Do Tense Geopolitical Factors Drive Crude Oil Prices?

Geopolitical factors are considered a crucial factor that makes a difference in crude oil prices. Over the last three decades, many political events occurred frequently, causing short-term fluctuations in crude oil prices. This paper aims to examine the dynamic correlation and causal link between geopolitical factors and crude oil prices based on data from June 1987 to February 2020. By using a time-varying copula approach, it is shown that the correlation between geopolitical factors and crude oil prices is strong during periods of political tensions. The GPA (geopolitical acts) index, as the real factor, drives the rise in prices of crude oil. Moreover, the dynamic correlation between geopolitical factors and crude oil prices shows strong volatility over time during periods of political tensions. We also found unidirectional causality running from geopolitical factors to crude oil prices by using the Granger causality test.


Introduction
This paper examines the dynamic correlation and causality between geopolitical factors and crude oil prices in different political environments. Although there are many studies discussing the impact of extreme political acts on crude oil prices [1][2][3][4], we use the GPR (geopolitical risk) index and its sub-indices, GPT (geopolitical threats), and GPA (geopolitical acts), constructed by Caldara and Iacoviello [5], Brent and West Texas Intermediate (WTI) to study the dynamic correlation between geopolitical relations (tense and moderate) and crude oil prices (see Appendix A for an explanation of the nomenclatures). Additionally, this study further examines whether there is a causality between geopolitical factors and crude oil prices. To the best of our knowledge, this is the first time that the GPR index is used to examine causality between geopolitical factors and crude oil prices.
The background and importance of this study are that international crude oil prices are always closely related to extreme political events. With the global spread of the COVID-19 virus in 2020, demand for crude oil has been hit hard, and international crude oil prices have fallen off a cliff since March. Many studies proved that geopolitical factors have an important impact on crude oil prices, such as following Gulf War and the 9/11 attack. For instance, Zhang et al. [6] and Dey et al. [7] examined the impact of extreme political events on crude oil prices and suggested that people consider the impact of extreme political events when predicting oil prices. The fluctuations of international crude oil prices will seriously affect the real economy and hinder the stability of the financial system, and may even lead to systemic risks in global financial markets [8][9][10][11]. Therefore, our study focuses on the relationship between geopolitical factors and crude oil prices, which has important implications for some Arab economies and/or for the major oil-exporting countries. the GPA index can significantly increase crude oil prices, although GPR (or GPT) and crude oil prices generally move together in the same direction (co-movement exist). Further, there is a unidirectional causality running from geopolitical factors to crude oil prices.
The theoretical and practical implications of this paper can be summarized as follows: First, this paper enhances the understanding of international crude oil price fluctuations by examining the dynamic correlation between the GPR index and crude oil prices. In particular, this paper analyzes the impact of war or terrorist activities on crude oil prices and provides new evidence that geopolitical factors affect crude oil prices. Second, an interesting finding from our analysis is that actual extreme political events will drive crude oil prices. The findings of this paper are relevant to stakeholders as they provide stakeholders with effective measures to prevent crude oil price fluctuations.
The structure of this study is as follows. Section 2 outlines the main method and data source. Section 3 shows the empirical results of the analysis and Section 4 is a further discussion on the relationship between geopolitical factors and crude oil prices. Finally, Section 5 concludes.

