Using structural break inference for forecasting time series

Gantungalag Altansukh,Denise R Osborn

doi:10.1007/s00181-021-02137-w

Gantungalag Altansukh, Denise R Osborn

Open Access

https://doi.org/10.1007/s00181-021-02137-w

Copy DOI

Abstract

Rather than relying on a potentially poor point estimate of a coefficient break date when forecasting, this paper proposes averaging forecasts over sub-samples indicated by a confidence interval or set for the break date. Further, we examine whether explicit consideration of a possible variance break and the use of a two-step methodology improves forecast accuracy compared with using heteroskedasticity robust inference. Our Monte Carlo results and empirical application to US productivity growth show that averaging using the likelihood ratio-based confidence set typically performs well in comparison with other methods, while two-step inference is particularly useful when a variance break occurs concurrently with or after any coefficient break.

Highlights

The pervasiveness of structural breaks in many macroeconomic time series is widely acknowledged (Stock and Watson 1996; Paye and Timmermann 2006) and they are an important source of a forecast failure (Hendry 2000; Hendry and Clements 2003)
This paper investigates the usefulness for forecasting of employing a wider range of information relating to structural break testing than implied by the use of a point estimate of the break date in the model’s coefficients
Our simulation results show that the Eo and Morley (2015) set is useful for this purpose and performs well relative to other methods, including those based on a point estimate of the break date and others that do not use any break information

Summary

Introduction

The pervasiveness of structural breaks in many macroeconomic time series is widely acknowledged (Stock and Watson 1996; Paye and Timmermann 2006) and they are an important source of a forecast failure (Hendry 2000; Hendry and Clements 2003). This paper considers a scenario in which a discrete and permanent change in model coefficients may occur during the sample period used for estimation. Various forecast methods and strategies are proposed in the literature to deal with such a possibility, and these can be broadly classified into those that employ an estimated break date and robust methods that treat the break date as unknown.

Objectives

Methods

Results

Conclusion