How We Validate a 9-Strategy USDJPY Portfolio Beyond a Single Backtest
How We Validate a 9-Strategy USDJPY Portfolio Beyond a Single Backtest
A profitable backtest is useful, but by itself it tells us very little about whether a trading system deserves to be used.
A strategy can look attractive for many reasons that have little to do with a durable trading edge:
- favorable market conditions,
- parameter selection,
- insufficient trade count,
- exposure to one specific regime,
- unrealistic execution assumptions,
- or simply chance.
For that reason, when developing the EdgeDriven Dollar Yen Portfolio, the objective was not to maximize one backtest curve.
The objective was to build a fixed multi-strategy USDJPY portfolio and examine how it behaved across different historical assessment periods, operating constraints and risk conditions.
The result is a nine-module USDJPY portfolio for MetaTrader 5.
This article explains the validation framework behind it.
1. Why use a portfolio instead of one strategy?
A single strategy usually depends on a relatively narrow set of market conditions.
A portfolio can combine several strategy modules that respond differently to price, momentum and volatility conditions.
The EdgeDriven Dollar Yen Portfolio contains nine fixed H2 strategy modules.
They are not nine completely independent sources of future profit. They all trade USDJPY and therefore share:
- instrument exposure,
- margin,
- execution conditions,
- periods of simultaneous loss.
The purpose of the portfolio is not to pretend that correlation disappears.
The purpose is to avoid relying entirely on one entry logic or one specific type of market behavior.
The portfolio membership and relative construction are fixed. The user does not select which modules are active based on recent performance.
2. Separate historical periods before looking at the final result
One of the easiest ways to overestimate a strategy is to keep adjusting it while repeatedly looking at the same data.
To reduce this problem, the portfolio assessment was divided into several historical periods.
Development
January 2014 – December 2022
This period was used for portfolio construction and candidate selection.
Validation
January 2023 – December 2024
Portfolio-level narrowing and priority setting were performed using Development and Validation information.
Final Historical Holdout Confirmation
January 2025 – June 2026
The selected portfolio structure was already fixed before this period was evaluated at portfolio level.
An important limitation should be stated clearly:
This is not claimed to be universally untouched or never-seen data for every individual strategy. Some component strategies had previous later-period exposure during earlier research or qualification.
For that reason, I prefer the term Final Historical Holdout Confirmation rather than presenting the period as something it is not.
Transparency about what a test can and cannot prove is part of the validation process.
3. Current historical reference results
The current production reference is Version 1.103.
Reference configuration:
- USDJPY
- H2
- USD 10,000 initial capital
- Hedging account
- Leverage 1:100
- BaseRiskPercent = 0.9
- ProductRiskScale = 1.0
- Compounding = false
- RestrictedNewsGuard = true
- MetaQuotes-Demo real-tick test mode
The resulting historical reference statistics are:
| Period | Net Profit USD | MT5 PF | Max Equity DD | Trades |
|---|---|---|---|---|
| 2014–2022 Development | USD 4,193.96 | 1.61 | 4.06% | 2,400 |
| 2023–2024 Validation | USD 663.73 | 1.48 | 1.66% | 555 |
| 2025–June 2026 Final Historical Holdout Confirmation | USD 524.74 | 1.56 | 1.62% | 431 |
| 2014–June 2026 Combined History | USD 5,382.43 | 1.58 | 4.06% | 3,386 |
The continuous history contains both profitable and losing periods.
Across the full history there were:
- 96 positive calendar months
- 54 negative calendar months
This matters because a trading system should not be evaluated only by its final net profit.
The path matters too.
4. Why trade count matters
A very high Profit Factor based on a small number of trades can be misleading.
This portfolio produced 3,386 historical trades in the combined reference test.
That does not guarantee future performance, but it provides substantially more observations than a system based on only a few dozen or a few hundred historical entries.
More trades do not automatically mean better quality.
They simply give us more evidence with which to evaluate:
- losses,
- drawdowns,
- changing market environments,
- execution behavior,
- and periods when the system does not perform well.
5. Risk is part of the system
I do not view position sizing as something separate from strategy validation.
The reference configuration allocates a nominal risk budget across the nine modules.
