ZenWave develops automated trading strategies and publishes the research and subsequent live results. Walk-forward analysis (WFA) evaluates parameters on later out-of-sample (OOS) periods; each report explains what that evidence supports and where it is limited.
Walk-Forward Analysis divides historical data into sequential in-sample (IS) and out-of-sample (OOS) periods:
The published studies use rolling windows: A1–A4, B1–B5 and C-long C1–C4. Some OOS periods overlap. They provide historical evidence rather than independent trials or proof that a final full-window configuration was always out of sample.
Initial research, validation, and deployment were conducted on USDJPY due to its liquidity profile, execution consistency, and identifiable macro regime behavior.
Additional markets may be evaluated under the same protocol.
Strategy parameters remain frozen during forward tracking. Operational fixes and protective actions are documented; changes to strategy logic require a separate research version.
Position sizing is derived mechanically from a predefined risk percentage and stop-loss distance.
Reported risk is calculated from stop-loss exposure relative to account equity. Actual exposure may differ due to minimum lot size granularity (0.01 lots), broker price rounding, and execution effects such as slippage. Parameters are frozen during live deployment.
No martingale. No grid logic. No hedge stacking. No loss-based sizing.
Stop-loss and take-profit required on every position (server-side).
Server-side execution only.
Each engine’s signals and positions are identified separately for monitoring.
The platform operates multiple engines with defined time horizons:
A and B also run together in ZenWave Book, the IC Global account targeting 0.25% risk per trade for each engine. It remains unchanged. A separate IC Trading Book is planned; its A/B/C risk weights are the next allocation study.
AI-assisted tools help generate candidates, implement tests and review results. The operator makes research and deployment decisions. Tool-assisted review does not independently establish the absence of leakage or overfitting.
Generates candidates and assists with feature exploration and implementation.
Helps inspect test boundaries, compare outcomes and flag possible implementation or research errors.
Engine selection, parameter acceptance, and risk allocation are approved by the operator.
Reports distinguish historical tests, research configurations and observed live results. The aim is to make the evidence and its limits inspectable:
Out-of-sample validation is distinguished from full-window configuration reruns. In-sample and rerun results provide research context, not an unseen live track record.
Slice periods, outcome metrics and equity curves are published in reports and supporting media. Sanitized exports do not expose every strategy parameter or establish the complete research search history.
Public cTrader account records provide balance, equity and trade-history context for live tracking. Public visibility is not an independent audit of the strategy or research process.
Review tools assist the operator. Their output is checked against source reports and does not certify that a strategy is free of errors or selection bias.
Production strategy changes are researched and versioned separately. Forward tracking records the behaviour of each frozen configuration and any documented operational intervention.