Black swan events and machine learning

Abstract

Black Swan events refer to hard-to-predict and rare events that have a low probability of occurrence but have widespread impacts whenever they occur, be it positive or negative. These events can either be in the form of natural disasters, political events or even as catastrophic financial market crashes. From the stock market crash in 1987 to the global financial crisis in 2008 and to the most recent COVID- 19 stock market crash in 2020 that has devastated the world economy, Black Swan events in finance have severe consequences on global socio-economy due to which not contemplating them for their rarity is no longer an option. The purpose of this research study is to use machine learning tools and statistical tools in order to detect, fit and predict such critically catastrophic financial market crashes in anticipation of capturing black swan events in future. For this, we have used the concepts of both crashes as “outliers” [4] as termed by the authors Didier Sornette and Anders Johansen and also crashes characterized as “drawdowns” [4] by author Emilie Jacobsson as guidance and then analyzed and fitted major financial market crashes using different statistical tools including an advanced non-linear generalized log-periodic power law model proposed by Sornette and Johansen, and also different machine learning tools including neural networks in order to predict the trend of financial market indices. Due to complex time series of financial market index, we found thresholds for different financial crash data exhibiting different behavior, based on which we detected historical financial market crashes using statistical and machine learning tools including neural networks. In order to find thresholds, we have considered the viewpoints of author Emilie Jacobsson and authors Didier Sornette and Anders Johansen. Using data from six major global financial market indices, we also cross validated our models in order to evaluate them. Using these thresholds, any supervised machine learning model can learn about financial market crashes.