Beyond the Candlesticks: An Integrated Approach to Intraday US Equity Markets Regime Classification and Price Forecasting

Abstract

This paper presents a comprehensive study on intraday regime classification of US equities, with a focus on intraday price forecasting using diverse machine learning models. In the dynamic landscape of financial markets, a deep understanding of intraday regimes becomes imperative, marked by distinct patterns and behaviors. Recognizing these regimes is essential for traders, investors, and risk managers, enabling them to make well-informed decisions in a timely manner. The research leverages a dataset comprising high-frequency intraday price and volume data for SPDR S&P 500 ETF Trust (SPY) and Apple Inc. (AAPL). Feature engineering is employed to extract pertinent information from market data, capturing both trend and volatility-related patterns. Pattern recognition leads to classification of intraday regimes by clustering, Hidden Markov Models. In addition to intraday regime classification, the study extends its focus to intraday price forecasting. Various machine learning algorithms, including Long Short Term Memory Recurrent Neural Networks and deep learning models, are explored for effective intraday price forecasting. Technical analysis indicators and regime classification are incorporated as features to enhance the predictive ability of the models. Performance evaluation of the proposed classification and forecasting models is conducted using metrics such as accuracy, mean absolute error, root mean squared error, R-squared, and L2 loss. A comparative analysis of the models is undertaken to identify the most robust and adaptive approach to intraday regime classification and price forecasting.