A0778
Title: A multivariate-index-driven anomaly detection system with supervised learning
Authors: Rogemar Mamon - University of Western Ontario (Canada) [presenting]
Abstract: A hybrid supervised learning system is developed to detect anomalies in multivariate time-series index data. The focus of the application is the determination of signs of possible crisis episodes that may wreak havoc on the financial market or economic stability. The proposed statistical-computing approach synthesises stochastic process modelling, hidden Markov filtering, Random Forest and XGBoost. Such an approach is capable of efficiently and accurately tracing simultaneously the financial stress indices (FSIs) of multiple countries and more importantly identifying anomalous FSIs behaviour that signals an impending financial instability. We show that our method is capable of dynamically making 6-step-ahead binary anomalous-normal classification predictions in a probabilistic sense for the benefit of industry practitioners and regulators. Our method, which also gives rise to an early-warning system, is benchmarked with other alternative financial-instability monitoring methods and its advantage is highlighted via various model validation measures.
