A simple method to detect extreme events from financial time series data

• Published in: Machine learning with applications Vol. 10; p. 100415
• Main Authors: Qu, Tingyu, Mei, Ko Wai, Doray, Arnold
• Format: Journal Article
• Language: English
• Published: Elsevier 01.12.2022
• Subjects: Airline industry; Extreme-event detection; Financial time-series data; Power-law model for tail-slope estimation; Tail risk
• ISSN: 2666-8270; 2666-8270
• DOI: 10.1016/j.mlwa.2022.100415

Stock prices and other financial data fluctuate dramatically as a result of extreme events. We offer a technique for automatically detecting and ranking these events with financial time series data. Our technique works by fitting the tail of a moving window’s return distribution to a power law. We find that rapid changes in the tail slope may be easily connected with major events in the real world when applied to airline stock prices. Our technique also displays a distinct periodicity in tail behavior, implying that it may be utilized to anticipate long-term financial market developments. We also compare our technique to standard deviation, proving that our model outperforms it in several aspects.