28/10/2026
Seminario 29 OTTOBRE 10:45 - 11:45
Sala dei Consigli - Dipartimento di Scienze Economiche e Statistiche (DiSES) (Edificio C)
Richard Gerlach - University of Sydney
Neural Network Financial risk forecasting via data scaling and global training
Abstract
The effects of data volume, model size, and architectural choices on the performance of neural networks (NNs) in financial forecasting is investigated using a dataset of over 10,000 stocks for volatility and tail risk forecasting. Results identify that performance gains from data scaling decisively outweigh those from model scaling and architectural choice. Additionally, a global training approach, which trains a single NN model on all time series, is demonstrated to perform very favorably when applying NNs to financial data.
Seminario Dottorale – Dottorato di Ricerca in Economia e Politiche dei Mercati e delle Imprese