Essays in statistics and econometrics

Abstract

Helped by cheaper data computation, companies make more use of sophisticated statistical analysis in decision making and economic management. In the dissertation I evaluate and develop statistical methods and apply them for economic applications, e.g. credit risk evaluation and commodity pricing. Recent developments in modern Monte Carlo methods have made statistical inference possible for complex non-linear and non-Gaussian latent variable models. It is typically computationally expensive to fit data to such dynamic models, due to a large number of unobserved parameters. However, the flexibility of the models has ensured a wide range of applications. This thesis mainly considers non-linear cases of a latent variable model class called state-space models. The main objective is Bayesian inference for all model parameters, based on the information in the observed data. The presented work considers the existing methods for dealing with latent variables, and propose modifications to some of the most promising methods. The performance of the proposed methods is investigated through applications on economic time series data. The thesis also includes research of a more applied nature, where an existing economic model for commodity prices is extended with a stochastic trend, to obtain a state-space model. It also contains applied economic research outside the latent variable domain, where different risk measures are compared in the context of credit risk regulation.