Skip to main navigationSkip to main content
The University of Southampton
CORMSIS Centre for Operational Research, Management Sciences and Information Systems

Financial Crisis Prediction with Machine Learning -- talk by Marcus Buckmann  Event

Time:
14:00 - 16:00
Date:
6 December 2018
Venue:
Building 54, Room 8031, School of Mathematical Sciences, University of Southampton, Highfield Campus, SO17 1BJ

For more information regarding this event, please email Konstantinos Katsikopoulos at K.Katsikopoulos@southampton.ac.uk .

Event details

Financial crises are extreme economic events and very challenging to predict. Part of the difficulty is their infrequent and inherently non-linear nature. We compare how well machine learning models and a logistic regression perform in predicting financial crises in a long-run macroeconomic datasets covering 17 advanced economies between 1870–2013. The machine learning models, in particular ensembles of decision trees, outperform the regression. To address the black-box critique of the machine learning models, we use the Shapley value framework. It identifies credit growth, especially on the global level, and the slope of the yield curve as the main indictors driving the prediction of all machine learning models. These insights could provide valuable input to the macroprudential policy process.

Speaker information

Marcus Buckmann, Bank of England, Marcus is a PhD intern at the Bank of England. He received his Master’s degree in Psychology at the Humboldt University, Berlin. His main research interests are machine learning in general and simple prediction models in particular. In his PhD research at the Max Planck Institute for Human Development in Berlin, Marcus uses computational methods to investigate decision heuristics. He compares them to methods from statistics and machine learning to understand the conditions under which the heuristics perform well. Using these insights, he develops simple prediction algorithms that are highly interpretable and easy to use but still compare well to more complex statistical models.

Privacy Settings