p.1

First Order Autoregression: Inference, Estimation, and Prediction

Guy H. Orcutt
Herbert S. Winokur, Jr.

Abstract

Monte Carlo techniques are used to study the first order autoregressive time series model with unknown level, slope, and error variance. The effect of lagged variables on inference, estimation, and prediction is described, using results from the classical normal linear regression model as a standard. In particular, use of the t and x^2 distributions as approximate sampling distributions is verified for inference concerning the level and residual error variance. Bias in the least squares estimate of the slope is measured, and two bias corrections are evaluated. Least squares chained prediction is studied, and attempts to measure the success of prediction and to improve on the least squares technique are discussed.