
MALI Platform
AI Stock Research
Forecast Engine and Public Track Record
MALI's forecasts are ranges, not targets: a central path, a band at the confidence you pick, and the probability of a rise at each step. Every daily call is hash-chained when it's made and scored against SPY in a record anyone can audit.
Plan: Every plan, Starter includedIn the app: MALI Lab → Forecast, Forecast Accuracy and Model Comparison; the public /track-record page
What you can do
- Forecast up to ten symbols over horizons from 15 minutes to 90 days, with a band at 60% to 99% confidence.
- See a central path, upper and lower bounds and the probability of a rise at each step, drawn as a forecast cone.
- Choose from a catalogue of time-series models, from GARCH and XGBoost to TimesFM and PatchTST, weighted by how they scored on held-out data.
- Apply one of five stress overlays, such as a sharp decline or an inflation shock, to the same forecast.
- Check stored forecasts against what happened: skill against a random walk, direction and band calibration by horizon step.
- Audit every daily call on the public track record: hashed and chained when made, scored against SPY over the same dates, with Brier scores and 95% intervals.
What it doesn't do
- The point forecast has measured no reliable directional edge, and the app says so; the value is in the range.
- The track record draws no conclusion until 30 calls have been scored.
- Nothing in it places an order.
Data: Daily and intraday prices, MALI's own record of every call
Questions
- How is a call scored?
- From the close when it was made to the first close on or after its due date, less SPY's return over the same dates. A bullish call counts only if the stock beat SPY.
- Can the record be edited after the fact?
- Each entry is hashed with the one before it when the call is made, so a later edit or deletion would break the chain. Anyone can recompute it.
- Which plans include the forecast engine?
- Every plan. The public track record needs no account at all.