Descriptive statistical information is needed to get common info about the data. In this respect, descriptive statistical information of data is presented in Table-1. Accordingly, the fact that mean and median values presented in Panel-A are the same or close to each other indicates that this distribution is symmetrical, which gives a priori information that the variables show a normal distribution. The fact that mean and median values of the variables are generally close, indicates that the distribution is symmetrical and normally distributed. As a matter of fact, since the probability values of the Jarque-Bera test are p > 0.05, it can be said that the variables excluding the findev show a normal distribution. lnpgdp has the highest standard deviation among the variables considered in the study. As a matter of fact, the biggest difference between the maximum and minimum values among the existing variables belongs to this variable. Correlation analysis results are shown in Panel-B. Accordingly, CO2 emissions in Turkey are positively related to all other independent variables.
In order to have more information about the structure of the variables, a general evaluation can be made about the stationary condition by examining the time path graphs given in Table-A1 in Appendix. Graphs of six different variables of the country generally tend to increase and decrease. Since most of the series is trending upwards or downwards, if the model is estimated without trend, it may not show some important characteristics of the data. For this reason, it has been tested whether data is stationary or not over the fixed and trended model. The results of unit root test are given in Table-A2.
According to ADF unit root test, all of the variables are stationary at the level of 1%, 5% and 10% significance I(1) at the first differences in the fixed and trended model. Since the series are stationary around the deterministic trend, they may be affected by the slope parameter or structural breaks in the constant term. Therefore, performing a traditional ADF unit root test without considering such structural breaks in the model may cause false results and decrease the predictive power of the model.
Therefore, the ADF unit root test with structural break is applied to determine structural break, as the majority of the series included trends. In addition, the unit root test with structural break is used with the presumption that country may have experienced important economic, social and political developments in the relevant period. According to the results of the ADF test with structural breaks, all series are stable at different significance levels, level I(0) and/or at first differences I(1).
At this stage, it is tried to determine whether there is a long-term connection among the variables or not through the ARDL bounds testing. This test is based on F-statistics or Wald test and the null and alternative hypotheses are given below. (Tursoy and Faisal, 2018).
H0: φ1 = φ2 = φ3 = φ4 = φ5 = φ6 = 0 (There is no cointegration)
H1: φ1 ≠ φ2 ≠ φ3 ≠ φ4 ≠ φ5 ≠ φ6 ≠ 0 (There is a cointegration)
If the test statistic goes over the upper limit, it is concluded that there is a cointegration relationship. If it falls down within the limits, the test is inconclusive, in other words, test evidence is insufficient. If it is smaller than the lower limit, a long-term relationship cannot be found (Pan and Mishra, 2018, Jalil and Mahmud, 2009). Limit test rsults are shown in Table 2.
According to the test results H0 is rejected and H1 hypothesis is accepted that there is a cointegration link among the series. From this point of view, it can be said that there is a long-term link among CO2 and findev, per capita income, freight carried by road and rail, and enco in Turkey. The short and long run coefficient estimates and diagnostic test results obtained with the ARDL model are shown in Table-3.
After determining the cointegration link among variables, it is first necessary to predict the long-term link between variables. The results of the model whose mathematical representation has been made below are presented in Panel-A in Table-3.
