Ridge Estimators for Distributed Lag Models

The paper explains how the Almon polynominal lag specification can be made stochastic in two different ways - one suggested by Shiller and another following the lines of Lindley and Smith. It is shown that both the estimators can be considered as modified ridge estimators. The paper then compares these modified ridge estimators with the ridge estimator suggested by Hoerl and Kennard. It is shown that for the estimation of distributed lag models the ridge estimator suggested by Hoerl and Kennard is not useful but that the modified ridge estimators corresponding to the stochastic versions of the Almon lag are promising. The paper has two empirical illustrations.