Using deep learning to characterise weak signals in global equity markets: a case study of COVID-19

This study examines the use of deep learning to identify and characterise anomalous events and their preceding weak signals in equity price data. Particular interest is placed on Gray Rhino events, indicated by the presence of progressively stronger signals prior. The market behaviour prior to and during the COVID-19 pandemic on G-20 equity markets provides a useful context to this end. Existing literature has examined the effects of the pandemic on these markets but has yet to provide conclusive insights into the development of the major equity crash. We compare the existing literature concomitantly to our rigorous application of event study methodology, identifying the presence and effects of signals prior to the market crash. In addition, we develop and deploy a novel Anomaly Characterisation Process (ACP). The ACP utilises an ARIMA time series model to transform equity price time-series for the extraction of relevant information, whereby subsequent fits of GJR-GARCH and deep-undercomplete-autoencoder models are deployed. Resultantly, measures of dispersion and atypicality are produced which allow for effective and clear characterisation of the degree of typicality of the equity prices and their movements. This innovative method demonstrates efficacy in detecting both point and contextual anomalies. When applied in the context of COVID-19, the findings suggest that different event types can be distinguished successfully with this novel approach through the identification of weak signals. Notably, these insights of the ACP in conjunction with those of the event study suggest that the COVID-19 market crash is consistent with a Gray Rhino event and not a Black Swan event. We briefly demonstrate that these insights can be used by market participants to improve risk-adjusted returns via ACP-informed risk-mitigation techniques.

Grantor dc:publisher.institution

Department of Finance and Tax

Year dc:date.issued

2024

Author dc:creator

• Clarke, Keegan G

Advisor dc:contributor.advisor

• Huang, Chun-Sung

Handle dc:identifier.uri

http://hdl.handle.net/11427/40838

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/40838