|
Stelios Arvanitis |
|
Some Recent Working Papers The following consist of some recent work concerning entirely properties of classes of indirect estimators. A New Class of Indirect Estimators and Bias Correction (with A. Demos) Abstract: In this paper we define a set of indirect estimators based on moment approximations of the auxilary estimators. We provide results that describe higher order asymptotic properties of these estimators. The introduction of these is motivated by reasons of analytical and computational facilitation. We extend this set to a class of multistep indirect estimators that have potentially useful higher order bias properties. Furthermore, the widely employed "feasibly biased corrected estimator" is an one optimization step approximation of the suggested on.
Stochastic Expansions and Moment Approximations for Three Indirect Estimators (with A. Demos) Abstract: This paper deals with properties of three indirect estimators that are known to be (first order) asymptotically equivalent. Specifically, we examine a) the issue of validity of the formal Edgeworth expansion of an arbitrary order. b) Given a), we are concerned with valid moment approximations and employ them to characterize the second order bias structure of the estimators. Our motivation resides on the fact that one of the three is reported by the relevant literature to be second order unbiased. However, this result was derived without any establishment of validity. We provide this establishment, but we are also able to massively generalize the conditions under which this second order property remains true. In this way, we essentially prove their higher order inequivalence. We generalize indirect estimators by introducing recursive ones, emerging from multistep optimization procedures. We are able to establish higher order unbiaseness for estimators of this sort.
On the Existence of Strongly Consistent Indirect Estimators when the Binding Function is Multivalued Abstract: In this paper we establish the definition of IE and their strong consistency, when the binding function is a compact valued correspondence under mild conditions. These results are generalizations of the analogous results in the relevant literature, hence permit a broader scope of statistical models. We provide some examples that concern linear models with weak instruments, and conditionally heteroskedastic ones.
|
