On modeling heterogeneity in linear models using trend polynomials
Project Author
Issue Date
2019-03-18
Authors
Michaelides, Michael
Spanos, Aris
Loading...
Embargo
This version of the article is available for viewing to the public after May 18, 2022.
First Reader
Additional Readers
Keywords
Linear model , t-Hererogeneity , Near-collinearity , Trend polynomial , Orthogonal polynomial
item.page.distribution
Abstract
The primary aim of the paper is to consider the problems and issues raised when the data exhibit time heterogeneity in the context of linear models. Ignoring time heterogeneity will undermine the reliability of inference and will give rise to untrustworthy evidence. Accounting for it using trend polynomials, however, is non-trivial because it raises several modeling issues. First, when the degree of the polynomial is greater than 4, or so, one needs to deal with the near-multicollinearity problem that arises. The second issue pertains to the type of polynomial that will adequately account for the time heterogeneity. Third, when the trend polynomials are treated as additional regressors, they will give rise to highly misleading statistical results. The paper investigates how different types of polynomials could deal with the near-multicollinearity and the modeling issues they raise, and makes recommendations to practitioners.
Description
Chair
Major
Department
Economics
Recorder
License
This is an open access article distributed under the terms of the Attribution-NonCommercial-NoDerivs 2.0 Generic (CC BY-NC-ND 2.0) Creative Commons license.
Citation
Michaelides, M., & Spanos, A. (2020). On modeling heterogeneity in linear models using trend polynomials doi:https://doi.org/10.1016/j.econmod.2019.05.008
Version
Final manuscript post peer review, without publisher's formatting or copy
editing (postprint)
Honors
Publisher
Elsevier