Crude oil price pattern recognition and prediction based on sub-series clustering

Crude oil prices exhibit pronounced time-varying fluctuations and recurring local patterns. This study proposes a novel sub-series clustering framework for crude oil price pattern recognition and forecasting. Key findings demonstrate that: (1) the proposed method substantially reduces prediction errors for WTI crude oil futures prices, with particularly strong improvements during periods of elevated volatility; (2) the newly developed Multi-scale Trend (MsT) measure provides an effective quantification of sub-series similarity by capturing price dynamics across multiple temporal horizons, leading to enhanced forecasting accuracy; and (3) the framework maintains robust performance across alternative sample splits and different choices of cluster-level prediction models, remains stable when incorporating exogenous variables, and exhibits strong generalizability across diverse commodity markets.