Regime timing did not beat holding the spread

The test

Eight hundred entry days on QQQ and SPY from April 2022 to September 2026, 30 to 45 days to expiry, longs at the ask and shorts at the bid, fees on every leg. On each day we priced nine structures: short put, put credit spread, call credit spread, iron condor, bear put debit spread, bull call debit spread, long straddle, buy-write and a put backspread. Regime features included trend, momentum, drawdown, realized volatility, VIX level and z-score, VIX term structure, VVIX and VIX minus realized.

Policies were trained only on trades that had settled before each test year, then tested on that year, at 10% of equity at risk per trade.

The result

Always holding the put credit spread grew 8.3% a year with a monthly Sharpe of 2.67, positive in all four test years. The best regime policy that could choose among structures, allocating by the trend and VIX ladder, grew 3.0%. Allocating by trend alone lost 2.4% a year, by VIX minus realized lost 6.0%, and by VIX level lost 18.3%.

An oracle that picked the best structure per regime with hindsight reached 21.1% a year, so regime structure exists. Nothing knowable at the close predicted it. Each model learned to buy bear put spreads when the trend was down from 2022 and lost that in 2023, then learned to buy bull call spreads from 2023 and 2024 and lost that in 2026.

The longer replay agreed

Extended back to 2016, with three bears in the sample (late 2018, early 2020, 2022), the timed spread-or-cash rules grew 3.0% to 4.9% a year against 4.7% for always on. Two all-structure allocators scored higher on average, but with year-to-year swings four to five times larger and losing years of 30% to 50%: their returns came from bull call spreads in bull years. That is leveraged long exposure, not regime skill.

On the structures themselves, over the same days: the put credit spread averaged +4.7% on dollars at risk with an 86% win rate; the bear put debit spread -21.5%; the long straddle -5.7%; the call credit spread -5.3%; and the put backspread, the cheapest always-hedged form, -3.6% a trade, without paying off once in four and a half years.

What it changes in practice

Be skeptical of any system that switches option structures by regime, especially one fitted on a sample containing a single bear market. A model learned from one bear is a model of that bear. If you want protection, a fixed stand-down rule priced as insurance (see the lesson on the 200-day average) is more honest than a classifier, and more independent bets do more for a book than more structures on the same index.