Abstract:
|
We consider estimation of a parameter of interest based on several possibly incorrect model assumptions. Correct model assumptions improve asymptotic efficiency whereas incorrect assumptions are suppressed and have no influence for large samples. This procedure minimizes the mean squared error (MSE) and, in contrast to previous methods, allows to estimate the parameter of interest using model assumptions from conceptually different classes of models (parametric, semi-parametric, non-parametric). Theorems on large sample properties and simulation studies explore the use of the proposed estimation procedure and highlight its benefits. An illustrative example shows how to implement the proposed approach.
|