Crude oil price shocks and hedging performance A comparison of volatility models原油价格冲击和对冲表现英文文献

Crude oil price shocks and hedging performance: A comparison of volatility models Dohyun Chun a, Hoon Choa, Jihun Kimb, a College of Business, Korea Advanced Institute of Science and Technology, Seoul, Republic of Korea b KB Research, 115, Yeouigongwon-ro, Yeongdeungpo-gu, Seoul, Republic of Korea a b s t r a c ta r t i c l ei n f o Article history: Received 13 December 2018 Received in revised form 8 April 2019 Accepted 4 June 2019 Available online 15 June 2019 JEL classifi cation: C22 G11 G17 G32 Froma practicalperspective,itiscrucial tohedgethecrudeoilpriceriskinperiods ofdramatic pricechange.In this study, we directly investigate the performance of crude oil hedge portfolios in the fi ve periods in which the largest oil price shocks in history occurred. We use stochastic volatility (SV), GARCH, and the diagonal BEKK model to es- timate the minimum variance hedge ratio of hedge portfolios. Our empirical results provide evidence that hedging strategies based on the SV model are able to outperform the GARCH and BEKK models in terms of variance reduc- tion. Our results are also consistently valid for various hedge horizons. Interestingly, although it is important to es- timate variance and covariance accurately when constructing minimum variance portfolios, we fi nd that reducing the mean squared and mean absolute errors does not guarantee superior hedge performance. 2019 Elsevier B.V. All rights reserved. Keywords: Crude oil prices Hedging strategies Minimum variance hedge ratio Stochastic volatility model Crude oil price shocks 1. Introduction Globalized oil markets have historically suffered from large price shocks. Conceptually, the demand for crude oil is relatively inelastic to thesupply;consequently,decreasesintheoilsupplyhaveledtodramatic price changes. An early study by Hamilton (1983) claims that substantial disruptions in oil production primarily result in historical oil price shocks becausethemajorityofcrudeoilisproducedbyarelativelysmallnumber offi rms.GeopoliticaleventssuchastheGulfWarinAugust1990(follow- ing Iraqs invasionof Kuwait) and the Iraq War that followedthe 9/11 at- tack have resulted in signifi cant disruptions of oil supplies from central globalproducers.Therefore,thepriceofoilrapidlyincreasedatthebegin- ning of these events. Recently, a demand-decrease shocksuch as the global fi nancial crisisor the rise of a substitutesuch as the shale gas boomhave led to substantial falls in the crude oil price. Hedging performance around these periods is particularly important to the participants in the oil industry. The crude oil market is now one of the most globally mature commodity markets. Almost 80% of all inter- nationaltransactionsincrudeoilnowinvolvewaterwaydelivery.Foron- goingbusiness,theparticipants inthe oil industrysuchasoilproducers, refi ningcorporations,oilphysicaltraders,andoilcorporationsarephys- icallyexposedtolongdeliveryandholdingperiods.Therefore,itisimpor- tant to select hedging strategies in periods of dramatic price changes because a price shock can result in a permanent change in the volatility of the oil price (Narayan and Narayan, 2007; Chen et al., 2016). In this study, we directly investigate the performance of a crude oil hedge portfolio during the fi ve largest historical oil market shocks: the GulfWar, theAsianfi nancialcrisis, theIraqWar, theglobal fi nancialcri- sis,andtheshalegasboom.Giventhatthesesampleperiodscapturethe largest oil price shocks in history, we can analyze the hedge perfor- mance while keeping the inherent characteristics of oil price shocks. In this work, we identify the oil price structure changes in terms of vol- atility and the Fama and French (1988) spread after the episodic events occur:thestandarddeviationoftheoilspotandfuturespriceandthefu- tures basis tend to increase after the price shock. Previous studies generally utilize regime-switching models to han- dle structural breaks in assessing the performance of oil hedging strate- gies. For example, Billio et al. (2018) show that regime-switching modelshaveanadvantagewhenhedgingduringa fi nancialcrisis.How- ever, the inherent properties of each shock can be ignored under the regime-switching framework because the volatile period after each shock is identically treated as a high-volatility regime typically. It is im- portanttotesthedgingstrategiesforeachpriceshockbecauseoilshocks Energy Economics 