Findev is defined by the financial development index has a statistically considerable impact on CO2. However, this effect is in the direction of increasing carbon emissions in Turkey. The financial ecosystem in Turkey transfers resources for expenditures or opportunities to increase carbon emissions to individuals/businesses through financial institutions and financial markets, which are its two sub-mechanisms. Gross Domestic Product (GDP) per capita has an increasing impact on CO2, and this effect is statistically significant. Accordingly, individuals in Turkey make consumption and investment expenditures in a way that will increase CO2 as their income increases. Although the increase in the amount of freight transported by rail has a reducing effect on CO2, this effect is statistically considerable. As a matter of fact, when the amount of freight transported in domestic transport modes is evaluated in terms of Turkey in 2019, it is seen that 92.6% was transported by road, 5.1% by rail, and 2.4% by air. In addition, while the amount of freight transported in transport modes in 2019 was compared with the year 2000, the share of road transport in Turkey decreased from 94–92.6%, and the share of railway transport decreased from 5.8–5.1%. It is observed that the share of air transport increased from 0.2–2.4%. Despite the increase in the amount of freight transported by road in Turkey, it reduces CO2 emissions. This negative effect obtained with the ARDL model was also confirmed in the cointegration regression models and it was proved to be a statistically significant result with the DOLS model. There is a broad consensus in the existing literature that the amount of cargo transported by road has an increasing effect on carbon emissions. At this point, this result needs to be explained for Turkey. It can be attributed to four reasons: First of all, the tightening of pollution and noise limits, especially in new vehicles today, greatly reduces local externalities related to loading ((Leonardi et.al. 2015), and accordingly, it provides less fuel consumption in new generation motor vehicles. As a matter of fact, approximately 47% of trucks and 34% of trucks in Turkey are under the age of 10 (TUIK, 2019). Secondly, it may be possible to change the type of fuel by switching to LPG, which is more environmentally friendly than gasoline and diesel. Thirdly, it can be said that, depending on the increase in awareness of logistics businesses/services, manufacturing companies have started to work with logistics companies in order to utilize from economies of scale within the scope of outsourcing. Accordingly, as a result of logistics companies increasing their vehicle capacity utilization rates (for example, ensuring that vehicles are returned in full by logistics companies), there may be a decrease in CO2 per load (logistics service CO2 emissions). Finally, infrastructure investments on the roads between production and consumption points (double roads, highways, tunnels, bridges, etc.) may have resulted in a reduction in CO2 due to the shortening of travel times (For example, a journey that used to take 8 hours between Istanbul and Izmir now takes only 3.5 hours). In addition, higher traffic fines (eg related to speeding and tachograph) and increased use of technology in the TSec (eg navigation) are among the reasons that reduce CO2.
In the next stage, a functional pattern of ARDL error correction model, which was developed to investigate the short-run dynamics between variables, is given below:
ECM refers the error correction term and the coefficient (λ) must be negatively signed and statistically considerable (Paul 2014). The (λ) indicates the rate of returning to long-run equilibrium after a short-run shock (Folarin and Asongu 2019). The results of estimation are shown in Panel-B in Table-3. Financial development and the level of road freight have similar effects on CO2 with long-term consequences. In Turkey, the enco has an increasing effect on CO2 in the short term, and this finding is in line with long-term results. The short-term impact of rail transport in Turkey is similar to the long-term findings. Moreover, it was observed that the deviations that occurred in the short term disappeared in the long term. As a matter of fact, the error correction coefficient has taken values in accordance with the theoretical expectations as expected (Narayan and Smyth, 2006). Accordingly, the effect of a short-term shock that causes deviations in the long-term balance may disappear in the coming period (year).
According to the diagnostic test results in Table-3 (Panel-C), it is understood that ARDL models do not have any autocorrelation (Breusch-Godfrey LM Test), changing variance (Breusch-Pagan-Godfrey) problem, and there is no modeling error (Ramsey Reset Test). In addition, it is understood that error term in the models developed has a normal distribution (Jarque-Together Normality Test). Diagnostic test results for ARDL models verify the stability, reliability, and validity of the generated models. Also, it was observed that long-run coefficients obtained with the ARDL bounds test according to CUSUM and CUSUM-Q graphs have not exceeded the critical values, so the coefficients belonging to the model are stable.
At this point, analysis with cointegration regression methods was performed in order to confirm the ARDL test results and thus to obtain robust empirical evidence for the results of the variables. Table-4 shows the long-term estimations of FMOLS, DOLS and CCR models for Turkey.
First of all, it should be noted that FMOLS and DOLS is the most functional model in terms of the significance of independent variables. It must be said that the findings from the cointegration regression analysis largely agree with the empirical findings from the ARDL bounds test. In this context, It was found evidence that findev has a positive effect in Turkey, which is consistent with the ARDL model findings. The cointegration regression model findings show that the increase in per capita income in Turkey increases the CO2, which is consistent with the empirical results from the ARDL analysis. Empirical results on the effect of enco on CO2 and the magnitude of this effect overlap with the ARDL model results, and strong evidence is obtained in this regard. The empirical findings related to the variables of the amount of freight transported by rail and road have significant reducing effect on CO2.