Measuring Non-Stationarity in the Parameters of a Linear Model with Applications to Asset Returns.
Abstract
The purpose of this paper is to introduce a technique for measuring non-stationarity in linear models. The technique is conceptually simple. It can be adapted easily to a wide variety of problems and improved upon in obvious ways (which we will mention). Its restriction to linearity is probably not important to the practitioner because the overwhelming majority of commonly used techniques are linear; Examples: the sample mean, regression, analysis of variance. The method produces a time path of a linear model's coefficients and it provides the capability to assess the statistical significance of the non-stationarity. Furthermore, this is done under very weak assumptions concerning the probability distribution that generated the data. No particular generating process need be assumed. Such a robust method will clearly not be optimal for every application. An investigator who knows the generating process, or is willing to assume that he knows it, will be able to find a specialized technique better adapted to his case; but for those with a lower degree of skill or of arrogance, the method to be described here has much practical value.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1977
- Accession Number
- ADA042137
Entities
People
- Melvin J. Hinich
- Richard Roll
Organizations
- Virginia Tech