In this paper we propose a new approach for estimating the unknown par
ameter in the stochastic linear regressive model with stationary ergod
ic sequence of covariates. Under mild conditions on the joint distribu
tion of the covariate and the error, the estimator constructed is show
n to be strongly consistent in two important special cases: (1) The se
quence of (variate, covariate) is independent identically distributed
(i.i.d.), and (2) the sequence of variates is a stationary autoregress
ive series. the asymptotical normality is also discussed under more as
sumptions on the distribution of the covariate. (C) 1997 Academic Pres
s.