Geopo Riza D Rangan Qiang J Aviral Litical Ris Demirer N Gupta Ji Kumar
Total Page:16
File Type:pdf, Size:1020Kb
University of Pretoria Department of Economics Working Paper Series Geopolitical Risks and the Predictability of Regional Oil Returns and Volatility Riza Demirer Southern Illinois University Edwardsville Rangan Gupta University of Pretoria Qiang Ji Chinese Academy of Sciences and University of Chinese Academy of Sciences Aviral Kumar Tiwari Montpellier Business School Working Paper: 2018-60 September 2018 __________________________________________________________ Department of Economics University of Pretoria 0002, Pretoria South Africa Tel: +27 12 420 2413 Geopolitical Risks and the Predictability of Regional Oil Returns and Volatility Riza Demirer*+, Rangan Gupta**, Qiang Ji*** and Aviral Kumar Tiwari**** *Department of Economics & Finance, Southern Illinois University Edwardsville, Edwardsville, IL 62026- 1102, USA. Email: [email protected] **Department of Economics, University of Pretoria, South Africa. Email: [email protected] ***Center for Energy and Environmental Policy Research, Institutes of Science and Development, Chinese Academy of Sciences, Beijing, 100190, China; School of Public Policy and Management, University of Chinese Academy of Sciences, Beijing, 100049, China. Email: [email protected] ****Montpellier Business School, Montpellier, France. Email: [email protected] Abstract This paper hypothesizes that global geopolitical risks (GPRs) can predict oil market return and volatility. For our purpose, we use a k-th order nonparametric causality-in- quantiles test, applied to a daily data set covering the period of 15th May, 1996 to 31st May, 2018 of six oil prices (the Nigerian Bonny Light, Brent, Dubai, OPEC, Tapis, and WTI). Our results indicate that the relationship between oil returns and GPRs is highly nonlinear and hence, linear tests of Granger causality cannot be relied upon. Based on the data-driven econometric method, we observe that GPRs have predictability for oil returns of the West African Bonny Light, OPEC and Tapis, while in terms of volatility, causality is observed for all oil prices barring the case of Dubai. In sum, the impact of GPRs is primarily on volatility of oil markets, but more importantly, the impact of GPRs is not uniform across the oil markets. Keywords: Geopolitical Risks, Oil Prices, Nonparametric Causality-in-Quantiles Test JEL Codes: C22, C32, Q41 + Corresponding author. 1 1. Introduction Oil market movements (in both return and volatility) are known to predict recessions (Hamilton, 1983, 2008, 2009, 2013; Elder and Serletis, 2010), inflation (Stock and Watson, 2003), as well as other commodity and financial markets (Gupta and Yoon, 2018). In addition, oil is indispensable for industrial, transportation, and agricultural sectors, whether used as feedstock in production or as a surface fuel in consumption (Mensi, et al., 2014). Naturally, accurate prediction of oil market movements is of importance to academics, investors and policymakers alike. Understandably, there exists a large literature (see Baumeister, 2014; Lux et al., 2016; Degiannakis and Filis 2017a, b; and Gupta and Wohar, 2017 for detailed reviews) aiming to predict oil price movements using various types of econometric methodologies (univariate and multivariate; linear and nonlinear), and predictors (macroeconomic, financial, behavioural, institutional). In this regard, more recent studies by Bloomberg et al., (2009), Fattouh (2011), Antonakakis et al., (2017a, b), Caldara and Iacoviello (2018), Cunado et al., (2018), Demirer et al., (2018), and Plakandaras et al., (2018) have related oil price movements with geopolitical risks (GPRs).1 These studies point out that, since GPRs affect the economic conditions of both developed and emerging markets, and oil prices are functions of the state of the economy, it is expected that oil market movements are likely to be affected by risks associated with geopolitical events through the oil- demand channel. In addition, with GPRs also affecting financial markets (Balcilar et al., 2018a), and with oil and financial markets closely connected, such risks can also affect the oil prices indirectly through asset markets. All the above-mentioned studies relating GPRs with movements in the oil markets, however, are restricted to either the West Texas Intermediate (WTI) oil price, or a measure of world oil price, via the U.S. Crude Oil Imported Acquisition Cost by Refiners. However, there is widespread evidence that the possibility of a global oil market via integration is at best, sample period- or regime-specific, and does not necessarily hold at all points in time (Ji and Fan, 2015, 2016; Jia et al., 2017; Bhanja et al., 2018). Given this, there is no guarantee that the impact of GPRs on WTI or a measure of global oil price can be generalized to other major crude oil prices like Bonny Light, Brent, Dubai, Organization of the Petroleum Exporting Countries (OPEC), and Tapis. This is an important consideration as regional oil markets can have specific risk and wealth implications for the