**4. Results**

Hypothesis H1 was first evaluated in order to determine the appropriate final model for the data sample used. In Tables 4 and 5 are presented the estimations of the ARDL model (1.1) through the three estimators for exports (EXP) and imports (IMP). In the basic vector, the variables of labor force (LABOR), the added value of agriculture, forestry, and fisheries (AGRI), and the expenditure of research and development (RD) were considered as factors influencing the exports/imports.


**Table 4.** Estimation of the Autoregressive Distributed Lag (ARDL) model (1.1) for exports.


**Table 4.** *Cont*.

Source: own processing. Note: Hausman test: MG vs. PMG: Since the p-value = 0.4865 is greater than all significance thresholds, the null hypothesis that the PMG estimator is the preferred model to form the relationship between variables was accepted. \*\*\* and \*\* indicate statistical significance at a threshold of 1% and 5%. In the above table, the following abbreviation were used: Dynamic fixed effects (DFE), pooled mean group (PMG), mean group (MG), exports (EXP), labor force (LABOR), gross added value of agriculture, forestry, and fisheries (AGRI), expenditure on research and development (RD), and error correction term (ECT).


**Table 5.** Estimation of the ARDL model (1.1) for imports.

Source: own processing. Note: Hausman test: MG vs. PMG: Since the p-value = 0.7162 is greater than all significance thresholds, the null hypothesis that the PMG estimator is the preferred model to form the relationship between variables was accepted. \*\*\* and \*\* indicate statistical significance at a threshold of 1% and 5%.In the above table, the following abbreviation were used: Dynamic fixed effects (DFE), pooled mean group (PMG), mean group (MG), imports (IMP), labor force (LABOR), gross added value of agriculture, forestry and fisheries (AGRI), expenditure on research and development (RD), and error correction term (ECT).

First of all, it was noticed that the Error Correction Term (ECT) is negative and strongly significant for all models. Thus, the modeling technique is justified and the specification of the models is validated. Moreover, for the export equation, the associated coe fficient varies between about 0.13 and 0.4, suggesting a low adjustment rate (similarly to the case for the import equation, where the adjustment rate varies between about 0.11 and 0.47). Second, the Hausman test, by which we discriminated between the estimator MG and PMG, suggests that the most appropriate of these is the PMG estimator for both the dependent variable exports and imports (p-value = 0.486 for exports, p-value = 0.716 for imports). Consequently, the working hypothesis H1 is accepted: The PMG estimator is the preferred one for modeling the relationships between variables. Countries have a common long-term tendency with respective short-term heterogeneities. The common long-term trend can be explained by the e fforts of all states to increase trade openness, ultimately stimulating the economic growth. For example, at the European Union level, trade policies focus on coordinating states towards a common trajectory, which involves increasing the trade flow. In tandem, the heterogeneities in the short term can be given by the di fferences in the commercial structures of the countries, both in terms of exports and imports of food products, or respectively of the national macroeconomic policies. As stated in the methodological section, to evaluate the validity of the other working hypotheses, the results of the PMG estimator were analyzed. Regarding the export model, the following were observed:


The results of the import model show the following:


The robustness analysis of the basic model of exports and imports was made by introducing the three additional variables closely related to the concept of sustainable development (environmental quality); namely, the forest area (FOREST), the energy consumption of fossil fuels as a share of total energy (ENG), and the renewable energy consumption (RENEW). Considering that the ARDL technique provides consistent results for mixed independent variables (i.e., first order (I(1)), as well as

stationary (I(0)), no prior analysis of the stationarity of these factors (additional ones included in the study) is required. However, it should be mentioned that due to the availability of data, the analysis period of the models that include the FOREST variable is 1996–2016, and for the other two variables, the analyzed period is 1996–2015. The results of the models estimated through the PMG estimator are presented in Tables 6 and 7. The additional independent variables were included one at a time in the vector of the basic variables, so that finally, in the last column (column (4)) of the two tables, all of the independent variables are considered. On the one hand, for the export model, it was observed that ECT is negative and strongly significant in all models, validating once again the chosen technique and specification of the models. Moreover, its magnitude is comparable from one model to another, also indicating a relatively low rate of adjustment to long-term equilibrium (see Table 6). Regarding the independent variables, overall, it can be observed that the statistical significance and the signs of the coe fficients of the variables in the basic vector do not change with the inclusion of the additional factors. Moreover, if the last model, where all of the exogenous factors were included (see column (4) of Table 6), is considered, it is to be noticed that the additional factors have a positive impact on exports in the long term. In contrast, in the short term, it is to be mentioned that only the consumption of energy and renewable energy significantly influences exports, the first variable in the positive sense and the second in the negative sense. Hypothesis H5 is accepted for exports in both the short and long terms. Hypothesis H6 is invalidated at the level of the analyzed sample, highlighting a direct link between the renewable energy consumption and exports. On the other hand, for the model of imports, the associated coe fficient of ECT is also strongly statistically and negatively significant, suggesting once again that the model is well specified and has a high accuracy. Contrary to the main model, it is observed that the inclusion of the variable of forest area or the inclusion of all variables in the equation (see column (1) and column (4) of Table 7) brings a significant gain to the long-term coe fficient associated with the variable of labor force (the sign of the associated coe fficient is positive). In addition, if the most complete model is analyzed (see column (4) in Table 7), it must be noted that the variables of the main vector (AGRI and RD) keep their positive sign for the associated coe fficient and, respectively, the high statistical significance. In addition, the FOREST variable has a positive influence on imports, while RENEW has a negative impact on the value of imports. For the ENG variable, the associated coe fficient does not have statistical significance. In the case of imports, both working assumptions related to sustainability factors are rejected (H5 and H6). The short-term coe fficients are statistically significant for the variables AGRI, RD, and ENG, and the positive sign illustrates that all of these macroeconomic factors contribute to the increase of the value of imports. In total, for the models of both exports and imports, the inclusion of additional factors in the main equation does not significantly affect the results of the initial models, so it can be said that that they have a high robustness.




**Table 6.** *Cont*.

Source: Own processing. Standard error in round brackets. \*\*\* and \*\* indicate statistical significance at a threshold of 1% and 5%. In the above table, the following abbreviation were used: Exports (EXP), labor force (LABOR), gross added value of agriculture, forestry, and fisheries (AGRI), expenditure on research and development (RD), forest area (FOREST), fossil fuel energy consumption (ENG) and renewable energy consumption (RENEW), and error correction term (ECT).



**Table 7.** *Cont*.


Source: Own processing. Standard error in round brackets. \*\*\* and \*\* indicate statistical significance at a threshold of 1% and 5%. In the above table, the following abbreviation were used: Imports (IMP), labor force (LABOR), gross added value of agriculture, forestry, and fisheries (AGRI), expenditure on research and development (RD), forest area (FOREST), fossil fuel energy consumption (ENG) and renewable energy consumption (RENEW), and error correction term (ECT).
