Multilayer feedforward networks are universal approximators
Neural Networks
Time-dependent series variance estimation via recurrent neural networks
ICANN'11 Proceedings of the 21th international conference on Artificial neural networks - Volume Part I
Heavy-tailed mixture GARCH volatility modeling and Value-at-Risk estimation
Expert Systems with Applications: An International Journal
| Hi-index | 0.00 |
We consider the generalization of the classical GARCH model in two directions: the first is to allow for non-linear dependencies in the conditional mean and in the conditional variance and the second concerns specification of the conditional density. As a tool for non-linear regression we use neural network-based modeling, so called recurrent mixture density networks, describing conditional mean and variance by multi-layer perceptrons. All of the models are compared for their out-of-sample predictive ability in terms of Value-at-Risk forecast evaluation. The empirical analysis is based on return series of stock indices from different financial markets. The results indicate that for all markets the improvement in the forecast by non-linear models over linear ones is negligible, while non-gaussian models significantly dominate the gaussian models with respect to most evaluation tests.