markets they serve and any external shock could have severe impacts on return and volatility dynamics in regional markets due to the market frictions in these markets that investors do not necessarily experience with the widely traded oil types like Brent or WTI. Against this backdrop, the objective of this paper is to analyze the impact of overall GPRs, and risks originating from geopolitical threats and actual attacks, on returns and volatility of the Nigerian Bonny Light, Brent, Dubai, OPEC, Tapis, and WTI oil markets. For our purpose, we use daily data covering the common sample period of 15th May, 1996 to 31st May, 2018, and conduct the predictability analysis based on the k-th order nonparametric causality-in-quantiles test recently developed by Balcilar et al., 1 Guo and Ji (2013), and Ji and Guo (2015) are somewhat related papers based on internet (Google) searches of events associated with the oil market, and its impact on oil returns and volatility. 2 (2018b). As indicated by Balcilar et al., (2018b), the causality-in-quantile approach has the following novelties: First, it is robust to misspecification errors as it detects the underlying dependence structure between the examined time series, which could prove to be particularly important as it is well-known (based on the literature discussed above) that oil returns display nonlinear dynamics with respect to its predictors – something that we show to exist formally via statistical tests in our case as well. Second, via this methodology, we are able to test not only for causality-in- mean (1st moment), but also for causality that may exist in the tails of the joint distribution of the variables, which in turn, is important if the dependent variable has fat-tails – a feature we show below to hold for oil returns. Finally, we are also able to investigate causality-in-variance and, thus, study impact on volatility. Such an investigation is important because, during some periods, causality in the conditional- mean may not exist while, at the same time, higher-order interdependencies may turn out to be significant. To the best of our knowledge, this is the first paper that evaluates the predictive power of various global geopolitical risks for different oil market returns and volatility based on a nonparametric causality-in-quantiles framework. An important issue to highlight at this stage is the realization that measuring geopolitical risks, which has traditionally been associated with terror attacks only, and hence modelled via a dummy, is much broader and hence, not straight-forward to capture and incorporate into time-series models involving continuous data. However, Caladara and Iacoviello (2018) has constructed indices of GPRs by counting the occurrence of words related to geopolitical tensions in leading international newspapers, and circumvent the above-mentioned issues. Given this, in our predictability exercise, we use the various GPRs indexes developed by these authors. In addition, it is also important to point out that, unlike the existing studies dealing with oil market movements and geopolitical risks based on low-frequency (monthly) data, we rely on daily data. This, we believe, is important, given that oil price is considered to act as a leading indicator for the economy, and hence, prediction of oil price movement, at a higher frequency would provide policymakers information about the future path of lower-frequency variables like output and inflation. The rest of this paper is organized as follows: Section 2 describes the econometric framework involving the higher- moment nonparametric causality-in-quantiles test. Section 3 presents the data and discusses the empirical results, with Section 4 concluding the paper. 2. Econometric Framework In this section, we briefly present the methodology for the detection of nonlinear causality via a hybrid approach as developed by Balcilar et al., (2018b), which in turn is based on the frameworks of Nishiyama et al., (2011) and Jeong et al., (2012). We start by denoting the six oil returns (considered separately) by yt and the predictor variable (in our case, the various GPR indexes considered one at a time, as discussed in the data segment in detail) as xt. We further let Yt1 (yt1,...,ytp ) , X t 1 (xt1 ,..., xt p ) , Zt (X t ,Yt ) and Fy |Z (yt ,Zt1) and Fy |Y (yt ,Yt1 ) denote t t1 t t 1 the conditional distribution functions of yt given Z t1 and Yt1 , respectively. Denoting Q (Zt1 ) Q (yt | Zt1 ) and Q (Yt1 ) Q (yt | Yt1 ) , we obtain 3 F {Q (Z )| Z } with probability one. As a result, the (non)causality in the yt |Zt1 t1 t1 -th quantile hypotheses to be tested are: H : P{F {Q (Y )|Z } } 1, (1) 0 yt|Zt1 t1 t1 H : P{F {Q (Y )|Z } } 1. (2) 1 yt |Zt1 t1 t1 Jeong et al. (2012) use the distance measure J {t E(t | Zt1) fz (Zt1)}, where t is the regression error term and f z (Zt1) is the marginal density function of Zt1. The regression error t emerges based on the null hypothesis in (1), which can only be true if and only if E[1{yt Q (Yt1) | Zt1}] or, expressed in a different way, 1{yt Q (Yt1)} t , where 1{} is the indicator function.