, of course, the disagreement between the United States and Turkey over the purchase of the Russian S-400 system by the latter and the ensuing US reaction to exclude Turkey from the F-35 programme. A further issue worth noting is the incident between Turkey and France. Following the encounter of the French frigate Courbet against several Turkish Navy units on June 10, off the Libyan coast. France protested about the Turkish ships having locked the frigate three times with their fire control radars, an act of hostility. The consequence was that on July 1, 2020, France announced that she was suspending her involvement in NATO operation Sea Guardian. The latest episode in the series of such incidents is the AUKUS agreement of September 2021 which France has called “a stab in the back” because it led to the cancellation by Australia of a

Such expressions of concern have been appearing mainly in the daily and weekly press since 2017 (e.g.

For a thorough analysis on this issue see Symeonidis and Zombanakis (

There are more such relevant time series provided by the Greek data sources that one may use (FIR violations, total number of aircraft used, armed aircraft etc.). However, we have opted for using the number of engagements between Greek and Turkish fighters rather than that of air space violations which has been the most popular choice in the literature thus far (e.g.

The “Pure Public Good Model” demands that defence benefits are nonrival and non-excludable among allies, as opposed to the “Joint Product Model” in the case of which the alliance defence activity produces both public and private outputs (

Suboptimality is defined as the degree of deviation from efficiency. This means that as the number of allied members increases, the resulting equilibrium is apt to be more suboptimal as free riding opportunities are enhanced through greater spill ins (

According to (Bazin and Kounertova, 2018, pp. 1-12), “The rise of populism and radical nationalism with authoritarian inclinations, further fuelled by hybrid, cyber, or information warfare coming from Russia, appears threatening NATO’s core values and will create frictions within NATO”

NATO support has been questioned since 1974 and the Cyprus crisis, following which Greece withdrew from the NATO military structure for a period of six years. Since then, however, despite the 25% GDP reduction due to the ten-year economic crisis and contrary to the IMF defence budget cut recommendations (International Monetary Fund [IMF],

The relative importance of property and human resources for both Greece and Turkey is thoroughly examined in Andreou and Zombanakis (2011, pp. 459-469). It is indicative of the disadvantageous position of Greece versus Turkey as regards personnel matters to point out that while the former faces an annual shortage of several thousand conscripts, the latter provides for conscripts to pay for a military service relief with the money thus earned supporting the country’s DIB.

For a thorough analytical explanation on deriving the demand for the defence spending function of an economy see Sandler and Hartley (1995, pp. 52-72). As this concerns the Greek case, the Turkish defence expenditure is threatened exogenously.

Aimed at verifying the result of earlier research on the insignificance of the income variable, we have performed a Wald test, as shown in

I(1) in the jargon of statistics, meaning that they wander over time.

The test was implemented in EViews 11.

We have decided to avoid extending the data series to years affected by the COVID-19 pandemic.

The number of lags is determined empirically during the estimation process. The chosen model is the one with the highest explanatory power. In our estimation, we use the Akaike criterion.

An engineer would describe this process as negative feedback.

Parameter estimates using the ARDL methodology were obtained using the EViews statistical package.

In the statistical jargon we are testing for ARCH effects, i.e., autoregressive conditional heteroskedasticity.

Attention is drawn to the fact that the elasticity is calculated using the chain rule, i.e. regarding changes of an explanatory variable (TRPOP) which is expressed in terms of log differences, i.e., rates of change in time.

Free riding policies must be related to whether the alliance is regarded as a “Pure Public Good” or a “Joint Product” model. Since the scope of this paper is to focus on the Greek-Turkish friction and its impact on the cohesion of the alliance, one must consider the fact that Turkey is more likely to regard NATO functioning as a “Joint Product” model, since its geostrategic interests extend to a wide variety of targets, ranging from Syria to Libya and several other African states. By contrast, Greece is more inclined to think of the alliance as a “Pure Public Good” model, as it focuses exclusively on the NATO’s geopolitical interests.

Given the annual frequency of the data, a coefficient of about 0.8 indicates that convergence has been attained to a rough 80% within a year, implying that the convergence procedure will be completed in about a year and a half.