Time-Varying Copula
Our motivation for selecting the time-varying copula model is that the correlation between the geopolitical risk and the price of crude oil in extreme environments requires special attention. The reason is that in a relatively stable political situation, geopolitical risks will not significantly change the trend of international crude oil prices, but the upper and lower tails of the distribution of geopolitical risk index (referring to the period of extreme political events) are more and more closely related to the fluctuation of crude oil prices.
Economists mostly use the wavelet analysis method, a family of the GARCH (generalized autoregressive conditional heteroscedasticity) model, and a copula approach to capture the correlation between two variables. The wavelet approach has great advantages in dealing with non-stationary information and it can deal with frequency and time information. Thus, the wavelet approach is very common when dealing with the correlation of frequency information of non-stationary time series [41]. The DCC-GARCH (dynamic conditional correlational-generalized autoregressive conditional heteroscedasticity) model decomposes the covariance matrix into a conditional standard deviation and correlation matrix through orthogonal basis decomposition, thereby obtaining the dynamic correlation coefficients between time series [42]. This process needs to satisfy the conditional heteroscedasticity (ARCH effect) of variables. The correlation of an extremely political environment (upper-lower tail correlation) is particularly important when describing the dynamic correlation between the GPR index and crude oil prices. However, neither the wavelet analysis method nor the family of the GARCH model can capture the tail correlation of the GPR index and the crude oil price distribution. To solve these defects, we used the time-varying copula approach to measure the correlation between geopolitical factors and crude oil prices.
In this paper, we focus on the dynamic correlation between the GPR index and crude oil price (X 1 and X 2 , respectively). According to Sklar's theorem, the bivariate joint distribution F X 1 X 2 (x 1 , x 2 ) for the GPR index and crude oil prices can be represented as a copula function after transforming marginal distributions into uniform distributions [43].
Therefore, the bivariate joint distribution of the GPR index and crude oil price can be expressed as: where u = F X 1 (x 1 ) and v = F X 2 (x 2 ), and C is a copula function that describes the correlation between the GPR index and crude oil price. The density function of the copula is C(u, v) = ∂u∂v . Thus, the bivariate joint probability density of the GPR index and crude oil price (X 1 and X 2 ) is as follows: where f X 1 (x 1 ) and f X 2 (x 2 ) are the marginal densities of the GPR index and crude oil price, respectively (X 1 and X 2 ). According to Sklar [43], an n-dimensional joint distribution can be decomposed into its n-univariate marginal distributions and an n-dimensional copula. Patton [44] showed that the upper-lower tail correlation of the GPR index and crude oil price is given for the copula as: where τ L and τ U ∈ [0, 1]. The normal (Gaussian) copula used in this study and its correlation parameters are briefly presented below. The normal copula density is given by: The normal copula shown in Formula (5) can measure the static correlation, but does not allow for time-varying correlation. Following Patton [44], for the Gaussian copulas we specify the linear correlation parameter as ρ t , in order to evolve in time according to an autoregressive (AR) moving average (MA) process, namely ARMA (1,q), as follows: where is the modified logistic transformation, ρ t ∈ (−1, 1), and the correlation parameter is characterized by the constant ψ 0 , by the autoregressive term ψ 1 and by the average product over the last q observations of the transformed variables, i.e., ψ 2 .

Marginal Distribution
The GPR index and crude oil price went smoothly after the first-order difference (stability test results can be provided upon request). Thus, this study adopted the widely used ARIMA (autoregressive integrated moving average) model. The ARIMA(p, d, q) model is an extension of the ARMA model. The ARIMA(p, d, q) model is transformed into a stationary series after d-order difference, which becomes an ARMA(p, q) model. The basic form of the ARIMA model can be written as: where d x t represents the GPR index (or crude oil price) x t after d-order differential conversion and ε t represents the random error of the t moment. At the same time, ε t is an independently distributed white noise series and obeys a normal distribution with a mean of zero and a variance of σ 2 . Given the GPR index (or crude oil price) x t , the GARCH(p, q) model can be written as: where ω represent a constant, ε 2 t−i represent the ARCH component and σ 2 t−j represent the GARCH component. The number of lags (p, q) is selected according to the akaike information criteria (AIC).

Estimation
This study used the two-stage maximum likelihood (ML) method used by Patton [44] when estimating Copula parameters. The following briefly introduces this parameter estimation method. The log-likelihood function can be written as: where α X 1 and α X 2 are the parameters of the marginal distribution of X 1 and X 2 , respectively, θ is the copula density parameter and ψ is the joint density parameter. The two-step inference for the margins procedure was adopted. First, the GPR index and the marginal distribution of crude oil prices are estimated as follows: where i = x 1 , x 2 . Second, the correlation coefficient between the GPR index and crude oil price is obtained by estimating the copula function: whereû t = F X 1 x 1,t ;α X 1 andv t = F X 2 x 2,t ;α X 2 .

