Investigation of Causality Relationships among COVID-19 Cases, ISE100 Index, Dollar, Euro, Gram Gold Prices and 2 Years Bond Rates: The Case of Turkey

The purpose of this research is to analyze such economic data during the outbreak of the COVID-19 in Turkey. The variable rates were taken from COVID-19 situations, ISE-100 index, Turkish lira dollar (TRY), TRY euro prices, TRY gram Gold and two year bond rates. General COVID-19 information was provided and certain financial indicators were investigated in COVID-19 (47 days). First of all, these variables were used as descriptive statistics and correlation matrix. For the purposes of stationarity testing, the first variables were stationary with Augmented Dickey-Fuller and Phillips-Terron Tests. The lag duration of the deployment model VECM was then calculated as the fourth lag with the highest information requirement. The co-integration relationship between the variables was calculated by the Johansen Cointegration Test. Thanks to this relationship, the variables have a long-term correlation. The Vector Fix Model (VECM) was chosen because it is co-integration. Inverse roots, autocorrelation and normality have been developed, which are essential assumptions to use the VECM (4) model; Therefore, the Granger Causality / Block Exogeneity Wald Test was applied to the variables of the VECM(4) model to define causality relationships between these variables. The results of this test have identified causalities for Turkey 2 years of government bond rates, Euro in TRY, Dollar prices in TRY and Gram in TRY


Introduction
The 2019-2020 coronavirus pandemic  is an incessant pandemic caused by extreme coronavirus acute disease (SARS-CoV-2). The pandemic was reported in 2019-2020. The first outbreak in December 2019 was recorded in Wuhan, China (World Health Organization, 2020a). On 11 January 2020, the first death was reported (Pharmaceutical Technology, 2020). The World Health Organization (WHO) on 30 January 2020 called the epidemic an International Public Health Emergency (PHEIC) and on 11 March 2020 it recognized it as a pandemic [(World Health Organization, 2020b;World Health Organization, 2020c]. Other signs of COVID-19 include acute infection of the air (ARI), exhaustion, rage, fever, or temperature level of 0.01 ° C and cough. Contacting with COVID-19 confirmed patients in particular (within 2 meters for more than 15 minutes) triggers COVID-19 disease (World Health Organization, 2020d).
After April 2020, COVID-19 does not occur in countries other than small countries / regions that receive data. Countries have taken stringent steps to tackle the COVID-19 outbreak, such as curfews and education disturbances. COVID-19 has spread, with several financial, political and social implications, in particular in the United States (USA) and numerous European countries. As at 25 April 2020 in the nation, COVID-19 had 2.868.539 cases, 201.502 cases, and the number of patients treated was 811.660. Although the overall death rate is 7.02% worldwide, 14.16% of the top 10 countries are the highest in France. However, although the average rate recovered is 28.30% globally, it is 93.13% among the top 10 countries in China with the highest recovered rate. The U.K., however. This has a recovery rate of 0.52 percent that is the lowest.
107,773 COVID-19 cases have been confirmed in Turkey which will be examined in this report. Among these, 25582 is recuperated, while 2,706 died. Turkey is the second lowest-fatality country in the top ten countries with 2.51 percent. Nevertheless, this is slightly below global average at a recovered rate of 23.74 percent. This may be one of the reasons that Turkey is one of the countries with the most recent (47 days) COVID-19 outbreak. The rate of recovery in other countries can increase over time (Johns Hopkins University, 2020

