Comparisons of estimators are made on the basis of their mean squared errors and their concentrations of probability computed by means of asymptotic expansions of their distributions when the disturbance variance tends to zero and alternatively when the sample size increases indefinitely. The estimators include k-class estimators (limited information maximum likelihood, two-stage least squares, and ordinary least squares) and linear combinations of them as well as modifications of the limited information maximum likelihood estimator and several Bayes' estimators. Many inequalities between the asymptotic mean squared errors and concentrations of probability are given. Among medianunbiasedestimators, the limited information maximum likelihood estimator dominates the median-unbiased fixed k-class estimator.

Under normality, we obtain higher-order approximations to the distributions of the periodogram and related statistics. Our approach is based on the theorem which decomposes the periodogram into the sum of two independent random variables. It is seen that this decomposition enables us to study fairly closely the higher-order properties of not only the periodogram, but also periodogram-based statistics such as the estimators of the spectrum and prediction error variance. Some of the approximation results are graphically presented together with the exact results based on simulations.

Exact expressions are derived for the density function, variance, and kurtosis of a linear combination of the elements of a two-stage estimator for the coefficients in a single equation of a SUR system. The estimator is the first iterate in the iterative generalized least squares procedure described by Telser [14]. Our results generalize all previously known results for this estimator and, in certain special cases, also generalize some earlier exact results for Zellner's unrestricted covariance matrix estimator, to which it reduces in these special cases.

This paper examines the sensitivity of the distributions of OLS and 2SLS estimators to the assumption of normality of disturbances in a structural equation with two included endogenous variables. The approach taken is that ofimposing Edgeworth distributed errors on the reduced form equations and deriving the pdf of the estimators via the technique of Davis [11]. The sensitivity of the pdf s to changes in the non-normality parameters, i.e., skewness and kurtosis i s examined via extensive numerical computations.

In the context of a linear dynamic model with moving average errors, we consider a heteroscedastic model which represents an extension of the ARCH model introduced by Engle [4]. We discuss the properties of maximum likelihood and least squares estimates of the parameters of both the regression and ARCH equations, and also the properties of various tests of the model that are available. We do not assume that the errors are normally distributed.

In the case of regression models, one robust estimation procedure which has recently emerged is that of functional least squares. The procedure is based on the use of characteristic functions for which the tail behavior is relected by the behavior of these functions at the origin. Its attraction is that it is applicable to situations where the distribution of the disturbances may be long-tailed and/or asymmetric.

This paper extends this theory to include a large class of regression models of importance in econometrics. Indeed the regression models considered here include lagged dependent variables and deterministic exogenous variables.

It is shown that the words ‘far weaker’ used four times by Singh and Ullafa [4] in describing their mixing condition in relation to ones of Robinson [1] for establishing the central limit theorem for nonparametric estimators, should in each case be replaced by theword ‘stronger,’ Singh and Ullah's [4] condition being equivalent to complete independence across time.