Given the substantial public debt of Greece, spending money earlier rather than later is clearly a burden.

Attaining long-run equilibrium between NATO members is expected to lead to efficiency maximisation, as pointed out in footnote 8. If mere free-riding practices are considered a threat to such an equilibrium (

Indirect costs are substantially higher in the sense that because of engagements, Greece has to accelerate its defence spending.

The hourly flying cost of an F-16 is €9,000 while that of a Mirage is €13,000. The weighted average of the two given the analogy of F-16s and Mirages in the Hellenic Air Force is €10,000 per aircraft for a rough estimate of a total of 1,500 FIR and ICAO violations as well as engagements per year.

For an extensive evaluation of the programme, see

Short run /adjustment parameters of equation 1a

Variable | Coefficient | Std. Error | t-Statistic | Prob. |

D (LGREQ (-1)) | 0.260895 | 0.103334 | 2.524776 | 0.0173 |

D(LTURDEF) | -0.414944 | 0.331387 | -1.252143 | 0.2205 |

D (LTURDEF (-1)) | -0.723694 | 0.262178 | -2.760315 | 0.0099 |

D (LTURDEF (-2)) | -0.362543 | 0.283925 | -1.276898 | 0.2118 |

D (LTURDEF (-3)) | -0.518470 | 0.303202 | -1.709982 | 0.0979 |

D(LSPILL) | 0.220635 | 0.479457 | 0.460177 | 0.6488 |

D (LSPILL (-1)) | 0.420360 | 0.460796 | 0.912247 | 0.3692 |

D (LSPILL (-2)) | -2.062025 | 0.456732 | -4.514739 | 0.0001 |

D (LSPILL (-3)) | -3.182232 | 0.496490 | -6.409460 | 0.0000 |

LDOGS (-2) | 0.052101 | 0.013027 | 3.999494 | 0.0004 |

α | -0.818133 | 0.107117 | -7.637769 | 0.0000 |

R-squared | 0.750361 | Mean dependent var | -0.044920 | |

Adjusted R-squared | 0.672349 | S.D. dependent var | 0.348491 | |

S.E. of regression | 0.199479 | Akaike info criterion | -0.170054 | |

Sum squared resid | 1.273337 | Schwarz criterion | 0.280485 | |

Log likelihood | 14.65617 | Hannan-Quinn criter. | -0.003909 | |

Durbin-Watson stat | 2.149285 |

a. Correlogram – Q Statistics | ||||||

Autocorrelation | Partial Correlation | AC | PAC | Q-Stat | Prob* | |

.*| . | | .*| . | | 1 | -0.084 | -0.084 | 0.3271 | 0.567 |

**| . | | **| . | | 2 | -0.266 | -0.275 | 3.6642 | 0.160 |

.*| . | | .*| . | | 3 | -0.100 | -0.166 | 4.1498 | 0.246 |

.*| . | | **| . | | 4 | -0.092 | -0.223 | 4.5654 | 0.335 |

. | . | | .*| . | | 5 | 0.013 | -0.132 | 4.5737 | 0.470 |

. | . | | .*| . | | 6 | -0.021 | -0.187 | 4.5978 | 0.596 |

. | . | | .*| . | | 7 | 0.066 | -0.068 | 4.8295 | 0.681 |

. | . | | .*| . | | 8 | -0.021 | -0.151 | 4.8542 | 0.773 |

.*| . | | **| . | | 9 | -0.117 | -0.236 | 5.6274 | 0.777 |

. |*. | | . | . | | 10 | 0.182 | 0.036 | 7.5616 | 0.672 |

. |*. | | . |*. | | 11 | 0.148 | 0.090 | 8.8836 | 0.633 |

.*| . | | . | . | | 12 | -0.119 | -0.049 | 9.7640 | 0.637 |

. | . | | . | . | | 13 | -0.029 | 0.056 | 9.8200 | 0.709 |

