Market weakness is a risk input, not a veto

Everyone writes the same rule first. If the index is weak, don't buy. It feels like the responsible thing a system should do, and it is the first line of defence anyone reaches for when a position goes against them. Wait for things to stabilise. Come back when the tape is healthy.

We wrote that rule into our own engine. Then we measured it, and it did the opposite of what we expected.

The rule everyone writes first

The version we built was the ordinary one. A market-level gate sitting above every individual decision: if the index reads weak, no new entries anywhere in the book, regardless of how good the individual setup looked. One switch, at the top, overriding everything below it.

It is an appealing design because it maps onto how losing feels. Losses cluster in weak tapes, so a rule that keeps you out of weak tapes seems like it must keep you out of losses. That reasoning has a hidden step in it, and the hidden step is the whole problem: it assumes that where losses cluster, poor returns also cluster. Those are separate claims. They can be measured separately, and they should be.

What we measured

The test was measured cold — run once, over a fixed window chosen before the result was known, with no second pass to find a friendlier configuration. The window was January 2016 through July 2026 — 2,649 S&P sessions, including the 2022 bear market and the recoveries on either side of it. For every session we recorded whether the gate would have blocked entries, then measured what actually happened over the following five sessions.

Two outcomes per session, not one. Forward return, which is what the gate claimed to improve. And maximum adverse excursion — the worst point the position passed through before the window closed — which is what the gate was really being asked to protect you from, whether or not anyone said so out loud.

The result, both numbers

The gate blocked 29.6% of all sessions in the window. Those blocked sessions went on to return +0.48% over the next five sessions, versus +0.24% for the sessions the gate allowed. It vetoed the better half of the tape, and by roughly a factor of two.

A structural version of the rule fared no better. Reading market weakness from trend structure rather than from a single day's move, and confining the test to 2022 — the worst stretch in the window, and the one where a veto should have earned its keep — the days flagged "correcting" returned -0.00% forward-five while the days flagged "healthy" returned -2.05%. Both figures are negative-to-flat, which is what a bear market looks like. The ordering is the point. The gate's own definition of a good day was the worse day.

An index-level veto was anti-predictive for return in every regime we tested. We did not find a version that worked. We stopped looking for one.

Why a real signal can point the wrong way

The reason is not subtle once you say it plainly. Weakness means the price is already lower.

A gate that blocks on weakness is a gate that blocks on discount. The discomfort and the discount are the same event, observed from two different angles: the tape feels worst precisely when other people have finished selling into it. Waiting for stability is waiting for the discount to close, and the measurement says you pay for that wait in forward return.

This is not a claim that down tapes are good, or that falling markets recover on any schedule. Plenty of the blocked sessions in that window sat inside genuinely painful stretches, and some of them kept falling. It is a narrower claim, and a testable one: conditional on the setup already qualifying on its own merits, the market being weak that day did not make the forward return worse. It made it slightly better, on average, across ten and a half years.

What it did predict: drawdown

The same signal that failed as a return filter succeeded, decisively, at the other job. Worst-case excursion over the following five sessions ran -2.63% on weak sessions versus -1.40% on healthy ones — close to double the pain along the way, on the same research replay, with the same caveat that it describes a measured history rather than a forecast.

So the signal was never noise. It carried real information. It was wired to the wrong output. It was answering "how rough will this ride be" while being asked "should I be in the car."

The permission channel and the risk channel

This is the general shape, and it has come up often enough in our own research that we now treat it as the default hypothesis whenever a promising filter fails.

Every input to a trading decision can be routed to one of two channels. The permission channel answers a binary: may this trade happen at all. The risk channel answers a continuous question: given that it happens, how much, with what leverage, behind what stop. A signal that fails in the first channel has not been refuted. It has been mislocated. The test that clears it is separate: does it predict return, or does it predict damage?

Most signals that feel protective predict damage. Damage is easier to forecast than direction, because damage is mostly volatility and volatility is persistent, while direction mostly is not. That asymmetry is why so many discretionary rules feel right and test badly. They are correct about the ride and wrong about the destination.

The concrete consequence in our engine: a weak market halves new position size and withdraws leverage entirely. It does not close the book. Tighten-only, never a boost, and never a switch. Where leveraged vehicles are involved that withdrawal matters more than the sizing does — leveraged ETFs can lose value rapidly, including total loss, and a rough ride is the exact condition under which their daily-reset mechanics hurt most.

The honest caveats

Two, and they both cut against the tidiness of the story.

First: a system with no market veto can still show you an empty board in a real correction, and ours does. That is not the market gate operating in disguise. It is the per-name entry bar — each individual candidate has to clear its own structure test, and in a broad correction few or none of them do. The distinction is worth holding onto, because the two produce similar-looking output for completely different reasons. One is a switch somebody flipped. The other is an arithmetic consequence of nothing qualifying. We wrote separately about why an empty board is a legitimate answer in cash is a position.

Second, on what this measurement is and is not. It was taken on the S&P 500 ETF's own daily series — the index as a single instrument, not a panel of its members — so it says nothing about which individual names did well on those days. That is a real limit. It means the study answers one narrow question well: did a market-level gate, applied to the whole book, help or hurt? It does not tell you how any particular stock behaved inside a weak tape, and we would not stretch it that far.

What to do with the instinct instead

The instinct to protect yourself in a weak tape is correct. It is the implementation that is usually wrong. "I will stop buying" is a permission-channel answer to a risk-channel problem, and it costs you the discount without measurably improving the outcome.

The defensible version is smaller size, no leverage, a stop you set before you entered and can actually live with, and the willingness to be wrong at a survivable scale. That is a worse feeling and a better structure. Most risk management is that trade.

To disclose our own bias: this is how our engine handles it, and the Coil Scanner shows the sizing damp and the leverage withdrawal on the board rather than hiding them in the logic. We publish our refuted experiments, and the market veto is one of them.

FAQ

Should I stop buying when the market is down?

We tested exactly that rule across 2,649 S&P sessions from 2016 to mid-2026, and it was anti-predictive for return: the sessions it blocked returned +0.48% over the next five sessions versus +0.24% for the sessions it allowed. That is a research replay on real historical rules, benchmarked bucket against bucket, not live results, and it does not predict the future. What weakness did predict reliably was drawdown, so treat it as a sizing input rather than a permission switch.

Should I wait for things to stabilize?

Waiting for stability is waiting for the discount to close. In our measurement the discomfort and the discount arrived together, and the calmer sessions did not go on to return more. The defensible version of the instinct is not to wait but to enter smaller, with a stop decided in advance.

Does a system with no market veto just buy through a crash?

No. A per-name entry bar can empty on its own with no market-level switch involved. In a genuine correction few or no individual names clear their own structure test, so the board thins out because of the names, not because a gate was flipped. Nothing here is investment advice.