My Note Yields 21.6% Annualized. Why Is Its Long-Run Expectancy Still Below QQQ?

Mid-July, after several down days had fattened option IV, I started building a structure in size: buy QQQ, simultaneously sell a call expiring in 35 days struck about 5% below spot. One real fill: bought at 693.08, sold the 660 call for 46.50 in premium, true net cost 646.58. As long as QQQ stays above 660 in 35 days (any drop smaller than 4.8%), I collect the full 13.42 coupon — 2.08% per period on capital, 21.6% annualized. I only start losing below 646.58, a 6.7% cushion.

I call them "notes": cushion plus fixed coupon, like a bond, except the coupon is equity-sized.

Then I asked my engine for the structure's mathematical expectancy. Answer: about +3% a year. Treasury territory.

I lost it. A 21.6% coupon, a 6.7% cushion, a seventy-something percent win rate — how does that collapse to 3%? If that's all there is, why not just hold QQQ? This post is the record of one long night of interrogating myself into a corner, then filling the gap layer by layer.

0. Conclusions first

1. Layer one: free protection = arbitrage, which doesn't exist

A note ≡ holding the stock + selling one call. So its expectancy differs from plain stock by exactly one term: the expected P&L of the short call itself.

The key: the market prices calls under a near-zero-drift distribution (market makers hedge direction away; they price volatility), while I believe QQQ drifts +10% a year. Re-valuing that call under my own worldview:

That 3.12 is the cushion's price tag. The 6.7% buffer, the ~76% win rate, the 60% volatility cut — all bought with those 3.12 per period. Flip it around and it clicks: if a structure existed with protection and undiminished expectancy, it would be free insurance — arbitrage — and market makers would buy it to extinction instantly. The only protection the market will sell you is protection that's fairly priced in expectation. You pay not in cash, but in upside above the strike.

So "expectancy must be below plain stock" has no mathematical escape hatch. The only remaining questions: lower by how much, and what do you get back?

2. Layer two: reconciling 21.6% with expectancy — the three-bucket decomposition

The full reading of 21.6% is "annualized if every period collects in full." Sort all 6,858 rolling 35-day windows across 27 years into three buckets:

Weighted average = +1.10% per period, +11.5% annualized.

$\underbrace{83.7\%\times 2.08\%}_{\text{winning bucket}} + \underbrace{5.3\%\times 1.15\%}_{\text{partial}} \underbrace{-\ 11.0\%\times 6.32\%}_{\text{landmine bucket}} = +1.10\%/\text{period}$

21.6% is the premium revenue rate; 11.5% is the margin after claims. In any insurance business those two numbers are necessarily one-high-one-low — and the low one is the long-run account.

Wait — how is 11.5% so much higher than the engine's 3%? Because the engine uses the market's pricing distribution (future volatility = the day's 29.6% IV), while the real historical world realizes only about 7.3% per 35 days (~24% annualized). Sell insurance charging for 29.6%, watch the world claim at 24% — the spread is the volatility risk premium (VRP), the profit margin of the insurance trade. It's also the entire economics of my entry rule — "only after a selloff, when IV is fat": underwrite only when premiums are overpriced. Run the same structure on calm-period terms and the coupon roughly halves; the historical replay collapses to ~+2% a year — below T-bills.

3. Layer three: 25 full years against QQQ — two columns

My hard order was: "it must not lose to QQQ, or I'd rather just hold QQQ." So replay my actual playbook: always in the market; coupon pinned at 20%+ annualized, taking the deepest cushion that clears it (fat IV buys a deep cushion, thin IV only a shallow one); after taking assignment, never lock in a loss. Plus a financed column — half my own capital, half borrowed at 4.25% (per-period return ×2, minus interest):

Over the full 25-year cycle the note beats QQQ in both columns, at roughly one-third the annual volatility. The 21.6% I couldn't let go of is genuinely realized in the "margin × full cycle" cell.

