1. Introduction
Accounting for, respectively, 19.9% and 16.4% of world oil consumption in 2021,Footnote 1 the USA and China are two key players in the oil market. While the major role of the USA has been established for many years, that of China dates to the mid-2000s when the boom in oil prices was mainly driven by growth in emerging markets and, primarily, China. This increasing role of the Chinese economy has been accompanied by tensions between the two countries, with significant impacts on the oil market. The US-China trade war provides an emblematic example with a succession of threats and tariffs that have affected oil market fundamentals, such as oil supply and demand. The present paper tackles this issue and aims to investigate the dynamics of the oil market by accounting for the potential asymmetric effects of US-China political tensions on oil prices depending on their level.
Identifying the factors contributing to the explanation of oil prices has been a long-standing topic of study.Footnote 2 Several contributions argue that exogenous political events, like terrorist attacks or wars, are the primary cause of oil price fluctuations (see, e.g., Hamilton, Reference Hamilton2003, Reference Hamilton2009a). These explanations are particularly interesting for analyzing oil price changes in the wake of the two oil price shocks of the 1970s. However, the literature evolved toward a more nuanced view where exogenous political events are only a part of the explanation. Various empirical studies have shown that market forces (global demand, global supply, inventories, precaution demand, speculative demand) also play an essential role in driving oil price fluctuations (Bodenstein et al. Reference Bodenstein, Guerrieri and Kilian2012; Lippi and Nobili, Reference Lippi and Nobili2012; Baumeister and Peersman, Reference Baumeister and Peersman2013; Kilian and Hicks, Reference Kilian and Hicks2013; Kilian and Lee, Reference Kilian and Lee2014; Kilian and Murphy, Reference Kilian and Murphy2014; Cross, et al. Reference Cross, Nguyen and Tran2022).
Our paper falls into this strand of the literature by considering the role of both political tensions between the USA and China and global market forces in explaining oil price fluctuations. The literature linking political tensions between the two countries and the oil market dynamics is inexistent, apart from Cai et al. (Reference Cai, Mignon and Saadaoui2022) which is the study closest to ours. Relying on the structural vector autoregressive (SVAR) methodology over the 1971–2019 period, the authors show that US-China political tensions pull down oil demand and raise supply at medium- and long-run horizons.
We go further than Cai et al. (Reference Cai, Mignon and Saadaoui2022) by considering possible asymmetric impacts of political tensions and market forces on the oil price dynamics, depending on both the level of real oil prices and the period. The effect of the explanatory variables may indeed change over time—especially if the period under study is long—but also according to the level reached by the oil price. As it is well known, the role of China as an international key player mainly starts in the mid-2000s, making it relevant to rely on time-varying schemes. Regarding asymmetric effects, it is worth mentioning that oil demand may increase if prices are low but does not necessarily decrease in high oil price regimes as there is no immediate substitution possibility. Similarly, the impact of political tensions on the oil market may differ depending on the oil price level.
To account for such asymmetric effects, we rely on the quantile autoregressive distributed lag (QARDL) model developed by Cho et al. (Reference Cho, Kim and Shin2015). This framework enables us to consider the existence of long-term, cointegrating relationships between oil prices and their determinants that can vary across quantiles, that is, according to the level of oil prices. Furthermore, such a specification allows for locational asymmetry because the estimated parameters can vary depending on the location of oil prices within their conditional distribution. In other words, compared to the usual ARDL method, the QARDL approach has the advantage of introducing potential asymmetries in various levels reached by oil prices. To go a step further, given that our period under study covers more than 65 years, we extend the QARDL model to a time-varying QARDL specification to account for the time-varying nature of the cointegrating relationship.
Our contribution to the existing literature is manifold. First, we do not focus exclusively on political tensions but also consider market forces to explain the dynamics of oil prices. In this respect, we go further than Qin et al. (Reference Qin, Hong, Chen and Zhang2020), who do not account for market determinants. Second, regarding political tensions, we focus on the US-China relations and their impact on the oil market. Although these countries are two key players, the influence of their relations on the oil market has not been investigated, except by Cai et al. (Reference Cai, Mignon and Saadaoui2022). Third, we account for potential asymmetries using an appropriate methodology, namely, quantile regressions. In this sense, we go further from Cai et al. (Reference Cai, Mignon and Saadaoui2022), who only consider linear effects, and from Qin et al. (Reference Qin, Hong, Chen and Zhang2020), who also rely on the quantile framework but do not investigate the role of US-China political tensions. Fourth, we use the US-China political relations index built by the Institute of International Relations at Tsinghua University to assess political tensions. From this viewpoint, we fill a gap in the literature which is mainly based—except Cai et al. (Reference Cai, Mignon and Saadaoui2022)—on the use of the geopolitical risk (hereafter GPR) index introduced by Caldara and Iacoviello (Reference Caldara and Iacoviello2018). Although it is of interest, GPR does not focus on the relationship between the USA and China; instead, it provides an overall picture of the geopolitical uncertainty for China in contrast to the US-China political relation index that focuses on the bilateral relationship between the two countries.
Our results show the existence of a quantile-dependent cointegrating relationship between oil prices and their determinants, namely, US-China political tensions, world oil demand, and global oil supply over the period ranging from January 1958 to March 2022. In particular, the effect of tensions between the two countries is exacerbated in times of high oil prices. Moreover, we find that this quantile-dependent cointegrating relationship is time varying across quantiles. This finding is particularly interesting as it highlights the increased role played by China in the oil market since the mid-2000s.
