Grid
Evaluate the configured combinations of parameter values. Useful when the search space is small enough to examine systematically.
THE RESEARCH WORKFLOW
Search broadly.
Validate chronologically.
A good experiment separates the decisions made using earlier data from the evaluation on later data. Studio keeps that sequence visible, configurable, and reproducible.
01A CLEAR SEPARATION
Give every part of the
history a purpose.
01 / TRAINING
Make the selection.
Define the strategy, parameters, costs, and search method. Use the training window to select the configuration.
02 / VALIDATION
Challenge the selection.
Examine the chosen setup on a later period. Keep this evaluation separate from the training result.
03 / OUT OF SAMPLE
Evaluate later history.
Reserve another chronological period for assessment. Changes made after inspecting it should be treated as new research decisions.
02PARAMETER OPTIMIZATION
Choose how to search.
Keep the same standard.
The search method changes how candidate parameters are explored. It does not remove the need for independent evaluation.
Random
Sample candidate configurations within defined ranges. Explore a larger search space within a bounded experiment budget.
Genetic
Use population-based selection and variation to search the parameter space. Treat the selected configuration as a candidate that still needs validation.
03WALK-FORWARD EVALUATION
Move through time.
Repeat the discipline.
Plan successive training and evaluation windows before running. Examine whether the selection process remains useful as the available history advances.
Read the research setup guideA walk-forward study repeats a chronological process. It should be interpreted as a series of linked experiments, with the window definitions and assumptions preserved.
FOLD 01
FOLD 02
FOLD 03
TrainingValidationOut of sample
The diagram explains time order. It does not represent a completed study or a performance claim.
04REPRODUCIBLE BY DESIGN
The evidence belongs
with the experiment.
Results are easier to review when you can recover the decisions and inputs that produced them.
- The strategy source and parameter values used for the run
- Dataset fingerprints, coverage, and clock conventions
- Capital, quantity, commission, and slippage assumptions
- Historical news and event availability settings
- Saved results, equity paths, and execution records
- Chronological window definitions and research configuration
INTERPRETATION MATTERS
An assumption should
never be invisible.
A backtest is an experiment on selected history under specified rules. Source coverage, quote construction, corporate actions, costs, and information availability can all change its meaning.
Historical results do not establish future returns. The purpose of this workspace is to make the experiment easier to examine and repeat.
Review data and timing assumptionsQUANTDEVELOPER STUDIO
Your next idea deserves
a better experiment.
Get to know the workflow before you begin.