Some decision rules for discriminating among alternative regression models are proposed and mutually compared. They are essentially based on the Akaike Information Criterion as well as the Kullback-Leibler Information Criterion (KLIC) : namely, the distance between a postulated model and the true unknown structure is measured by the KLIC. The proposed criteria combine the parsimony of parameters with the goodness of fit. Their relationships with conventional criteria are discussed in terms of a new concept of unbiasedness .

[This paper is concerned with the so-called k-class estimators of structural parameters in a simultaneous system. The structural equation being estimated is assumed, as is common in other small-sample investigations, to consist of two endogenous variables; and the number of the exogenous variables (included or excluded) as well as the number of equations in the system are arbitrary so long as the identifiability condition of the estimated equation is satisfied. Moreover, we assume that the system contains no lagged endogenous variables and disturbance terms of each period are independently distributed as multivariate normal. The exact finite-sample moments of the k-class estimators are evaluated for 0 @ extless k extless 1. For k extgreater 1 it is proved that the estimator does not possess even the first-order moment. The exact moment functions are expanded in terms of the inverse of the noncentrality (or concentration) parameter. This expansion sheds more light on the comparative study of alternative k-class estimators. Numerical calculations of the mean square error and the bias for some specific cases are also given for illustrative purposes.]

[The preliminary test estimator (or predictor) is studied in the context of linear normal regression models. The estimator (or predictor) obtained after the preliminary test is known to be inadmissible with respect to squared error loss under certain assumptions. Nevertheless, it is still widely used in practical application of regression analysis, particularly in econometrics. The object of the present paper is to tabulate the optimal significance points in the preliminary test for practical use. The optimality is based on the minimax regret principle. It is shown that if, as is usual, we take the significance point equal to the 5 per cent point or 10 per cent point, the risk is extremely large for some parameter value. The optimal significance point of the preliminary t test decreases slightly as the degrees of freedom increase,but it is nearly constant; i.e., it lies between 1.370 to 1.380 if d.f. is greater than 6.]

T. W. Anderson, Takamitsu Sawa, Distributions of Estimates of Coefficients of a Single Equation in a Simultaneous System and Their Asymptotic Expansions, Econometrica, Vol. 44, No. 6 (Nov., 1976), p. 1329

[The limited information maximum likelihood and two-stage least squares estimates have the same asymptotic normal distribution; the ordinary least squares estimate has another asymptotic normal distribution. This paper considers more accurate approximations to the distributions of the so-called "k-class" estimates. An asymptotic expansion of the distribution of such an estimate is given in terms of an Edgeworth or Gram-Charlier series (of which the leading term is the normal distribution). The development also permits expression of the exact distribution in several forms. The distributions of the two-stage least squares and ordinary least squares estimates are transformed to doubly-noncentral F distributions. Numerical comparisons are made between the approximate distributions and exact distributions calculated by the second author.]

T. W. Anderson, Naoto Kunitomo, Takamitsu Sawa, Evaluation of the Distribution Function of the Limited Information Maximum Likelihood Estimator, Econometrica, Vol. 50, No. 4 (Jul., 1982), pp. 1009-1027