We tested betting on the decline. It did not work.

The instinct

When a market falls, the obvious move suggests itself immediately. If prices are going down, buy the thing that goes up when prices go down. Inverse ETFs exist for exactly this instinct, and the leveraged ones — SQQQ, SOXS, SPXU — promise to pay three times as much attention to it.

The instinct is not stupid. Declines are real, and watching an account bleed while doing nothing is unpleasant. The question is whether a rules-based system can identify a decline early enough, hold the right instrument long enough, and get out cleanly enough to end up ahead of the boring alternative.

We tried to answer that with a measurement rather than an opinion. This page is the result.

What we tested, and how

The design was frozen before any data was pulled. That matters more than the result. A study you can rewrite after seeing the numbers is not a study, it is a story.

The structure: a market-weakness gate computed on the underlying index, and a decision about what to hold while that gate is on. Three universes — semiconductors, the Nasdaq-100 and the S&P 500. Three gates of increasing severity, from "price below its 50-day average" up to "price below its 200-day average". Two position sizes in the inverse product, a volatility-equivalent third of notional and full notional. And two entry rules: enter the moment the gate turns on, or wait for a bounce and short into strength.

Three universes × three gates × two sizes × two timing rules is 36 configurations. Every one is reported below, winners and losers alike.

Three things about the method are worth stating plainly, because they are the difference between a fair test and a flattering one:

• The traded instrument was the real inverse ETF at its real historical price, never a synthetic short constructed from the index. Real prices carry the daily-reset decay and the reverse splits. A synthetic short would have made the sleeve look far better than it was.
• Costs were charged at 0.15% round trip on every inverse leg, and on the daily rebalancing turnover as well. The benchmarks were allowed to trade free. We deliberately handicapped the idea we wanted to work.
• The bar was not zero. Beating zero is trivially achievable by staying in cash. The bar was to beat both cash and a defensive macro basket — equal-weight bonds, gold and a dividend fund — over the identical set of days.

That last point is the one most hedging arguments skip. In a weak tape a long-only system is not idle. It is in cash, and it can rotate defensively. Those are the real competitors, and they are harder to beat than nothing.

The result

Zero of the 36 configurations passed. Not one beat the macro basket over the weak-tape days pooled. Not one beat cash. The best cell in the entire grid won in 40% of individual weak-tape episodes, well short of the 70% consistency the design required in advance.

The semiconductor book under the fastest gate shows the shape of it. Over the study window that ran from the start of 2016, that gate was on for 761 days across 90 separate episodes. Pooled over those identical days, cash returned 0% by construction, the defensive macro basket returned +26.4%, the inverse ETF at a third of notional returned −79.6%, and the inverse ETF at full notional returned −99.9%. On the full-period curve the same ordering held: 24.2% annualised for the macro version, 21.4% for the cash version, 4.4% for the third-notional inverse, and −34.5% for the full-notional inverse, whose worst drawdown reached −99.6%.

The most instructive test was the one we called the decay trap, and it was defined before we looked. Take the sleeve's gain during the worst year in the window, then compound it through the choppy years that followed. Did the bear-market winnings survive? For the third-notional inverse, the answer was 0.67 — a third of the capital gone across the cycle. The macro basket came out at 1.07 over the same construction. The inverse sleeve did make money in the bad year. It handed all of it back and more in the years that were merely uneven.

Why it fails: the mechanics

Half the answer is arithmetic and it is not controversial. A leveraged inverse ETF targets a multiple of one day's move and resets each night so it can make the same promise tomorrow. Two equal-and-opposite moves in the underlying do not cancel out for the holder; they leave you poorer. Over a choppy stretch that ends flat, the fund bleeds. We work through the compounding in full in leveraged-ETF decay, explained honestly, and the same math runs in both directions — the inverse cousin decays exactly like the long one.

This is why the full-notional cells were catastrophic while the third-notional cells were merely bad. Decay scales with exposure. It does not go away at smaller size, it just takes longer to show.

Why it fails: the timing

The other half is worse, because you cannot size your way out of it. Mechanical triggers that identify a decline fire after the decline.