The rest of the paper is organized as follows. Section 2 reviews the related literature. Section 3 describes the QARDL methodology and data and provides some preliminary analysis. The estimation results are displayed and discussed in Section 4. Section 5 concludes the paper.
2. Literature review
In this literature review, we focus on studies that explore the followingFootnote 3: (i) the relative importance of market and political forces in driving the real oil price dynamics, (ii) the existence of time-varying patterns in the oil market, and (iii) the consideration of asymmetric dynamics. The first set of studies is directly linked to our work since it relies on a quantitative measure of political tensions to estimate the influence of political events on the dynamics of the real price of oil. The second set of investigations, which explores the existence of time-varying patterns in the oil market, is also worth mentioning since the Chinese role on the international scene—and thus on the oil market—has been primarily at play since the mid-2000s (Cross, et al. Reference Cross, Nguyen and Zhang2022). Turning to the third set of studies dealing with asymmetries in the oil market, the underlying idea is that the impact of political tensions could differ depending on the oil price quantiles. Overall, this section provides an overview of studies that consider time-varying patterns and asymmetries in the relationship between the real price of oil and political tensions.
Cai et al. (Reference Cai, Mignon and Saadaoui2022) are the first attempt to disentangle the causal effects of market forces (oil supply and demand) and exogenous political events relying on a quantitative measure for political tensions between the USA and China.Footnote 4 Using an SVAR model with monthly data over the period spanning from January 1971 to December 2019, they show that bilateral political tensions have a causal impact on oil demand and supply. During the whole period, a deterioration of the political relationship between the USA and China induces a decrease in oil demand and an increase in oil supply. As explained by the authors, the uncertain climate resulting from these conflicting relations between the two countries deteriorates oil demand, reflecting threats of a global economic downturn. Since the Organization of the Petroleum Exporting Countries (OPEC) has no control over prices in this case, it is not inclined to reduce its production. The authors also show that these effects may change according to the current state of the political relations, suggesting a time-varying pattern for these causal mechanisms.
Before Cai et al. (Reference Cai, Mignon and Saadaoui2022), various contributions have explored the quantitative impact of political tensions and GPRs on the oil price dynamics.Footnote 5 Chen et al. (Reference Chen, Liao, Tang and Wei2016) use the political risk associated with OPEC countries as a measure of political tensions, thanks to a transformation of the well-known International Country Risk Guide (hereafter ICRG) index. Their SVAR analysis shows that the two main contributors to oil price fluctuations are political risk shocks and demand shocks over the January 1998–September 2014 period. Interestingly, they find that political risk shocks in the Middle East positively influence oil prices, whereas political risk shocks in North Africa and South America have no impact. Lee et al. (Reference Lee, Lee and Ning2017) extend Chen et al. (Reference Chen, Liao, Tang and Wei2016)’s paper to the G7 countries and find that political risk shocks in the USA have a different impact on the world economy given the size of the US economy and the status of the dollar in the international monetary system.
Beyond the use of the ICRG index, Miao et al. (Reference Miao, Ramchander, Wang and Yang2017) and Perifanis and Dagoumas (Reference Perifanis and Dagoumas2019) proxy the GPR factors with the number of terrorist attacks coming from the Global Terrorism Database.Footnote 6 Specifically, Miao et al. (Reference Miao, Ramchander, Wang and Yang2017) examine the predictability of crude oil prices using daily data from January 4, 2002, to September 25, 2015. Relying on Least Absolute Shrinkage and Selection Operator methods, they find that market (demand, supply, speculation) and geopolitical (captured by the number of terrorist attacks in the Middle East and North America (MENA) region) factors are the most important determinants of oil prices. Perifanis and Dagoumas (Reference Perifanis and Dagoumas2019) investigate the long-run relations between various determinants of oil prices—demand, supply, speculation variables, US shale oil production, and the number of terrorist attacks in the MENA region—over the 2008–2017 period with monthly data. They show that oil prices are mainly driven by fundamentals such as consumption, OPEC production, or US shale oil production but do not find a significant impact of indicators measuring political instability such as the number of terrorist attacks in oil-producing countries.
Along with studies using quantitative measures of political tensions, Caldara and Iacoviello (Reference Caldara and Iacoviello2022) introduce the GPR in the empirical literature. GPR is a monthly index generated by running automated text searches on the electronic archives of 11 North American and British newspapers, available since 1985.Footnote 7 Assuming that the GPR index only contemporaneously reacts to its own shocks, the authors offer an interesting distinction between geopolitical acts and geopolitical threats. Indeed, they find that geopolitical acts reduce uncertainty and produce minor economic effects. Besides, geopolitical threats increase uncertainty (especially for firms) and may have larger economic consequences than the occurrence of conflicts, as underlined by Bloom et al. (Reference Bloom, Bond and Van Reenen2007). In their empirical investigation, the authors show that an increase in the GPR index leads to a short-lived decrease in oil prices of around 7% after 3 months over the 1985–2016 period.
