A Neural Stochastic Volatility Model

Rui Luo, Xiaojun Xu, Weinan Zhang, Jun Wang

Abstract:

In this paper, we show that the recent integration of statistical models with recurrent neural networks provides a new way of formulating volatility models that have been popular in time series analysis and prediction. The model comprises a pair of complementary stochastic recurrent neural networks: the generative network models the joint distribution of the stochastic volatility process; the inference network approximates the conditional distribution of the latent variables given the observable ones.
Our focus in this paper is on the formulation of temporal dynamics of volatility over time under a stochastic recurrent neural network framework. Our derivations show that some popular volatility models are a special case of our proposed neural stochastic volatility model. Experiments demonstrate that the proposed model generates a smoother volatility estimation, and largely outperforms a widely used GARCH model on several metrics about the fitness of the volatility modelling and the accuracy of the prediction.

TL;DR: A novel integration of statistical models with recurrent neural networks providing a new way of formulating volatility models.

Conflicts: cs.ucl.ac.uk, apex.sjtu.edu.cn

Keywords: Deep learning, Supervised Learning