.*| . | | .*| . | | 14 | -0.171 | -0.167 | 11.761 | 0.626 |

. |*. | | . |*. | | 15 | 0.105 | 0.090 | 12.520 | 0.639 |

. | . | | . | . | | 16 | 0.054 | 0.004 | 12.726 | 0.693 |

. | . | | . |*. | | 17 | 0.048 | 0.093 | 12.898 | 0.743 |

. | . | | . |*. | | 18 | 0.060 | 0.096 | 13.182 | 0.781 |

.*| . | | . | . | | 19 | -0.174 | -0.039 | 15.624 | 0.682 |

. | . | | . | . | | 20 | 0.006 | 0.046 | 15.627 | 0.740 |

b. Correlogram Squared Residuals | ||||||

Autocorrelation | Partial Correlation | AC | PAC | Q-Stat | Prob* | |

. |*. | | . |*. | | 1 | 0.149 | 0.149 | 1.0262 | 0.311 |

. | . | | . | . | | 2 | -0.024 | -0.047 | 1.0537 | 0.590 |

. | . | | . | . | | 3 | -0.009 | 0.002 | 1.0576 | 0.787 |

. | . | | . | . | | 4 | -0.017 | -0.017 | 1.0719 | 0.899 |

. | . | | . | . | | 5 | 0.037 | 0.043 | 1.1404 | 0.950 |

. | . | | . | . | | 6 | 0.046 | 0.033 | 1.2504 | 0.974 |

**| . | | **| . | | 7 | -0.209 | -0.224 | 3.5883 | 0.826 |

. | . | | . | . | | 8 | -0.019 | 0.056 | 3.6088 | 0.891 |

.*| . | | .*| . | | 9 | -0.117 | -0.146 | 4.3847 | 0.884 |

. | . | | . |*. | | 10 | 0.029 | 0.082 | 4.4348 | 0.926 |

. |** | | . |*. | | 11 | 0.230 | 0.210 | 7.6398 | 0.745 |

. | . | | .*| . | | 12 | -0.030 | -0.106 | 7.6957 | 0.808 |

. | . | | . |*. | | 13 | 0.012 | 0.089 | 7.7049 | 0.862 |

. |*. | | . | . | | 14 | 0.088 | 0.024 | 8.2220 | 0.877 |

. | . | | . | . | | 15 | -0.055 | -0.063 | 8.4281 | 0.905 |

.*| . | | .*| . | | 16 | -0.075 | -0.131 | 8.8274 | 0.920 |

. | . | | . | . | | 17 | -0.045 | -0.020 | 8.9783 | 0.941 |

.*| . | | . | . | | 18 | -0.141 | -0.062 | 10.516 | 0.914 |

. | . | | . | . | | 19 | 0.048 | 0.051 | 10.700 | 0.934 |

.*| . | | .*| . | | 20 | -0.189 | -0.169 | 13.699 | 0.845 |

c. Heteroskedasticity Test: Breusch-Pagan-Godfrey | ||||

Null hypothesis: Homoskedasticity | ||||

F-statistic | 0.776885 | Prob. F(13,29) | 0.6768 | |

Obs*R-squared | 11.10702 | Prob. Chi-Square (13) | 0.6019 | |

Scaled explained SS | 3.289561 | Prob. Chi-Square (13) | 0.9967 | |

Test Equation: Dependent Variable: RESID^2 Method: Least Squares | ||||

Variable | Coefficient | Std. Error | t-Statistic | Prob. |

C | 0.145940 | 0.082453 | 1.769991 | 0.0872 |

LGREQ (-1) | 0.004978 | 0.020394 | 0.244107 | 0.8089 |

LGREQ (-2) | -0.001124 | 0.019551 | -0.057508 | 0.9545 |

LTURDEF | -0.042604 | 0.062723 | -0.679229 | 0.5024 |

LTURDEF (-1) | -0.003258 | 0.065066 | -0.050071 | 0.9604 |

LTURDEF (-2) | 0.013804 | 0.065435 | 0.210964 | 0.8344 |

LTURDEF (-3) | 0.065469 | 0.072239 | 0.906291 | 0.3723 |

LTURDEF (-4) | -0.086550 | 0.057277 | -1.511079 | 0.1416 |

LSPILL | 0.025303 | 0.086740 | 0.291715 | 0.7726 |

LSPILL (-1) | -0.132959 | 0.133440 | -0.996398 | 0.3273 |

LSPILL (-2) | 0.191612 | 0.124029 | 1.544897 | 0.1332 |

LSPILL (-3) | -0.060564 | 0.127879 | -0.473606 | 0.6393 |

