The First Step Away from Linear Regression

This paper is aimed at suggesting an alternative approach to a rolling window Linear Regression. Typically, while doing time series regressions, running an ordinary least square regression is too restrictive. This can happen if we believe that over time our regression coefficients change due to structural reasons. A usual approach is to employ a rolling regression, however this amounts to discarding information in the past. A cleaner approach is to use a state space model and fit it's parameters using Maximum Likelihood Estimation