We measured this directly on semiconductors, cataloguing every sustained downward leg and then asking four plain triggers — three consecutive down closes totalling −4% or worse, a break of the 20-day average on a down day, a −5% five-session slide, and a gap-down open — how late each one arrived. The fastest trigger fired a median of three sessions in, by which point the median leg had already given up 7.4% from its high. The median move still available after it fired was positive: +0.7%. It was not signalling a decline. It was signalling a low.

The false-positive rates finish the argument. Of the times those triggers fired, 77%, 92% and 75% respectively landed outside any real downward leg at all. A tool that is mostly wrong and, when right, arrives near the bottom, is not a hedge. It is a machine for selling weakness.

We then built the trading logic anyway, out of stubbornness: 48 combinations of trigger, exit rule and entry timing, costed. None had positive expectancy. Separately we went hunting for something that could see a top coming rather than confirm one, scoring 126 candidate detectors against a catalogued set of real market tops. The best of them flagged about half the tops while arming itself roughly 34 times a year in periods where nothing was happening. Nothing there was worth wiring to money.

What the same research kept

The study was not a waste, because the benchmark it was measured against won. Two things survived and both are now load-bearing.

Cash. Over the weak-tape days, cash was ahead of every one of the 36 inverse configurations. Doing nothing outperformed every attempt to be clever, which is a result even if it is an ungratifying one. We treat that seriously rather than sheepishly — see cash is a position for why an allocation to nothing is still a decision, and why it has a real cost in a rising market.

The defensive macro book. The plain bonds-gold-dividends basket beat both cash and every inverse cell in the semiconductor test above, and it did so without ever needing to be right about direction. It does not require a top call. It does not decay. It just holds different things.

There is also a durable finding buried in the failure: the anatomy of a decline. Legs have a typical length, the bounce lags by a couple of sessions, and the value of noticing early falls off sharply. We kept that ruler. We use it to measure how fast the system reduces risk, not to decide when to bet against the market.

What we do instead

Stated plainly: the engine is long-only. There is no short sleeve and no inverse-ETF sleeve. When market weakness shows up, the response is to stop buying, hold cash, and allow a rotation into the defensive book. Weakness affects position size and leverage. It is never treated as permission to bet on the downside, because the measurement above says that bet loses to sitting still.

If that sounds unambitious, it is. The alternative on offer was a sleeve that lost to cash in every single configuration we pre-registered.

Disclosure of bias: we sell the software this research feeds. Coil is long-only partly because this study came back empty, and the absence of a short sleeve is a design decision we would rather explain than hide.

The caveat we owe you

This is our test of our configurations, over one window, with our cost assumptions and our benchmarks. It is not a proof that no hedge can ever work. Options carry a defined, capped premium rather than open-ended decay, and behave differently from anything measured here. Professional risk desks hedge constantly and are not being foolish.

What we can say is narrower and, we think, more useful: the specific retail-accessible version of this instinct — buy a leveraged inverse ETF when a mechanical weakness signal fires, hold it while the signal persists — failed every pre-registered bar we set for it, in all 36 forms, against benchmarks we handicapped in its favour. If someone sells you that strategy, ask them what they measured it against. Zero is the wrong answer. Cash is the right one.

FAQ

Should I buy SQQQ or SOXS to hedge a falling market?

We cannot tell you what to do with your money, and nothing here is investment advice. What we can tell you is what our own measurement found: across 36 pre-registered configurations, measured from 2016 onward on real inverse-ETF prices with round-trip costs charged, none beat cash over the same defensive windows and none beat a plain defensive macro basket. That is a research replay of historical rules, not live results, and it does not predict the future.

Why do inverse ETFs lose money even when the market falls?

Two effects stack. The instrument resets its leverage daily, so a round trip through volatility costs you even if the index ends where it started. And the signal that tells you to buy it arrives late — in our measurement the fastest trigger fired only after a median 7.4% had already been lost from the prior high, with a median remaining move of +0.7% after it fired.

What does Coil do in a falling market instead?

It stops buying, holds cash, and can rotate into a defensive macro book. Cash is treated as an allocation with its own real cost, not as a refuge. There is no short sleeve, because the research did not support one.