Authors:

Jennifer L. Castle, Climate Econometrics and Calleva Project, Magdalen College

Jurgen A. Doornik, Climate Econometrics and Nuffield College

David F. Hendry, Climate Econometrics and Nuffield College

About the paper

Time series data often exhibit breaks in trend which are difficult to predict and affect forecasting afterwards. This paper proposes a real-time approach to rapidly detect trend breaks and subsequently update forecasts. A sequence of increasingly large same-sign 1-step-ahead forecast errors as the forecast origin advances signals a sudden trend shift. Impulse indicators are included as intercept corrections to offset successive forecast errors, then replaced by a deterministic process which can take varying functional forms, tested using encompassing tests. Analysis and simulations for detecting and then forecasting sudden trend changes after just 2 or 3 periods confirm its feasibility for dynamic systems, illustrated by modelling UK money demand following the 1984(3) Finance Act which led to a sharp break in trend.

Data

All the required files for the reproducibility check can be found here. The package was first assembled and submitted to the Editor with the initial version of the paper on 12 February 2026.

The data file UKM1NL.xlsx provides UK quarterly data on UK M1 and relevant variables from this paper.

The batch file called M1XLSX25.fl runs the single equation and system models, so also creates the graphs for their forecasts.

The ModellingTrendBreaksSims24.ox file runs all the simulations.

The batch and ox files require a licensed version of PcGive Professional (PcGive 16 and OxMetrics 9)