Abdel-Latif and El-Gamal (Reference Abdel-Latif and El-Gamal2020) use the GPR index to investigate the interactions between oil prices, financial liquidity, and GPRs. They estimate a quarterly global VAR over the 1979–2017 period for 53 countries, arguing that financial liquidity and GPRs are endogenous to the USA. They show that one-standard-deviation shocks to the GPR index (i) induce a persistent and significant increase in oil prices of around 4% and (ii) harm investment, especially in the MENA region for commodity exporters (like Saudi Arabia, Bahrain, or Qatar). Besides, the impact on investment is more mixed for other countries which are not commodity exporters.
Baumeister and Peersman (Reference Baumeister and Peersman2013) explore the role of time-varying elasticities for oil demand and supply in explaining the coexistence between reduced oil production volatility and larger oil price volatility. They recall that several reasons may explain some gradual time variation in the parameters. Firstly, spot markets have gradually become increasingly important relative to long-term oil contracts (Hubbard, Reference Hubbard1986). Secondly, investments in the oil sector require a long-time span, and the response to price incentives may be gradual (Hamilton, Reference Hamilton2009b; Smith, Reference Smith2009). Capacity constraints in oil production may thus exhibit some time-varying patterns depending on the historical episode under scrutiny (Kilian, Reference Kilian2008). Thirdly, the quest for a substitute for oil production took place over an extended time period (Dargay and Gately, Reference Dargay and Gately2010). Efforts toward energy conservation are reflected in the slow variation in the energy share over time, and the effect of oil prices on consumption varies smoothly (Edelstein and Kilian, Reference Edelstein and Kilian2009).
Along with studies that acknowledge time-varying schemes in the oil market (see, e.g., Herrera et al. (Reference Herrera, Karaki and Rangaraju2019) for a brief survey), some authors find that the relationship between oil prices and geopolitical events could also exhibit time-varying patterns (Noguera-Santaella, Reference Noguera-Santaella2016; Monge et al. Reference Monge, Gil-Alana and de Gracia2017; Song et al. Reference Song, Chen, Hou and Yang2022). Using monthly data from September 1859 to March 2013, Noguera-Santaella (Reference Noguera-Santaella2016) examines 32 major geopolitical events (from the 1861–1865 American Civil War to the Arab Spring started in December 2010) to estimate their impact on oil prices and their volatility. Relying on both AR(1) and GARCH(1,1) models along with dummy variables accounting for the geopolitical events, he found that 6 out of 32 geopolitical events have had an impact on real oil prices, all of them occurring before 2000. Moreover, 20 out of 32 geopolitical events have affected volatility, with 17 events occurring before 2000. During the 2000s, market forces appear to be the main driver of oil price fluctuations.
Monge et al. (Reference Monge, Gil-Alana and de Gracia2017) investigate the statistical properties of real oil prices before and after important geopolitical events. Using monthly data between January 1946 and November 2014, they examine the unit root properties of six geopolitical events previously identified in the literature (the 1973 Yom Kippur War followed by the Arab oil embargo in 1973/1974, the 1978/1979 Iranian Revolution, the 1980–1988 Iran-Iraq War, the 1990/1991 Persian Gulf War, the 2002 Venezuelan crisis and the 2003 Iraq War, and the 2011 Libyan uprising). Using fractional unit root tests, they show that the real price of oil is stationary but follows a long-memory process (the order of integration being equal to
$ d = 0.78$
in the “best” model). After establishing the presence of two structural breaks in the oil price series (in October 1973 and October 1990), they estimate the order of integration for the three sub-periods and find that the null of mean reversion is rejected, even when nonlinear time trends are accounted for. Considering a window size of 120 months centered around each of the six major geopolitical events, they find some evidence of time-varying patterns as the persistence in the series is stronger when the window size increases, as witnessed by the monotonic increase in the integration order,
$d$
.
Following the contributions of Coleman (Reference Coleman2012) and Miao et al. (Reference Miao, Ramchander, Wang and Yang2017), Song et al. (Reference Song, Chen, Hou and Yang2022) explore the time-varying interactions between oil prices and terrorist attacks. They distinguish between the number of terrorist incidents per month and their brutality, proxied by the number of fatalities per attack each month. Using time-varying causality tests based on a bivariate VAR over the January 1995–December 2018 period, they show that terrorist incidents Granger-caused oil prices between (i) January 2001 and March 2003, (ii) October 2008 and September 2008, and (iii) February 2015 and January 2016. During the first episode, the influence of terrorist incidents on oil prices was negative, while being positive for the two following periods.
Finally, two contributions are worth mentioning as they rely on quantile regressions to explore asymmetries and the heavy tail behavior of oil prices, together with the impact of GPRs and political uncertainty (Qin et al. Reference Qin, Hong, Chen and Zhang2020; Apergis et al. Reference Apergis, Hayat and Saeed2021).Footnote 8 Using quantile regressions with daily data over the period spanning from June 28, 1990, to October 31, 2018, Qin et al. (Reference Qin, Hong, Chen and Zhang2020) investigate the asymmetric effects of GPRs on energy (including oil) returns and volatility. They reject the stability of the impact of GPRs on energy across quantiles, with GPRs harming crude oil returns in the case of a bearish market. They also show that geopolitical threats positively impact crude oil volatility from the quantile 0.3 but have no significant effect for lower quantiles. Unfortunately, they do not control for market forces (energy demand and production) in their regressions.
