Deep learning for volatility forecasting in asset management

Abstract Predicting volatility is a critical activity for taking risk- adjusted decisions in asset trading and allocation. In order to provide effective decision-making support, this paper we investigate the profitability of deep Long Short-Term Memory (LSTM) Neural Network forecasting daily stock market using panel 28 assets representative Dow Jones Industrial Average index combined with factor proxied by SPY and, separately, 92 belonging NASDAQ 100 index. The plus data are from January 2002 August 2008, while December 2012 November 2017. If, on one hand, expect that evolutionary behavior can be effectively captured adaptively through use Artificial Intelligence (AI) flexible methods, other, setting, standard parametric approaches could fail optimal predictions. We compared forecasts generated LSTM approach those obtained widely recognized benchmarks models field, particular, univariate such as Realized Generalized Autoregressive Conditionally Heteroskedastic (R-GARCH) Glosten–Jagannathan–Runkle Multiplicative Error Models (GJR-MEM). results demonstrate superiority over popular R-GARCH GJR-MEM when condition high volatility, still producing comparable predictions more tranquil periods.