Stability Of Regression Models

This book begins with a discussion of the danger of model instability which has received scant attention in the assessment of regression models; this leads to their frequent unsatisfactory performance when used in forecasting or in prediction. A main reason is a well specified regression model with statistically significant parameter estimates of relevant variables can still suffer from structural instability which comes from two sources: (i) a statistical source due to measurement error and variation in the bias due to unknown or known but unexpected exogenous factors, and (ii) the numerical source due to the ill condition of the data matrix. The two sources can jointly amplify the destabilizing force on the parameter estimates, then the model predictions or forecasts. This book exposes major tests for detecting instability of regression models with sufficient discussion and guidance to help the reader select the appropriate tests and apply them. An illustrative example provides a step by step guidance on how to implement those tests in TROLL, the econometric package produced by the MIT (USA). The book is prepared by Ngoc Luan Ho Trieu who holds a Bachelor of Economics and a Bachelor of Art from the University Of Western Australia; and his post-graduate qualifications include a Graduate Diploma in Economics, a Graduate Diploma in Econometrics, and a Master of Economics of Development (Australian National University).