UnFEAR: Unsupervised Feature Extraction Clustering with an Application to Crisis Regimes Classification
IMF Working Papers, November 25, 2020
Source details
- Canonical URL
- UnFEAR: Unsupervised Feature Extraction Clustering with an Application to Crisis Regimes Classification
Other formats
Bibliographic details
- Authors: Jorge A Chan-Lau
- Published: November 25, 2020
- Series: IMF Working Papers
- DOI: https://doi.org/10.5089/9781513561660.001
Summary
- Introduces unFEAR, Unsupervised Feature Extraction Clustering, to identify economic crisis regimes.
- Given labeled crisis and non-crisis episodes and the corresponding features values, unFEAR uses unsupervised representation learning and a novel mode contrastive autoencoder to group episodes into time-invariant non-overlapping clusters, each of which could be identified with a different regime.
- The likelihood that a country may experience an econmic crisis could be set equal to its cluster crisis frequency.
- unFEAR could serve as a first step towards developing cluster-specific crisis prediction models tailored to each crisis regime.
Methodology and Technical Features
- Unsupervised representation learning is used to extract features from crisis and non-crisis episodes.
- A novel mode contrastive autoencoder is central to grouping episodes into time-invariant non-overlapping clusters.
- Clustering output: non-overlapping clusters that can be associated with distinct crisis regimes.
- Uses labeled crisis and non-crisis episodes and their corresponding features values as input.
Applications and Implications for Crisis Prediction
- Cluster crisis frequency can be used to set the likelihood that a country may experience an economic crisis.
- Enables development of cluster-specific crisis prediction models tailored to each crisis regime.
- Addresses the biased label problem via unsupervised feature extraction and clustering.
Subject Areas and Keywords
- Subject: Early warning systems, Economic and financial statistics, Financial crises, Financial statistics, Machine learning, Technology
- Keywords: autoencoder, biased label problem, clustering, crisis data points, crisis frequency, crisis observation, crisis prediction, crisis risk, deep learning, Early warning systems, Global, machine learning, unsupervised feature extraction, WP
Publication and Metadata
- Pages: 24
- Volume: 2020
- DOI: https://doi.org/10.5089/9781513561660.001
- Issue: 262
- Series: Working Paper No. 2020/262
- Stock No: WPIEA2020262
- ISBN: 9781513561660
- ISSN: 1018-5941
- Author: Jorge A Chan-Lau
- Publication date: November 25, 2020
- Series title: IMF Working Papers
Source: UnFEAR: Unsupervised Feature Extraction Clustering with an Application to Crisis Regimes Classification, IMF Working Papers (Working Paper No. 2020/262).
Content in this bundle
- Working Paper