Using monthly data between 2001 and 2019, Apergis et al. (Reference Apergis, Hayat and Saeed2021) examine the existence of asymmetries in the impact of US partisan political uncertainty on oil prices, thanks to a QARDL model and a partisan conflict index built by Azzimonti (Reference Azzimonti2018). In their main regressions, they put emphasis on the growth channel to explain such a potential asymmetric effect. Higher uncertainty puts a strain on economic growth and reduces oil demand, which in turn provokes a decrease in oil prices. The authors find (i) a cointegrating relationship only for quantiles greater than 0.5 and (ii) a positive impact of US partisan political uncertainty on oil prices from the quantile 0.5 as well. This result shows that political uncertainty matters only in a bullish oil market.
Overall, the previous literature shows that (i) political events are important in explaining oil price dynamics, (ii) time-varying patterns are at play in the oil market, and (iii) there are asymmetries in the impact of political tensions and uncertainty on oil prices. Although the aforementioned papers focus on one of these features, the current study takes all of them into account simultaneously, allowing for a complete and meaningful investigation of political tensions on oil prices. Compared to the literature that focuses on the impact of political tensions on the oil market, our contributions are several. First, we go further than Cai et al. (Reference Cai, Mignon and Saadaoui2022)—the paper closest to ours—who do not account for potential asymmetries in the impact of political tensions on oil prices. Second, we focus on two fundamental market players, namely, China and the USA, using an appropriate measure of political relations between them—the US-China political relations index. In this sense, we depart from the previous literature, which mainly relies on the GPR index, which can be influenced by factors other than those specifically linked to the USA. Third, we account for asymmetries in the impact of political tensions on oil prices using a QARDL specification, which allows us to obtain quantile-varying cointegrating relationships depending on the level of oil prices. Finally, we acknowledge the existence of time-varying patterns by extending the QARDL model to a time-varying specification. To the best of our knowledge, the present paper is the first to address all those features in a single specification.
3. Methodology and data
3.1. Methodology
To assess whether asymmetric effects between oil prices and their determinants are at play, we rely on the QARDL framework developed by Cho et al. (Reference Cho, Kim and Shin2015). Such a specification allows us to estimate the coefficients at various quantile levels and, in turn, capture the asymmetric relationship between integrated series at different quantiles.
The QARDL model is specified as follows:
\begin{equation} wti_{t} = \alpha _*(\tau )+\sum _{j=1}^{p} \phi _{j*}(\tau ) wti_{t-j}+\sum _{j=0}^{q} \boldsymbol{\theta }_{j*}(\tau )' \textbf{X}_{t-j}+U_t(\tau ) \end{equation}
where
$wti_t$
denotes the real price of oil at time
$t$
,
$\textbf{X}_t$
is the matrix of explanatory variables,
$U_t(\tau )$
is the error term, and
$\tau \in (0,1)$
stands for the quantile level. In this specification, the quantile level of all variables is based on the quantile level of real oil prices.
Following Cho et al. (Reference Cho, Kim and Shin2015), equation (1) can be rewritten in the error-correction model (ECM) or form (i.e., the QARDL-ECM representation) as follows:Footnote 9
\begin{equation} \Delta wti_{t} = \alpha _*(\tau )+\zeta _*(\tau )(wti_{t-1}-\boldsymbol{\beta }_*(\tau )'\textbf{X}_{t-1})+\sum _{j=1}^{p-1} \phi _{j}^*(\tau ) \Delta wti_{t-j}+\sum _{j=0}^{q-1} \boldsymbol{\theta }_{j}^*(\tau )' \Delta \textbf{X}_{t-j}+U_t(\tau ) \end{equation}
where for
$j=1,\ldots,p-1$
:
\begin{equation} \phi _{j}^*(\tau )=-\sum _{h=j+1}^{p} \phi _{h*}(\tau ) \end{equation}
and
\begin{equation} \boldsymbol{\theta }_{j}^*(\tau )=-\sum _{h=j+1}^{p} \boldsymbol{\theta }_{h*}(\tau ) \end{equation}
with
$\boldsymbol{\theta }_{0}^*(\tau )=\boldsymbol{\theta }_{0*}(\tau )$
.
$\zeta _*(\tau )$
is the quantile error-correction coefficient given by
\begin{equation} \zeta _*(\tau )=\sum _{j=1}^{p} \phi _{j*}(\tau )-1 \end{equation}
and
$\boldsymbol{\beta }_*(\tau )$
denotes the vector of quantile long-run parameters.
As shown by equation (2), the short- and long-run parameters are quantile dependent, meaning that the QARDL coefficients can be affected by the shock
$U_t(\tau )$
at each point of time and, thus, can vary across quantiles. Cho et al. (Reference Cho, Kim and Shin2015) derive the full asymptotic theory for this QARDL-ECM specification and show that the estimators of the short- and long-run coefficients asymptotically follow a (mixture) normal distribution.