LSPILL (-4) | -0.041366 | 0.092890 | -0.445322 | 0.6594 |

LDOGS (-2) | -0.005998 | 0.005652 | -1.061209 | 0.2974 |

R-squared | 0.258303 | Mean dependent var | 0.029612 | |

Adjusted R-squared | -0.074182 | S.D. dependent var | 0.034193 | |

S.E. of regression | 0.035439 | Akaike info criterion | -3.584760 | |

Sum squared resid | 0.036421 | Schwarz criterion | -3.011346 | |

Log likelihood | 91.07234 | Hannan-Quinn criter. | -3.373303 | |

F-statistic | 0.776885 | Durbin-Watson stat | 2.030562 | |

Prob(F-statistic) | 0.676826 |

d. Histogram Normality Test

Short run/adjustment parameters of equation 1’a

Variable | Coefficient | Std. Error | t-Statistic | Prob. |

D (LGREQ (-1)) | 0.263102 | 0.104990 | 2.505961 | 0.0183 |

D(LTREQ) | 0.406270 | 0.123577 | 3.287585 | 0.0027 |

D (LTREQ (-1)) | -0.302666 | 0.122071 | -2.479434 | 0.0194 |

D (LTREQ (-2)) | -0.113919 | 0.115486 | -0.986424 | 0.3324 |

D (LTREQ (-3)) | -0.318634 | 0.109261 | -2.916266 | 0.0069 |

D(LSPILL) | 0.333725 | 0.448425 | 0.744217 | 0.4629 |

D (LSPILL (-1)) | 0.324692 | 0.416054 | 0.780408 | 0.4417 |

D (LSPILL (-2)) | -2.001707 | 0.435707 | -4.594159 | 0.0001 |

D (LSPILL (-3)) | -2.674004 | 0.482169 | -5.545782 | 0.0000 |

LDOGS (-2) | 0.081285 | 0.013554 | 5.996990 | 0.0000 |

α | -0.794982 | 0.096704 | -8.220775 | 0.0000 |

R-squared | 0.760174 | Mean dependent var | -0.044920 | |

Adjusted R-squared | 0.685229 | S.D. dependent var | 0.348491 | |

S.E. of regression | 0.195519 | Akaike info criterion | -0.210156 | |

Sum squared resid | 1.223285 | Schwarz criterion | 0.240384 | |

Log likelihood | 15.51834 | Hannan-Quinn criter. | -0.044010 | |

Durbin-Watson stat | 2.205929 |

a. Correlogram – Q Statistics | ||||||

Autocorrelation | Partial Correlation | AC | PAC | Q-Stat | Prob* | |

.*| . | | .*| . | | 1 | -0.109 | -0.109 | 0.5502 | 0.458 |

***| . | | ***| . | | 2 | -0.367 | -0.384 | 6.9198 | 0.031 |

. | . | | . | . | | 3 | 0.053 | -0.053 | 7.0541 | 0.070 |

. | . | | .*| . | | 4 | 0.037 | -0.123 | 7.1222 | 0.130 |

. | . | | . | . | | 5 | 0.027 | 0.019 | 7.1586 | 0.209 |

.*| . | | .*| . | | 6 | -0.092 | -0.132 | 7.5987 | 0.269 |

. | . | | . | . | | 7 | -0.011 | -0.023 | 7.6053 | 0.369 |

. | . | | .*| . | | 8 | 0.026 | -0.075 | 7.6431 | 0.469 |

.*| . | | .*| . | | 9 | -0.083 | -0.120 | 8.0312 | 0.531 |

. |*. | | . |*. | | 10 | 0.170 | 0.141 | 9.7255 | 0.465 |

. | . | | . | . | | 11 | 0.007 | -0.019 | 9.7286 | 0.555 |

.*| . | | . | . | | 12 | -0.108 | 0.014 | 10.457 | 0.576 |

. | . | | . | . | | 13 | 0.066 | 0.046 | 10.740 | 0.633 |

.*| . | | .*| . | | 14 | -0.125 | -0.163 | 11.775 | 0.624 |

. |*. | | . |*. | | 15 | 0.125 | 0.138 | 12.857 | 0.613 |

. | . | | .*| . | | 16 | -0.046 | -0.142 | 13.009 | 0.672 |

. | . | | . |*. | | 17 | -0.007 | 0.130 | 13.012 | 0.735 |

