Status: Per-family deep reference (companion to the consolidated Fusion AI Studio & ONNX User Guide) Audience: first-time AI Studio user Answer family: Future Log Return Prediction style: Regression
Future Log Return asks:
“What is the expected log-scaled move from the current close to the close N bars ahead?”
This family is very close in spirit to Future Return.
The difference is:
Future Return uses a raw return-style moveFuture Log Return uses a log-return moveSo the trader meaning is still directional, but the scale is slightly different.
Example:
Target_LogReturn10Meaning:
Target_LogReturn10[t] = log(Close[t+10] / Close[t])
Plain-English translation:
The result is still:
This family is still a signed directional return forecast.
That means the main idea is the same as Future Return:
positive = expected upside
negative = expected downside
near 0 = weak directional expectation
The difference is not the long/short meaning.
The difference is the scale.
So the right mental model is:
“This behaves like a return forecast, but on a log-return scale.”
This family uses the normal Step 1:
Fixed-Horizon Windows (bars)If Step 1 says:
5,10,20
Fusion can create:
Target_LogReturn5Target_LogReturn10Target_LogReturn20Then Step 2 chooses one of those columns for one model.
For Future Log Return:
Look-ahead Windows mattersThis is a regression family.
The correct interpretation is:
positive = expected upside
negative = expected downside
near 0 = weak directional expectation
This output is:
So 0 is a real centerline here.
This is the main question a user will ask.
Future ReturnFuture Log ReturnSo the practical difference is not:
The practical difference is:
If the user wants the simplest mental model, Future Return is usually easier.
If the user wants a return-style target with smoother scaling, Future Log Return is a good alternative.
0 Means HereFor Future Log Return, 0 means:
That is the same centerline idea as:
Future ReturnATR-Normalized ReturnIt is not like:
Future Volatility, where 0 is the floorUp / Down, where 0 means not upDirection (3-Class), where 0 means neutral classThis family can be useful when the user wants:
This is a reasonable alternative when:
Future Return feels too rawOpen 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_LogReturn5Target_LogReturn10Target_LogReturn20In Step 2:
Prediction Style: RegressionAnswer Column Source: Generated Answer ColumnsGenerated Question Family: Future Log ReturnExample:
10 barsNow the model is learning:
Target_LogReturn10Pick the inputs that may help answer:
“Will the next N bars produce an upside or downside move?”
Good first feature types often include:
Good first rule:
Choose:
Good first algorithms:
LightGBM if availableRandom ForestThink of the output like this:
positive = expected upside
negative = expected downside
near 0 = weak directional expectation
The sign matters more than the exact plotted shape.
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 Log Return5 or 10 barsAlgorithm: LightGBMRandom ForestThis gives a clean return-style directional workflow with a log-return scale.
It is still directional. The main difference is scale, not the long/short meaning.
It does not.
0 is anything other than the centerlineHere, 0 still means no expected move.
The sign is usually easier to interpret than the exact raw number.
These are the product improvements this guide exposes.
Suggested note:
“Future Log Return is still a signed directional return forecast. The main difference from Future Return is the log-return scale, not the long/short meaning.”
Suggested compact display:
Future Log Return
signed | centered on 0 | regression
Suggested note:
“If you want the easiest first mental model, start with Future Return or ATR-Normalized Return. Choose Future Log Return when you specifically want a return-style target with smoother scaling.”
Future Log Return means:
“Predict the signed log return from the current close to the close N bars ahead, where positive means expected upside, negative means expected downside, and 0 is the centerline.”