Three Bayesian econometric studies on forecast evaluation

A complete theory for evaluating forecasts has not been worked out to this date. Many studies on forecast evaluation implicitly relied on assumptions that are not supported by data, e.g., the assumption of homoskedastic and uncorrelated errors, forecaster homogeneity, etc. In this dissertation, I apply Bayesian methods to analyze various aspects of forecast evaluation. The overall objective is to better evaluate forecasts in terms of bias, efficiency, and information content by accounting for the structure of forecasts and directly addressing various critical econometric issues that are ignored by previous studies. Three related studies have been undertaken to address three issues. My first paper studies forecasts' bias and inefficiency after accounting for forecast error correlations. My second paper studies forecasts' bias and inefficiency after accounting for forecasts' hierarchical structure. My third paper proposes new measures of forecasts' information content of actual variables. Although the three papers in this dissertation studies specific data sets, the employed methods could be easily applied to forecasts with similar structures.