Variables and Data Source
This study sought to investigate the basic question raised earlier regarding the relationship between geopolitical risks and crude oil prices. For this purpose, the GPR measured by Caldara and Iacoviello [5] can be used as a proxy indicator of geopolitical risk. Monthly data for geopolitical risk index are available for download from the website https://www.matteoiacoviello.com/gpr.htm. There are two developed variants of the GPR indices, the GPT and GPA indexes, which we used in our study to investigate the relationship between geopolitical risk and crude oil prices. The GPT index is constructed by searching articles that include words in the groups directly mentioning risks, while the GPA index searches only for the groups directly mentioning adverse events. In addition, crude oil price can be explained by Brent and WTI spot prices based on most of the literature [45][46][47]. We obtained monthly crude oil price data through the US Energy Information Administration (http://www.eia.doe.gov). According to the availability of crude oil price data, the time dimension of the data selected in this study is from June 1987 to February 2020. Table 1 shows the descriptive statistical results of the variables involved in this study. Table 1 reports the descriptive statistics of all variables. From all observations (overall panel), the average GPT value was 88.421, which is greater than the average GPA value of 72.689. This result shows that risk events are more directly mentioned in newspapers relative to adverse events. The kurtosis values of the three indices representing geopolitical risk are very large, indicating that Energies 2020, 13, 4277 6 of 19 their distribution is more peaked than the Gaussian distribution and shows non-normal characteristics.
The statistical values of Brent and WTI are very similar, indicating that whether Brent or WTI represent the international crude oil price, there will be no significant difference in the empirical results. The tension panel presents descriptive statistics for the periods of geopolitical tension and the stabilization panel is for the periods of geopolitical stabilization. For detailed rules on the division of periods of geopolitical tension and stabilization, please refer to Section 4.1 of this paper. At the same time, we also plotted geopolitical risk indexes (GPR, GPT and GPA) and international crude oil prices (BRENT and WTI), as shown in Figure 1.  Table 1 reports the descriptive statistics of all variables. From all observations (overall panel), the average GPT value was 88.421, which is greater than the average GPA value of 72.689. This result shows that risk events are more directly mentioned in newspapers relative to adverse events. The kurtosis values of the three indices representing geopolitical risk are very large, indicating that their distribution is more peaked than the Gaussian distribution and shows non-normal characteristics. The statistical values of Brent and WTI are very similar, indicating that whether Brent or WTI represent the international crude oil price, there will be no significant difference in the empirical results. The tension panel presents descriptive statistics for the periods of geopolitical tension and the stabilization panel is for the periods of geopolitical stabilization. For detailed rules on the division of periods of geopolitical tension and stabilization, please refer to Section 4.1 of this paper. At the same time, we also plotted geopolitical risk indexes (GPR, GPT and GPA) and international crude oil prices (BRENT and WTI), as shown in Figure 1.  Since 2015, the correlations between geopolitical factors and international crude oil prices has been continuously strengthened. As shown in Figure 1, the GPR index (including GPT and GPA) and crude oil prices reached a peak at the same time around 1991. During this period, geopolitical factors and international crude oil prices presented similar historical trends. Subsequently, around the 9/11 event, the GPR index (including GPT and GPA) reached its highest values. However, crude oil prices experienced only small fluctuations during this period. Between 2007 and 2015, crude oil prices showed extreme volatility, especially at the end of 2008. Crude oil price dynamics during this period may have little relevance to adverse political events. After 2015, the frequency of adverse events became higher and higher, and crude oil prices also showed a trend similar to the GPR index. Therefore, we initially believe that the relationship between geopolitical factors and international crude oil prices has continued to strengthen since 2015.