Turkey Economy with COVID-19 Disease
In most countries economic stagnation can occur, especially during epidemics or wars, and economic crises can also occur after these stagnations. The main purpose of the analysis was to analyze the relationship of some of the economic indicators in Turkey during COVID-19. In this link, daily (CASES), COVID-19 is considered as the ISE 100, which holds the 100 highest shares in market value and volume traded in Borsa Istanbul, Turkish Lira dollar prices (USD / TRY), Turkish Lira euro prices (EUR / TRY), Turkish Lira gold price per gram (GAU / TRY) and Turkey bond prices for 2 years (2years Bond) were analyzed daily. Their findings include: A recent report in the Wall Street Journal notes that a decline of more than 12% in the Dow Jones industrial average on 16 March 2020 was the second-worst of 124 years. The phenomenal volatility is not entirely explained by these factors. "In general, cycles of high volatility are correlated with economic and political instability (Gormsen & Koijen, 2020). In the case of Figure 1 after the first appearance of COVID-19 in Turkey, ISE 100 values were analyzed based on this example. In the first 2 weeks, COVID-19 was seen to have decreased considerably. But the index almost returned to its former level later on. The exchange rate is a conventional vector of crisis contagion. In the late 90s, for example, the Asian crisis involved companies and countries that borrowed in one currency and received income in another. For example, a sudden currency exchange rate devaluation almost immediately bankrupted several Thai companies. The value of the dollar's profits does not meet the interest and loan servicing dollar cost criteria. No indication of this mechanism has been given to date (Baldwin & Mauro, 2020). In addition, the lessons from this crisis led to considerably lower cross-currency borrowing. In Turkey, the parity of USD / TRY and EUR / TRY is examined on the basis of this quotation. The upward trend has been observed in the USD / TRY and EUR / TRY since the first COVID-19 appeared on March 10th. "This is a more widely-desired currency, as the volumes grow and gold markets worldwide will be trading more than one of the world's most liquid currencies. In the short term the volatility in gold will remain extremely volatile in light of the uncertainties surrounding cu, the Chairper and Chief Executive Officer at GoldSeek.com told MarketWatch. Another element under consideration in this analysis is gold, which reached its highest value in the epidemic since February 2013 (the most recent increase in gold prices). The Gold Price per gram has risen in value against the Turkish Lira, provided that the first COVID-19 case has been identified. GDS denotes bonds issued by the Treasury Undersecretariat on the domestic market for domestic borrowing. Upon the completion of the payment period and on maturity, the borrowed state shall pay the GDS holder of the balance owed. Such GDS are considered government bonds with a maturity of 1 year and more (Borsa İstanbul, 2020). As of March 10, Figure 4 indicates the 2-year bonds' interest rate. It is noted that, particularly after 20 April, interest rates decreased.

Material
Data were collected regularly between 10 March 2020 and 25 April 2020 when COVID-19 was first detected. Data is obtained from the official website of the Turkish Ministry of Health (Turkish Ministry of Health, 2020). The Eviews 9 Software carried out all the analyzes applied to these 47-day results. Eviews is a Windows Statistics Package developed by Quantitative Micro Software (QMS). Eviews is the most important statistical package. This can be used for data processing, drawing, statistical analysis [Agung, 2008;Agung, 2011], modeling analysis (Ju et al., 2009), forecasts and simulations as econometric software. Often commonly used in political, macroeconomic and simulatory analysis and sales. Eviews is primarily used for time series-oriented econometric analysis in contrast to other applications, such as EXCEL, SAA, SPSS. The relationship between the m variables, where the cross members in the matrix are equal, is described by a matrix correlation. It is a rectangular, symmetrical mxm dimension matrix derived from the covariance matrix of variance. The same data is present in both matrices but since it is clearer and straightforward to compare the variables in the matrix (Horn & Johnson, 1985). The corresponding matrices are simpler.

Statistical Analysis and Findings
The results of the matrix for the correlation in Table 3 show that between EUR / TRY and GAU / TRY variables the most positive correlation was 92,77 percent. The highest negative relationship among the ISE100 and the 2-year Bond variables was 13.64%; however, in Table 3, the relationship levels of all the variables can be interpreted in the same way. Dickey-Fuller (ADF) test is one of the many common ones. The ADF test (Dickey and Fuller, 1979) means the first difference between a variable y, the exogenous variable(s) and k, the first differences that were lagged at their lagging level: Where Y_t is the variable in the t cycle, T refers to the time trend, while the Δ is the operator of the differences, ε_t is an error disturbance with the zero mean and variance 2, and k represents a lag in ADF equation. The number of lags of the ADF test is reduced. Due to the increased number of lags, the power of this test to reject the null of a unit root is decreased and additional parameters must be estimated and freedom loss must be reduced (Hosseini et al., 2011). The Phillips-Perron test is another root test process. The following equation (Günaydın, 2004) is used to evaluate the PP test: In equation 2, the word Δ refers to the initial processor of the difference, any time pattern, error time term, series, and N refers to the delay factor defined by the criterion of knowledge to solve the consequent dependence of the error conditions. The PP test is a test that allows for poor dependency and heterogeneity between the Error Conditions for the Dickey and Fuller tests (Öztürk & Pehlivan, 2020). The negative side of the PP test is that the sample diameter is error skewed (Egeli H.A. & Egeli, H., 2008).
Stationarity of variables was examined through increased Dickey-Fuller and Phillips-Perron tests among the root unit tests. Stationarity of the variables As in Table 3 and  Table 4, the results of these tests are. All experiments indicate that the variables were not stationary. The outcome was the same. For this reason all variables have first differences in order to ensure stationary variables. When all p-values are less than Alphanumeric Journal Volume 8, Issue 1, 2020 0.05, the hypothesis of H1a null has been dismissed. Convenient terminology for the use of variables were therefore given.
 H1a: The variable is non-stationary and has a unit root.
 H1b: The variable isn't non-stationary and hasn't a unit root.

