Modeling and Forecasting Intraday Volatility of Crude Oil Futures

Abstract

This paper examines the intraday 5-minute volatility dynamics of crude oil futures using Market-by-Order (MBO) data from the front-month contract. Within the context of crude oil futures, we report the following key findings: (1) Realized Volatility (RV) serves as a reliable proxy for latent volatility; (2) 5-minute RV displays pronounced intraday seasonality and long-memory characteristics; (3) Deseasonalized 5-minute RV is still long-memory but simultaneously highly stationary, a surprising observation. (4) Fractional differencing tends to eliminate both high- and low-frequency components of the 5-minute deseasonalized RV, making the commonly used AutoRegressive Fractionally Integrated Moving Average (ARFIMA) model less effective. This has profound implications on how we understand volatility. (5) Eventually, we adopt a simpler AutoRegressive Integrated Moving Average (ARIMA) approach and present the results.