Marginal Distribution Model Results
The marginal models (Formulas (7)-(9)) are estimated by taking different combinations for the lags values, ranging between zero and four, and by selecting the most appropriate ARIMA(p, d, q) -GARCH(p, q) specification with skewed t-distribution according to the AIC values (the best model is the one which minimizes the AIC value). The results of AIC show that the best model for fitting the marginal distribution is ARIMA(1, 1, 2) − GARCH(1, 1). Table 2 presents the parameter estimates for the marginal distribution models.
The results showed that most of the parameters of ARIMA(1, 1, 2) − GARCH(1, 1) were statistically significant. This shows that the ARIMA(1, 1, 2) − GARCH(1, 1) model is suitable for fitting the marginal distribution of the GPR, GPT, GPA, Brent and WTI series. In addition, the parameters α and β of the GARCH model were statistically significant for all series (the parameter β that fits the marginal distribution of GPA was an exception), which shows that GPR, GPT, BRENT and WTI all have a volatility clustering effect. As α + β was close to 1, this indicates that the shock was quite persistent to the GPR, GPT, GPA, Brent and WTI series [48,49]. We performed autocorrelation and partial autocorrelation tests on the standardized residuals obtained by fitting the marginal distribution of the GPR, GPT, GPA, BRENT and WTI series. The results showed that the ARIMA(1, 1, 2) − GARCH(1, 1) model has basically eliminated the autocorrelation, partial autocorrelation and the conditional heteroscedastic effect of the GPR, GPT, GPA, Brent and WTI series (the results are available upon request from the authors). Through the use of the ARIMA(1, 1, 2) − GARCH(1, 1) model to filter the data, a residual series with no sequence correlation and no heteroscedasticity was obtained. Finally, we used probability integral transformation to make these residual series obey the 0-1 distribution. Table 2. Parameter estimation of the ARIMA-GARCH models for GPR indices and crude oil prices. Notes: ARIMA-GARCH represent autoregressive integrated moving average-generalized autoregressive conditional heteroscedasticity model. GPR, GPT, GPA, BRENT, and WTI represent geopolitical risk index, geopolitical threats index, geopolitical acts index, Brent, and WTI spot prices, respectively. The standard errors of the parameters are in parenthesis. *, **, and *** represent 10%, 5% and 1% significance, respectively. The sample period is from June 1987 to February 2020.

Dynamic Correlation between GPR and Crude Oil Prices
Based on the time-varying copula model, this study estimated the dynamic correlations between the full sample GPR index and crude oil prices from June 1987 to February 2020. Figure 2 presents the dynamic Kendall's τ coefficients of the GPR-Brent pair and GPR-WTI pair. The dynamic Kendall's tau coefficients were used to assess the strength of dynamic correlations between the GPR index and crude oil prices.

Dynamic Correlation between GPR and Crude Oil Prices
Based on the time-varying copula model, this study estimated the dynamic correlations between the full sample GPR index and crude oil prices from June 1987 to February 2020. Figure 2 presents the dynamic Kendall's τ coefficients of the GPR-Brent pair and GPR-WTI pair. The dynamic Kendall's tau coefficients were used to assess the strength of dynamic correlations between the GPR index and crude oil prices. . During these extreme events, the correlation between the geopolitical risk index and crude oil prices reached its peak. This means that before and after extreme events, the correlations between geopolitical factors and crude oil prices were enhanced. It could be the case that some extreme geopolitical events are severe enough to have repercussions on and crude oil prices were enhanced. It could be the case that some extreme geopolitical events are severe enough to have repercussions on global economic policy uncertainty through trade linkages, international capital flows, and confidence channels [11,50,51].
To highlight the advantage of time-varying copula in calculating the correlation between geopolitical factors and crude oil prices, this paper also gives the results of DCC-GARCH and wavelet coherence analyses, as shown in Figures 3 and 4, respectively. dynamic conditional correlations coefficients even showed opposite trends. The reason may be that the DCC-GARCH model needs to estimate a large number of parameters when calculating the correlation, so that the results are prone to unavoidable errors. However, the dynamic correlation coefficient shown in Figure 2 is calculated by the time-varying copula, which is a non-parametric approach. The copula approach can effectively measure the correlation between the upper and lower tails of two distributions. In other words, the copula accurately measured the correlation between geopolitical factors and crude oil prices in extreme environments. Figure 4 shows the results of wavelet coherence between the GPR index and crude oil prices. Wavelet coherence analysis provides information on the correlation between the GPR and crude oil prices at different frequencies. Similarly, the DCC-GARCH and wavelet coherence analyses failed to capture the asymmetric correlation between the upper and lower tails of the GPR and crude oil price distribution. Therefore, this study used a time-varying copula to better describe the dynamic correlation between the GPR and crude oil prices.

