FORECASTING OIL PRICE VOLATILITY: THE ROLE OF MIXED-FREQUENCY-DATA (MIDAS) MODEL

摘要

The increase in crude oil price volatility observed particularly around the period of the global financial crisis has created new interest into how to improve volatility forecasts. It is of primary importance for energy researchers, firms, financial market participants and policy makers to have models available that can provide accurate forecasts of oil price volatility. Furthermore, in the literature, high oil price volatility periods have also caused new debate on the potential drivers ( Kilian and Murphy, 2014; Van Robays, I., 2016; Degiannakis and Filis, 2017). Comprehending oil price volatility determinants is also crucial due to the fact that oil price volatility by itself could have a negative impact on economic activity (Jo, 2014; Van Robays, I., 2016). At the same time, the literature has experienced increased emphasis on the link between oil and financial markets. The critical question for this new link is whether financial market information can assist in forecasting oil price volatility. Therefore, this paper focuses not only on forecasting oil price volatility but also its potential determinants in the context of when they are most likely to exert their predictive power. The study of Degiannakis and Filis (2017) is the first study to extract information from other markets to help improve oil price volatility forecasts. However, their study forecasts daily oil price volatility whereas this paper forecasts on a monthly basis. The rationale for this is that policy makers, firms and oil companies are more interested in the longer run forecasts and a wider range of oil benchmarks and, in consequence, this paper uses both Brent and WTI volatility. In more detail, this study investigates whether higher frequency (weekly) financial indices (both their volatilities and returns) help to improve the lower frequency (monthly) crude oil price realized volatility, using the MIDAS model. In this way, we assess whether there is useful predictive information for monthly oil price volatility in higher frequency data from financial markets. Using MIDAS models, we examine a large group of financial and economic variables which include either the returns and the volatilities of other asset class indices such as the Baltic Dry Index (proxy for global macroeconomic conditions), the Commodity Future Price Index (proxy for commodities market conditions), the MSCI World Index, the MSCI World Energy Index, S&P GSCI Index, S&P GSCI Energy Index (as proxies for global financial conditions), and Trade-weighted US exchange rate (proxy for global foreign exchange market).The use of these asset classes is motivated by Degiannakis and Filis (2017) who show that daily oil price volatility forecasts are improved when the information of asset volatilities that belong to these asset classes is incorporated to the forecasting models. Our data period is from January 3, 1995, to March 31, 2017. Thus, to forecast monthly oil price volatility, our initial sample size is 119 months and the last 148 months are used to evaluate the out-of-sample volatility forecasts. Since 2005/2006 onwards oil prices started to reach much higher price levels which would change the behaviour of volatility. Thus we ensure our out-of-sample covers this more turbulent period.