3.2. Data and preliminary analysis
We consider monthly data ranging from January 1958 to March 2022. For the oil-market-related variables, we rely on the real price of oil (
$wti$
), world oil demand (wip), and global oil supply (
$\textit{gop}$
). Our dependent variable,
$wti$
, is the WTI spot price deflated by the US consumer price index. Turning to world oil demand,
$\textit{wip}$
, we use the industrial production index measured by Baumeister and Hamilton (Reference Baumeister and Hamilton2019) by considering 23 Organisation for Economic Co-operation and Development countries and 6 major emerging economies (Brazil, China, India, Indonesia, the Russian Federation, and South Africa). Global oil supply corresponds to monthly oil production measured in thousands of barrels per day obtained from the US Energy Information Administration’s (EIA) Monthly Energy Review. All those variables are expressed in logarithmic terms and retrieved from Christiane Baumeister’s website.Footnote 10 We update the data until March 2022, following the methodology presented in Baumeister and Hamilton (Reference Baumeister and Hamilton2019).
Our main variable of interest is the US-China political relation index (
$\textit{pri}$
), which allows us to assess the effects of US-China political tensions on the oil market.Footnote 11 This index is divided into 6 sections, ranging from −9 to 9, which classify political relations as confrontation (−9), rival, disharmonious, common, harmonious, and friendly (9).
Figures A1 and A2 in Appendix display the evolution of our four considered variables, and Table A1 reports some basic descriptive statistics for the first-differenced series. All variables exhibit high kurtosis levels and depart from Gaussianity, as shown by the Jarque-Bera test. This property adds support for using quantile methods to provide robust inference.
Indeed, as recalled by Koenker and Xiao (Reference Koenker and Xiao2004), usual unit root tests may be characterized by poor power performance under departures from the Gaussian case, especially for distributions with heavy tails like ours. To overcome this limit, we apply Koenker and Xiao (Reference Koenker and Xiao2004)’s quantile unit root tests, which consist in examining the unit root property in each quantile separately. By providing a detailed look at the dynamics of the series, these tests allow us to detect possible asymmetries, that is, to consider different adjustment mechanisms toward the long-run equilibrium value—mean-reverting behavior—at different quantiles.
Briefly speaking, Koenker and Xiao (Reference Koenker and Xiao2004)’s quantile unit root test consists in extending the usual Augmented Dickey-Fuller (ADF)-type regression:
\begin{equation} y_t=\alpha + \rho y_{t-1}+\sum _{j=1}^{p-1}\phi _j \Delta y_{t-j}+u_t \end{equation}
as follows:
\begin{equation} Q_y(\tau \vert I_{t-1})=\alpha (\tau ) + \rho (\tau ) y_{t-1}+\sum _{j=1}^{p-1}\phi _j(\tau ) \Delta y_{t-j}+Q_u(\tau ) \end{equation}
where
$Q_y(\tau \vert I_{t-1})$
denotes the
$\tau$
-th quantile of
$y_t$
conditional on the past information and
$Q_u(\tau )$
is the
$\tau$
-th quantile of
$u_t$
.
$\rho (\tau )$
measures the speed of mean reversion of
$y_t$
within each quantile
$\tau$
, and the test consists of testing the unit root null hypothesis, that is,
$\rho (\tau )=1$
.
The results displayed in Tables B1–B4 in Appendix indicate that all series are not constant unit root processes, that is, there is an asymmetry in persistence. Specifically, the autoregressive coefficient,
$\rho (\tau )$
, augments when we move from lower to higher quantiles for the real oil price and the US-China political relation index. This result is particularly interesting as it shows that these two series are more stationary during low oil price episodes and conflicting relationships between the two countries than during regimes of high prices and weak tensions. In other words, during high oil price periods and friendly relationship episodes, political tensions and the price of oil itself tend to be more persistent. This finding is consistent with the fact that political tensions are likely to be exacerbated in times of a bullish oil market. At the 5% significance level, the unit root hypothesis is not rejected for quantiles higher than 0.4 for
$\textit{wti}$
and 0.1 for
$\textit{pri}$
contrary to lower quantiles, illustrating asymmetric adjustment dynamics of both series.
Turning to oil demand and supply, the results in Tables B2 and B3 in Appendix indicate that the unit root hypothesis is not rejected at lower quantiles, while it is the case at higher quantiles. This result is logical as it suggests that when oil demand and supply are high, they tend to exhibit a mean-reverting behavior.
Overall, this preliminary analysis illustrates the relevance of the quantile framework by highlighting asymmetry phenomena.Footnote 12 Furthermore, the existence of different behaviors of the series in terms of persistence depending on the quantiles justifies investigating the dynamics of their relationship at various quantiles through the use of QARDL models.
4. Empirical results
4.1. Quantile ARDL error-correction model
Our preliminary analysis suggests that the impact of the explanatory variables on oil prices may be heterogeneous across quantiles, that is, may depend on the location of oil prices within their conditional distribution. To address this hypothesis and assess the stability of the long-term relationship across the quantiles, we estimate the QARDL-ECM given by equation (2) with
$\textbf{X}_t=(\textit{wip}_t,\textit{gop}_t,\textit{pri}_t)$
.Footnote 13
Table 1 and Figure 1 report the estimation results of the long-term part of the QARDL-ECM—that is, the dynamic trends of the estimated coefficients associated with the error-correction term and the variables in levels (long-run cointegrating coefficients). In addition to the quantile estimates of the four parameters of interest, Figure 1 provides the 90% confidence intervals against quantile indices ranging from 0.05 to 0.95.Footnote 14
As shown, the error-correction term is significantly negative for quantiles greater than 0.4 and increases in absolute terms when moving to higher quantiles. In other words, the higher the oil price, the stronger the adjustment speed toward the long-term equilibrium. This result highlights the existence of asymmetries as mean reversion is at play when oil prices are high, whereas there is no cointegration at lower quantiles in line with Apergis et al. (Reference Apergis, Hayat and Saeed2021)’s conclusions. Such asymmetric behavior is consistent with the fact that the negative impact of an oil price increase on the economy is stronger than the positive effect of a fallFootnote 15—some authors (e.g., Mork, Reference Mork1989) have even shown that an oil price decrease has no impact on economic activity.
