Docs  /  AI Studio Targets  /  Future Return

AI Studio Answer Guide: Future Return

Status: Per-family deep reference (companion to the consolidated Fusion AI Studio & ONNX User Guide) Audience: first-time AI Studio user Answer family: Future Return Prediction style: Regression

1. The Plain-English Question

Future Return asks:

“How much higher or lower is price expected to finish after N bars, measured as a percent change from the current close?”

This is one of the clearest first answer families because it is easy to picture:

current close  ----------------->  close N bars later
      |                                   |
      |----------- percent move ----------|

If the future close is higher, the answer is positive. If the future close is lower, the answer is negative.

2. What Fusion Is Actually Building

Example:

Meaning:

Target_Return5[t] = (Close[t+5] - Close[t]) / Close[t]

Plain-English translation:

That future percent return becomes the answer column for training.

3. What The Model Output Means

If the trained model outputs:

Very important

Future Return is:

So 0 is a meaningful centerline here.

That is different from:

where 0 does not mean the same thing.

4. What Look-ahead Windows Means For Future Return

If Step 1 says:

Fixed-Horizon Windows (bars): 1,5,10,20

then Fusion can create:

This does not mean one model uses all of them.

It means:

So the workflow is:

Step 1: build several possible Future Return answers
Step 2: choose one horizon for this model

5. Does The Flat-Zone Threshold Affect Future Return?

This is one of the biggest user questions, and the answer should be explicit:

No, not directly.

The Step 1 neutral-band / flat-zone setting is for direction-style label creation, especially:

It is not a built-in training dead-zone for Future Return.

If you later use a Future Return model output in user-defined conditions like:

that neutral zone is your workflow rule, not the Step 1 direction neutral-band setting.

6. When Future Return Is A Good Choice

Use Future Return when you want the model to answer:

It is a good first family when:

7. When Future Return Is Not The Best Choice

It is usually not the best first choice when you really mean:

8. Best First Setup For A New User

Export

Step 1

Step 2

Step 3

This gives the cleanest first learning loop.

9. Step-By-Step Future Return Walkthrough

Step A: Prepare the raw dataset

Open Export and make sure:

Step B: Create the prepared dataset

In AI Studio Step 1:

Now Fusion creates future answer columns such as:

Step C: Choose the actual training question

In Step 2:

Example:

Now the model is learning:

not:

Step D: Choose the features

Now choose the inputs the model is allowed to use.

These should be:

Good first rule:

Step E: Train

Choose:

Then start training.

Step F: Read the output

After training, think of the model output like this:

positive value = expected upside
negative value = expected downside
near zero      = weak directional expectation

10. How To Use It In User-Defined Conditions

The simplest first condition pattern is:

positive-side condition when model output > 0
negative-side condition when model output < 0

But that is only a first pass.

A better workflow condition is often:

positive-side condition when model output > tested positive threshold
negative-side condition when model output < tested negative threshold
neutral condition when model output is near zero

That neutral area is a user-defined threshold decision, not a built-in Future Return training label setting.

11. What Not To Do

Do not assume raw Close is the same thing

Future Return is future-looking by design.

Raw Close in Manual Answer Column is usually same-row Close, not future close.

Do not assume 0.0 means the same thing across all answer families

For Future Return, 0 is the centerline. For other families, it may mean something else entirely.

Do not assume the best R2 always means best trading result

Fusion now surfaces more trader-meaningful metrics too, but Future Return models still need to be judged by actual trading usefulness, not just one fit metric.

12. Current Product Notes From The Future Return Manual Pass

These are the changes the UI would still benefit from.

A. Look-ahead Windows needs an inline note

Suggested user-facing note:

“For fixed-horizon answer families, this creates several possible future answer columns in the prepared dataset. Step 2 later chooses one of them for the model.”

B. Direction Neutral Band needs family scope

Suggested user-facing note:

“This affects direction-style label creation such as Direction (3-Class). It does not create a flat zone for Future Return regression models.”

C. Step 2 should reinforce the chosen resolved answer column

For example:

“Current training answer: Target_Return5”

That would make the workflow more concrete for new users.

D. Future Return should be shown in the catalog with a simple centerline icon

Something visually like:

Future Return
signed | centered on 0 | regression

That would help users immediately understand why 0 matters here.

13. One-Sentence Rule To Remember

Future Return means:

“Future Return trains on the percent move from the current close to the close N bars later; positive and negative values are signed target outputs, and zero is the centerline.”

← All target families
NinjaTrader® is a registered trademark of NinjaTrader Group, LLC. Fusion Indicators LLC is an independent software vendor and is not affiliated with, endorsed by, or sponsored by NinjaTrader.