How Tradient works
A plain-english guide to the data, the models, and how to use what you see.
What Tradient does
Tradient processes 2,125 signals from CFTC positioning, macroeconomic data, and cross-asset prices across 48 instruments — every day. It scores each signal against historical data, combines them into a directional bias per asset, and tells you whether that bias is fresh or already played out.
The data
CFTC / COTWeekly positioning reports from the US regulator — who is long, who is short, and how extreme that is historically.
Macro (FRED)Interest rates, CPI, unemployment, industrial production and more from major economies.
Cross-asset pricesGold, oil, crypto, volatility (VIX), yield curves — price signals that historically predict FX and equity moves.
79,162 correlationsEvery signal is tested against every asset across 5-day, 10-day, 20-day, and 30-day horizons. Only statistically significant correlations survive.
How the bias is built
1. Score each signal
Every signal gets a percentile rank (0–100) based on its current reading vs the last year of history. A score of 90 means it's near its highest level in a year.
2. Weight by predictive power
Signals are weighted by their information coefficient (IC) — how well they historically predicted the asset's direction. Weak predictors count less.
3. Composite score
The weighted average of all signal scores produces a single composite number. Above zero = bullish bias, below zero = bearish bias.
4. Confidence
How many of the 8–12 models agree on direction — expressed as a percentage. 70%+ is considered reliable.
Reading the dashboard
Direction
BULLISH or BEARISH — the current fundamental bias for that instrument.
Confidence %
Backtested OOS hit rate for this asset's signal system. This is what the model achieved on data it had never seen. Not a guarantee, a track record.
Expected Move
Estimated price move over the next 20 days based on current signal strength. Derived from historical correlation between score and actual returns.
Window State
Whether the entry window is still open. A signal can be fundamentally correct but the move may already be done. See below.
The entry window
Not all active biases are equally actionable. A bias that's been running for 60+ days has likely already produced most of its price move — even if the fundamentals haven't changed. Tradient labels each signal with a window state:
● ActiveSignal is fresh (< 14 days). Highest historical hit rate: 63%.
◑ ClosingSignal is 15–30 days old. Still valid but act soon.
◔ WatchSignal is 31–60 days old. Move may be partially complete.
○ DevelopingSignal is building but not at an extreme yet.
◌ StaleSignal is 60+ days old. Primary move likely already happened.
Source: analysis of 28,789 OOS episodes. Window state is based on the current episode duration, not the total fundamental signal age.
Reading the instrument detail
Price Targets
Conservative, expected, and extended price levels based on signal strength and historical move distributions. Each has an OOS hit rate — conservative is higher-probability, extended is lower.
Active Episode
When a signal crossed the entry threshold, an episode opens. It shows return since entry, how long it's been held, and the asset's OOS hit rate for episodes like this one.
Signal Map
A visual breakdown of which signals are at extremes and in which direction. Signals cluster into factors (COT, macro, rates, etc). More red/green extremes = stronger bias.
What Tradient is not
- —Not a trading signal service. It gives you the fundamental context — the "why". You still decide the entry timing.
- —Not guaranteed. A 70% hit rate means 30% of the time it's wrong. Size positions accordingly.
- —Not real-time. The CFTC data updates weekly. Macro data updates monthly. Price signals update daily.
- —Not advice. This is analytical output from a quantitative model, not personalized financial advice.