Why the ML Market Map Describes, It Doesn't Predict

Learn why unsupervised models like clustering and PCA map market structure but never give buy or sell signals.

Category: Machine Learning

In plain English

SurgeFlow's ML "market map" is like sorting a big box of buttons into piles by how they look, not by how good they are. The computer looks at lots of stocks and groups together the ones that behave alike, then draws a picture of the piles. It is a description of what the data looks like right now. It does not say which button is lucky, and it never tells you what to buy or sell.

What it measures

"Unsupervised" means nobody told the model the right answer ahead of time; it just finds patterns by similarity. It measures structure: which stocks cluster together, which one or two summary lines (called PCA components) capture the most spread in the data, and which stocks look statistically unusual today (an "anomaly"). Example: imagine 10 stocks. If 6 of them rose about +1% on heavy volume and 4 fell about -1% on light volume, the model might put them in two clusters of 6 and 4. That split describes today's behaviour; it is not a forecast that the 6 will keep rising. - Cluster = a group of stocks that moved similarly - PCA component = a summary line capturing the biggest spread, not a real economic force - Anomaly = unusual compared to peers today, which is neither good nor bad

How SurgeFlow uses it

This underpins the ML tab (/ml), a once-a-day snapshot of the US market built only from data that already exists (prices, volume, and similar features). The pipeline scales the features, compresses them with PCA, then groups stocks using several clustering methods and keeps the grouping they agree on. It is purely descriptive: it organises the market into a readable map. It produces no scores, no rankings of "better" stocks, and no trade signals.

How to read it on the site

Open the ML tab under Research and you will see stocks placed into clusters, plus any names flagged as unusual for the day. Read a cluster as "these moved alike recently," and read an anomaly as "this stood out from its peers today." Use it as a lens to ask better questions, then check other tabs to investigate. - A cluster tells you who is moving together, not who will win - An anomaly is a prompt to look closer, not a recommendation - Cross-check a name on Price, Factors, Fundamentals, or News before drawing conclusions

What it cannot tell you

This is the honest core: the map describes structure, it does not predict returns or give buy/sell signals. PCA components are math summaries, not real economic factors, and they have no inherent meaning. Cluster numbers are just labels with no permanent identity, the clusters themselves change as the sample changes, and the number of clusters (K) is a choice a human made, not a fact about the world. Treat everything here as a starting point for questions, never as advice. - Cluster #3 today is not the same "thing" as cluster #3 tomorrow - Re-run on different data and the groups can shift - Probability-style outputs (e.g. from a GMM) are model estimates, not the odds a stock goes up - Similar stocks sitting close together is not cause and effect - Statistically unusual is not the same as a good or bad opportunity - SurgeFlow is research and education, not investment advice