What Is Steve Cohen Buying Next? Predicted 13F Moves Before They File

By the time a 13F lands, the trade inside it is old. The filing tells you what Point72 owned at the end of last quarter, and it can arrive up to 45 days after that quarter closed. The “news” you are reading describes a position that has been on the books for weeks, sometimes months.

Prior Moves estimates the next likely buy before the filing is public, using a model trained on this one investor’s own history. There is a separate model for each tracked investor, and the Steve Cohen model is trained only on Point72’s filings (34 quarters of history) plus roughly 290 public signals. Style, in one line: Multi-strat (kept out of the consensus mirror).

The current top predicted buy

Right now the model’s top-ranked next buy for Steve Cohen is ETHA (Ishares Ethereum Trust Etf). Names the model put in that position were the investor’s actual next buy 1.6% of the time on walk-forward holdout, against a board base rate of 0.62%. The live number, the full candidate list, and the running record of hits and misses are on the Steve Cohen mirror, updated each filing cycle.

What “predicted next buy” means

It is a model’s estimate of where this investor’s pattern points, stated as a probability, ahead of the paperwork that would settle the question. It is never a claim that Steve Cohen has bought the name, or will, or that you should. When the next 13F files, you can check the prediction against the real filing yourself. That test runs in the open on the track record page.

The honest numbers

Across the combined consensus mirror, the backtest shows a raw edge of about +2.2 points per quarter versus the S&P 500 over 35 complete quarters (2017Q3 to 2026Q1), net of costs. That edge is directional and it is not statistically significant: the t-statistic is about 1.31, the 95% confidence interval runs from -0.7 to +5.8, and a wild bootstrap puts the probability of seeing it by chance at 0.29. Two things weaken it further and belong right here. Most of the average is one quarter, 2026Q1, which posted a +47.5 point edge. And factor adjustment does not rescue it: controlling for market, size, value and momentum leaves +2.73 points per quarter, still not significant at t=1.45, and dropping that one quarter takes it to +0.75 at t=0.68. A raw spread is not a result. The live paper account is -2.87% against the S&P at +1.72% over 50 trading days as of 12 August 2026, far too short to conclude anything and quoted because it is the current number. The model’s accuracy on the board it actually publishes is a measured hit rate: the top-ranked name is the investor’s next new buy 1.7% of the time against 0.6% for a random name on the same board. The older AUC figure of 0.659 grades a different and much narrower exam, ranking names already inside a filing, and a one-line rule taken from the previous filing scores a perfect 1.000 on that exam, so we no longer quote it as next-buy accuracy. We publish the misses alongside the hits because hiding them would make every other claim here worth less.

See the live mirror

• Live Steve Cohen mirror: current predicted buys, probabilities, hits and misses.
• The full top picks board across all tracked investors.
• How the model works: methodology.