At the reference setting:
BaseRiskPercent = 0.9
This is divided across nine nominal module entry budgets before broker rounding.
That does not mean the portfolio can lose a maximum of 0.9%.
Realized losses can differ because of:
- gaps,
- slippage,
- simultaneous positions,
- pending-order fills,
- execution conditions,
- minimum lot constraints.
The EA therefore does not increase position size simply to force a trade when the broker minimum volume would exceed the permitted sizing budget.
If a trade cannot be sized within the allowed risk, it can be skipped.
That behavior is intentional.
6. Why the News Guard is part of the historical evidence
Scheduled macroeconomic events can materially change spread, execution quality and short-term price behavior.
The current production EA therefore includes a Restricted USD News Guard.
Historical Strategy Tester runs use the frozen restricted-news windows used during development.
In live operation, the EA uses the MetaTrader 5 Economic Calendar to identify the corresponding future restricted USD events.
During an active restricted window — five minutes before through five minutes after the event — the EA:
- blocks new entries,
- requests cancellation of its own pending orders,
- requests closure of its own open positions.
The protection is not absolute.
It still depends on:
- terminal connectivity,
- incoming quotes,
- trading permissions,
- market sessions,
- broker execution.
A News Guard cannot eliminate gap risk or guarantee a specific exit price.
Its purpose is to reduce scheduled-event exposure under defined conditions.
7. Why production-code consistency matters
A research process is only useful if the production EA reflects the assumptions tested during development.
During final product QA, the News Guard implementation was checked again to ensure that:
- historical testing used the intended frozen news windows,
- live operation could retrieve future events from the MT5 Economic Calendar,
- failure of live calendar synchronization resulted in a fail-closed state for new entries.
After restoring the development-consistent News Guard to the production EA, the historical reference evidence was regenerated using the same production version.
This is an important distinction.
I do not want to show one set of historical assumptions and then operate a materially different EA in live trading.
8. What these results do not prove
Historical testing cannot prove future profitability.
The results above also do not establish:
- future Profit Factor,
- future monthly return,
- a maximum future drawdown,
- broker-independent execution,
- independence between the nine modules,
- protection from extreme gaps or slippage.
Backtests are evidence, not guarantees.
Even a long historical test can still be affected by:
- data quality,
- spread assumptions,
- execution modeling,
- market structure changes,
- unseen future regimes.
The goal is not to eliminate uncertainty.
The goal is to make the remaining uncertainty visible.
9. Why I prefer evidence over a single equity curve
An equity curve is easy to understand visually, but it is only the end result of many assumptions.
When evaluating an automated strategy, I would rather ask:
- How many trades produced the result?
- Was the portfolio structure fixed before the later assessment period?
- How did Validation differ from Development?
- What happened during losing months?
- How was risk calculated?
- What happens around scheduled macroeconomic events?
- Does the production EA implement the same controls used in testing?
- What are the known limitations?
Those questions are less exciting than a large percentage return.
But they are much more useful when deciding whether an EA deserves real capital.
10. Next step: Forward monitoring
Historical evidence is only one stage.
The next step is a dedicated Forward monitoring account in live market conditions.
The purpose of the Forward record will not be to prove that future returns must match the historical test.
It will be to observe:
- real execution,
- live strategy participation,
- drawdown behavior,
- broker conditions,
- News Guard operation,
- and the difference between historical modeling and actual trading.
That record will be disclosed separately from historical simulation results.
Conclusion
The EdgeDriven Dollar Yen Portfolio was not selected because one backtest looked attractive.
The current evidence consists of:
- nine fixed USDJPY strategy modules,
- 3,386 historical trades,
- Development / Validation / Final Historical Confirmation periods,
- combined historical MT5 PF 1.58,
- modeled maximum equity drawdown of 4.06% at the reference configuration,
- a restricted USD News Guard,
- explicit operating and risk limitations.
None of this guarantees future profitability.
It simply provides more information with which to make a rational decision.
That is the principle behind EdgeDriven Algo:
Edge, Backed by Evidence.
The EdgeDriven Dollar Yen Portfolio is available on the MQL5 Market for readers who want to review the product details and reference evidence.
Historical simulations are not live or Forward results. Leveraged trading can cause substantial losses.
