Deep Network for Volatility Forecasting: Applications and Comparisons.

Realized volatility plays a very important role in various financial areas and beyond, as it plays an essential role nowadays in risk management and investment strategy development. In this thesis, different volatility forecasting models are examined and compared using the S&P500 index, taking into account a time series covering the period between 1st January 2012 and 31st December 2024. In particular, traditional econometric models such as ARCH and GARCH, and models based on Deep networks such as CNN, LSTM and Transformer will be compared. The purpose of this analysis is to identify the best model for predicting future volatility, through the measurement of mean square error (RMSE), in order to identify the most effective approach. The results obtained from this analysis show that models based on Deep networks are more accurate compared to traditional models, with the CNN model leading the ranking. In contrast, the traditional ARCH and GARCH models show greater difficulty in adapting to dynamic market changes and non-linear changes in volatility.