*4.1. Bootstrap Full-Sample Causality Test*

In our study, we relied on the complete Granger non-causality test, namely the bivariate autoregressive vector (VAR). Given the sensitivity of this method of analysis for a certain period, we set out to use the variables for a given period of time, so that the results express a causality between ME and GDP, not a constant causal relationship [39].

In the VAR model, the time series data are assumed to be stationary in the Granger causality test. Therefore, according to the statistical data, including the likelihood ratio (LR), the Wald test and the Lagrange multiplier (LM) tests will not be in the definitive estimation in VAR models. The modified test is proposed to estimate the processing variables of the augmented VAR model, applying when the time series is I (1) [40,41].

*RB* and *LR* can be explained by the VAR framework for two variables as follows:

$$Y\_t = \phi\_0 + \phi\_1 Y\_{t-1} + \dots + \phi\_p Y\_{t-p} + \varepsilon\_{t,\*} \ t = 1, 2, \dots, T \tag{1}$$

By splitting *yt* into two sub-vectors, *yt* = (*y*1*t*, *y*2*t*) , thus the above equation can be rewritten as follows:

$$
\begin{bmatrix} GDP\_{1t} \\ ME\_{2t} \end{bmatrix} = \begin{bmatrix} \phi\_{10} \\ \phi\_{20} \end{bmatrix} + \begin{bmatrix} \phi\_{11}(L)\phi\_{12}(L) \\ \phi\_{21}(L)\phi\_{22}(L) \end{bmatrix} \begin{bmatrix} GDP\_{1t} \\ ME\_{2t} \end{bmatrix} + \begin{bmatrix} \varepsilon\_{1t} \\ \varepsilon\_{2t} \end{bmatrix} \tag{2}
$$

where φ*ij*(*L*) = *p* +1 *k*=1 <sup>φ</sup>*ij*,*kL<sup>k</sup>*, *i, j* <sup>=</sup> 1, 2 and *<sup>L</sup>* is the lag operator defined as *Lkxt* <sup>=</sup> *xt*−*k*.

Starting with Equation (2), we analyze the causality hypothesis between ME and GDP, taking into account certain restrictions, such as for *k* = 1, 2, ..., *p*, we consider that ME does not produce effects. Similarly, the inverse causal hypothesis is tested for *k* = 1, 2, ..., *p*. Following the above analysis, we mention that it is not conclusive whether ME has an impact on GDP and vice versa.

### *4.2. Parameter Stability Test*

If the Granger test is not conclusive, the final results can be considered null, and the causal correlations are considered unstable [39]. Therefore, before using the rolling window estimation, the parameter constant over a short period of time was tested using Mean-F, Exp-F and Sup-F tests [42] and the long-term parameter stability of the VAR model above, applying the Lc Test of Nyblom [43] and Hansen [44]. Considering these data, when the basic variables have been cointegrated, the VAR model at the first analysis may be wrongly predicted, unless the error correction is allowed. For clearer and more precise conclusions, the fully modified Ordinary Least Squares estimator (FM-OLS) proposed by Phillips [45] and Hansen [44] was used. In these analyzes, testing the constant parameters in the VAR model is compared with any possible alternative a unique structural change.

### *4.3. Rolling-Window Granger Estimation*

In this analysis, the test used is adopted based on a modified estimation of the bootstrap [46]. Thus the complete sample is transformed into a sequence of variables Tl, i.e., τ-l + 1, τ-l, ..., *T* for τ = l, l + 1, ..., *T*. Non-test. Modified causality based on RB is then estimated in each sub-sample, but not in the complete sample. By calculating the *p*-values of the bootstrap of the observed LR statistics that are carried out through *T*-l sub-samples, some changes can be observed between ME and GDP.

The causal relation index is observed by calculating the effect value, and the GDP on the impact on the ME is mentioned as the average of all derived bootstrap estimates from the formula, with *N*−<sup>1</sup> *b p <sup>N</sup>*−<sup>1</sup> <sup>φ</sup>ˆ<sup>∗</sup> 21,*<sup>k</sup>* with *Nb* the number of bootstrap repetitions. Similarly, the impact of MEs on GDP is measured by the formula *N*−<sup>1</sup> *b p <sup>N</sup>*−<sup>1</sup> <sup>φ</sup>ˆ<sup>∗</sup> 12,*k* . Both φˆ<sup>∗</sup> 21,*<sup>k</sup>* and <sup>φ</sup>ˆ<sup>∗</sup> 12,*<sup>k</sup>* are bootstrap estimates from the VAR models in Equation (2). The 90% confidence intervals are provided, in which the lower and upper bounds are the same as the 5th and 95th quantiles of φˆ<sup>∗</sup> 21,*<sup>k</sup>* and <sup>φ</sup>ˆ<sup>∗</sup> 12,*<sup>k</sup>* respectively [46].

#### **5. Data and Empirical Results**

In this study, we base annual data from 1980 to 2018 to examine the nexus between ME and GDP in Romania. The dataset of GDP is present in current prices (purchasing power parity, billions of international dollars), which is sourced from the International Monetary Fund (IMF). The ME data, in millions of US dollars, current prices, converted at the exchange rate for the given year, is from SIPRI Arms Industry Database. The trend of these two variables is shown in Figure 1.

