Econometrica: Mar, 1978, Volume 46, Issue 2

Discrete Parameter Variation: Efficient Estimation of a Switching Regression Model

https://doi.org/0012-9682(197803)46:2<427:DPVEEO>2.0.CO;2-I
p. 427-434

Nicholas M. Kiefer

An efficient estimator for regressions in which the parameter vector can take any of several values is devised. It is shown that although the likelihood function is unbounded, the likelihood equations have a consistent root. An initial consistent estimator is provided. One Newton step provides efficient estimates. Applications to nonlinear models and contaminated normal models are suggested.