Dynamic Correlation between GPR Sub-Indices and Crude Oil Prices
The GPR index can be further decomposed into two sub-indices, that is, whether the news vocabulary of geopolitical risks is related to potential risks or actual adverse events. After verifying the dynamic correlations between the GPR index and crude oil prices, this study estimated the dynamic correlation between the GPT indices and crude oil prices under the full sample. The dynamic Kendall's τ coefficients of the GPT-Brent pair and GPT-WTI pair are depicted in Figure 5 (Upper). At the same time, this study also estimated the dynamic correlation between the GPA indices and crude oil prices under the full sample. The dynamic Kendall's τ coefficients of the GPA-Brent pair and GPA-WTI pair are depicted in Figure 5 (Lower).  Figure 3 shows the results of dynamic conditional correlations between the GPR index and crude oil prices calculated by the DCC-GARCH model. The dynamic conditional correlations of the GPR-Brent pair and GPR-WTI pair showed a large deviation between 1990 and 2005. These two dynamic conditional correlations coefficients even showed opposite trends. The reason may be that the DCC-GARCH model needs to estimate a large number of parameters when calculating the correlation, so that the results are prone to unavoidable errors. However, the dynamic correlation coefficient shown in Figure 2 is calculated by the time-varying copula, which is a non-parametric approach. The copula approach can effectively measure the correlation between the upper and lower tails of two distributions. In other words, the copula accurately measured the correlation between geopolitical factors and crude oil prices in extreme environments. Figure 4 shows the results of wavelet coherence between the GPR index and crude oil prices. Wavelet coherence analysis provides information on the correlation between the GPR and crude oil prices at different frequencies. Similarly, the DCC-GARCH and wavelet coherence analyses failed to capture the asymmetric correlation between the upper and lower tails of the GPR and crude oil price distribution. Therefore, this study used a time-varying copula to better describe the dynamic correlation between the GPR and crude oil prices.

Dynamic Correlation between GPR Sub-Indices and Crude Oil Prices
The GPR index can be further decomposed into two sub-indices, that is, whether the news vocabulary of geopolitical risks is related to potential risks or actual adverse events. After verifying the dynamic correlations between the GPR index and crude oil prices, this study estimated the dynamic correlation between the GPT indices and crude oil prices under the full sample. The dynamic Kendall's τ coefficients of the GPT-Brent pair and GPT-WTI pair are depicted in Figure 5