Variables
No  The correct lag period for the VECM model was created in table 5. The fourth lag time with LR test statistics, FPE, and Akaike Information Criterion (AIC) were picked. Since the criterion indicated by "*" is the 4th lag duration. All exams on VECM(4) were carried out in accordance with this test. The co-integrated relations were calculated in the 4th and 5th models according to the findings in Table 6. Since the lowest-error model should be selected from the defects, the lowest AIC (30.62101) value has been selected. The quadratic fifth model is therefore ideal for research with intercept and pattern. Cointegration relationships have been investigated using this model.   Engle and Granger (1987) claimed that the co-integration between the variables was sufficient to apply the error correction model (Engle & Granger, 1987) . The structure of equation in the model is: Here values of and are not stationary, but the series that is stationary once the first difference, i.e. I(1) series, has been made. When you take and rearrange the 1 st difference of the equation, takes the form in Formula (4). Where, It is expressed as = ′. It expresses two matrices with and ′ ( ) dimensions and rank (Göçer et al., 2013) .
represents the adaptation rate, that is, the coefficient of error correction term, ′ is the long-term cointegration coefficient matrix and is the rank of the matrix (Tarı ve Yıldırım, 2009). If the rank is equal to 1, it is concluded that there is 1 cointegration relationship between the variables and if it is greater than 1, there is a cointegration relationship as much as the value of rank. Trace and maximum eigenvalue statistics are checked to see if there is a cointegrated relationship between the series (Akpolat & Altıntaş, 2013). Table 7 shows the Johansen Cointegration Test results under the value of trace statistics. 'At most 2', the Trace statistic and eigenvalue statistics are greater than their respective critical value at 0.05 level and its probability is also less than 0.05 level, which over again leads to the rejection of the null hypothesis relating with one cointegration equation. At most 2 shows two cointegration equations in the selected Alphanumeric Journal Volume 8, Issue 1, 2020 variables, which indicates the Trace statistics value and eigenvalue statistics value are lesser than their critical values at 0.05 level. In addition, due to cointegration, there is a long-term relationship among variables. A VECM (Vector Error Corrigation Model), which adapts to short term fluctuations of variables and deviations from balance (Andrei, D.M. & Andrei, L.C., 2015), is a suitable estimation technique if one or more of the cointegrating vectors is detected. VECM analysis will remove the question of fake regression between dependent and explicative variables. The VECM is therefore the following (Sevüktekin & Nargeleçekenler, 2010): In the model, ECTt-1 refers to error correction term, refers to correction coefficient and represents the number of equations. The statistical significance of the error correction coefficient ( ) indicates the deviation from the long-term balance. The size of the coefficient shows the speed of approaching long-term equilibrium (Gujarati, 2004).
In addition to the VECM(4) model review, whether the model has a stationary structure should be considered. It must be evaluated if the reverse roots of the AR function are in the circle of the array. The results of this study are provided in Figure  5 for a stationary structure evaluation and the own values of the obtained coefficient matrix have to be in the unit circle in order to use the VECM(4) model. Based on this information, the VECM(4) model is determined to be stationary since all reverse roots are in the circle of units in Figure 5. In other words, there is no root beyond the circle of the unit. One conclusion that the VECM(4) model would be suitable is that the sequence does not pose a problem with autocorrelation. Whereas the LM test results of Table 8 have been reviewed, it has been determined that the 2 nd 3 rd and 4 th p-values are above 0.05. The H2a null assumption can not therefore be refused. It is concluded that the VECM(4) series does not have an autocorrelation problem.
 H2a: There is no lagging serial association at the 95% confidence stage.
 H2a: There is a lagging serial correlation at a trust point of 95%. Recently, the model VECM(4) must be believed to be distributed according to the normal distribution. Table 9 and Table10 contain the results of the VECM Residual Standardization Test used to check this distribution. The H3a null hysperthesis could not be dismissed, and this assumption was also made, as p-values for skevity, Kurtosis, and test statistics of Jarque-Bera (joint) are higher than 0,05 point. The causality of these variables has been investigated in the VECM(4) model. In addition to the explanatory power of the lag of such variables (Ahmed, 2011) the Granger Causality / Bloc Wald Test is used to evaluate whether the lagged value is sufficient to describe the dynamics of certain variables within the multivariate system. The regression of Y is monitored in the Granger Test by its own delays and X delays. It also monitors a regression in conjunction with X own delays and Y delays. Y is the dependent variable and X is the stand-alone p lag variable. Therefore it is possible to decide whether the causality is unilateral, where either X Granger causes Y but Y does not cause X, X Granger causes Y or Y causes X bi-directionally. The H5a null hypothesis is introduced for every variable to be evaluated as a dependent variable. The regression of vector Y, for example, (Garcia & Rodrigues, 2019) has been used.
The Exogeneity Forest Test of Granger Causality / Block was carried out to explore relationships of causality between variables. The test results are as shown in Table  11. Since other p-values in the table are below 0.05, the above hypothesis of H4a can be dismissed. Many causal associations were also found. For a model in which the dependent variable 2-Years Bond 1 is, the causes of the 2-years Bond 1 is the variable ISE100 1 and the variable GAU / TRY

