Semiarium i matematisk statistik 
tisdagen den 23 maj 13.15 i MH:227

Kim Nolsøe
Andersen Consulting

Titel:
Estimating functions -- a general framework

Abstract:
In this presentation the main focus will be on estimating
functions. First it will be shown that LS (Least Squares) WLS
(Weighted Least Squares) and ML (Maximum likelihood) can be viewed
as estimating functions. It will be discussed what a "reasonable"
measure of an optimal estimation function is. This measure of an
optimal estimating function takes both the variance and the bias of
the estimator into account. Some examples will be shown to demonstrate
the problem in connection with especially WLS. The focus will be on
observations which are from the Poisson distribution, the log-normal
distribution and an AR(1) process with ARCH(1) distributed noise, and
some general conclusions will be drawn. The ideas are extended to the
multidimensional case with correlation between observations. For
discretely observed stochastic differential equations it will briefly
be demonstrated how to build estimating functions, also when the process
of interest is indirectly measured.