SimStock : Representation Model for Stock Similarities

Abstract

In this study, we introduce SimStock, a novel framework leveraging self-supervised learning and temporal domain generalization techniques to represent similarities of stock data. Our model is designed to address two critical challenges: 1) temporal distribution shift (caused by the non-stationarity of financial markets), and 2) ambiguity in conventional regional and sector classifications (due to rapid globalization and digitalization). SimStock exhibits outstanding performance in identifying similar stocks across four real-world benchmarks, encompassing thousands of stocks. The quantitative and qualitative evaluation of the proposed model compared to various baseline models indicates its potential for practical applications in stock market analysis and investment decision-making.