81 (2019) 11321147 Corresponding author. E-mail address: jihunkim79 (J. Kim). /10.1016/j.eneco.2019.06.002 0140-9883/ 2019 Elsevier B.V. All rights reserved. Contents lists available at ScienceDirect Energy Economics journal homepage: have different effects on oil price dynamics according to the underlying causes (Kilian, 2009). To evaluate hedging strategies, we adopt the most popular and rep- resentativevolatilitymodelsthathavebeensuccessfullyusedtoanalyze oil return volatility, such as stochastic volatility (SV), GARCH, and diag- onal BEKKmodels.TheSV model, whichwehave themostinterestin,is distinct from the GARCH framework because it allows randomness in the conditional volatility. Researchers have shown that the SV model can successfully identify features of variance in asset returns. Although previous studies show that the SV model fi ts well with oil price volatil- ity, to the best of our knowledge, the SV model has not been used to es- timate the minimum variance hedge ratio (MVHR) in the crude oil market.1Our results provide evidence that a hedging strategy that is based on the SV model is able to outperform GARCH and BEKK model- basedstrategies inhistoricalpriceshockperiods.However, thediagonal BEKK modelis not superior totheothermodels even thoughitutilizes a time-varying dynamic correlation structure. Our study specifi cally focuses on the hedging performance of the MVHR rather than the forecasting performance of variance estimation. Hedgingthecrudeoilpriceriskwithafuturescontractisrelatedtotheis- sues of constructing a minimum-variance hedge portfolio. Theoretically, hedging portfolio performance requires the optimal hedge ratio to be specifi ed. In MVHR estimation, time-varying variance and covariance play prominent roles. However, the literature on portfolio optimization demonstrates that accurate volatility estimation does not always result in a better minimum variance portfolio performance. Thus, although vol- atilityisanimportantcomponentwhenconstructingtheMVHR,volatility forecasting accuracy must be distinguished from out-of-sample hedging performance. We fi nd that reducing the mean squared error (MSE) and mean absolute error (MAE) does not guarantee superior hedge performance. Our results can be used practically by oil market participants. The crude oil industry participants are often able to detect an oil shock in ad- vance. However, evaluating the hedge performance during stable periods can be less effective because hedge performance that is based on the stable period maybe inconsistent with that during the oil shock period (Abosedra and Laopodis, 1996 and Alizadeh et al., 2004). Therefore, we suggest that an oil hedge portfolio that is based on the SV model is more effective when preparing for upcoming shocks. Consequently, our results canbe extended horizontallybyapplyingour methodto otherassets,and they can be extended vertically by using advanced volatility models. Thispaperisarrangedasfollows.InSection2,wereviewtheliterature on time-varying volatility models and oil market hedge performance. In Section3,weexplainthemethodologiesusedinthisstudy.Wealsointro- duce the MVHR, variance reduction, and volatility forecasting frame- works in this section. In Section 4, we describe our data set and the characteristics of the crude oil price before and after historical events. In Section 5, we provide the empirical results. The hedge performance and predictivepowerofvolatilityareanalyzedaccordingtothesampleperiod andthevolatilitymodel.Thetheoreticalneedfortheintroductionofafu- tures market and anecdotalevidenceshowingthe positiveeffects of a fu- tures market are discussed. Finally, in Section 6, the contributions, implications, and limitations of this study are described. 2. Literature review Johnson (1960) measures the risk of portfolio with return variance and suggests that the MVHR can be used to analyze the hedging effec- tiveness, and subsequentstudiesinvestigatethe propertiesandfeasibil- ity of MVHR (Ederington, 1979; Myers, 1991; Chen et al., 2013; Wang et al., 2014). The estimation of MVHR is determined based on the volatility model. Although early studies assume a time-invariant vari- ance by using linear regression, it is now widely accepted that the vola- tility of fi nancial and macroeconomic variables including variance and covariance have a time-varying property (Ederington, 1979; Figlewski, 1984). Baillie and Myers (1991) show that estimating a static hedge ratio may be