#### 4.1. Summary Statistics, Correlations and Stationarity Examination

The descriptive statistics of the variables are provided in

Table 2. The distributions of all stock market returns, as well as most of included commodities are negatively skewed. Thus, negative returns are more prevalent than positive returns, supporting a greater likelihood for very high losses. Kurtosis shows the thickness of the tail and highlights a high level of risk for selected stock markets, especially Spain and Italy. In addition, except EUR/CNY and Natural Gas Futures Contract 1, the Jarque–Bera test provides evidence that selected series are not normally distributed.

Figure 1 shows the evolution of the number of new cases due to COVID-19, whereas

Figure 2 reveals the progress of the number of new death due to COVID-19. There is noticed that USA registers the highest figures in this regard.

Figure 3 shows the evolution of stock market returns amongst the explored period. There is reinforced the significant volatility, especially for FTSE MIB on March 9, 2020 and March 12, 2020, as well as for Dow Jones Industrial Average on March 16, 2020. In the first two months of 2020, DAX declined by 10.2 percent, CAC 40 dropped by 11.2 percent, whereas FTSE 100 plunged 12.7%. In the same vein, Dow Jones throw down by 11 percent and S&P 500 by 8.6 percent. The Bucharest Stock Exchange also encountered instabilities and registered a decay of 8.6 percent [

89]. Capelle–Blancard and Desroziers [

90] contended that prior to February 21, stock markets disregarded the pandemic, but over February 23–March 20, the reaction to the rising number of diseased people was strong. As such, Mazur, Dang and Vega [

79] emphasized that the failure of stock quotes in March 2020 marked one of the major financial market collapses in history. Baiardi, et al. [

91] developed a three-regime switching model and concluded that in 2020 the most common state for the Dow Jones Industrial Average was turbulent.

Figure 4 reveals the evolution of oil futures. There is noticed the sharp decline registered on 21 April 2020.

Figure 5 shows the progress of Philadelphia Gold/Silver Index returns. Therewith, high volatility is prevailing.

Table 3 reveals the correlations among selected variables. There are acknowledged high negative correlations (below −0.7) between the number of new cases and new deaths due to COVID-19 in Italy and crude oil, WTI, as well as NYMEX light sweet crude oil. In case of the number of new cases and new deaths due to COVID-19 in China, there are not recorded high correlations with the included measures. Therewith, high positive correlations (over 0.7) are registered amongst the stock market returns, except SSE 100 (China).

Non-stationary variables lead to inadequate results, which means insignificant results. The verification of the stationarity of the selected data is performed through ADF stationarity test. This test is most commonly used to confirm the stationarity of a data series.

Table 4 shows the results of the ADF test at the level and in the first difference, as well as the level of integration of the stock indices.

The outcomes of ADF test provide support that all covered stock indices are stationary at the first difference, showing an integration order of I(1), except the stock market index from the Shanghai Stock Exchange. We also notice that the indicators related to the evolution of COVID-19 for the most affected regions, China and Italy, show a mixed integration order (I(0)and I(1)).

#### 4.2. Cointegration Analysis and Long-term Relationships

After studying the stationary of the data series and due to the mixed results, we conclude that the ARDL model is the most appropriate for exploring the linkages between variables. Further, the purpose is to assess whether new cases and new deaths due to COVID-19 in China and Italy, along with Chinese and Italian stock market returns, several commodities, and currencies are related to the Romanian stock market as measured by BET index return and Romania 10-year bond yield.

The ARDL (autoregressive distributed lag) model is used especially when the variables I(0) and I(1) are integrated. For the accurate choice of the ARDL model that would allow us to research the relationships that are established between variables, it is imperative to choose the correct number of lags. Therefore, we will analyze the Akaike information criteria (AIC) to select the optimal lags for the variables included in the ARDL model.

We will apply the criteria graph, which will indicate the suitable lags for the ARDL model and the lowest value is preferred.

Figure 6 shows the results of criteria graph for the ARDL model that takes into account the number of new cases and new deaths in China, both for the BET stock index return and for the Romanian Government bond (10Y).

According to the results, in total, 1,562,500 ARDL model specifications were considered for each of the four cases given the information related to COVID-19 in China. The top 20 results are presented in the criteria graph.

Further,

Table 5 summarizes the selected lags for the model Romania and COVID-19 (China) according to criteria graph out of

Figure 6.

Figure 7 shows the results of criteria graph for the ARDL model that takes into account the number of new cases and new deaths in Italy, both for the BET stock index return and for the Romanian Government bond (10Y). Likewise, in case of Italy, in total, 1,562,500 ARDL model specifications were considered for each of the four cases.

Table 6 exhibits the selected lags for the model Romania and COVID-19 (Italy) in line with criteria graph out of

Figure 7.

The results reported in

Table 7 and

Table 8 provides the ARDL bound test for cointegration. If the F-statistic is greater than the upper bound, then the variables comprised in the model are cointegrated and a long-run relationship befall. With reference to new cases in China models (see

Table 7), the F-statistic for BET_R (18.06988) and RO_BOND (4.523219) models is greater than the upper bound of bounds value at 5%, which is suggesting that long-run relationship occur between the variables. The same result is achieved in the case of new deaths in China models, where the value of the F-Statistic is greater than the upper bound critical value. Hence, the null hypothesis is rejected, meaning that the variables in the model are cointegrated.

Regarding Italy, in all four estimated ARDL models the existence of cointegration is confirmed (see

Table 8) since the F-statistic is significantly higher than the critical values in I(0) and I(1). Consequently, the examined variables are cointegrated and will move together in long-run.

Further, we will analyze the results of the long-term linkages between selected measures.

