Volatility spillovers among oil and stock markets in the US and Saudi Arabia

ABSTRACT

In this article, we use high frequency data and an identification via changes in volatility approach to assess the volatility spillovers among oil and the US and Saudi Arabian stock markets. We document the existence of asymmetry in contemporaneous spillover effects. Particularly, during the times when oil’s trading hours overlap with the US and Saudi Arabian stock markets, the volatility spillover from oil to the stock markets is higher than the other way around. We highlight the importance of taking into consideration the information present during continuous trading hours of oil, especially during simultaneous trading hours with the stock markets. We compare our findings based on our structural VAR with those of a traditional reduced-form VAR, and observe that contemporaneous and intraday effects are necessary to be taken into account since the indirect transmission of volatility occurs through them.

KEYWORDS:

• Contemporaneous spillovers
• volatility spillovers

JEL CLASSIFICATION:

• C32
• G10
• Q40

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Financialization means that oil prices are not only determined by the supply–demand structure of the oil market but are also importantly affected by changes in financial market conditions (Wan and Kao 2015). This is due to the increased participation in oil and commodity markets of investors, who are looking to achieve greater portfolio benefits, rather than commercial traders, who use derivatives markets to hedge against price fluctuations (Basak and Pavlova 2016).

² The GCC consists of the following countries: Saudi Arabia, United Arab Emirates, Kuwait, Qatar, Oman and Bahrain.

³ An interesting extension would be to conduct our analysis for other GCC countries and determine how those are affected by the instantaneous spillovers. However, the lower liquidity in these markets may be problematic when constructing volatility measures based on high frequency data.

⁴ The existence of these indirect effects could be an explanation for the mixed results in literature about the relationships between oil and the SA stock market (Jouini and Harrathi 2014; Arouri, Lahiani, and Nguyen 2011).

⁵ For instance, consider an investor with exposure to the SA stock market and who is worried about volatility spilling over from the oil market. If intraday we observe that the correlation between the SA stock market and oil volatility is high, an investor may be worried about spillover. However, if the contemporaneous effect only runs from the stock market to the oil market then hedging will be ineffective. In addition, knowledge of the intraday lead–lag effects will be useful as well, as this dynamics would be ignored if one were to focus only on daily data. Finally, knowledge of the indirect effects can be useful as well, as an indirect spillover from oil to SA volatility through US volatility could simply be hedged by a hedge on oil.

⁶ The literature generally uses three different econometric approaches that rely on identification through heteroskedasticity. The first approach is based on Rigobon (2003) and uses different regimes to capture non-proportional shifts in heteroskedasticity. Studies that have implemented a modified version of this approach include Ehrmann, Fratzscher, and Rigobon (2011) and Ehrmann and Fratzscher (2017). The second approach is based on Rigobon and Sack (2003) who model the heteroskedasticity in residuals through a GARCH model. This approach has been implemented by, e.g. Badshah, Frijns, and Tourani-Rad (2013). The third approach is that of Lütkepohl (2013), which we implement in this article. This approach uses a mixture of Normals to capture heteroskedasticity in the residuals. As Rigobon (2003) notes, identification of the structural parameters is robust to misspecification of the process to model the heteroskedasticity, so in theory results should be identical regardless of the econometric procedure used. We opt for the approach by Lütkepohl (2013) because it is the most appropriate procedure for the data we have. The procedure of Rigobon (2003) based on volatility regimes may result in regimes being selected that do not overlap with the time of the day when the contemporaneous effects are observed, and this may make identification cumbersome. The econometric procedure based on a GARCH model might work well with daily data, but devising an appropriate GARCH model for the intraday data we use is not trivial. The approach of Lütkepohl (2013) works very well with our data, as heteroskedasticity, as well as structural parameters can all be estimated in one single estimation (unlike the Rigobon (2003) approach, which requires a two-step procedure).

⁷ See also Kao and Fung (2012) who define the trading day based on the 24-hour GLOBEX trading in examining the volume–volatility relationships for the Japanese Yen futures, Eeuro FX futures and E-mini S&P 500 futures.