Moment based estimators such as Generalized Method of Moment estimators (GMM) are routinely used in estimation of economic models. This project analyzes the small sample properties of such estimators. Approximate formulas for small sample bias and variance are sued to evaluate different estimation methods. Bias corrections based on these formulas are obtained to improve the small sample behavior of existing procedures.
Typically, economic models imply an infinite number of moment conditions. A selection rule has to be adopted to decide which moments to match. An early literature on efficient GMM estimation established that in prinicple all available moments should be matched asymptotically in order to achieve the most efficient use of the information contained in the data. More recent research on the other hand, has found that inclusion of more and more moment conditions usually is associated with bad small sample performance.
The research will include extensive simulation experiments that will shed light on the quality of higher order approximations and the performance of different refinements in small samples. Ultimately this line of research is intended to give applied researchers guidance in choosing the most appropriate statistical tools for a given modeling situation.
