Essays on Parameter Heterogeneity and Model Uncertainty

The choice of a particular model in quantitative economic analysis reflects the economic question analyzed, jointly with related economic theory and the specific structure of the given data being analyzed. The degree to which economic theory or the data dominates the analysis is an important strategic decision that the researcher has to face. In the first strategy, the model is based mainly on a priori economic theory. Several contributions in the economics literature, in particular those that occurred in the period just after the second World War, are based on this strategy, suggesting explicit links between economic theory, mathematics and statistics (see e.g. the contributions of the Cowles Foundation for Research in Economics at Yale University. In the second strategy, which became more popular during late nineteen seventies and early nineteen eighties, modeling is based more on the data information, see e.g. Sims (1980). In the time series context, the advantages of this data-based approach are addressed and it is mentioned that economic theory often does not provide precise information on functional relationships between variables. A good survey of this approach is given by Zellner and Palm (2004). These latter authors conclude that the use of data information for discovering and repairing the defects of proposed models are of crucial importance. Common practice in empirical research is to combine these strategies in a meaningful way, i.e. the constructed model is based on economic theory and the data information at the same time. This combination of strategies is motivated by two arguments: On the one hand, data information may not be informative enough. On the other hand, too strong assumptions may affect the reliability of results and the forecasting performance. This thesis considers the relatively more data-based approach in analyzing economic relationships and provides alternative methods to avoid very strong assumptions in the analysis. This thesis consists of two parts. The first part develops new econometric models with a sufficient degree of flexibility to accommodate various forms and degrees of heterogeneity in (the relations among) economic variables. The second part considers model uncertainty issues providing new tools for evaluating to what extent one (or more) model is suitable to the economic data at hand.

• View at Worldcat