Determining NBER recession points using machine learning

  • Luca Lavarini (Student)

Student thesis: Master's Thesis

Abstract

The financial crises cause significant challenges due to their profound impact on the economy and the inherent difficulty in predicting such events. Successfully forecasting a financial crisis could offer remarkable advantages, enabling preemptive measures to mitigate its adverse effects. Previous research has highlighted the importance of various indicators in predicting economic downturns, including the inverted term spread, real GDP, and unemployment rates. Additionally, machine learning methods have shown potential in identifying non-linear patterns among these variables, making them valuable in forecasting NBER recessions. In this study, we evaluated several machine learning classification and non-linear regression algorithms such as Support Vector Machine, K-Nearest Neighbours, Decision Tree, Extreme Gradient Boosting, Adaptive Boosting, Random Forest, Extra Trees, and Categorical Boosting other than traditional time series models like ARIMA and AR. The best forecast of the NBER recession points from 0 to 12 months ahead was obtained by inputting the best machine learning models9 prediction as one of the exogenous variables of an ARIMA(1,0,1). The forecasts obtained were especially effective between t + 0 and t + 4, with real GDP being the most relevant macroeconomic feature. Additionally, one version of the forecast was better suited to predict market troughs than official NBER recessions. Future research could extend this work by exploring the impact of different types of recessions, developing models tailored to emerging markets, or training models on specific big debt crises, such as using data from the 2008 financial crisis to forecast recessions similar to Japan9s 1990 economic downturn.
Date of Award17 Oct 2024
Original languageEnglish
Awarding Institution
  • Universidade Católica Portuguesa
SupervisorJonathan Tepper (Supervisor)

Keywords

  • Machine learning
  • NBER recession
  • Business cycle forecast

Designation

  • Mestrado em Finanças (mestrado internacional)

Cite this

'