[2I6-GS-10] AI application: Finance

[2I6-GS-10-05] Anomaly Detection in the Stock Market Using Graph Based Entropy and Inter-Domain Linkage

〇Yoshiyuki Nakata¹, Takaaki Yoshino¹, Toshiaki Sugie¹, Kakeru Ito¹, Kaira Sekiguchi², Naijia Liu², Yukio Ohsawa² (1. Nissay Asset Management Corporation, 2. Univ. of Tokyo)

Keywords:Graph-based entropy, cluster analysis, anomaly detection

In the stock market, it is said that the group of stocks that are traded differ depending on the phase of the business cycle. In this study, we developed a method to calculate risk index values based on the degree of bias in the selling of stocks that are part of the stock market toward stocks related to long-term business cycles. Specifically, we clustered stocks with strong relationships based on the correlation coefficient matrix of long-term returns and quantified the occurrence of bias in short-term selling on each cluster using Graph Based Entropy and Inter domain linkage techniques. A comparative study was conducted on the properties of existing risk indicator values (volatility, liquidity, and correlation) for three equity indexes, TOPIX 500, S&P 500, and STOXX Europe 600. It was shown that the developed indicator values reacted in phases where the existing indicator values do not, and may be able to detect risks that are not captured by the existing indicator values.