統計学輪講（第９回）

日時    2008年 6月 3日(火)    15時00分〜16時40分
場所    経済学部新棟３階第３教室
講演者  国友 直人 (経済)
演題    Many Instruments, Heteroscedasticity and Asymptotic Optimality
of the LIML Estimation

概要

For the estimation of structural coefficients in the simultaneous
equations models, Anderson,
Kunitomo and Matsushita (2007) have given an asymptotic optimality of
the LIML (Limited
Information Maximum Likelihood) Estimator when there are many
instruments (with many incidental
parameters) under the homoscedasticity (or weak heteroscedasticity)
assumption.  When there are
persistent heteroscedasticities with many instruments, a particular
modification of the LIML
(called MLIML) has often an asymptotic optimality.  The result is
closely related to the efficiency
bound with/or without incidental parameters and it also has some
implication for recent
microeconometric applications including the testing problem of
structural coefficients.
A part of my talk will be based on recent works with T.W.Anderson
(Stanford University)
and Y. Matsushita (University of Tokyo).