Abstract
Living in the age of information, economic agents have access to a substantial amount of publicly available information on which they can base their decisions. Unfortunately, in most of the cases the key information required by the economic agents such as financial expectations or economic activity is not precisely observed. Instead, statistical measures and predictions of those must be derived from the observed macroeconomic and financial variables. A popular approach in practice is to employ unobserved components models to extract the required information from the observed data. These models assume an underlying process of the unobserved variables, denoted as states, to capture the evolution of the observed data that depend on these latent states. In the first part of the thesis, a very popular class of these models, viz. dynamic factor models, is employed to extract the common information and indicators of macroeconomic conditions. These factors are used to make accurate predictions of future stock returns and their volatilities. In the second part, another popular class of unobserved components models, viz. dynamic mixture models, is employed to extract the categorical information in some key macroeconomic and financial variables. Together with this, the information diffusion among different variables is modeled using these categorical representations. In the third part of the thesis, both approaches from the previous parts are combined. A specific class of dynamic factor models is proposed for modeling the dynamic evolution of a large dataset of bond yields, where the density of the factors is estimated using Bayesian semiparametric techniques along with other model parameters.
Original language | English |
---|---|
Awarding Institution |
|
Supervisors/Advisors |
|
Award date | 26 Jan 2012 |
Place of Publication | Rotterdam |
Publication status | Published - 26 Jan 2012 |
Research programs
- EUR ESE 31