Short strangle, backtested: the high win rate that still loses

A short strangle is the classic "high probability" income trade: sell an out-of-the-money Put and an out-of-the-money Call, collect two premiums, and win as long as the stock stays between the strikes. It wins most months - which is exactly why it seduces people. It is also undefined risk on both sides: the short Call loss has no ceiling, and the short Put runs almost to zero. So the only question that matters is what the rare losing months cost. We backtested it on 14 years of real S&P 500 option chains - seven different strikes, every leg at the real bid/ask mid.

Two findings survive every way we sliced the data. One: the win rate is real and the profit is a mirage. A monthly 16-delta SPY strangle won 78.4% of months from 2010 to 2023 and made 0.76% a year, while simply holding SPY made 10.73%. Two: it does not matter which strike you sell. Across seven deltas the returns are statistically indistinguishable - there is no "best" strangle to optimize toward, only different ways to lose to buy-and-hold while carrying unlimited risk.

The scorecard: dollars per contract (2010–2023)

The same trade at seven strikes, run on the same 167 monthly cycles, every leg filled at the real bid/ask mid and held to expiration. Because a strangle has no natural collateral, the honest way to score it is in dollars per contract, not a percentage:

†How many average winning months a single worst month wipes out. ★16-delta is the common tastytrade default. *20-delta merely printed the highest total on this one sample - see why that means nothing below.

Read down the win-rate and total columns together. Win rate falls steadily from 95.8% (5-delta) to 61.4% (40-delta) - but total profit barely moves, from $1,552 to $2,963. The far-OTM 5-delta wins 95.8% of the time and earns the least; one crash still erases about 110 of its winning months. Selling further out buys a prettier win rate and nothing else - the drawdown just gets shallower because the credit was smaller to begin with.

Does the strike you pick matter? The statistics say no

It is tempting to look at that table, see 20-delta on top ($2,963), and call it the sweet spot. That would be a mistake - and it is worth seeing exactly why, because it is the trap in every strategy backtest.

A single month's strangle P/L swings enormously (a typical 16-delta month is about ±$730; March 2020 alone was -$8,343). Over 167 months those swings add up to a standard error - the "give or take" on the 14-year total - of roughly $6,729 to $12,605 per strike. So each total in the table is really "that number, give or take about $12,605." The entire spread between the best and worst strike is only $1,411 - about 8.9× smaller than the noise on any one of them. You cannot rank numbers whose differences are dwarfed by their own error bars.

The proper test makes it starker. Because all seven strikes trade the same months, they crash and rally together - so to compare two of them you look at their month-by-month difference, where the shared market move cancels out. Do that and 20-delta versus 25-delta is a $352 gap against a $1,448 error (a t-statistic of 0.24); 20-delta versus 16-delta, 0.55. No pair anywhere in the sweep clears a t of 0.6 - you need about 2 to call a difference real. And the "winner" is not even stable: 20-delta tops the full sample, but drop 2020 and 25-delta takes the lead instead.

The tail: where a year of income goes to die

Here are the five worst months for the headline 16-delta strangle. Notice the scale against the $197 average winner:

The March 2020 crash is the whole story in one row: SPY fell from $333 to $229 in a single expiry cycle, blew straight through the 315 Put, and turned a $262 credit into a -$8,343 loss. That one month erased about 42 average winning months. And it could have been far worse - the put stopped losing only because the crash did; the short Call in an equivalent melt-up has no such floor.

Growth of $100,000: undefined risk, T-bill returns

To compare against buy-and-hold you need a capital base. We use the most generous honest one: fully cash-secure the short Put like a cash-secured put, with the short Call riding on the account's margin. On that basis, here is $100,000 in all seven strangles against simply owning SPY:

Gold = buy & hold SPY. Red = the seven strangles (5–40Δ). They cluster so tightly you can barely tell them apart - and all crawl along the bottom while SPY compounds.

Buy-and-hold turned $100,000 into $413,000. Every strangle, at every strike, finished between $103,311 and $112,572 - 0.23% to 0.85% a year against SPY's 10.73%, at drawdowns of 23.2% to 40.7%, while carrying unlimited risk the whole time. You took the worst risk profile in options - a capped, tiny gain against an uncapped, enormous loss - to underperform the thing you were selling options on.