inappropriate if the joint distribution of spot and futures prices vary over time. Empirical studies suggest that time-varying MVHR outperforms constant MVHR, and the variance model and sam- ple selection are important in constructing time-varying MVHR. Floros and Vougas (2004) and Salvador and Arag (2014) show that in terms of hedge performance time-varying MVHR outperforms constant MVHR. The commodity futures market provides effective instruments to hedgethecommodityspot.Thefuturespriceisknowntoleadandaffect the spot price in general. The arbitragers (who make an arbitrage profi t from the difference between futures price and its fundamental price), speculators (who make an arbitrage profi t from the difference between spot and futures price), and hedgers (who are the real commodity holders) prefer trading futures than a real commodity because of the lower transaction cost and the ease of shorting. Consequently, new in- formation is fi rst refl ected in the futures price, and then the spot price is affected. Silvapulle and Moosa (1999) analyze the linear and non- linearcausalitybetweenWTIspotandfuturespriceusingdataspanning from 1985/1/2 to 1996/7/11. The linear causality test suggests that the futures price leads the spot price, while Baek and Brocks (1992) non- linear causality test detects the bi-directional effect between them. For two distinct periods (1991/10 to 1999/10, and 1999/11 to 2007/10), Bekiros and Diks (2008) adopt the non-linear causality test from Diks and Panchenko (2006) considering cointegration. They show that the linear causality disappears after controlling cointegration, while the non-linearcausalityremainsevenaftercontrollingfortheGARCHeffect. Thisnon-linearbi-directionaleffectistime-variant.Furthermore,Chang and Lee (2015) adopt wavelet analysis to investigate the time- and frequency-varying crude oil spot and futures relationship. They con- clude that there is a signifi cant lead-lag relationship in the short-run, while it is absent in long-run. Time-varying variance and covariance play prominent roles in MVHR estimation. The empirical success of the GARCH model sug- gests that the GARCH effect accounts for a large portion of the varia- tion in fi nancial asset returns. The great success of the parsimonious model stems from the mean reversion and volatility clustering in the fi nancial time series. Abosedra and Laopodis (1996), Morana (2001) and Bina and Vo (2007) show that the crude oil return has a heavy- tailed distribution and the GARCH model fi ts the data well. Conse- quently, many studies have applied the GARCH framework to inves- tigate the oil return volatility process. For example, Hou and Suardi (2012) show that nonparametric GARCH models outperform para- metric GARCH models in predicting Brent and WTI oil price return volatility. Kang et al. (2009) examine the persistency and long mem- ory property of Brent, Dubai, and WTI oil price volatility using vari- ous models, and they conclude that the CGARCH and FIGARCH models perform best in predicting this volatility. Mohammadi and Su (2010) compare the forecasting performance of the GARCH, EGARCH, APARCH, and FIGARCH models for the oil spot price in 11 countries, and they conclude that the APARCH model signifi cantly outperforms the others. Lanza et al. (2006), Manera et al. (2006), and Chang et al. (2010) apply a time-varying conditional correlation structure to model the dynamic covariance process of the oil spot and futures. Chang et al. (2011) and Billio et al. (2018) estimate the conditional volatility of oil spot and futures return under the GARCH framework and calculate an MVHR of a hedge portfolio. The SV model is a classic econometric approach for modeling time- varying volatility. Taylor (1982) fi rst introduces the discrete version of the SV model and Taylor (1986) derives its basic statistical properties. Following this seminar paper, Ghysels et al. (1996), Shephard (1996), and Jacquier et al. (2002) analyze the economic properties of the 1 One of the most generally used hedging strategies is to minimize the variance of the hedge portfolio returnsthe ratio of futures to hedge is called MVHR (Johnson, 1960). 