On Estimation and Forecasting in Two-Factor Models

We study the joint estimation of bivariate latent factors and parameters in the extended Schwartz-Smith model. Two latent factors represent the correlated OrnsteinUhlenbeck processes. Kalman and nested particle filters are used for joint estimation of parameters and latent factors. Measurement errors may follow multivariate Gaussian, Laplace, and generalised hyperbolic distributions, allowing for serial and intercorrelations. Model performances are assessed using simulations and EU Allowance futures multivariate data. For comparative analysis, a hybrid ARIMA-MLS-SVR model is considered. The theoretical study of bias and mean-squared error of finite time span estimators for parameters of the Ornstein-Uhlenbeck process is carried out.<p></p>