Status: Per-family deep reference (companion to the consolidated Fusion AI Studio & ONNX User Guide) Audience: first-time AI Studio user Answer family: Future Volatility Prediction style: Regression
Future Volatility asks:
“How much price movement is likely over the next N bars?”
This family is different from most of the other regression targets in AI Studio.
It is not trying to predict:
It is trying to predict:
“How much movement or risk should I expect?”
Example:
Target_Volatility20Meaning:
Fusion builds a future volatility label from the next set of forward returns.
Current trainer formula:
Target_Volatility20[t] = rolling std of future returns across bars t+1..t+20
Plain-English translation:
So the label is asking about expected movement size, not expected move direction.
This family predicts how much, not which way.
That is the most important thing to understand.
A higher value means:
“The next N bars are expected to be more active / more volatile.”
A lower value means:
“The next N bars are expected to be quieter.”
It does not automatically mean:
This family uses the normal Step 1:
Fixed-Horizon Windows (bars)If Step 1 says:
5,10,20
Fusion can create:
Target_Volatility5Target_Volatility10Target_Volatility20Then Step 2 chooses one of those columns for one model.
For Future Volatility:
Look-ahead Windows mattersThis is a regression family.
The correct interpretation is:
higher value = more expected future movement
lower value = less expected future movement
This output is usually:
0So if you see values like:
0.00020.0007that can be completely normal.
0 Means HereFor Future Volatility, 0 is the floor, not the centerline.
That means:
0 mean low expected movement0 mean more expected movementIt does not mean:
This is one of the biggest misunderstandings users can have with this family.
This family can still be very useful, just not as a direct entry signal.
It is often useful for:
Plain-English examples:
Users naturally want to turn every ONNX output into:
above 0 = long
below 0 = short
That is not correct here.
This family is not a directional forecast.
It is a:
So the right way to use it is usually:
not:
These two families can feel similar because both can involve “how much movement.”
But they answer different questions.
ATR-Normalized ReturnAsks:
“How far up or down do I expect price to move?”
That is directional and zero-centered.
Future VolatilityAsks:
“How much movement do I expect, regardless of direction?”
That is not directional and usually positive-only.
So if the user wants a centerline long/short score, ATR-Normalized Return is usually the right family, not Future Volatility.
Open Export and make sure:
OHLCV is checkedIn AI Studio Step 1:
Look-ahead Windows, for example 5,10,20Create New CaptureNow Fusion creates answer columns such as:
Target_Volatility5Target_Volatility10Target_Volatility20In Step 2:
Prediction Style: RegressionAnswer Column Source: Generated Answer ColumnsGenerated Question Family: Future VolatilityExample:
20 barsNow the model is learning:
Target_Volatility20Pick the inputs that may help answer:
“How active is the next period likely to be?”
Useful first feature types often include:
Good first rule:
Choose:
Good first algorithms:
LightGBM if availableRandom ForestThe right way to think about the output is:
higher output = more expected movement
lower output = less expected movement
Then decide how to use it, for example:
high volatility forecast = allow breakout setups / wider stops / smaller size
low volatility forecast = prefer mean reversion / smaller targets / normal size
The exact trading use depends on your workflow.
If you want a clean first experiment with this family:
Look-ahead Windows: 5,10,20Prediction Style: RegressionAnswer Column Source: Generated Answer ColumnsGenerated Question Family: Future Volatility10 or 20 barsAlgorithm: LightGBMRandom ForestDo not deploy this first as a direct long/short signal.
Use it first as:
0 as a centerlineFor this family, 0 is the floor, not the midpoint.
> 0 and < 0 as long/short rulesThis family does not directly predict direction.
Positive here just means more expected future movement.
This family is about expected activity, not expected directional edge.
These are the product improvements this guide exposes.
Suggested note:
“Future Volatility is a movement/risk forecast. Use it as a filter, sizing input, or regime signal rather than a direct long/short trigger.”
0 is the floorSuggested compact display:
Future Volatility
positive-only | 0 = floor, not centerline
Suggested note:
“If you want a zero-centered long/short style regression target, ATR-Normalized Return is usually a better first choice.”
Future Volatility means:
“Predict how much movement is likely over the next N bars, where higher values mean more expected activity and near-zero is the floor, not a bullish/bearish centerline.”