**Figure 1.** The trend of ME and GDP.

To test the data stationarity test, we performed the Augmented Dickey–Fuller test (ADF, 1979), the Phillips–Perron test (PP, 1988) and the Kwiatkowski Phillips Schmidt Shin test (KPSS, 1992). Table 1 mentioned the conclusions in which it was found that both ME and GDP are stationary processes in the first differences, which suggests that both are a process I (1).


**Table 1.** Unit root tests.

Source: Authors' calculations. Notes: \*, \*\* and \*\*\* denote significance at 10, 5, and 1 percent, respectively.

We choose the optimal lag-lengths of ME and GDP 2 based on the Schwarz Criterion (SC). Therefore, the next Table 2 shows the full-sample causality results. We can infer that there is a relevant relation running from ME to GDP while GDP has no significant impact on ME over the full-sample perspective. This finding is basically consistent with Bra¸soveanu [47].

Taking into account the above, we argue that the estimation of a complete causality neglects the unknown structural changes. The structural changes in the ME and the GDP consequently have unstable and meaningless causal relationships of the analyzed sample. Therefore, we use Sup-F, Mean-F, Exp-F, and Lc tests to investigate whether this result is supported by the constancy of the parameters, and the results are presented in Table 3. The results suggest that there is a strong change once in ME, GDP and VAR system at 1% level. The conclusion is also rejected considering the *p* values of the second and third lines, which indicates equations from ME, GDP and VAR system could evolve gradually. The results of the Lc test do not demonstrate the consistency of the parameters for the VAR model.


**Table 2.** Full-sample Granger-causality tests.


**Table 3.** Stability tests.

Source: Authors' calculations Notes: We calculate p-values using 10,000 bootstrap repetitions. \*, \*\* and \*\*\* denote significance at 10, 5 and 1 percent, respectively.

We will take into account the results obtained from the estimation of the model based on the rolling window, which gives us a greater accuracy of the data. Considering the null hypothesis, we concluded that there was no causal relationship between ME and GDP. According to Pesaran and Timmerman [48], we choose a period of 15 years in adopting the estimation of the rolling window so that we have greater clarity and accuracy of the data. Moreover, different time periods will be used, such as 18–20-year tests and the impact of ME on GDP is estimated and vice versa, the results coinciding with those following the 15-year analysis. From this, we deduce that the conclusions are the same regardless of the period. Figure 2 shows the starting system of p values of LR statistics using GDP as dependent variable in Romania. Figure 3 shows the estimates of the bootstrap test for the sum of the running coefficient, measuring the impact of ME on GDP. Exceeding the zero value of the prominent line represents a positive impact, otherwise, the effect is negative.

Specifically, the null hypothesis is rejected from 1996 to 2006. Figure 3 indicates that in 1996–1999 and 2002–2004, the ME exerted a negative effect on the GDP, while in the sample of 1999–2002 and 2004–2006, the relationship between the two series was positive. From 1996 to 1999, the economy in Romania experienced a continued three-year decline [49]. The decline has been largely accounted for the severe reduction in fixed investment and private consumption. The Kosovo crisis had a modest impact on increasing ME. However, parliament still approved a tough austerity budget, including much higher excise taxes on fuels and property taxes. In this situation, military spending may adversely affect investments, savings, human capital, infrastructure programs, and market-oriented technological innovation. This verifies the ME can impede sustainable economic development by crowding-out private investment.

Starting with 1999, the decrease in purchasing power and the deterioration of the economic well-being in the country, led to several demonstrations among the population in the country. One of these was triggered by the miners' dissatisfaction with the economic situation and the employment prospects in the Jiu Valley, the unemployment rate [50]. The government increases ME can generate economic benefits because it provides security, which promotes a stable social and economic environment. It also contributes to improving the educational level of the workers and may act as a stabilizing influence in society by expenditure on defense training. As Bras, oveanu [47] has mentioned,

the war, corruption, security and defense policy of a state strongly impact the sustainable development of a society. At that time, Romania made all the necessary steps to join the EU. On 18 June 1999, a new national security and security strategy was adopted by the Supreme Defense Council of the country where the idea of EU membership was one of the main priorities for Romania. Increasing the ME shows the improvement of comprehensive national power, which is prone to be accepted by the EU. Overall, the rise in ME in this period (1999–2002) stimulated economic development.

**Figure 2.** ME does not Granger cause GDP.

The causality from ME to GDP was negative in 2002–2004. Having inherited Soviet-era equipment, Romania could postpone major military purchases. Combined with reductions in defense budgets, Romania has reduced spending on military maintenance, operations and training for the majority of the armed forces. For example, Romanian Chief-of-Staff General acknowledges that 70 percent of Romania's air force pilots were not operational due to budgetary constraints [51]. Since mid-2000, GDP has remained robust, inflation and unemployment have been steadily declining, and it is imperative to revitalize the economy in Romania [52].

