K-class estimators: The optimum normalization for finite samples

Frequently, in estimating an equation, one is interested in a particular set of normalized coefficients. There is still a normalization problem, however, for if an equation contains m₁ endogenous variables, there are m₁ ways to estimate the same coefficients. The purpose of this article is to determine the optimum normalization for finite samples. Restricting the analysis to k-class estimators and employing Kadane’s small sample technique [9], let (Equation presented) denote the correlation coefficient between the ith endogenous variable and the disturbance. Then, measuring “endogenousness” by (Equation presented), this article shows that subject to several important qualifications, one should normalize on the most endogenous variable.