Anomalijų aptikimas finansų rinkose /

Information that allows better understanding of the situation in the financial markets and a timely response to an unusual behaviour is extremely valuable for both the private sector and public organizations. Anomaly detection algorithms are widely used when there is need to detect abnormal behavior of an object in a real time, however the application of such algorithms in financial markets has not yet been widely studied. The aim of this thesis is to explore the algorithms used for anomaly detection and difficulties related to use of such algorithms, find an algorithm suitable for a stock market application and test it using real market data. In order to achieve this goal, potentially suitable methods were examined and an algorithm that allows flexible experimentation with different configurations was developed. The experiment used trading data of S&P500 financial sector companies from 1990 to 2019. LSTM neural networks and LASSO regression were used for data modeling. In the analyzed situation, the algorithm using Long Short Term Memory neural networks showed better results than the algorithm using the Least Absolute Shrinkage and Selection Operator regression, allowing a much earlier detection of an unusual period. The algorithm using Long Short Term Memory neural networks began to indicate an abnormal situation later than the VIX indicator, however, it was more stable during the normal period which meant less false-positive signals.

Grantor dc:publisher

Institutional Repository of Vilnius University

Year dc:date

2020

Author dc:creator

• Chrunov, Michail,

Contributors dc:contributor

• Raudys, Aistis

Statement dc:rights

• info:eu-repo/semantics/openAccess

Language dc:language

lit

Repository record dc:identifier

https://repository.vu.lt/VU:ELABAETD81705591&prefLang=en_US

OAI identifier oai:identifier

oai:vu.lt:elaba:81705591