Incorporating Liquidity and Order Flow Imbalances for Intraday Stock Volatility Prediction

Intraday market volatility forecasting is the essence of profitable trading and risk management in times of high market volatility. Classical econometric models rely on historical data and sound economic theory, which renders them incapable of fully comprehending the complex evolution of liquidity depth, bid-ask spreads, and order flow imbalances that occur. in high-speed markets. As a result of these limitations, low liquidity or uncertain price conditions make these predictions less reliable, exposing traders and financial institutions to poor risk decisions and missed opportunities for gain as market conditions are capable of changing rapidly. These shortcomings pose an increasingly pressing need for adaptive, dynamic modeling approaches, which the present research responds to. To address these issues, we present the Stacked Light-GBM Ensemble, a novel approach with three tuned Light-GBM models for varying hyperparameters: small learning rate, shallow depth, and strong regularization. The three models are mixed in a meta-layer to formulate a unified prediction model. The method employs rolling volatility measures, liquidity variations, and market micro-structure to enhance short-term price forecasting. Confirmed with time-series cross-validation using adaptive rolling windows, the model performs much better under times of volatility. Through the incorporation of depth, volume, and order imbalances from the order book, it provides accurate answers to intraday high-frequency behavior. Tensor-RT optimizes the system's performance in real-time to ensure real-time performance consistency during trading sessions. The integration of macroeconomic variables and real-time data feeds transcends the limitations of conventional models. The innovation enhances risk management processes to enable traders to navigate turbulent markets with greater certainty and precision.