Table 1. Quantile regressions for variables in levels
Note: The number of observations is equal to 771.
$ECM$
is the value of the error-correction coefficient. Wald tests for symmetric quantiles and slope equality strongly reject the null hypotheses of symmetry between quantile and slope equality, respectively. Source: authors’ calculations.
One usual explanation of this asymmetry relies on the time required to set up additional production capacities: investment is not immediate, while the decline in the profitability of oil-consuming firms is rapid. Furthermore, according to Hamilton (Reference Hamilton1988), adjustment costs—due to sectoral imbalances, coordination failures between firms, etc.—may lead to an asymmetric response to the oil price change. Indeed, a rise in the price of oil slows down economic activity (directly and indirectly). In contrast, a fall can have both positive (direct) and negative (indirect) effects, which tend to compensate for each other. It should be noted that the price of petroleum products may also contribute to the asymmetric relationship between the price of oil and economic activity, as gasoline prices increase more quickly when the price of oil rises than they fall when the price decreases. Finally, Bernanke et al. (Reference Bernanke, Gertler and Watson1997) put forward a possible role for monetary policy in explaining the asymmetry phenomenon: while in the case of a rise in the oil price, monetary authorities pursue a restrictive policy to fight inflation, they do not react when the price falls. This difference in the reaction of monetary authorities to a rise and a fall in the oil price provides a way to explain the asymmetry phenomenon.
Turning to oil demand, the associated coefficient decreases as the oil price rises but remains significant all the time. When prices are high, the impact of demand on prices is smaller than for the low quantiles, but the effect remains significant. This result is logical because the elasticity of demand to oil prices is very low in the short run. Thus, even if prices are high, the economy’s needs for oil are still present, and there is no possibility of—short-term—substitution.
Regarding the coefficient related to global oil supply, it is always significant and increases when moving to higher quantiles. It is negative for the low quantiles and becomes positive from the 0.6 quantile onward. When prices are low, an increase in production tends to lower prices, leading producers to generally decrease their production to raise prices. When prices are high, increased production positively impacts prices, but the effect tends to decrease in the case of a strong bullish market. Producers do not reduce their production when prices are high (but may do so when prices reach very high levels), which is consistent with the low elasticity of demand and the fact that they benefit from rising prices.
Finally, the impact of political tensions is always significant. This is not surprising as, as stressed above, China and the USA are two key players in the oil market. The existence of tensions between the two countries is thus likely to impact the dynamics of oil prices. A recent example is the period of the Trump administration, during which the relationship between the two countries has been tense due to various events such as the trade war, Hong Kong’s judicial independence, intellectual property issues, and the international status of Taiwan. As recalled by Cai et al. (Reference Cai, Mignon and Saadaoui2022), the trade war between China and the USA caused a fall in Chinese production due to (i) the increase in tariffs on imported Chinese goods imposed by the USA and (ii) the increase in tariffs on a list of some imported US products decided by China. This fall in Chinese production led to a fall in consumption and demand for oil. Combined with strong supply, this fall in oil demand has led to significant volatility in crude oil prices. Such an episode, among many others, illustrates that political tensions between China and the USA significantly impact the oil market.
Figure 1. QARDL error-correction model.
Note: This figure reports the estimated parameters (solid middle line) using all available observations for different quantile levels (0.05, 0.10, . . ., 0.95) with 90% confidence intervals (outer dotted lines). Source: authors’ calculations.
As shown in Table 1, the impact of political tensions is positive and augments with the price of oil: the higher the quantiles, the greater the influence of political tensions. This finding is particularly interesting as it indicates that the effect of political tensions between China and the USA is exacerbated in times of high oil prices—confirming the quantile unit root test conclusions. Specifically, when US-China relations improve, oil demand and supply tend to rise significantly, reflecting the favorable outlook for global economic activity and, in turn, boosting oil prices. This result is also in line with Qin et al. (Reference Qin, Hong, Chen and Zhang2020) who have shown that the impact of GPRs varies across quantiles.
To assess the robustness of our findings to potential endogeneity issues, we complement our analysis by using smooth instrumental variable quantile regressions (Chernozhukov and Hansen, Reference Chernozhukov and Hansen2005; Kaplan and Sun, Reference Kaplan and Sun2017; Kaplan, Reference Kaplan2022). The corresponding results are reported in Figure 2, in which the explanatory variables are instrumented by their first-month (12-month) lags. As shown and as expected for the long-run relationship, the conclusions are similar to those obtained with the quantile regressions for variables in levels, illustrating the robustness of our results.
