A Model Combining LightGBM and Neural Network for High-frequency Realized Volatility Forecasting

The financial market is a nonlinear and frequently changing complex dynamic. Volatility, as one of the important indicators to measure the return of financial assets, occupies an indispensable position in the field of financial measurement. With the development of machine learning and massive data technology, there is an increasing demand for volatility prediction. In this paper, an ensemble learning model mainly based on the LightGBM algorithm and supplemented with a neural network is constructed. The model achieves the prediction of high-frequency realized volatility using ultra-high frequency stock market data and through the method of moving windows in finance. The superiority of the LightGBM-NN model is verified by comparing it with the single LightGBM model. The LightGBM-NN model produces less error and has higher accuracy, precision, and F1 score. The lightGBM-NN model has advanced the application of LightGBM in the field of financial measurement, which brings new ideas on how to handle the massive data efficiently and fast in the stock market.