Status: Per-family deep reference (companion to the consolidated Fusion AI Studio & ONNX User Guide) Audience: first-time AI Studio user Answer family: Regime Prediction style: Classification
Regime asks:
“What kind of market state are we in right now: low volatility, normal volatility, or high volatility?”
This family is not a directional model.
It is a:
Example:
Target_RegimeCurrent trainer meaning:
0 = low volatility
1 = normal volatility
2 = high volatility
Current data-generator rule:
Target_Regime uses volatility quantiles:
0 = low
1 = normal
2 = high
Plain-English translation:
This family does not use future bars in the same way as the forward targets.
It is a state classification target, not a forward return label.
This family predicts market state, not trade direction.
That is the most important thing to understand.
It is not mainly answering:
It is answering:
“What volatility regime am I in?”
That makes it useful as:
This family now has its own dedicated Step 1 control.
In the Regime Classification group (“Group C”) of the Step 1 capture settings, enable Volatility Regime (low / normal / high). That exposes two optional controls:
Like Future Volatility, it is a market-state family, so it does not use the directional Look-ahead Windows the forward-return targets do.
Regime is a built-in generated answer family. Enabling the Volatility Regime checkbox in the Regime Classification group at capture time produces a low / normal / high volatility-state label.
So the user should think of it as:
“a built-in market-state label you switch on in the Regime Classification group at capture time”
This is a classification family.
The correct trader meaning is:
0 = low volatility
1 = normal volatility
2 = high volatility
The model outputs an encoded class id (0 / 1 / 2). The AI Studio model card and this guide tell you what each id means, so you read it as the regime label directly:
0 → low volatility, 1 → normal volatility, 2 → high volatility
not:
treat the raw number like a centered oscillator
0 Means HereFor Regime, 0 means:
It does not mean:
That is one of the biggest user traps with this family.
This family is often very useful because many strategies behave differently in:
Examples:
This is why Regime is better thought of as a helper model than a direct signal model.
Users naturally want to convert every classifier into:
one class = long
another class = short
That is not what this family is for.
The classes are:
They describe the market environment, not the trade direction.
These two families are related, but they are not the same.
Future VolatilityAsks:
“How much movement is likely over the next N bars?”
That is a regression forecast.
RegimeAsks:
“What volatility state am I in?”
That is a classification/state model.
So:
Future Volatility gives a numeric movement forecastRegime gives a low / normal / high bucketOpen Export and make sure:
OHLCV is checkedBecause this is a state model, features describing current market condition are often more useful than purely directional features.
In AI Studio Step 1:
Create New CaptureWith the Volatility Regime checkbox on, Fusion includes:
Target_RegimeThere is no separate “Regime horizon” field — unlike the forward-return families, Regime is a current-state label, so it has no look-ahead window.
In Step 2:
Prediction Style: ClassificationAnswer Column Source: Generated Answer ColumnsGenerated Question Family: RegimeNow the model is learning:
Target_RegimeThere is no separate horizon picker for this target in the same way as the forward families.
Pick the inputs that may help answer:
“What kind of volatility state am I in?”
Useful first feature types often include:
Good first rule:
Choose:
Good first algorithms:
LightGBM Classifier if availableRandom Forest ClassifierThink of the output like this:
0 = low-volatility regime
1 = normal regime
2 = high-volatility regime
Then decide how your strategy should behave in each state.
Examples:
low regime = smaller targets / more mean-reversion bias
normal regime = normal rules
high regime = breakout bias / reduced size / wider thresholds
The specific policy is your trading design, not the label itself.
If you want a clean first experiment with this family:
Prediction Style: ClassificationAnswer Column Source: Generated Answer ColumnsGenerated Question Family: RegimeAlgorithm: LightGBM ClassifierRandom Forest ClassifierUse the model first as:
not as a standalone entry trigger
0 as a short signalHere, 0 means low volatility.
2 as a long signalHere, 2 means high volatility.
It is a market-state model, not a directional model.
It is a market-state family. Enable it in the Regime Classification group at capture time, but read its output as a volatility regime (low / normal / high), never as a long/short signal.
These are the product improvements this guide exposes.
Earlier drafts of this guide noted that Regime lacked a dedicated Step 1 field. That has since shipped: the Regime Classification group with a Volatility Regime (low / normal / high) checkbox plus optional Volatility Period / Volatility Column controls. This guide was updated accordingly on 2026-06-28.
Suggested compact display:
Regime
class output | 0 = low | 1 = normal | 2 = high
Suggested note:
“Use Regime as a filter or market-state model rather than a direct long/short entry target.”
Regime means:
“Classify the current market state as low, normal, or high volatility, and use that state as trading context rather than as a direct directional signal.”