Romania became a member of NATO in 2004. The Land Forces have overhauled their equipment, which participates in a peacekeeping mission in Afghanistan and Iraq, together with the other NATO countries. They are forced to develop more professional elements within their armed forces that are more suited to deployment abroad in multinational military operations, Cottey et al. [51]. Furthermore, in order to complete preparations for EU accession, Romania makes efforts to improve legitimation and regulation on the military. ME may be considered a tool of fiscal policy and can be increased to stimulate demand. Moreover, the trade of EU exports to Romania has significantly increased [52]. Therefore, we can conclude that Romania achieves expansion of aggregate demand and exports through the fiscal policy of ME. The accession to NATO and planning to join the EU also provide a more open and stable economic environment. This contributes to the positive relationship between ME and GDP in 2004–2006.

In Figure 4, we observe that the null hypothesis is not considered rejected for all the analyzed periods, which means that the GDP has no significant effects on the ME in Romania. ME can be regarded not necessarily as a purely economic problem, but rather as a strategic political, social, economic and psychological effect [53]. In such studies, Gleditsch and Njølstad [54], Intriljgator [55] have also mentioned such correlations between the two variables and strategies. Romania has done its utmost to join NATO and the EU at the end of the Cold War, thus supporting US operations in Iraq and Afghanistan [56].

**Figure 4.** The null that GDP does not Granger cause ME.

#### **6. Conclusions and Discussion**

In this paper, we have tried to investigate the relationship between ME and GDP in Romania since 1980. In order to obtain conclusive results on the correlation between the two variables, the Granger non-causality test with full sample and rolling window estimation was used in the analysis. The results show that the ME would have both positive and negative effect on GDP in Romania. In specific, the impact of ME on GDP was negative during 1996–1999 and 2002–2004. It can be inferred that during the periods of turbulence in neighboring countries, the increase of ME would crow out private

investment and consume, which is harmful to GDP. From 1999 to 2002 and 2004–2006, GDP was positively associated with ME. We can conclude that, in the period of domestic turmoil and participation in NATO, rising ME contributes to stabilize the domestic environment and thus stimulate the sustainable development of the economy. This finding is in accordance with the Dunne et al. [11], who propose that the positive effect of ME on GDP is particularly significant in less developed countries, where war and turmoil are major obstacles to sustainable development. In specific, raising ME can protect society from violence and invasion of other countries or groups. Moreover, training in the armed forces can improve the quality of human capital, which makes them more competitive when they are employed. Therefore, the improvement of the employment rate is associated with higher economic output. However, the empirical result indicates that GDP does not have a significant effect on ME, which is contrary to the result of Su et al. [29]. It can be attributed that the ME in Romania is lack of independence, which would be affected by the policy of NATO, EU and US.

Therefore, the implication for the policymakers can be summarized as: The government should adjust its ME according to the military or political environment at home and abroad to promote economic development. Excessive ME in peacetime will squeeze out private investment and civilian resources, which is harmful to sustainable development. While in the period of chaos, increasing ME can maintain a safe social order and increase economic output. Moreover, mastering the independence of military policy should be considered by authorities. The analysis proposal could contribute both to the decision-making at the governmental level and the allocation of military sector funds in the current context, as well as to the foundation of a political decision-making process aimed at increasing the efficiency of the expenditures by the executive power of a state.

Considering the nonlinear nexus between ME and sustainable economic development, future research can further explore if there exists threshold value, before and after which, ME will have a different effect on output. In addition, the limitation of this article is that the conclusion is confined to the context of Romania, hence, it cannot be extended to the general situation. A future study could provide a more general analysis of this topic.

**Author Contributions:** Conceptualization, R.T., O.R.G., Z.-Z.L., O.R.L and A.A.G.; data curation, R.T., O.R.G., Z.-Z.L., O.R.L. and A.A.G; formal analysis, R.T., O.R.G., Z.-Z.L., O.R.L. and A.A.G.; funding acquisition R.T., O.R.G., Z.-Z.L., O.R.L. and A.A.G.; investigation, R.T., O.R.G., Z.-Z.L., O.R.L. and A.A.G.; methodology, R.T., O.R.G., Z.-Z.L. and O.R.L.; project administration, R.T., O.R.G., Z.-Z.L., O.R.L. and A.A.G.; resources, R.T., O.R.G., Z.-Z.L. and O.R.L.; software, R.T., O.R.G., Z.-Z.L. and O.R.L.; supervision R.T., O.R.G., Z.-Z.L. and O.R.L.; validation, R.T., O.R.G., Z.-Z.L., O.R.L and, A.A.G.; visualization, R.T., O.R.G., Z.-Z.L. and O.R.L.; writing—original draft, R.T., O.R.G., Z.-Z.L., O.R.L and, A.A.G.; writing—review & editing, R.T., O.R.G., Z.-Z.L., O.R.L. and, A.A.G. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** This work was cofinanced from the European Social Fund through Operational Programme Human Capital 2014–2020, project number POCU/380/6/13/125015 "Development of entrepreneurial skills for doctoral students and postdoctoral researchers in the field of economic sciences".

**Conflicts of Interest:** The authors declare no conflict of interest.
