← Our researchSeptember 30, 2026 · AI models

Can Google’s forecasting AI call the market?

Across 7,127 forecasts, TimesFM got the direction right 52.7% of the time — less than simply assuming stocks go up.

Forecasts
TimesFM52.7%
Always “up”58.8%

The question.

TimesFM is Google’s open forecasting model. Can it tell where a stock goes next, or how wide the range should be?

52.7%direction right, vs 58.8% for “always up”
+0.001information coefficient, vs +0.018 for momentum
74.5%of outcomes inside its 80% range
11 of 15years its range was worse than a simple rule

How we tested it

1The data41 large US stocks plus the S&P 500, a forecast every 21 trading days from 2012 to 2026.
2The questionDirection and range over the next 20 trading days, against simple rules anyone can apply.
3The extra roundQuarterly and annual revenue for hundreds of companies, and next month’s volatility.

What we found.

  • On returns, zero-shot forecasting is close to a coin flip, in line with recent academic work.
  • Its ranges are too narrow: they miss more often than they should.
  • Where it helps is next month’s volatility, where it beat simple averages.
Can Google’s forecasting AI call the market? · Finaxion