Conclusion
This research analyzed the relationship between economic data in Turkey during the COVID-19 epidemic. On 10 March 2020, the first detection of COVID-19 was achieved for 47 days. The data is till 25 April 2020 daily. The six variables produced from these daily data are the number of cases of COVID-19 perday, the ISE100 stock index in the Turkish Lira, Turkish Lira dollar prices, Turkish Lira gram gold prices and the 2-year bonds. The following variables are the number of cases of COVID-19 per day. Such variables were initially investigated by descriptive statistics and correlation matrix. The maximum correlation between Turkish Lira and Turkish Lira gold Gram prices was found in the correlational matrix at 92.77 percent. Because of the time series of all variables, their standardity has been checked. The 1st variable differences were found Alphanumeric Journal Volume 8, Issue 1, 2020 to be stationary, according to the findings of both Augmented Dickey-Fuller and Phillips-Perron experiments. Variables suitable for causality analysis were therefore achieved through VECM.
The deficiency duration of the mounted VECM model has been calculated to be the fourth deficiency with the most knowledge criterion. For the VECM(4) model, reverse roots of the AR polynomial are analyzed in the unit circle. As a consequence of this analysis, the unit circle held all the reverse roots. And it is important to test certain theories. Due to the Autocorrelation LM Test the VECM(4) model was found to have no autocorrelation problem. In addition, as a result of VECM residual normality checks, the model was found to show Normal distribution. The VECM(4) model guaranteed these premises, thus testing causality relations.
The Granger Causality / Block Exogeneity Wald Test was applied to variables in the VECM(4) model for evaluating causality relationships between the variables. As a consequence of this study, causalities were calculated for dependent variables, including government bond rates for Turkey for two years, TRY euro prices, TRY dollar prices, and TRY gram gold prices. Turkey 2 years of government bond rates have causalities that are ISE100 and gram gram gold prices at Seek, with two different variables. Seek Euro prices have causalities that are ISE100 stock index and COVID-19 cases a day with two independent variables. TRY dollar prices have cause-related causalities of ISE100, COVID-19 cases per day, and Turkey, 2 years government bond rates with three independent variables. And ultimately, TRY's gold prices are causal, with 3 different variables, ISE100 inventory index, COVID-19 cases per day and TRY dollar values. In such causality relationships the independent variables affect the dependent variables. During the COVID-19 era, although no variables affect the regular number of cases and ISE100 stock index, causalities affecting other variables have been reported.