Predicting Stock Price Volatility using a CNN-BiLSTM-AM Model

This study presents a novel approach to forecasting stock price volatility by combining Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks with a self-attention mechanism to enhance model accuracy. The CNN-BiLSTM model leverages CNN's spatial feature extraction capabilities alongside BiLSTM's bidirectional memory for analyzing temporal dependencies. To further refine model performance, Optuna was employed for hyperparameter tuning. The study utilizes daily stock data from NVIDIA, spanning 2,520 trading days from May 2014 to May 2024, with key features including first-order differenced prices, change, turnover, and volume. The data is split into 75% for training, 10% for validation, and 15% for testing. The model processes features extracted from the past 10 days to predict future volatility. Experimental results demonstrate that the CNN-BiLSTM-AM model significantly improves forecasting accuracy, providing valuable insights into financial time series analysis, investment strategies, and risk management.