Date/Time
Date(s) - 09/02/2024
11:00 am - 12:00 pm

Location: HH 406

Speaker: Joseph Voskamp

Title: Financial Crisis Prediction using Machine Learning

Abstract:  We consider the problem of predicting when financial crises will occur. We look at the methods from “Credit growth, the yield curve and financial crisis prediction: evidence from a machine learning approach (Joseph, 2020) and their use of machine learning in financial crisis prediction. They look at 5 different machine learning models and investigate whether they can achieve better results than logistic regression. We apply the methods to updated datasets and aim to apply the methods to different types of financial crises. The Jorda-Schularick-Taylor (JST) macro history database is used as the main data set, which contains (country, year) observations of different variables including whether country suffered (1) or did not suffer (0) a financial crisis that year. We then use definitions of different types of financial crises from “This Time is Different” (Reinhart and Rogoff, 2009). In their work, they investigate banking crises, inflation crises, and currency crises. Using the (updated) datasets from This Time is Different, we redefine where the 0s and 1s go in the crisis variable and apply the methods from (Joseph 2020).

We show that the extremely randomized trees machine learning method can outperform logistic regression in financial crisis prediction. We also employ the SHapley Additive exPlanantions (SHAP) approach from “A Unified Approach to Interpreting Model Predictions” (Lundberg and Lee 2017), as a means of using shapley values to uncover the ‘black box’ nature of machine learning models.