ATLASForecast track record: how often were ATLAS ranges right?
ATLAS publishes how its past forecasts turned out. In testing on about 148,000 predictions, its 50% range held 50-51% of the time and its 90% range 89-91%.
What it does
A forecast is only useful if you can check it. The track record page compares what ATLAS said (ranges and a probability of rising) with what actually happened, for the same predictions every user sees.
It also reports the misses. Calibration is not about being right every time, but about the stated odds matching reality over many cases: when we say 90%, nine in ten should land inside.
Worked example with live data
Real ATLAS numbers, prices as of Oct 26, 2026.
What the table says: among about 148,000 past predictions, ranges ATLAS called 50% landed inside 50-51% of the time and ranges called 90% did so 89-91% of the time. Today the 90% range for AAPL is $316 to $384; that statement can be checked against this record.
| Range ATLAS stated | Share of outcomes that landed inside |
|---|---|
| Likely range (middle 50%) | 50–51% |
| Wide range (90%) | 89–91% |
| Tariff shock week, April 2025 | 52% |
How ATLAS calculates it
The backtest used 500 US stocks (100 from each market-cap quintile) and 431 stocks from 49 exchanges, with month-end starting points from September 2023 to September 2026 and horizons from one week to one year. Volatility and drift were recomputed using only data available at each starting date.
The first version of the model had ranges that were too narrow for US stocks in the short term and too wide outside the US. We fixed that with three changes: the long-run drift for every stock, a volatility blend, and a fat-tailed curve. A cross-check on two halves of the period confirmed the improvement.
Limits you should know
The direction forecast, the chance of rising, has little skill beyond the base rate, and the model says so. A shock that hits all stocks at once, like the April 2025 tariff week, left only 52% of outcomes inside the 90% range, because no per-stock model can foresee it.
Where it sits in the plans
The track record is available on every plan.
Run it on your own numbers
Create an account to use this tool with your stocks, your amounts and your horizon.
Get started →Plans from €3.99 a month.
Frequently asked questions
How accurate are stock price forecasts?
Direction is hard to forecast: most models land close to a coin flip. What can be measured is calibration, whether stated ranges contain the real outcome as often as claimed. ATLAS publishes that.
What does calibration mean?
A forecast is calibrated when events it gives a 90% chance happen about 90% of the time. It is different from being accurate on every call.
Where did the model fail?
During shocks that move the whole market at once, for example April 2025, when only 52% of outcomes were inside the 90% range.
How was the model tested?
On about 148,000 predictions for US and world stocks from September 2023 to September 2026, with inputs limited to data available at each starting date.
Related tools
- Stock probabilityFor any stock, ATLAS estimates the probability that it closes higher in 21 trading sessions (about one month) and the range where it should land. Most answers sit close to a coin flip, and that is the honest result.
- Monte Carlo simulationA Monte Carlo simulation runs thousands of random but realistic price paths and reads the outcomes as a distribution, so you see how wide the possibilities are instead of one guess.
- Fair value calculatorATLAS estimates the fair value of a stock from its cash flows and the valuations of comparable companies, and shows a reasonable range and a confidence level next to the price.
Stocks used in this example
Open any of them to see the full public page, with fair value, odds and a one-year chart:
Not investment advice. ATLAS estimates come from statistical models and public data; they can be wrong and past results do not predict future ones. Disclaimer