In this article published in the Journal of Financial Data Science, coauthors from MSCI and MIT develop and describe a data-driven system using AI tools to capture market perception and, in turn, group companies into peer clusters.

We observed historically notable crowding across factors, industries and stocks through the first half of 2021. Examining crowding from multiple perspectives and incorporating multiple data elements provides investors a more holistic view. 

It is not etched in stone that relationships between factor exposures and returns must be linear. We found machine-learning algorithms could identify nonlinear relationships and be used to construct a factor showing significant explanatory power. 

Hedge funds held heavy, relatively illiquid short positions in crowded stocks before the January 2021 short squeeze. We created a risk factor that added explanatory power and may provide insights into potential short squeezes and other risks.

How does one identify or quantify a bubble? We propose a framework for assessing the “bubbliness” of stocks and portfolios, rooted in the idea that bubbles are driven by the same forces, and share characteristics with crowded trades.

It’s clear that some companies were better positioned to take advantage of a remote work environment than others. We built a hypothetical “remote-operation capacity” factor to seek to measure the effect on different firms.

Peer selection plays a central role in many aspects of finance — including valuation, financial and competitive analysis, risk and portfolio modeling and ESG scoring and alpha generation. One common method of peers identification is industry classification, such as the Global Industry Classification Standard (GICS®). Here we ask if it’s possible to identify and measure relationships between companies from a systematic analysis of various types of data, including fundamentals, news and the...