We tested 12 machine-learning models on the stock market. None of them worked, and we published everything.

Description and use

Testing 12 machine‑learning models on the stock market revealed no significant edge over random; higher scores correlated with worse performance, while volatility models performed slightly better.

Problem

Determine whether machine‑learning models can predict next‑month stock performance better than random.

Outcome

None of the 12 ML model families achieved significant predictive skill; higher scores correlated with worse performance in the sealed year. Volatility prediction models performed modestly better.

Workflow

• 1. Build a score (0‑100) from five measures for ~100 US stocks and Nifty 50.
• 2. Train each model on historical data using walk‑forward testing.
• 3. Lock the last 12 months as a sealed final year; freeze rules before opening.
• 4. Compare model rankings against random rankings.
• 5. Count every configuration and correct for multiple testing.
• 6. Evaluate performance via correlation and t‑tests.
• 7. Publish all results and formulas publicly.

Evidence

We built a free site, quantooo.com, to do the opposite: test ideas properly, in public, and publish the result whether it is good or bad.

Source

The question was simple: do stocks with a higher score do better over the next month than stocks with a lower one?

Source

Only the past. Every model was trained on earlier data and judged on later data it had never seen (walk-forward testing).

Source

The score: in development (2020–2025) its link with next-month results was +0.0085, statistically indistinguishable from random (t = 0.46). On the sealed final year it was −0.0585 (t = −2.47): higher-scored stocks actually did worse.

Source

Predicting turbulence did a bit better. A volatility model for the S&P 500 had lower forecast error than the standard HAR model on 2025–26 (QLIKE 0.816 vs 1.047, p = 0.047). In an exploratory follow‑up, the decades‑old GARCH model did best of all.

Source