Generalized dynamic linear models for financial time series (Q2722286)

A general class of conditionally Gaussian dynamic linear models is considered in this paper. It is discussed how they can provide a flexible tool for modelling financial time series. These models generalize the standard dynamic linear model in various directions. In particular, they allow for stochastic regressors, dependence of the state of previous observations, and stochastic covariance structure of the error terms. It is referred to the extended model as the generalized dynamic linear model. In this paper conditionally Gaussian state-space models are considered and some directions in which they can be extended for better modelling financial time series with stochastic volatility are shown. The basic point is that they provide a flexible and fairly simple class of models for financial data. Also, it is discussed how some popular models for financial time series with stochastic volatility, such as GARCH models, can be regarded in the framework of conditionally Gaussian state-space models. The model is illustrated by two examples with real data sets. The first is an exchange rate volatility series modelled by an autoregressive model with GARCH disturbances. The second is an exchange rate volatility series modelled by a seasonal model with stochastic regressors.