Overall, our findings show the existence of location asymmetries between lower and medium-to-higher quantiles for the four key coefficients, with a quantile-dependent cointegrating relationship between oil prices and their determinants. Following Xiao (Reference Xiao2009) and Cho et al. (Reference Cho, Kim and Shin2015), such quantile-dependent cointegration may come from the fact that the underlying relationship between non-stationary series (for some quantiles) can vary over time because of heterogeneous shocks arriving at different dates. As argued by Cho et al. (Reference Cho, Kim and Shin2015), the quantile cointegrating framework is particularly suitable in such conditions as the quantile coefficients can be considered random parameters—randomness coming from a common shock arising at each point of time.
Figure 2. Smoothed instrumental variable quantile regression.
Note: This figure reports the estimated parameters (dotted middle line) using all available observations for different quantile levels (0.05, 0.10, . . ., 0.95) with 95% confidence intervals. The explanatory variables are instrumented with their 1-month lags. Similar results are found when we use 12-month lags. Source: authors’ calculations.
Figure 3. Time-varying QARDL error-correction model.
Note: The parameters are estimated using the rolling window method, and each window has a size of 320 observations. The first date on the horizontal axis is August 1984. The number of out-of-sample observations is 452. Source: authors’ calculations.
4.2. Time-varying QARDL-ECM model
Given that our period under study covers more than 65 years and to account for the time-varying nature of the cointegrating relationship, we extend our previous analysis by estimating a time-varying QARDL-ECM. This specification enables us to cumulate the benefits of time-varying parameter regressions to explain oil prices, with the ability of quantile regressions to model flexibly the whole distribution of oil prices. To this end, we re-estimate our specification (equation (2)) using the robust rolling estimation procedure proposed by Cho et al. (Reference Cho, Kim and Shin2015).Footnote 16 This allows us to investigate whether location asymmetries are monotonic over the whole period. Specifically, we re-estimate our QARDL-ECM, moving the estimation window forward one quarter at a time until the sample ends. The corresponding results are displayed in Figure 3, which shows the time-series plots of the rolling quantile estimates of the four parameters together with 90% confidence intervals.Footnote 17
Since our previous findings show evidence of cointegration for quantiles greater than 0.4, we focus on the results for
$\tau$
equal to 0.5 and 0.75. Figure 3 clearly shows that the rolling quantile estimates of the four coefficients were not highly time varying before the mid-2000s, locational asymmetry being weaker in the earlier periods of our sample. This is consistent with the fact that China was not a key player on the international scene before that date, explaining its negligible impact on the oil market. Important time-varying patterns in the parameters are observed after the mid-2000s due to the increasing role of China worldwide, highlighting the relevance of the time-varying approach.
This is confirmed by the strong time-varying evolution of the error-correction term after the mid-2000s. As shown, the speed of adjustment tends to increase after 2005, before stabilizing and decreasing since the mid-2010s, with an average adjustment speed that was higher for
$\tau =0.75$
from 2010 to 2015. In the recent period, the adjustment speed tends to be lower. The cointegrating relationship has been effective since the 2010s for
$\tau =0.5$
and around 2005 for
$\tau =0.75$
. This is consistent with the fact that no long-term equilibrium relationship was at play before China became a key player on the international scene and, in turn, on the oil market.
From this period, the oil-demand effect on prices becomes positive, with the corresponding coefficient following an upward trend. This clearly illustrates the major impact of Chinese demand on the price of oil: the rise in oil prices was driven by Chinese growth—and, therefore, Chinese demand. The rolling quantile estimates of the demand coefficient display a strong upward time-varying pattern. In other words, the demand effect on the oil price has increased over time, in line with China’s growing weight. As expected, a decreasing trend is observed in 2020−2021 due to the coronavirus disease 2019 pandemic and the subsequent reduction in economic activity.
Turning to oil supply, its coefficient has followed a downward trend since the 2007−2008 global financial crisis. This evolution can be explained by various factors, such as the spectacular shale oil and gas boom since 2009 that profoundly disrupted the oil market and OPEC’s supremacy. During the 2015−2020 period, prices sharply fell due to the battle for market share and global price fixing between key players such as Russia and Saudi Arabia. The negative estimated coefficients associated with oil supply during that period are thus consistent with those important changes affecting global oil production.
As shown in Figure 3, the impact of political tensions between China and the USA was relatively stable and positive before 2005, especially for
$\tau =0.75$
. The effect increased in the mid-2000s, that is, when China started to play a major role at the global level. The most interesting result is that this impact strongly changed and became negative in 2015 during the trade war with the USA. The deterioration of political relations between the two countries created uncertainty that has strongly affected the oil market, pulling prices down. At the end of the period, once the effect of the shock has passed, the impact of political tensions recovers a “normal” pattern, illustrating a “new normal” regime.
Overall, our findings show that the cointegrating relationship between the price of oil and its determinants is both quantile dependent and time varying across quantiles. This result is particularly interesting as it clearly highlights the increased role played by China in the oil market since the mid-2000s.
5. Conclusion
This paper assesses the role of political tensions between the USA and China and global market forces (oil demand and supply) in explaining oil price fluctuations. We pay particular attention to the potential existence of asymmetries between oil prices and their determinants depending on the level reached by oil prices, as well as possible time-varying effects across the January 1958−March 2022 period.
We show that oil prices and their determinants are not constant unit root processes, meaning there is an asymmetry in persistence. In particular, during periods of high oil prices and friendly relationships between China and the USA, political tensions and the price of oil itself tend to be more persistent.
