ABSTRACT
This chapter introduces the concept of regression quantiles. It presents a side-by-side analysis of a time series using a standard and a robust approach, describes how tests of significance may be performed with the latter and illustrates how confidence limits on forecasts may be empirically calculated. Robust estimation of time series models has been a subject of limited investigation. Many of the concepts and results discussed here extend directly to autoregressive time series analysis. The time series chosen for a representative analysis is the Personal Saving Rate, recorded on a quarterly basis from 1955 to 1984. Taken from the Business Conditions Digest, this time series is seasonally adjusted prior to publication. In the context of multiple regression, the use of a least squares criterion for the estimation of parameters can be justified in several ways.
