What is it about?
This research introduces a new way to understand how different stocks are related to each other in the stock market. Instead of relying on traditional financial metrics, we developed a method that uses AI to learn these relationships by looking at how stock prices move together over time. Just like how people might notice that technology companies often move together in the market, our system learns these patterns automatically, without subjective bias. The method uses a technique called contrastive learning - it learns by comparing stocks that behave similarly versus those that behave differently. When tested, our approach was better at grouping similar companies together and finding stocks that could help balance investment portfolios compared to traditional methods.
Featured Image
Photo by Markus Spiske on Unsplash
Why is it important?
Understanding relationships between stocks is crucial for many financial decisions, from building diversified investment portfolios to running statistical arbitrage techniques like pairs trading. However, traditional methods often rely on oversimplified measures or require extensive manual analysis by experts. Our approach offers a more sophisticated and automated way to uncover these relationships, making it easier for both financial professionals and individual investors to make more informed decisions. This could lead to better investment strategies and risk management tools that are more accessible to a broader range of users.
Perspectives
Traditional ways of grouping similar companies, like the Global Industry Classification Standard (GICS), rely heavily on subjective human judgments. While these classifications are useful, they can miss important relationships - for instance, a technology company and a retail company might be more financially connected than their different industry labels suggest. Until now, attempts to use data to understand these relationships have been relatively simple, like computing Pearson correlation. While there's been extensive research using machine learning to predict stock prices, understanding the deeper relationships between companies has remained surprisingly manual and subjective. Our approach offers a middle ground: it can automatically discover complex relationships between companies that might not be obvious from traditional industry groupings, while still being systematic and data-driven. This could complement existing classification systems by revealing connections that human experts might miss, leading to more nuanced and comprehensive ways of understanding how companies relate to each other in the modern economy. The real advantage here isn't in replacing human judgment, but in augmenting it with insights drawn from patterns in vast amounts of market data that would be impossible to analyse manually.
Rian Dolphin
Read the Original
This page is a summary of: Contrastive Learning of Asset Embeddings from Financial Time Series, November 2024, ACM (Association for Computing Machinery),
DOI: 10.1145/3677052.3698610.
You can read the full text:
Contributors
The following have contributed to this page
