Man Group Explores AI for Financial Time Series Forecasting

Man Group PLC's Oxford Man Institute of Quantitative Finance has released research employing Generative Adversarial Networks (GAN) to forecast financial time-series data. The study, presented by postgraduate student Milena Vuletić, introduces a novel approach called Fin-GAN for predicting and classifying financial time series.

Moving beyond traditional forecasting methods that often rely on point estimates, the Fin-GAN methodology generates full conditional probability distributions for return forecasts. This allows for the incorporation of uncertainty based on past historical values and positions GANs within a supervised learning framework suitable for classification tasks.

Numerical experiments conducted on equity data demonstrated the effectiveness of the Fin-GAN approach. The research indicates that this method achieved higher Sharpe Ratios when compared to conventional supervised learning models such as LSTMs (Long Short-Term Memory) and ARIMA (Autoregressive Integrated Moving Average).

This research forms part of Man Group's ongoing collaborations with academic institutions aimed at advancing quantitative finance. The development suggests potential for improved investment strategies and risk management tools that account for market volatility through advanced AI techniques.