We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
RIDGE REGRESSION UNDER ALTERNATIVE LOSS CRITERIA.
- Authors
Lin, Karl; Kmenta, Jan
- Abstract
The ordinary ridge regression (ORR) estimator with a given k is a linear estimator which is biased but which, for values of k in a certain interval, has a smaller mean square error than the ordinary least squares (OLS) estimator. Since the interval of dominance of ORR over OLS depends on the true values of the regression parameters, the advantage of ORR over OLS is, for practical purposes, illusory. The various interpretations of the ORR estimator offered, however, indicate that if there is some prior knowledge about the parameter space of coefficient vector, and if this knowledge is sufficiently sharp, the ORR estimation provides a convenient and simple way of incorporating such knowledge in estimation and of reducing the size of the mean square error. The results of Monte Carlo experiment indicate that, in general, the ORR estimators do out-perform the OLS estimator very substantially when the degree of multicollinearity is medium or high, even when a loss criterion other than that of mean square error is used.
- Subjects
RIDGE regression (Statistics); ESTIMATION theory; REGRESSION analysis; LEAST squares; MATHEMATICAL models of economics; MONTE Carlo method; MATHEMATICAL statistics
- Publication
Review of Economics & Statistics, 1982, Vol 64, Issue 3, p488
- ISSN
0034-6535
- Publication type
Article
- DOI
10.2307/1925948