And this base flatters the strangle. Cash-securing the Put reserves nothing for the naked Call: in a sharp rally a real margin account would face a margin call and be forced to buy the Call back at a loss, or post more capital, exactly when it hurts most. The smooth-looking line hides a financing risk the chart cannot show.

The tiny edge rests on two crisis years

Whatever small profit the strangle made was not spread evenly - it was concentrated in the premium-rich aftermath of crises. Watch what happens to each strike's total when you remove one crisis year at a time:

Excluding 2020 raises the totals (removing the crash removes the big loss) - but excluding 2022 instead collapses every strike toward zero, and pushes the far-OTM 5-delta negative (-$62). The positive 14-year result depended on both 2020's rebound premium and 2022's realized-bear premium. That is the real distinction worth keeping: a short strangle survives a slow grind (in 2022 SPY fell about 19% yet the strangle made +$1,842) and dies in a fast gap (the 2020 crash, -$7,100). It feels safe for years - right up until the one fast crash that defines its whole record.

What this means for how you trade

• A high win rate is bait, not a moat. The 5-delta won 95.8% of months and earned the least of any strike. Judge an income trade by its worst month and its total, never its win rate.
• Stop optimizing the delta. The seven strikes are statistically indistinguishable on return - the whole spread is 8.9× smaller than the noise. A backtest that hands you an "optimal" strike is fitting randomness; it will not repeat.
• The undefined risk is the whole point, and it never goes away. The short Call is unlimited; the short Put runs to near-zero. If you can't define and survive your worst case, you can't size the trade. Prefer a defined-risk iron condor if you want the same range bet with both tails capped.
• If you still sell strangles, do it only on broad, liquid indices, manage early (roll or close near 21 DTE, take profits around 50%), and size for a multi-times-credit loss. Model the exact trade in the short strangle calculator first.

Caveats - read these

• Held to expiration, no management. This is a deliberate worst-case baseline. Managing at 21 DTE or 50% profit would trim the tail (and the win rate) - real traders should not hold a losing strangle into a crash. The direction is clear; the exact numbers would improve.
• The percentage returns assume the Put is fully cash-secured. That is not how strangle margin actually works - a real account posts far less, which scales both the return and the risk up sharply. The dollar-per-contract figures are base-free and are the honest core; read the percentages as an idealized, generous framing.
• Mid fills flatter the wider strikes. We fill at the bid/ask mid; at realistic fills you cross a wider spread on the fatter near-money credits, so part of the "flat across strikes" shape is a fill-model artifact, not the market. It pushes toward, not away from, the no-best-strike conclusion.
• One 40-delta cycle is missing (166 vs 167). An out-of-the-money-only strike filter dropped a single 40-delta month that was actually a small winner, so the 40-delta total is very slightly understated - immaterial to any conclusion here, but disclosed.
• Settled at intrinsic; no commissions or early-assignment. European-style settlement and the odd early assignment on a deep-ITM leg would make the real result modestly worse, never better.
• One window, one underlying. SPY, 2010-2023, monthly ~30-DTE cycles at fixed deltas. A different index, a single stock (which can gap far harder), or a different era would shift the numbers. Past performance is not predictive. Educational, not advice.

Method: real OptionsDX end-of-day SPY option chains, 2010-01-15 to 2023-12-15 (167 monthly third-Friday cycles). Each cycle sells a short Put and short Call at the 5-, 10-, 16-, 20-, 25-, 30- and 40-delta strikes chosen by each option's real delta, priced at the real bid/ask mid, held to expiration, no early management, and settled at intrinsic against the real SPY close on the expiration trading day. Percentage returns compound each cycle's dollar P/L over the short-Put strike (cash-secured base). Buy-and-hold is SPY price return over the same window. The "no measurable best strike" verdict uses the paired-difference standard error (the month-by-month difference between two strikes × the square root of the sample), the correct test for settings run over the same periods. Every figure regenerates from the underlying chains and is independently re-derived from the raw files (7/7 delta aggregates plus raw-cycle re-derivations verified); none are hand-entered.