Change-Point Detection in Business Cycles using Machine Learning Algorithms
2022
Masoudnia, Mohammadreza | Masoudnia, Mohammadreza | Pérez Quirós, Gabriel | Carnero, M. Angeles | Universidad de Alicante. Departamento de Fundamentos del Análisis Económico
Turning points in business cycles are defined as the onset of a recession or an expansion which are quite difficult to be predicted. In this thesis, we approach the problem of turning (change) point detection as the viewpoint of binary classification task. Due to the small ratio of changes to total data (as the number of recessions is relatively low), we face heavily class-imbalance challenge in this problem. We explore a wide variety of machine learning-based solutions for this problem: from base classifier to the multi-step classifier ensemble algorithm as well as a feature selection step. We examined the proposed classification methods on Canadian large dataset. Among different examined methods, the hybrid ensemble method including data sampling followed by a feature selection and multi-step ensemble can predict the Covid19 recession’s changepoints precisely with all the time series available one month ago. Some robustness checks such as the effect of window size on the model performance are also provided. Moreover, excluding the financial crisis from the training set, the method 8 is still able to predict the changepoints in the case of financial crisis precisely, however, in the case of the Covid-19 recession, they were detected one-period late, suggesting importance of financial crisis’ data in detecting Covid-19 change points.
Show more [+] Less [-]AGROVOC Keywords
Bibliographic information
This bibliographic record has been provided by Universitat d'Alacant