Exploring the Effective Technical Indicators of LSTM for WTI Price in Different Regimes

Abstract

This study aims to explore the effectiveness of technical indicators in predicting WTI crude oil prices and investigate their applicability in different oil price regimes. By collecting monthly WTI data, we analyzed the closing prices using the Markov switching regression (MS-Regression) modeling approach. The study reveals that the market conditions can be divided into two distinct periods: low oil price regime and high oil price regime. For each of these periods, we employed LSTM regression models with the 30 least correlated technical indicators as features to predict the closing prices for the subsequent period. The MSE, RMSE, MAE, and MAPE were calculated for the test sets of both regimes. For Regime 1, the results were 43.409, 6.588, 4.745, and 10.299%, respectively. For Regime 2, the corresponding values were 120.872, 10.994, 8.521, and 9.371%, respectively. The results demonstrate that the models perform more accurately in predicting price fluctuations during the high oil price regime. Furthermore, through feature importance analysis, we identified the effectiveness of SMA_20, MIDPOINT_14, and BBANDS_middle_20_2 indicators in both low and high oil price regimes. Additionally, BBANDS_upper_20_2 and CCI_14 exhibit better predictive capabilities during the low oil price regime, while NATR_14 and RSI_14 prove more effective during the high oil price regime. These findings contribute to a better understanding of the role of technical indicators in predicting WTI crude oil prices in different oil price regimes.