Methodology
How the Therm Score is built and validated
The Therm Score combines multiple standard technical factors of an asset's behavior — trend, recent activity, relative volatility, and market participation — into a single 0-to-100 indicator, recalculated every day. The exact formula and its internal parameters aren't published; what we do publish, in full detail, is how it was validated before being shown to any user.
Statistical normalization, not fixed thresholds
Instead of using generic thresholds that are the same for every asset, each component of the Therm Score is compared against the asset's own recent history. This makes the indicator comparable across assets with very different behavior — a stable, low-volatility company and a high-growth stock aren't measured by the same yardstick.
Validation process
Every improvement to the Therm Score goes through this same process before it's published.
✓
Non-overlapping historical backtest
Observations are drawn from windows that are independent of each other, not consecutive — this avoids artificially inflating the sample size with data that actually shares almost all of its information.
✓
Aggregation by market episode
Results are grouped by calendar period, not by individual signal — a month with many simultaneous signals can't dominate the aggregate result over a month with just one.
✓
Strict train/test separation
Any parameter adjustment is decided by looking only at the training segment, split by date — a threshold is never chosen by searching for the best result in the test segment.
✓
Net of transaction costs
Every published result is discounted for an estimated transaction cost — a signal that only works gross, without friction, isn't considered validated.
✓
Tested on a diversified basket before scaling
Before being applied to the full covered universe, every improvement is first tested on a basket of assets across multiple sectors, to reduce the risk that the result depends on a specific sector or regime.
What triggers the informational badges
⚠Caution: a combination of factors that, in historical validation, has shown more erratic behavior than the rest of its own range.
↓Weak Trend: the asset shows a sustained, deteriorated underlying trend — a context in which the Therm Score loses reliability.
◔Limited Data: the asset doesn't yet have enough trading history for statistical normalization to reach its usual reliability.
What can't be claimed
- —This isn't a validated trading strategy on its own — it doesn't account for market impact, liquidity, or sector concentration at scale.
- —It relies exclusively on historical data — no past result guarantees future behavior.
- —The validation period covers a specific stretch of the market, not every possible regime.
Data and update frequency
The Therm Score is recalculated every day after market close, using market data from institutional sources. Prices and volumes used correspond to the daily close — this isn't a real-time indicator. See How to Read It for a guide to the zones, badges, and horizons.
This page summarizes the validation process for informational purposes. The Therm Score and its associated statistical perceptions don't constitute personalized investment advice or a recommendation to buy or sell. Past performance doesn't guarantee future results.