1133D. Chun et al. / Energy Economics 81 (2019) 11321147 model. Under the SV model framework, conditional variance is calcu- lated from the latent variable ht, which follows the AR(1) process. An unobservable ht signifi es that the SV model parameters cannot be esti- mated well using maximum likelihood estimation. As an alternative, the Markov Chain Monte Carlo (MCMC) procedure of Tanner and Wong (1987) can be used to estimate SV model parameters, and Kim et al. (1998) improve on the MCMC approach to practically estimate the parameters of the SV model. Hol (2003) shows that of the various volatility models, the SV model has the best predictive power in terms of major stock market indices volatility. Studies show that the SV model fi ts commodity return volatility well, including crude oil spot and futures. Trolle and Schwartz (2009) apply an unspanned SV model to commodity derivative pricing and fi nd evidence of an unspannedfactor.Vo(2009)appliesaregime-switchingSVmodeltoin- vestigate crude oil price volatility, and Vo (2011) applies a multivariate SV model to analyze risk prediction information in both the stock mar- ket and the oil futures market. Many studies of the MVHR suggest that it is important to consider cointegration between spot and futures to improve hedge perfor- mance (Kroner and Sultan, 1993; Park and Switzer, 1995; and Kenourgios et al., 2008). Haigh and Holt (2002) apply the BEKK model to estimate the time-varying hedge ratio for WTI, heating oil, and unleaded gasoline futures. Alizadeh et al. (2004) hedge the marine bunker prices of Rotterdam, Singapore, and Houston with the crude oil and petroleum futures of the New York Mercantile Ex- change (NYMEX) and the International Petroleum Exchange, using VECM and BEKK models. Jalali-Naini and Manesh (2006) estimate the hedge ratio between the WTI spot and the futures price using various volatility models. They confi rm that the mean and standard deviations of the MVHR depends on time and the perceived risk in- creases as the expiration of the futures extends. Lien and Shrestha (2007) compare the error correction and other hedge ratios for 23 commodities including crude oil, based on hedging effectiveness. They conclude that the error correction hedge ratio outperforms for the short hedge horizon. Chang et al. (2011) construct a hedge portfolio using the MVHR from the CCC, VARMA-GARCH, DCC, BEKK, and diagonal BEKK models and conclude that the diagonal BEKK model performs best in terms of variance reduction. Structural break is also an important issue when testing the vola- tility structure, as explained by Lamoureux and Lastrapes (1990) and Pan et al. (2017). Wilson et al. (1996), Lee et al. (2010), Salisu and Fasanya (2013), Silva (2015), and Zhang and Zhang (2015) fi nd evi- dence of structural breaks in crude oil and futures prices. Several studies show that a price shock results in a permanent change in the volatility process of oil prices. Narayan and Narayan (2007) show that several geopolitical regime shifts in the oil market bring oil price shocks and these have an asymmetric and permanent effect on the volatility of oil. Chen et al. (2016) investigate the effect of OPEC on the crude oil price, and fi nd that the decision of OPEC affects the price signifi cantly. Hedge performance around these periods is particularly important to the participants in the oil industry. Fong and See (2002) apply a regime switching model to consider sudden changes in crude oil returns and volatility. They show that regime shift is signifi cant in the crude oil futures return and dominates the GARCH effects, implying that the regime switching model performs better than non-switching models in out-of-sample prediction. Pan et al. (2017) apply a regime switching GARCH-MIDAS model to examine the link between macroeconomic fundamentals and oil price volatility. They show that the regime switching model outperforms other models in forecasting WTI and Brent oil price volatility. Billio et al. (2018) apply the Bayesian multi-chain Markov-switching GARCH framework for dynamic hedging on the crude oil futures market. They fi nd that this framework is a good fi t with the conditional variance dynamics of crude oil spot and futures, resulting in better hedge performance before and during a fi nancial crisis. 3. Methodology 3.1. MVHR and variance reduction The crude oil holders can establish various hedging strategies with futures contracts. The crude oil hedge portfolio return at time t (RH,t) can be represented as Eq. (1). RH;t RS;ttRF;t;1 where RS,tand RF,tdenote time t oil spot and futures returns, calculated as the logarithmic differences of spot and futures prices, respectively. t is thetimet hedge ratio, and represents the amountoffutures contracts that the hedgers should