. |*. | | . | . | | 18 | 0.140 | 0.068 | 14.525 | 0.694 |

**| . | | .*| . | | 19 | -0.242 | -0.189 | 19.230 | 0.442 |

.*| . | | .*| . | | 20 | -0.094 | -0.136 | 19.978 | 0.459 |

b. Correlogram Squared Residuals | ||||||

Autocorrelation | Partial Correlation | AC | PAC | Q-Stat | Prob* | |

. |** | | . |** | | 1 | 0.240 | 0.240 | 2.6524 | 0.103 |

. |** | | . |** | | 2 | 0.263 | 0.218 | 5.9141 | 0.052 |

. |*. | | . | . | | 3 | 0.083 | -0.020 | 6.2498 | 0.100 |

.*| . | | .*| . | | 4 | -0.067 | -0.153 | 6.4756 | 0.166 |

. | . | | . |*. | | 5 | 0.045 | 0.081 | 6.5800 | 0.254 |

. | . | | . | . | | 6 | -0.020 | 0.016 | 6.6004 | 0.359 |

.*| . | | **| . | | 7 | -0.196 | -0.246 | 8.6596 | 0.278 |

.*| . | | . | . | | 8 | -0.088 | -0.027 | 9.0907 | 0.335 |

.*| . | | . |*. | | 9 | -0.081 | 0.087 | 9.4614 | 0.396 |

. | . | | . | . | | 10 | -0.037 | 0.003 | 9.5398 | 0.482 |

. | . | | . | . | | 11 | 0.012 | -0.046 | 9.5484 | 0.571 |

. | . | | . | . | | 12 | -0.039 | -0.014 | 9.6438 | 0.647 |

.*| . | | .*| . | | 13 | -0.108 | -0.093 | 10.398 | 0.661 |

. | . | | . | . | | 14 | 0.017 | 0.035 | 10.417 | 0.731 |

.*| . | | .*| . | | 15 | -0.092 | -0.081 | 11.007 | 0.752 |

. | . | | . | . | | 16 | -0.007 | -0.004 | 11.011 | 0.809 |

. | . | | . | . | | 17 | -0.049 | -0.031 | 11.187 | 0.847 |

. | . | | . | . | | 18 | -0.063 | -0.013 | 11.496 | 0.872 |

. | . | | . | . | | 19 | -0.003 | 0.004 | 11.497 | 0.906 |

. | . | | .^{*}| . | | 20 | -0.048 | -0.075 | 11.686 | 0.926 |

c. Heteroskedasticity Test: Breusch-Pagan-Godfrey | ||||

Null hypothesis: Homoskedasticity | ||||

F-statistic | 0.822708 | Prob. F(14,28) | 0.6405 | |

Obs*R-squared | 12.53280 | Prob. Chi-Square(14) | 0.5636 | |

Scaled explained SS | 4.554527 | Prob. Chi-Square(14) | 0.9911 | |

Test Equation: Dependent Variable: RESID^2 | ||||

Method: Least Squares | ||||

Variable | Coefficient | Std. Error | t-Statistic | Prob. |

C | 0.071166 | 0.094623 | 0.752095 | 0.4583 |

LGREQ (-1) | 0.004145 | 0.023968 | 0.172955 | 0.8639 |

LGREQ (-2) | -0.017095 | 0.021541 | -0.793616 | 0.4341 |

LTREQ | -0.004496 | 0.027195 | -0.165338 | 0.8699 |

LTREQ (-1) | -0.007570 | 0.038563 | -0.196304 | 0.8458 |

LTREQ (-2) | 0.011505 | 0.037085 | 0.310245 | 0.7587 |

LTREQ (-3) | -0.013822 | 0.034238 | -0.403693 | 0.6895 |

LTREQ (-4) | -0.015139 | 0.025123 | -0.602609 | 0.5516 |

LSPILL | -0.074364 | 0.093096 | -0.798786 | 0.4311 |

LSPILL (-1) | -0.004747 | 0.134701 | -0.035244 | 0.9721 |

LSPILL (-2) | 0.047551 | 0.128318 | 0.370573 | 0.7137 |

LSPILL (-3) | 0.063257 | 0.140134 | 0.451401 | 0.6552 |

LSPILL (-4) | -0.093604 | 0.102459 | -0.913574 | 0.3687 |

D(LTRPOP) | 2.313702 | 3.823015 | 0.605203 | 0.5499 |

LDOGS (-2) | -0.004551 | 0.007723 | -0.589248 | 0.5604 |

