Bad volatility is not always bad: evidence from the commodity markets

ABSTRACT

Using exchange-traded fund (ETF) options data, we examine the predictive power of variance risk premium on returns of four commodities: crude oil, natural gas, gold and silver. We also analyze the predictive power of upside and downside variance risk premiums using a decomposition model conditional on the direction of the underlying market movement. We find that both the undecomposed and decomposed variance risk premiums are able to predict commodity prices. The decomposed variance risk premiums, however, outperform the undecomposed premium. The importance of upside and downside variance risk premiums differs across markets, related to hedging demand. In energy markets, both upside and downside premiums have strong predictive power, while in precious metal markets, only the upside premium is predictive.

KEYWORDS:

• Commodity markets
• upside and downside variance risk premiums
• asymmetric risk
• prediction

JEL CLASSIFICATION:

• C53
• G11
• G14
• G17

Acknowledgments

We thank the participants at the 2018 International Conference on Energy Finance, China, the 2018 Greater China Area Finance Conference, China, and the 2018 Conference of Asia-Pacific Association of Derivatives, South Korea, for helpful comments.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

¹ The idea of variance risk premium decomposition comes from the findings that positive and negative shocks to market fundamentals have asymmetric impact on prices (see e.g. Guo, Wang, and Zhou 2014; Patton and Sheppard 2015; Segal, Shaliastovich, and Yaron 2015).

² Studies such as Gruber, Tebaldi, and Trojani (2015) and Ait-Sahalia, Karaman, and Mancini (2018) emphasize that variance risk premiums with various maturities may contain different information of market risks.

³ In this study, the term ‘good’ and ‘bad’ indicate that the components are related with upward and downward market movements, respectively.

⁴ The daily price changes are classified into positive and negative using a suitable threshold. In our study, we set the threshold to zero.

⁵ We take intraday frequency at a 5-minute interval; we also treat the period from the close of trading day $t$ to the open of the next day as the last 5-minute period.

⁶ USO stands for United States Oil Fund, and it is one of the largest and most liquid oil ETF; UNG stands for United States Natural Gas Fund; GLD stands for SPDR Gold Shares; SLV stands for iShares Silver Trust.

⁷ First launched in 2008, commodity ETF options are relatively new. There are currently two energy and two precious metal ETF options which are frequently traded in the market.

⁸ Specifically, we remove options under the following conditions: options with a zero close bid or ask, close bid greater than ask, zero trading volume, and options which violate the standard no-arbitrage conditions.

⁹ All figures are scaled up by 1000 for presentation.

¹⁰ https://fred.stlouisfed.org.

¹¹ The default spread is the difference between the yields on Baa- and Aaa-rated corporate bonds. The term spread is the difference between the yields on the 10-Year Treasury bond and the 3-Month Treasury bill.

¹² The group of control variables are selected following the work of Bali and Peng (2006) and Bali and Engle (2010), which found that macroeconomic variables such as the Federal Funds Rate, default spread and term spread have predictive power for future market returns as they convey information on general economic condition.

¹³ In addition, Baur and McDermott (2010) show that the value of gold tends to rise in response to negative market shocks as investors are likely to look for a haven in response to severe market shocks suffered over a short period. One representative example is the surge of gold price during the time of the intensification of Global financial crisis.

¹⁴ Clark and West (2007) develop an adjusted version of the Diebold and Mariano (1995) and West (1996) statistic, which is subsequently labelled as the MSPE-adjusted statistic.