Dynamic Correlation between GPR Sub-Indices and Crude Oil Prices
The GPR index can be further decomposed into two sub-indices, that is, whether the news vocabulary of geopolitical risks is related to potential risks or actual adverse events. After verifying the dynamic correlations between the GPR index and crude oil prices, this study estimated the dynamic correlation between the GPT indices and crude oil prices under the full sample. The dynamic Kendall's τ coefficients of the GPT-Brent pair and GPT-WTI pair are depicted in Figure 5 (Upper). At the same time, this study also estimated the dynamic correlation between the GPA indices and crude oil prices under the full sample. The dynamic Kendall's τ coefficients of the GPA-Brent pair and GPA-WTI pair are depicted in Figure 5 (Lower). The dynamic correlation coefficient of the GPA and crude oil prices was not always negative. Figure 5 shows the difference in dynamic correlations between the GPT/GPA index and crude oil prices. Specifically, the dynamic correlation between the GPT index and crude oil prices was negatively correlated in all samples and maintained at around −0.05. The main reason is that the international crude oil market has entered an era of diversified pricing since 1985 [52]. After this, The dynamic correlation coefficient of the GPA and crude oil prices was not always negative. Figure 5 shows the difference in dynamic correlations between the GPT/GPA index and crude oil prices. Specifically, the dynamic correlation between the GPT index and crude oil prices was negatively correlated in all samples and maintained at around −0.05. The main reason is that the international crude oil market has entered an era of diversified pricing since 1985 [52]. After this, long-term contract prices began to be linked to spot prices and futures prices. At the same time, spot prices were increasingly affected by futures prices. When the news vocabulary of geopolitical risks related to potential risks rises, speculators of crude oil futures sell off crude oil futures by evaluating the potential downside or upside risks, causing crude oil futures prices to fall [53]. This process further affects international crude oil prices.
Unlike the GPT index, the dynamic correlation between the GPA index and crude oil prices This means that geopolitical risks related to actual adverse events led to an increase in international crude oil prices during the periods of the Gulf War, the 9/11 incident, the Iraq invasion, the 2008 global financial crisis, the European subprime mortgage crisis, escalation of the Syrian Civil War, Russia's annexation of Crimea, the Paris attacks and North Korea tensions. The main reason is that actual military actions have seriously affected the supply and demand of international crude oil [54]. Geopolitical risks related to actual adverse events have led to a sudden drop in crude oil production in some Arab countries [55], exacerbating market concerns about crude oil supply shortages. At the same time, tense geopolitical relations have increased the cost of crude oil extraction and transportation. This series of chain reactions together raised international crude oil prices.

Dynamic Correlations in Different Political Environments
Geopolitical risks (related to potential risks or actual adverse events) and international crude oil prices may have a varying dynamic relationship in different political environments. Examining the heterogeneity of the dynamic correlation between geopolitical risks and international crude oil prices, geopolitical relations tend to be tense or moderate, which is beneficial to investors or policymakers in oil-producing countries to assess potential risks. Therefore, it is also an important contribution to analyze the dynamic correlation between geopolitical risk and international crude oil prices in different political environments. To test the heterogeneity of the dynamic correlation between geopolitical risk and international crude oil prices, we divided geopolitical relations into periods of tense and moderate tendencies based on the trend of the GPR index. Specifically, the observation that GPR index after the first-order difference is greater than 0 fell into the set "tense geopolitical relation period", and less than 0 fell into the set "moderate geopolitical relation period". Based on this, this study further estimated the dynamic correlation between geopolitical risks and crude oil prices under 184 sub-samples of tense geopolitical relation periods and 208 sub-samples of moderate geopolitical relation periods. Figure 6 presents the dynamic correlations between geopolitical risks and crude oil prices in different political environments.
observation that GPR index after the first-order difference is greater than 0 fell into the set "tense geopolitical relation period", and less than 0 fell into the set "moderate geopolitical relation period". Based on this, this study further estimated the dynamic correlation between geopolitical risks and crude oil prices under 184 sub-samples of tense geopolitical relation periods and 208 sub-samples of moderate geopolitical relation periods. Figure 6 presents the dynamic correlations between geopolitical risks and crude oil prices in different political environments.  The dynamic correlation between geopolitical risks and crude oil prices showed strong fluctuations during tense geopolitical relation periods. Figure 6 shows that, during tense geopolitical relation periods, the maximum value of the dynamic Kendall's τ coefficient of the GPR-Brent pair was −0.107. All dynamic Kendall's τ coefficients of geopolitical relations and the crude oil prices, except the GPR-Brent pair and GPT-WTI pair, had a larger standard deviation during tense geopolitical relation periods than during moderate geopolitical relation periods.
In general, the correlation coefficient in periods of geopolitical tension had greater fluctuations than the correlation coefficient in periods of geopolitical moderation. This shows that the correlation between geopolitical risks and crude oil prices in different political environments is significantly different. In a tense political situation, the correlation between geopolitical risks and crude oil prices was more likely to change direction, and it is also prone to strong positive and strong negative correlations. In addition, the standard deviation of the dynamic correlation coefficient also reflects the same fact-the dynamic correlation between geopolitical risks and crude oil prices showed strong fluctuations during tense geopolitical relation periods.

