Two Sigma shares regime‑detection framework

Two Sigma’s regime-detection framework is not a new 2026 release. The firm published the note, “A Machine Learning Approach to Regime Modeling,” on October 6, 2021, and it remains available on Two Sigma’s research site and Venn resource pages. The paper was written by Alex Botte and Doris Bao and presents a machine-learning method for identifying market regimes from time-series factor data. ### So what did Two Sigma actually publish? The 2021 note is a short research paper focused on market regime modeling rather than a trading system or a live product launch. Two Sigma says the paper offers “a data-driven approach” to regime modeling and applies that approach to the firm’s Factor Lens, a framework it uses to describe broad drivers of portfolio risk and return. (twosigma.com) Alex Botte and Doris Bao describe regimes as periods when market behavior is relatively persistent. The paper says investors often want to know what regime they are in, how regimes change over time, and how portfolios may behave across those environments. ### What is the core machine-learning idea? The headline method is a Gaussian Mixture Model, or GMM. (twosigma.com) Two Sigma says it uses that machine-learning technique to cluster observations in factor space and infer different market environments from the data rather than defining regimes by hand. The Factor Lens is the input layer for that exercise. (twosigma.com) Two Sigma has said elsewhere that the lens was designed to be holistic, parsimonious, orthogonal and actionable, using broad liquid proxies to capture the main macro and style drivers in institutional portfolios. In the regime paper, those factors become the variables used to separate one market state from another. ### Why would a quant or trader care about regime detection? Financial markets do not behave the same way across time. Two Sigma says persistent shifts in market conditions can affect how portfolios respond, which is why identifying regimes matters for allocators and model builders. For trading applications, the practical use is straightforward even if the paper is written for a broader investment audience: a regime model can be used to decide when signals should be emphasized, downweighted, or stress-tested against a different backdrop. (twosigma.com) That inference follows from the paper’s stated goal of understanding how portfolios react in various regimes and from Venn documentation showing the same regime-modeling approach is used in extreme-scenario analysis and stress testing. (twosigma.com) ### Is this a how-to guide or a research template? The note reads more like a framework than a cookbook. Two Sigma says it presents the machine-learning approach, shows historical results, discusses the model’s output for the current environment at the time, and then outlines practical use cases for allocators. That makes it useful as a template for research projects. (twosigma.com) A quant reader can lift the structure — define a factor set, fit a clustering model, map historical periods to inferred states, and then test portfolio or signal behavior by state — even though the paper does not publish a turnkey trading strategy. That characterization is an inference based on the paper’s described methodology and use cases. ### How does this fit with Two Sigma’s broader research? Two Sigma has published several related pieces around the Factor Lens and regime-aware analysis. The 2018 Factor Lens paper set out the factor framework; a 2019 paper extended that work into factor-return forecasting; and the 2021 regime-modeling note applied machine learning to identify market states using those factors. (twosigma.com) Venn, Two Sigma’s portfolio analytics platform, also references the same regime-modeling research in its stress-testing materials. That suggests the note was meant to feed into portfolio construction and scenario analysis as much as into pure academic discussion. ### Where can readers check the original document? (twosigma.com) The October 6, 2021 article page on Two Sigma’s site links to the research, and a PDF version is hosted on the company’s domain. Venn’s resource page also hosts the same paper and describes it as foundational research for its extreme-scenario analysis. As of May 24, 2026, the document remains publicly accessible through those Two Sigma and Venn pages, where Alex Botte and Doris Bao are listed as the authors. (venn.twosigma.com) (twosigma.com 1) (twosigma.com 2)