We estimate a QARDL-ECM to account for such asymmetric behavior. Our main findings show that (i) the higher the oil price, the stronger the adjustment speed toward the long-term equilibrium, (ii) the effect of US-China political tensions is exacerbated during high oil price periods, and (iii) significant time-varying patterns in the parameters associated with oil prices’ determinants are observed.
Overall, our results highlight the increasing role of China in the oil market since the mid-2000s. Given that US-China political tensions are accentuated during a bullish oil market, and due to their significant impact on the oil market, special attention must be paid to the diplomatic relationships between the two countries. Limiting conflictual relationships helps to mitigate high price pressures, which is crucial in the current context of increased global inflation. As an extension, it would be interesting to investigate whether political tensions between China and countries other than the USA or between the USA and nations other than China help to explain oil price fluctuations.
Acknowledgments
We would like to thank the editor, Fredj Jawadi, and three anonymous referees for helpful remarks and suggestions. We are grateful to Jin Seo Cho and Sang Woo Park for sharing their MATLAB codes and precious suggestions and Qi Haixia for providing us with the updated database regarding the US-China political relation index. We also thank the organizers and all the participants at the 7th International Workshop on “Financial Markets and Nonlinear Dynamics” held in Paris on June 1–2, 2023, especially Gilles Dufrénot, Benk Szilard, and Timo Teräsvirta for helpful the comments and suggestions. Finally, we thank Frédérique Bec and Gilles de Truchis for helpful comments and discussions during the international workshop on “Non-stationarity, cyclostationarity and applications” held in Nanterre on June 5–7, 2023.
Appendix A. Descriptive statistics
Figure A1. Oil market fundamentals.
Note: This figure represents the evolution of the real price of oil (
$wti$
), world oil demand, and global oil supply. Data sources: Christiane Baumeister’s website.
Table A1. Descriptive statistics on first-differenced series
Source: authors’ calculations.
Figure A2. US-China political relations index.
Note: This figure represents the evolution of the US-China political relation index (
$pri$
). Data sources: Institute of International Relations’ website.
Appendix B. Quantile unit root tests
Table B1. Quantile unit root tests for the oil price
Note: We use 12 lags in the quantile unit root tests. The null is the presence of a unit root for the specified quantile
$\tau$
.
$\hat \rho (\tau )$
is the estimate of the largest autoregressive root at each quantile,
$\hat \rho (\textit{OLS})$
is the usual ordinary least squares (OLS) estimate of the autoregressive root, and
$ADF(\tau )$
is the quantile unit root test statistic. Source: authors’ calculations.
Table B2. Quantile unit root tests for the oil demand
Note: We use 12 lags in the quantile unit root tests. The null is the presence of a unit root for the specified quantile
$\tau$
.
$\hat \rho (\tau )$
is the estimate of the largest autoregressive root at each quantile,
$\hat \rho (\textit{OLS})$
is the usual OLS estimate of the autoregressive root, and
$ADF(\tau )$
is the quantile unit root test statistic. Source: authors’ calculations.
Table B3. Quantile unit root tests for the oil supply
Note: We use 12 lags in the quantile unit root tests. The null is the presence of a unit root for the specified quantile
$\tau$
.
$\hat \rho (\tau )$
is the estimate of the largest autoregressive root at each quantile,
$\hat \rho (\textit{OLS})$
is the usual OLS estimate of the autoregressive root, and
$ADF(\tau )$
is the quantile unit root test statistic. Source: authors’ calculations.
Table B4. Quantile unit root tests for the political tensions
Note: We use 12 lags in the quantile unit root tests. The null is the presence of a unit root for the specified quantile
$\tau$
.
$\hat \rho (\tau )$
is the estimate of the largest autoregressive root at each quantile,
$\hat \rho (\textit{OLS})$
is the usual OLS estimate of the autoregressive root, and
$ADF(\tau )$
is the quantile unit root test statistic. Source: authors’ calculations.
Appendix C. Symmetry tests
Figure C1.
$p$
-values of Wald statistics using rolling estimation.
Note:
$p$
-values of
$\boldsymbol{W}_{\boldsymbol{n}}(\beta )$
test statistics. The figures show the estimated
$p$
-values of the Wald tests, where
$\boldsymbol{W}_{\boldsymbol{n}}^{(1)}(\beta )$
tests
$\beta _*(0.25)=\beta _*(0.5)$
;
$\boldsymbol{W}_{\boldsymbol{n}}^{(2)}(\beta )$
tests
$\beta _*(0.5)=$
$\beta _*(0.75)$
;
$\boldsymbol{W}_{\boldsymbol{n}}^{(3)}(\beta )$
tests
$\beta _*(0.25)=\beta _*(0.75)$
; and
$\boldsymbol{W}_{\boldsymbol{n}}^{(4)}(\beta )$
tests
$\beta _*(0.25)=\beta _*(0.5)=\beta _*(0.75)$
. The coefficient
$\beta _1$
corresponds to the long-run elasticity of oil demand,
$\beta _2$
to the long-run elasticity of oil supply, and
$\beta _3$
to the long-run elasticity of political tensions. The parameters are estimated using the rolling window method, and each window has a size of 320 observations. The first date on the horizontal axis is August 1984. The number of out-of-sample observations is 452. Source: authors’ calculations.
