Phase 3 — The Validation Gate: Does the 2nd-Candle Edge Survive Out-of-Sample, Walk-Forward & Monte-Carlo?

Data basis: DAX / FTSE / NQ / Dow M5 + M1, Jan 2018 – Feb 2026, ~2,090 trading days per index. Ruleset: 2nd-candle break + Time-BE-60 exit (from the exit study), fixed, no in-sample optimisation, net of spread (1 pt + slippage). No trading recommendation.

In the algo pipeline this is Phase 3 — the validation gate. Entry, exit and filters were fixed in the exit-logic study (Phase 2). Now only one question counts: does the edge survive on data that did not contribute to the selection? Tested with four independent methods — out-of-sample, walk-forward, Monte-Carlo, deflated Sharpe. Because the ruleset is fixed (no parameter optimised in-sample), these tests are clean.

The strict gate requires four conditions simultaneously: (1) OOS PF ≥ 1.3, (2) ≥ 80% profitable walk-forward windows, (3) recovery ratio ≥ 2 (total-R / 95% shuffle-MaxDD), (4) bootstrap P(profitable) > 95%.

Why Time-BE-60 and not Time-BE-30? In the exit study (Phase 2) the total-R optimum was at ~30 min. Here we deliberately validate the more conservative BE-60: a later break-even intervenes less often, sits closer to the raw entry edge and preserves less of an in-sample optimum. BE-30 yields higher total-R but sits nearer the fitted peak — the stricter test demands the more robust variant, not the most profitable one.

1. Out-of-sample (locked data)

In-sample = 2018–2022, out-of-sample = 2023–2026 (never used for rule selection).

Index	IS PF (18–22)	OOS PF (23–26)	Years profitable
NQ	1.50	1.52	9/9
DAX	1.50	1.27	8/9
DOW	1.51	1.16	7/9
FTSE	1.50	0.84	5/9

In in-sample all four looked identical (PF ~1.50). Out-of-sample separates the wheat: NQ holds (1.52), DAX weakens (1.27), FTSE collapses to 0.84. The in-sample PF alone is worthless.

2. Walk-forward (rolling)

45 rolling 1-year windows (quarter step). Each window is its own mini-OOS test of the fixed ruleset.

Index	Windows	Profitable	Median PF	Min PF	Max PF
NQ	45	100%	1.53	1.00	1.99
DOW	45	91%	1.37	0.81	1.85
DAX	45	76%	1.29	0.68	2.41
FTSE	45	49%	1.00	0.65	1.84

NQ passes in 100% of all 45 windows (even the worst still PF 1.00). DOW 91%, DAX only 76% (< 80% → gate missed), FTSE essentially a coin flip (49%).

3. Walk-forward (anchored)

Cumulative PF in an expanding window — shows whether the edge degrades over time.

NQ and DOW start high (cum. PF 2.1–2.4 in the early years) and converge to 1.50 / 1.37 — the edge weakens but stays stable above 1.3 for NQ. DAX settles at 1.30, FTSE crawls from 0.68 to only 1.01 — no viable edge.

4. Monte-Carlo (shuffle + bootstrap)

Two methods with different meaning:

• Shuffle (5,000 permutations): varies the order → how bad can the drawdown get purely from ordering bad luck? (The total sum is order-invariant — a shuffle alone therefore always gives P=100%.)
• Bootstrap (5,000 draws with replacement): varies the sample → distribution of the total result and true P(profitable).

Index	Shuffle MaxDD q95	Bootstrap total q05	Bootstrap median	P(profitable)
NQ	35 R	+263 R	+381 R	100%
DOW	44 R	+159 R	+275 R	100%
DAX	56 R	+106 R	+237 R	99.9%
FTSE	118 R	−112 R	+13 R	56.5%

The decisive value is the bootstrap P(profitable): NQ/DOW 100%, DAX 99.9% — robust edges. FTSE only 56.5% (coin flip) and a 5% quantile of −112 R — no tradeable edge.

5. Deflated Sharpe (multiple testing)

The exit was selected from 37 variants in-sample — that inflates any best value. The deflated Sharpe (Bailey/López de Prado) corrects for this. Under this strict bar the per-trade Sharpe does not clear the significance threshold — an honest caution: the high in-sample PF is partly a selection effect. The robust evidence is not the backtest PF but the OOS persistence + walk-forward consistency above.

6. Gate verdict

Index	OOS PF≥1.3	WF prof≥80%	Recovery≥2	Bootstrap P>95%	GATE
NQ	✓ 1.52	✓ 100%	✓ 10.8	✓ 100%	✅ PASSES
DAX	✗ 1.27	✗ 76%	✓ 4.2	✓ 99.9%	❌ just fails
DOW	✗ 1.16	✓ 91%	✓ 6.3	✓ 100%	❌ fails (PF)
FTSE	✗ 0.84	✗ 49%	✗ 0.1	✗ 56.5%	❌ clearly fails

Only NQ passes all four criteria. DAX is borderline (fails only the 1.3 bar and 80% WF), DOW fails on OOS PF, FTSE fails across the board.

7. Exit robustness: is the edge in the entry or the exit?

A single validated exit does not answer the decisive question: is the edge a property of the entry (2nd-candle break) — or just luck with this one exit? So the same gate (OOS + walk-forward + Monte-Carlo) runs on three structurally different exits: BE-60 (pure risk management), ONR-edge + BE60 and Asia-edge + BE60 (two level take-profits). This is not a re-selection — BE-60 stays the main system — but a declared stress test with a fixed trial count.

Index	BE-60	ONR-edge + BE60	Asia-edge + BE60	Gate passes
NQ	1.52 ✓	1.44 ✓	1.42 ✓	3/3
DOW	1.16	1.12	1.12	0/3
DAX	1.27	1.18	1.13	0/3
FTSE	0.84	0.89	0.93	0/3

(OOS PF 2023–2026; ✓ = passes the full four-criteria gate)

Finding: NQ passes the complete gate under all three exits (OOS PF 1.42–1.52). This is the real proof — the edge survives a full swap of the exit mechanism and therefore sits in the entry, not in one specific exit rule. DAX, DOW and FTSE fail under every one of the three exits — no exit rescues an entry without an edge. This rules out that NQ's result is a BE-60 artifact.

Conclusion

1. The in-sample PF lies — all four looked identical in-sample (1.50), only NQ survived every OOS test.
2. Four independent tests agree: OOS, walk-forward and bootstrap Monte-Carlo all show the same ranking NQ > DOW ≈ DAX > FTSE.
3. Only NQ passes the strict gate. DAX is borderline tradeable, FTSE is out.
4. The edge is exit-robust. NQ passes the gate under three structurally different exits — the edge sits in the entry, not in one specific exit rule.

Only on this validated core does Phase 4 (position sizing/Kelly) follow → Position sizing & Kelly. The only truly clean final test remains the live forward test, since the OOS block stems from the same dataset (2018–2026).

Limitations: spread modelled (1 pt), slippage estimated — real only in forward testing. OOS from the same overall dataset; true OOS = live from today. Deflated Sharpe assumption-sensitive. Single-source data (Dukascopy). News events not covered.

---

📄 Full study as PDF · Part of the series: 2nd opening candle · Exit logic (Phase 2) · Position sizing (Phase 4)

Disclaimer: Historical statistics are no guarantee of future market behaviour. Not investment advice. Trading involves risk of loss up to total loss.