Table 9 shows the outcomes regarding the long-run causal connections among variables for the model Romania and COVID-19 (China)—new cases. The short-run estimates of ARDL approach are presented in

Table S1. In the first model, the number of new infection cases from China have no effect on the BET index return. However, a decrease of crude oil price leads to a higher uncertainty, consistent with Salisu, Ebuh and Usman [

23], suggesting the necessity for policymakers to diminish fears in financial markets. In addition, the exchange rate negatively influences stock market return in the long-run. The Philadelphia Gold/Silver Index coefficient is positive and significant at the 5% level of significance. Hence, the coefficient of XAU_R indicates that an increase of one unit in Philadelphia Gold/Silver Index leads to over 0.2983 units increase in BET index return in the long-run. The error correction term or adjustment speed provides evidence regarding the rate of convergence to equilibrium, being highly statistically significant. The adjustment speed of −1.017783 shows that deviations from the long-term equilibrium in BET index return are corrected the following day by approximately 101.7783 percent. However, the short-run results show no impact of new infection cases of COVID-19 from China on the BET index.

Regarding the second model from

Table 9, similar to the first model, the new infection cases from China does not influence Romania 10-year bond yield in the long-run. Unlike the previous model, the RO_BOND is negatively affected by XAU_R and indicates that an increase of one unit in Philadelphia Gold/Silver Index leads to over 0.3718 units decrease in RO_BOND return in the long-term. Besides, in the long-run, the return of stock market index SSE 100 negatively influences Romania 10-year bond yield. The coefficient of the error correction term is highly statistically significant. Hence, the Romanian 10-year bond will reach equilibrium with a speed of 185.3068 percent in next day. As well, the short-run results strengthen the lack of impact regarding new infection cases of COVID-19 from China on RO_BOND.

Table 10 reveals the outcomes of the long-term connection amongst variables for the model Romania and COVID-19 (China)—new deaths. The short-run results are shown in

Table S2.

The empirical findings reveal that the impact is stronger in this case as compared to the model that depends on the number of new cases in China due to COVID-19 (see

Table 9). However, both models shows that the number of new deaths in China due to COVID-19 has no influence on the BET index return, respectively, on the Romania 10-year bond yield, neither in the short-term, nor in the long-term. Therefore, both research hypotheses are rejected for Chinese COVID-19 figures, similar Topcu and Gulal [

86] which established that emerging European countries experienced the lowest influence of the outbreak.

Table 11 and

Table 12 reveals the results of serial correlation and heteroscedasticity tests for the models Romania and COVID-19 (China)—new cases and Romania and COVID-19 (China)—new deaths. The results support that the models are free from autocorrelation and heteroscedasticity.

In the case of models that take into account the effects of new cases and new deaths in Italy, unique relationships are identified between the selected variables, as opposed to the models that explored the impact of coronavirus from China.

Table 13 exhibits the outcomes of the long-term causal associations between variables for the model Romania and COVID-19 (Italy)—new cases. The short-run outcomes are exhibited in

Table S3. In the long-run, the results of the first model show the lack of any effect from the number of new cases of COVID-19 in Italy on BET index return. In contrast, the return of Milan stock market index FTSE MIB has a positive long-term impact on the BET index return. As well, the short-run results reveal no impact of new infection cases of COVID-19 from Italy on the BET index return. In contrast to COVID-19 figures from China, in case of Italian new cases of coronavirus, the first hypothesis is still rejected, but the second hypothesis is confirmed.

Moreover, in the second model, several statistically significant relationships are identified. There is found a positive impact of the number of new cases in Italy on the Romania 10-year bond yield in the long-term. In addition, a natural gas futures contract has a positive effect on RO_BOND, while the WTI Oil and Philadelphia Gold/Silver Index has a negative impact in the long-run. Another outstanding outcome is that new infection cases of COVID-19 from Italy negatively influence RO_BOND in the short-run, consistent with Sène, Mbengue and Allaya [

36]. Therefore, the related uncertainty triggered by the health emergency may determine investors to get rid of their securities.

Table 14 exposes the findings towards long-run linkages between variables for models related to Romania and COVID-19 (Italy)—new deaths. The results of short-run estimates are presented in

Table S4.

The first model out of

Table 14 exhibits that the number of new deaths from Italy have no effect on the BET index return in the long-run. The Philadelphia Gold/Silver Index coefficient is positive and significant at the 5% level of significance. Hence, the coefficient value of XAU_R indicates that an increase of one unit in Philadelphia Gold/Silver Index leads to over 0.1574 units increase in BET index return in the long-term. However, the short-run results show a negative impact of new deaths cases of COVID-19 from Italy on the BET index return, in line with Okorie and Lin [

58] which underlined a transitory contagion effect in the stock markets due to novel coronavirus. In addition, Erdem [

55] claimed that the index returns decline and volatilities rise due to corona crisis. Hence, the first hypothesis is confirmed.

The second model shows a negative effect of the new deaths’ cases from Italy on the Romania 10-year bond yield in the long-run. In addition, the Philadelphia Gold/Silver Index and the OK crude oil future contract negatively influence RO_BOND in the long-term. Besides, in the long-run, the returns of the stock market index FTSE MIB has no impact on the 10-year Romanian bond. Nevertheless, in the short-run, results show a negative impact of new deaths cases of COVID-19 from Italy on the RO_BOND. Therefore, the second hypothesis is established.

Table 15 and

Table 16 exhibit the outcomes of Breusch—Godfrey Serial correlation LM test and Breusch–Pagan–Godfrey heteroscedasticity test for the models Romania and COVID-19 (Italy)—new cases and Romania and COVID-19 (Italy)—new deaths. Hence, the models are not threatened by autocorrelation and heteroscedasticity.