Functional classification and dynamic prediction of cumulative intraday returns in crude oil futures

Abstract

Predicting the movement of crude oil futures is crucial for investors to identify opportunities and for the efficient functioning of financial markets. This paper applies functional classification and dynamic prediction (FCDP) to forecast the cumulative intraday returns of INE and Brent crude oil futures for every 5 min, to address the characteristics of high-frequency data and complex change patterns of crude oil futures market. The primary goal of FCDP is to classify trading days with the same volatility pattern employing probabilistic functional classification based on the artificial distinction between the daytime and nighttime trading hours, and then to generate forecasts using the classification results. The robustness test compares the forecasting results of FCDP with ARIMA, OLS and RW methods. According to the findings, FCDP outperforms other methods when it comes to making predictions concerning crude oil futures. Due to the FCDP's incorporation of the concept of dynamic updating, which is not reflected in conventional prediction methods, the prediction error gradually lowers as the amount of seen data increases. INE and Brent crude oil futures market trading patterns at night are more intricate than they are during the day. This study supports flexible trading strategy adjustments for both investors and policymakers.