B

Engine B — Validation

Historical researchB1–B5B5 in production

Published historical tests for the medium-horizon USDJPY strategy: results across five slices, the weaker 2023 period, and the limits of selecting B5 for production.

Past performance, including backtested and simulated results, does not guarantee future returns. Trading leveraged instruments involves risk of loss.

Historical evidence · live validation ongoing

Profitable OOS slices

5/5

Overlapping periods; not independent trials

OOS profit factor range

1.16–2.19

Across the published slice reports

Largest OOS equity DD

39.50%

At the sizing used in these backtests

All five published out-of-sample (OOS) runs closed profitably. Profit factors vary substantially across slices, with the weakest result in B3. These results support continued evaluation while showing that performance depends on the period tested.

B5 is the production choice following review of IS/OOS behavior. Selection informed by OOS results reduces their independence: B5’s historical result should not be treated as an untouched test of the selection decision. Subsequent observed live results provide additional evidence.

B

Walk-Forward Timeline

Each slice pairs an in-sample (IS) development window with a later OOS test window. Research begins in July 2018; the published OOS periods begin in 2021.

SliceIS PeriodOOS PeriodOOS Result
B1Jul 2018–20202021Profitable
B22019–20212022Profitable
B32020–20222023Profitable · lowest PF
B42021–20232024Profitable
B52022–20232024–2025Profitable · selected

B4 and B5 both include 2024. The five profitable slices are overlapping historical results, not five independent trials or a single non-overlapping OOS track record.

B3: profitable, with a much narrower margin

Profit factor falls from 2.47 IS to 1.16 OOS in 2023. OOS equity drawdown reaches 28.44%. This weaker period matters when interpreting the stronger slices.

B5’s OOS profit factor is 1.74, with 26.63% balance drawdown and 39.50% equity drawdown. Its selection does not establish that it is the least fragile configuration or that future regimes will resemble this test.

Engine B Walk-Forward Slices

In-Sample01/07/2018 - 01/01/2021
Out-of-Sample01/01/2021 - 01/01/2022
Sample (IS/OOS)Win RateProfit FactorBalance Sharpe*Balance DDEquity DDMean trade returnTrades
In-Sample16.4%1.371.4930.3%33.8%0.31%584
Out-of-Sample21.3%1.602.5518.3%27.0%0.58%277

*Balance Sharpe is an annualized return-to-volatility estimate based on closing balances, carrying quiet weekdays forward and assuming a zero risk-free rate. It excludes daily unrealized P/L. Mean trade return is net P/L divided by each trade’s pre-trade balance, averaged across recorded trades; it is not a risk-normalized R multiple.

Structural Diagnostics (IS / OOS)

Behavioral Structure

Runs Zi

IS

-3.99

OOS

-2.67

Observations Below Highi

IS

93.7%

OOS

89.6%

Drawdown Stress

Ulcer Indexi

IS

10.65

OOS

7.10

Across-slice statistics and their limits
Out-of-sample slice variation

Descriptive variation across the published slices, including overlapping periods. These are not independent samples or confidence intervals; low variation alone does not establish robustness.

Performance metrics

MetricMeanStDevCVMin / Max
Profit Factor (OOS)1.780.390.2191.16 / 2.19
Balance Sharpe (OOS)2.320.870.3750.75 / 3.20
Balance DD (OOS)20.24%5.59%0.27613.64% / 27.05%

Structural metrics

MetricMeanStDevCVMin / Max
Runs Z (OOS)-2.321.370.592-4.02 / 0.16
Ulcer Index (OOS)8.002.940.3685.01 / 13.45
Observations Below High (OOS)90.52%2.96%0.03387.54% / 96.15%
B

Reading the evidence

The published exports document simulated outcomes under their recorded settings. They do not independently establish the absence of overfitting or selection bias, or verify parameter-neighborhood stability. A positive OOS result is useful evidence, not a guarantee of future profitability.

Drawdowns describe the risk experienced at the sizing used in each test. Balance drawdown excludes open-position losses; equity drawdown includes them. Neither is a limit on future loss.

Watch the original WFA review →
B

Monte Carlo · Historical Resampling

Source: B5 live-condition backtest · 4y (01/01/2022 - 01/01/2026) · 1,088 balance-return samples · 10,000 simulated paths.

The input includes both B5’s IS and OOS periods. “Live-condition” describes the backtest settings, not observed live trades. This simulation is conditional on that historical sample and is not a forecast of live results.

Each simulated path independently resamples historical balance returns with replacement. This removes their original time ordering and does not preserve regime or loss clustering, model new execution costs, or establish an edge independently of the source backtest. Drawdowns use closing balances and exclude intraday equity losses.

The blue FPT line is a continuous live-condition backtest from account inception, normalized to the simulation’s starting capital. It shows a simulated closing-balance path, not observed live returns or a comparison at matched deployment risk.

Equity Path Simulation

10,000 paths · 1,088 trades resampled
Live-condition FPT Source backtest Median 25th–75th 5th–95th

Median Max Drawdown

26.1%

95th pctl: 38.0%

Original Max DD

28.5%

Backtest (single path)

Median Final Multiple

×606

5th–95th: ×85.2 – ×4,788

Original Final Multiple

×614

Backtest (single path)

Resampled Paths Above Each Drawdown Threshold

>20%

90.3%

>30%

26.5%

>40%

3.5%

>50%

0.2%

1,088 trades · 10,000 paths · bootstrap with replacement · seeded PRNG for reproducibility. These frequencies describe the resampled trades, not calibrated probabilities of future drawdowns.