Live paper-trade tracking

ID: S012 Slug: put-selling-capitulation-volregime-filter-v2-stage2 Stage: 2 (walk-forward validation) Run date: 2026-06-01 Outcome: PASSED Stage 2 — all 10 gates clear → ready for Stage 3 (paper trade) discussion Headline metric: OOS CAGR +5.00% / +3% Lineage: Direct continuation of S012 Stage 1 (PASSED 2026-05-31 — per-trade EV +1.19%, win rate 87%, year-concentration 22.2%, h1/h2 time-stationary at +1.20%/+1.18%). Stage 2 gates pre-registered 2026-06-01 — FROZEN before the walk-forward driver was written. Tastytrade-anchored tiered friction replaces Stage 1's %-of-premium proxy.

What we tested

Stage 1 confirmed the per-trade economics of S012's tightened filter were strong in-sample. Stage 2 asks the harder questions: does the in-sample edge generalize to held-out time windows? Does it survive realistic Tastytrade trading costs? Is the filter knife-edge or robust? Does the strategy survive a year (2022) with very few signals?

The walk-forward design splits 2019-2026 into 8 separate annual test folds. Each year is treated as held-out from the prior years. Because S012 is a rules-based strategy (no parameters tuned per fold), the walk-forward is a strict held-out test — we cannot adjust to a fold's regime.

The friction model was upgraded from Stage 1's %-of-premium proxy to the same Tastytrade-anchored tiered model used in S011 Stage 2: $1.30 round-trip commission ($1.14 open + $0.14 close) plus per-leg slippage by underlying liquidity tier ($1.50 for SPY/QQQ-class, $2.50 for mega-caps + other ETFs, $5.00 for top-1000 mid-caps). All-in round-trip cost: $4.30 to $11.30.

The parameter sweep covers filter multipliers across 5 levels (explicit sensitivity test from Stage 1's recommendation), sizing budgets across 3 uniform-notional levels, and 2 stop-loss configurations (30 cells total). S012 uses uniform fixed-fraction-notional sizing — NOT S011's risk-weighted sizing. The mechanisms are orthogonal by design: S011 bounds tail loss per trade, S012 refuses to play in volatile regimes.

The 10 pre-set criteria — frozen before any walk-forward code ran — required (1) baseline OOS CAGR ≥ +3%, (2) ≥ 6 of 8 OOS folds positive, (3) worst fold ≥ −10%, (4) OOS Sharpe ≥ 0.5, (5) OOS max drawdown ≤ 25%, (6) baseline still positive at 2× friction (≥ +1.5%), (7) ≥ 4 of 5 filter levels positive (filter robustness — S012's primary mechanism gate), (8) all 3 sizing levels positive, (9) year-concentration ≤ 35% of positive baseline dollar P/L, (10) no single combo above +18% CAGR (knife-edge detector).

What we found

All 10 gates passed.

Year-by-year OOS performance (baseline params)

(See baseline_folds.parquet for the full table.)

Why this matters / what surprised us

Every single OOS year was positive. 8 of 8 folds passing is the strongest possible robustness result, mirroring S011's perfect score. But where S011 achieved this by capping tail-losses per trade (sizing), S012 achieved it by refusing to play in volatile regimes (filter). These are genuinely different mechanisms with the same end-state — both are real.

2022 was the cleanest demonstration of the mechanism working. The filter aggressively excluded 2022 trades: only 14 trades survived the calm-regime threshold (vs 1,323 in the calmer 2023). All 14 won. The filter did exactly what it was designed to do — sit out the regime where the mechanic breaks. S008's catastrophic year (the −$2.8M 2022 in in-sample Stage 1 without sizing or filter) is essentially neutralized here without ever needing a stop-loss.

The max drawdown is 2.0%, an order of magnitude tighter than S011's 18.2%. This is the most striking number in the run. S011 takes a 19%-drawdown ride to its +9% CAGR; S012 gets +5% with a 2% ride. The Sharpe ratios reflect this: S012 at 2.01 vs S011's 1.42. The trade-off is clean: S012 produces lower headline CAGR but with vastly better risk-adjusted return. Different investor preferences will choose differently.

The filter is monotonic, not knife-edge. The Stage 1 concern that the multiplier might be knife-edge-sensitive (a specific value that just happens to work) is conclusively dispelled. The OOS results across the 5 sweep levels are monotonic in CAGR (looser filter → more trades → higher CAGR, but also higher drawdown). The baseline level happens to be the Sharpe-optimal point ex post. This is fortunate — it would have been concerning if a stricter or looser threshold had dominated by both measures.

Sizing scales close to linearly. Across the 3 sizing levels the CAGR scales close to linearly, drawdown rises but stays modest, and no leverage-blowup risk because positions are uniformly-sized notional commitments, not risk-weighted to a worst-case loss. Even at the most aggressive sizing the worst single OOS year is positive.

Year-concentration came in at 33.9%, just under the 35% gate. 2021 contributes a third of positive dollar P/L. This is the only number where S012 didn't dominate — the gate passes but with little headroom. The interpretation: 2021 was a uniquely calm-vol year, the filter fired more trades than usual (1,969 — twice 2024's count), and the dollar P/L compounded. This is consistent with the filter's design; it's not a sign of the strategy breaking, but it should be monitored in Stage 3 paper trade.

The strategy stays profitable under doubled friction (+3.43%). Same robustness as S011 to the friction parameter. The lower headline CAGR gives less buffer than S011, but the gate threshold was set appropriately lower (+1.5% vs +2.0%).

What this doesn't tell us yet

1. Real options pricing is still simulated. Like S011 Stage 2, the options were priced from a Black-Scholes / realized-vol proxy model rather than actual historical chain prices. Stage 3 paper trade uses live data and will be the first real-world stress test.

2. The 1,969-trade 2021 year is a tail event in trade-count distribution. If 2021's calm-vol regime re-occurred, the strategy could over-commit capital. Stage 3 will reveal whether the live signal frequency tracks expectations or starts deviating.

3. The 2022-style behavior of "almost no trades" needs to be tolerable in paper trade. A subscriber expecting daily alerts could find a 14-trade year disappointing even though it's correct strategy behavior. Stage 3 messaging matters.

4. S012 doesn't ladder with S011 cleanly. The two mechanisms are orthogonal but the strategies overlap on Variant B base signals. A combined version was already tried (S013) and failed in Stage 1 (drawdown was tighter but signal frequency too low to satisfy gates). The two strategies should run as parallel paper books, not as a combined book.

What happens next

S012 promotes to Stage 3 (paper trade). The lab now has two Stage-3 candidates running parallel:

• S011 — risk-weighted sizing, no filter. OOS CAGR +9.05%, DD 18.2%, Sharpe 1.42. Already running paper trade.
• S012 — calm-regime filter, uniform sizing. OOS CAGR +5.00%, DD 2.0%, Sharpe 2.01. Stage 3 setup pending.

Stage 3 design considerations specific to S012:

1. Parallel tracks: same scheduler timing (23:30 NL nightly) as S011, since S012 also fires off daily close.
2. Two sizing tracks mirroring S011's default + aggressive variants.
3. Filter level kept fixed at the Stage 2 baseline — Stage 2 confirms it isn't knife-edge, no sweep-tuning in Stage 3.
4. No stop-loss — Stage 2 confirms it isn't needed.
5. Stage 3 gates pre-registered before paper-runner code is written.