R-squared | 0.291460 | Mean dependent var | 0.028448 | |

Adjusted R-squared | -0.062809 | S.D. dependent var | 0.037687 | |

S.E. of regression | 0.038853 | Akaike info criterion | -3.389408 | |

Sum squared resid | 0.042267 | Schwarz criterion | -2.775036 | |

Log likelihood | 87.87227 | Hannan-Quinn criter. | -3.162847 | |

F-statistic | 0.822708 | Durbin-Watson stat | 2.186478 | |

Prob(F-statistic) | 0.640504 |

d. Histogram Normality Test

Greece: The Insignificance of the Income Variable Dependent Variable LGREQ

Variable | Coefficient | Std. Error | t-Statistic | Prob. |

LTURDEF | 0.480999 | 0.322072 | 1.493454 | 0.1458 |

LSPILL | 1.490759 | 0.275273 | 5.415570 | 0.0000 |

DLGDP | 0.620561 | 1.129768 | 0.549282 | 0.5869 |

C | -3.611879 | 0.356617 | -10.12816 | 0.0000 |

EC = LGREQ - (0.4810*LTURDEF + 1.4908*LSPILL + 0.6206*DLGDP -3.6119) | ||||

LDOGS (-2) | 0.078645 | 0.013715 | 5.734325 | 0.0000 |

α* | -0.755135 | 0.097530 | -7.742606 | 0.0000 |

Redundant Variable Test | ||||

Null hypothesis: DLGDP | ||||

Specification: LGREQ (-1) LGREQ (-2) LTURDEF LSPILL | ||||

LSPILL (-1) LSPILL (-2) LSPILL(-3) LSPILL(-4) DLGDP LDOGS(-2) C | ||||

Redundant Variables: DLGDP is not significant | ||||

Value | df | Probability | ||

t-statistic | 0.563461 | 30 | 0.5773 | |

F-statistic | 0.317489 | (1, 30) | 0.5773 | |

F-test summary: | ||||

Sum of Sq. | df | Mean Squares | ||

Test SSR | 0.014029 | 1 | 0.014029 | |

Restricted SSR | 1.339672 | 31 | 0.043215 | |

Unrestricted SSR | 1.325643 | 30 | 0.044188 |

According to the latest (August 2019) Ahvalnews report, Turkey can employ or plans to employ the following in the Aegean and Eastern Mediterranean:

Ada-class corvettes, Tuzla-class patrol ships and Kılıç-class fast patrol boats, which were either designed or constructed in Turkey.

The new domestically produced TB-2 Bayraktar armed drones, recently acquired by the Turkish Navy.

The Turkish Navy has also modernised the Gabya and Barbaros-type frigates and strengthened its naval air force with new helicopters, maritime patrol aircraft and unmanned aerial vehicles. 4. Construction is underway of the TCG Anadolu, the first of two multi-purpose amphibious assault ships, the TCG Istanbul, the first vessel of four Istif-class frigates, and the construction of the first three of six Reis-class submarines equipped with air-independent propulsion systems and TCG Ufuk intelligence ships.

The first TF-2000 air-defence warfare destroyer is scheduled to be put into service in 2027. The design process of the seven-ship project is still in progress, with final tests being conducted on important components designed for the ship - the ÇAFRAD Multipurpose Phase Index Radar and Atmaca Navy Missiles.

A cruise missile for use against land targets, the Gezgin, is still in the development phase.