Granger Causality Test
From the above empirical results, it can be concluded that geopolitical factors have a significant correlation with international crude oil prices. However, this cannot infer a causal relationship between geopolitical factors and international crude oil prices. To reveal a causal linkage between geopolitical factors and crude oil prices, this study implemented the Granger causality tests to test whether there is a causal relationship between geopolitical factors and crude oil prices. The causality tests were implemented for the GPR index, GPT index, GPA index, Brent and WTI. In this study, the Granger causality tests were implemented within a first-order to sixth-order lag. The results of the Granger causality with first-order lag are shown in Table 3.
For the full sample period, there was unidirectional Granger causality running from geopolitical factors to crude oil prices. Table 3 shows that during the full sample period, the GPR index was the unidirectional Granger causality of Brent and WTI, with a 5% significance level. In addition, the GPT index was the unidirectional Granger causality of Brent and WTI, with 5% and 10% significance levels, respectively. However, there was no Granger causality between the GPA index and crude oil prices. This shows that during the full sample period, geopolitical risks and geopolitical threats can directly cause fluctuations in international crude oil prices. By contrast, the correlation between geopolitical actions and crude oil prices may be facilitated by a third factor. In addition, research presented by Wen et al. [56] showed that oil prices and stock markets have non-linear causal crossings. The reason may be that fluctuations in the financial market have led to rising oil prices. Our results also showed that, in the sub-sample period, there was no Granger causality between geopolitical factors and crude oil prices, whether during tense geopolitical relation periods or moderate geopolitical relation periods. However, it can be seen from Table 4 that only Brent was a unidirectional Granger causality of GPT during the full sample period. Specifically, at a 10% significance level and a 2-month lag, there was unidirectional Granger causality running from Brent to GPT. By contrast, the null hypotheses "Crude oil prices do not Granger cause geopolitical factors" or "Geopolitical factors do not Granger cause crude oil prices" cannot be rejected. Such a result of Granger causality occurred simultaneously in the full sample period, tense geopolitical relation periods, and moderate geopolitical relation periods. In addition, the Granger causality tests within third-order to sixth-order lags were not significant. Therefore, we do not show these results in the text.
was not always negative. The results found by examining the dynamic correlations between the sub-indices and crude oil prices suggested that the dynamic correlation between the GPT index and crude oil prices was negatively correlated in all samples. However, the dynamic correlation between the GPA index and crude oil prices met a positive value during the period when extreme political events happened. This means that GPA and crude oil prices were positively correlated during this period. Third, the dynamic correlation between geopolitical factors and crude oil prices showed strong volatility during periods of political tension. All coefficients of dynamic correlation between geopolitical factors and the crude oil prices, except the GPR-Brent pair and GPT-WTI pair, had larger amplitudes during periods of political tension. Besides, all dynamic correlation coefficients, except the GPR-Brent pair and GPT-WTI pair, also had larger standard deviations during periods of political tension. The results also revealed that there was unidirectional Granger causality running from geopolitical factors to crude oil prices in the full sample period.
The political implications of these findings are that (1) the level of attention to traditional oil exporting countries needs to be strengthened. In the future, the Middle East will remain the world's main oil exporting region. According to the "World Energy Outlook 2035" released by British Petroleum (BP), the oil production of OPEC (organization of the petroleum exporting countries) countries in the Middle East will increase by 7 million barrels per day by 2035. The stability of the geopolitical situation in this region will directly affect the trend of international oil; (2) the serious challenges posed by international terrorism to international oil prices require close attention. As far as the current situation is concerned, terrorist attacks will not significantly change the general direction of the global oil market. However, as terrorism spread globally, the fluctuations in international oil prices caused by terrorist incidents have become more apparent. At the same time, the occurrence of terrorist incidents will lead to an increase in global risk sentiment, which will surely cause panic in the global investment field.
This study opens new avenues for future research. As a first extension, further research may examine the risk spillover from geopolitical factors to crude oil prices, such as the empirical framework of Liu et al. [57]. As a second extension, future research could detect the mediating variables or the moderating variables in the correlation between geopolitical factors and crude oil prices.