ETF buy-write on RSI-2 pullbacks (buy-write pillar)

ID: S032 Slug: buy-write-rsi2-etf-v1 Run date: 2026-07-01 Stage: Stage 1 (modeled screen — forward-pending) Outcome: CLEARED the modeled Stage-1 screen — all 6 cells pass. → advances to forward-paper (Stage 3) with real chains. This is a MODELED result (Black-Scholes fed with flat ATM IV), which overstates the premium of low-delta calls, so it is an optimistic bound. It is not a live or promoted claim — nothing goes live until forward evidence clears the gate. Headline metric: SORTINO +0.10 to +0.20 vs the matched naked long / gate > 0

Positioning — honestly. Low-alpha by design, and risk-reducing, not risk-free. This harvests the index variance-risk premium: a few points less drawdown and a small risk-adjusted edge over simply owning the ETF, paid for with capped upside. It is not low-risk in absolute terms — below the thin premium cushion the ETF's full downside (crashes included) is retained; the call is income, not tail protection. A modelled screen passed; the real premium is being validated forward in paper before any live claim. Skip stages: 2 — no historical option-chain data exists to walk-forward the option leg; a survivor is validated forward in paper, not on a close proxy. Pillar: buy-write (sibling of the wheel). Trial 1–6 of 6 in the v1 matrix — see ledger.

What we tested

The buy-write harvests the equity variance-risk premium on the call side: hold a long position and write a call against it. The premium is not a discount — it lowers cost basis by a few percent while the position keeps almost the full downside below that cushion. The real edge is that implied volatility sits, on average, above realised; the buy-write monetises that gap in exchange for capped upside and no left-tail protection beyond the premium. The payoff is left-skewed by construction (many small capped wins, rare large losses), so it is judged tail-first and on risk-adjusted value, never raw return.

Frozen design (buy-write-rsi2-etf-v1.md, S032): a 2 × 3 matrix — underlying {DIA, QQQ} × short-call delta {40, 30, 20}. Entry is the S027 pullback trigger (RSI-2 < 10 AND close > 200-DMA) so the buy-write is directly comparable to owning the ETF on the dip. We write a ~30-DTE call at the cell's target delta, hold single-cycle to expiry (no roll, no early management), layer entries weekly with a 4-position cap, and charge friction (30 bps on the stock leg, 1% of premium on the write). The option leg is modeled — Black-Scholes fed with the underlying's volatility index (VXD for DIA, VXN for QQQ). The matched baseline is the identical position without the written call — it deploys exactly the same capital, day for day (provably the same as a buy-and-hold that only commits the strategy's capital: the book is on average just ~17% invested within its own sleeve, cash the rest), so buy-write vs baseline isolates purely the effect of writing the call.

"Single-cycle" means the whole position is closed at expiry — both the ETF and the call, whether the call finishes in- or out-of-the-money. It is an episodic ~21-day swing trade, not a perpetual holding, which is exactly what makes it comparable to the matched naked long (same entry, same exit) and why the book sits in cash between signals. The intuitive "income" version — keep the ETF and roll a fresh call each cycle — is a distinct, path-dependent structure (needs a frozen roll rule and an open-ended terminus); it is a deferred variant on the roadmap (a new S-NNN, forward-validated), not this test.

What the entry looks like: the candlestick chart below (What we tested — the recipe) shows the last year of DIA with every RSI-2 signal day marked — short, sharp pullbacks inside an uptrend (price above the dashed 200-day average). Those marked days are the ~16 moments in a year a buy-write is opened; the book sits idle otherwise.

The gate (pre-registered, tail-aware): the buy-write's per-trade Sortino must exceed the matched naked long's (a raw-return gate would auto-fail a covered call in up-markets for the wrong reason, and a high win rate never passes alone), and max drawdown on the concurrent-position book must be ≤ the naked long's and ≤ an absolute 50% ceiling.

What we found

All six cells pass both legs of the gate.

The mechanism is exactly what a covered call should do to a mean-reverting entry: the premium shifts the whole per-trade return distribution up, turning small losses into small gains and shrinking the magnitude of the losses that remain — so downside deviation falls faster than the mean, and Sortino rises. The short call also gains value inside drawdowns, cushioning the book, so max-DD comes in below the naked long rather than above.

Year by year (strategy vs same-capital buy & hold)

Per calendar year: the buy-write book's return, the number of trades opened, and the same-capital buy & hold (the matched naked long — identical capital, no call). The pattern is the honest covered-call signature: the buy-write cushions the down and choppy years (2011 DIA −0.5 vs −1.6; 2020 −4.2 vs −5.0) and lags the strong up-years where the cap surrenders the big moves (2017 DIA +4.5 vs +6.2; 2023 QQQ +7.3 vs +9.3; 2024 QQQ +8.0 vs +8.6). Note QQQ opened zero trades in 2001–02: with the close below its 200-DMA through the dot-com collapse, the close > 200-DMA filter kept the strategy out entirely.

DIA · buy-write Δ30 vs same-capital buy & hold, by year

QQQ · buy-write Δ30 vs same-capital buy & hold, by year

Same-capital buy & hold = the matched naked long (identical day-by-day capital, no call). Buy-write Δ30 shown (highest Sortino lift); modelled, net of costs.

Why this is a bounded, not a finished, result

Three honest limits keep this a Stage-1 screen and not a claim:

The premium is modeled with flat ATM IV, so the low-delta cells are overstated. Real index chains are skewed — out-of-the-money calls trade at lower implied vol than at-the-money — so a real Δ20 or Δ30 write would collect less premium than this model pays it. The near-ATM Δ40 cell is the least distorted, and it passes on both underlyings; that makes the direction (writing the call helps risk-adjusted) robust, but the magnitude, and the ranking that puts Δ30/Δ20 on top, are partly a modelling artifact. Note the modeled buy-write mean return sits above the naked long — a covered call should not beat its underlying on mean return over a sample with strong up-legs, which is itself a tell that the flat-IV premium is generous. Read this table for its shape, not its decimals.

It is sample-poor and directional. One underlying per cell, a couple hundred trades each — enough to reject a bad idea, not enough to walk-forward-mine. That is why a survivor moves early to forward paper rather than into a deeper backtest of a proxy.

No left-tail protection. The buy-write's worst case is still owning the ETF through a crash, minus the small premium cushion. The gate's drawdown leg passed here because the entry (a pullback in an uptrend) and the ~21-day holds avoid the worst holding windows in this sample — not because the structure protects the tail. A collar/protective-put variant (deferred, a new S-NNN) is the tool for that, not this one.

What happens next

Per the pillar roadmap, a cell that clears the modeled screen validates forward in paper via Tastytrade with real chains — real IV, real skew, real fills — where the flat-IV overstatement is replaced by what the market actually pays. No historical option data is needed forward. Only if the forward evidence holds does anything approach a live claim.

This is now live: the two lead cells (DIA Δ30, QQQ Δ30) are being forward-tracked on real chains — see the forward-paper record → (pre-registration: lab/hypotheses/buy-write-rsi2-etf-v1-forward.md). Until then this is what it is: the first idea to clear its Stage-1 bar, on a modeled screen that flatters it, now owed a fair forward test.

Modeled screen. Flat-ATM IV overstates low-delta premium. Not live, not promoted. English per repo convention.