The Sector Playbook, Audited - rotation, two gates and two hedges on one honest scoreboard

The retail sector playbook has three moves: rotate into whatever led lately, step aside when the screen turns scary, and park the sidelined money somewhere shiny. We ran the whole playbook through one sealed machine: nine sector funds, twenty honest years, four studies that each move exactly one part. This is the scoreboard, and the reason the market keeps winning anyway.

The design rule for this series was that nothing would be clever, because the playbook we were testing is not clever, it is popular. The universe is the nine original Select Sector SPDRs, a fixed list that partitions the S&P 500 and has traded unchanged since 1998, so nothing could be cherry-picked and nothing could vanish. The signal is the academic recipe verbatim: trailing twelve-month return with the most recent month skipped. Selection is the top third of the menu. The gate thresholds were fixed months before these studies existed, the exposure ladder was priced by an earlier series and reused untouched, and every run charges ten basis points a trade, credits every distribution on its ex-date, and answers to SPY rebuilt with its own dividends, the honest form of the index. Each study then moves one thing and only one thing: first how many sectors to hold, then which gate to obey, then where the sidelined money lives, then which crisis asset gets that job. Four sealed walks, one machine, no tuning anywhere. What follows is what the playbook is actually worth.

The first study asked whether picking sectors works at all. Every quarter the machine held the three funds with the best trailing year and refreshed. Twenty years of that compounded 9.4 percent, while simply owning all nine funds paid 10.0 and the index itself paid 10.45. The rotation lost to laziness and lost to the market, exactly as the academic post-mortems of sector rotation said it would. Its character shows best in one pair of years: up 6.3 percent in 2022 while the market lost 18.6, because it happened to be sitting in energy and defensives, then up 3.3 in 2023 while the market roared 26.7, because it was still sitting there. Trailing-year selection arrives late to every party and leaves late too. It is also worth saying plainly why the index is such brutal opposition: SPY spent these two decades compounding a handful of giant technology names, and any strategy built from equal-weight sector funds underweights exactly those names by construction. The do-nothing arm gave up half a point a year to the index before any strategy was even applied. That handicap hangs over everything below, and pretending otherwise would be marketing.

The second study held the rotation fixed and made the gates fight. Both arms step exposure down the same ladder when their gate speaks; one listens to fear, the VIX at fixed bands, the other listens to trend, the index against its own 200-day average. The trend gate won and the result certifies: 8.0 percent against 6.1, ahead in fourteen windows of twenty, with the fear gate better in only one resampled path in eighty. The mechanism is the oldest ghost in this research programme, now priced in a second asset class: volatility stays elevated long after prices turn, so the fear gate sits half-invested through recoveries, giving up ten points in 2009 and again in 2021 while the trend gate was already back in. Every gate costs return in a rising market, that is what drawdown protection is made of. But the trend gate is the first overlay this shop has tested that is actually risk-efficient: it improved the Sharpe ratio over the ungated rotation while cutting the worst drawdown from 48 percent to 32. Fear is expensive. Trend is nearly free. That single sentence is worth the study.

The third study asked where the sidelined money should live. When the trend gate de-risks, the freed capital normally sits in cash earning nothing, by the engine's own stated rule. We taught the tester one new trick, built for this study and shipped to every user: the de-risked slice can buy a named asset instead, as a real position, marked daily and charged like any holding. Gold against cash came out 8.9 percent to 7.8, gold ahead in eleven windows against five, with a four-in-five probability the edge is real. A lean, not a certification, and we say so. The texture favours gold honestly: its wins cluster in the recovery years the gate half-sits-out, when gold itself was rallying. The composed arithmetic is the practical takeaway. The trend gate costs about 1.4 points a year against the ungated rotation; the gold sleeve claws back 1.1 of them. Half a point a year, all-in, for a worst drawdown of 35 percent instead of 48. That is the honest price of sleeping through the next 2008, and it is the cheapest we have ever measured it.

The fourth study gave bitcoin the same job, on the only sample bitcoin has: eleven windows, 2015 through 2025, because the earlier prints belong to the Mt. Gox era of thin, later-discredited venues and we refuse to mark a book against them. The bitcoin sleeve earned slightly more than the gold sleeve, 11.1 percent against 10.3, a gap well inside the noise of eleven windows. Everything else about it failed the job description. A stress sleeve exists to hold value when the book retreats; the bitcoin sleeve fell 30 percent in 2018 and 33 percent in 2022, the two years the gate was busiest, while the gold sleeve lost 4 and 3. Sharpe of 0.56 against gold's 0.71, worst drawdown 48 percent against 35. Its wins are crypto bull years, which is return, not refuge. On this record bitcoin is not the new gold; it is leverage wearing a hedge's name. So why does the whole playbook fail? Because nine big, correlated, heavily-watched baskets carry almost no cross-sectional edge; because trailing-year selection is a lagging answer to a leading question; because every defense buys drawdown with return, and twenty years of this particular market punished anyone holding anything but its biggest names. What survived is modest and real: a trend gate that de-risks nearly free, and a gold sleeve that softens what the gate costs. The return itself was never in the sectors. Where it actually lives, our earlier stock-level series already showed, and that is the next piece of work.

How to read this note

Every study is a two-arm comparative walk on the nine original Select Sector SPDRs, sealed one window at a time: hypotheses registered before each window was scored, twenty one-year windows from 2006 in the first three studies, eleven from 2015 in the fourth. One variable moves per study; every other knob is a literature spec, a shipped default, or a value frozen by an earlier study. Re-selection is quarterly, the platform's shipped cadence; the rotation literature is monthly, and the delta cuts both ways, fewer costs, slower signal, so it is disclosed rather than tuned. All runs are total returns with ex-date distributions, ten basis points per one-way traded dollar, against SPY rebuilt to total return. One engine-verification rehearsal per study is disclosed and counted. During the third study the sleeve's fail-loud gate refused to run three times and exposed a real defect in our own price cache, which was serving a fragmented slice of gold's history as if complete; the cache was repaired, every hedge window was re-verified against the full record, and the walk resumed. We mention this because a study that cannot fail loudly cannot be trusted to succeed quietly. No search record is claimed for any of the four designs, and nothing is corrected for a search that was never recorded.