Bear years are where it crushes — and why I run it (selected years):

Three rows demand honesty:

1. In 2008 and 2022 the margin column did worse than QQQ — IV was calm before the crash, cushions were thin, and leverage doubled the losses too. The cushion is not crash insurance; its home turf is grinding bears and flat years.
2. Melt-up years like 2023 are a mathematically guaranteed loss: a fully covered note caps each period at 2.08%, so even a perfect record tops out at $(1.0208)^{10.43}\approx$ +24%/yr (about +43% financed). When QQQ prints +56%, no strike, no IV, nobody's selection catches it. Anyone claiming a fully covered strategy never loses to a raging bull is lying to you.
3. So the lever for "never lose to QQQ" is coverage ratio, not terms: don't cover 100% of the position. Whatever fraction you cover is the fraction of melt-up you donate. My actual book is layered — deep-cushion notes locking coupons, shallower covered calls struck above spot keeping upside, and an uncovered long slice to catch the right tail.

4. Layer four: the honest words about financed notes

The margin column looks great, but two sentences belong in plain sight.

First: in the model's world, a financed note is negative carry. The market's fair expectancy for the structure (~3%) can't pay 4.25% interest. The whole trade is positive only because of VRP plus disciplined fat-IV entries. In other words: a cash note asks "how much do I earn?"; a financed note asks "does VRP keep paying?" — interest changes the trade's nature, not just its hurdle. Fortunately the break-even bar is low: interest eats only about a fifth of the full coupon, and ~84% of historical periods collect in full.

Second: financing's true cost isn't on the interest bill — it's the period when leverage meets bucket C. In 2008 the single worst window was −26% in cash terms, beyond −50% financed. That demands your normal leverage leave enough room that the worst window can't kill you. My rules: open new notes only when IV is fat and the coupon clears 20% annualized; roll next notes only with capital released by expiries — never add debt to open a note; a hard cap on total leverage.

5. Aside: do institutions run this?

Researching this produced a delightful fact: the payoff I'm running by hand is an industry with roughly ten trillion dollars outstanding.

• "Discounted floor + fixed coupon" is exactly the core of bank-issued FCNs / reverse convertibles / autocallables: Swiss structured products turned over nearly CHF 200bn in 2024 (reverse convertibles the #1 product), Hong Kong's equity-linked products sold over HK$1.7tn a year, US structured note issuance hit a record +46% in 2024. Europe's discount certificates — cap set below spot — are the one-to-one mathematical equivalent of a deep-ITM buy-write.
• Private-bank clients levering FCNs with lombard credit is standard practice in Asia — the institutional mirror of "financed notes"; its cost is also on record: the margin-call wave of March 2020.
• On the academic side, AQR decomposes covered calls into equity beta + short volatility and explicitly endorses using leverage to restore beta to 1; the same paper warns in black and white that covered calls' downside beta exceeds their upside beta and the left tail is fat — leverage magnifies both — so "maximizing Sharpe with leverage may not be the most prudent approach."

The self-run version skips the issuer's fee layer and bank credit risk, with transparent collateral and the freedom to roll; the price is bearing execution, hedging, and the left tail yourself. The institutions' two lessons map one-to-one: cushions don't stop crash tails, and leverage is for restoring beta — not for amplifying a return engine.

6. Closing: making peace with that "I want to die" moment

Looking back, that night's meltdown came from three misaligned lenses, each worth writing down:

1. Quoted speed ≠ long-run mean. 21.6% is the speed of winning periods; the mean including landmines is necessarily lower. Whenever you see "X% annualized," ask: is that the every-period-wins speed, or the mean after claims?
2. Model expectancy ≠ historical expectancy. The model prices at market IV (~3% expectancy); the historical world was calmer (replay: +11.5–14.5%). The spread is VRP — the actual product of this business, but it belongs only to those who underwrite when premiums are expensive.
3. Single-year verdicts ≠ full-cycle verdicts. Grade any capped structure against QQQ's strongest six and a half years ever and it fails; grade it across 25 full years and it wins, at a third of the volatility.

Last, the psychological ledger — for someone like me who can't sit through drawdowns in naked stock, this is the structure's real value: 35-day periods, most months settling on schedule, and drops under 5% don't touch me at all. It converts "enduring volatility," which I'm bad at, into "waiting for expiry," which I'm good at. The two or three points of expectancy I give up are the price of exactly that.

---

Every number in this post comes from reproducible scripts: QQQ adjusted daily closes 1999→2026 plus VXN, rolling 35-calendar-day replay windows; note terms are real July 2026 fills. The replay assumes no friction and atomic 35-day periods; adjacent windows overlap; QQQ's 27-year record carries survivorship glow. Not investment advice.