Forecasting Daily Stock Volatility: A Comparison between GARCH and Recurrent Neuro-networks

Abstract

Forecasting the volatility of financial derivatives and securities returns has always been the core of financial research. Accurate volatility forecast is integral to financial risk management, which is vital for investors and supervision authorities. Traditional time series models (e.g., ARCH and GARCH models) have been famous tools for volatility forecasting. However, those models cannot capture non-linear correlations, and prediction capability is unsatisfactory. In this paper, we use multifactorial deep learning algorithms RNN, LTSM, and GRU with sentiment data to predict the future volatility of Apple (AAPL), then compare the accuracy of prediction with the GARCH model. According to the analysis, the prediction accuracy of LSTM and GRU significantly improved compared with the GARCH model. These results shed light on guiding further exploration of stock volatility prediction using deep learning algorithms. Besides, it advises investors to choose efficient stock price volatility forecasting and risk management tools.