Abstract
A truncated least squares method (TLSM) is proposed for the on-line parameter estimation of linear regression models. The method depends on the truncated data which is a collection of the last M observations. Here, M is any number greater than N which is a number of unknown parameters in a regression model. With a suitable initial setting, the algorithm always satisfies the normal equation of the TLSM. Under the ideal circumstance where there is no uncertainty, the algorithm converges to its true value in a finite steps M. So, if M equals to N, the TLSM gives a minimum steps estimator. The case M=N coincides with the orthogonal projection adaptive algorithm proposed by the author.
