Correction, bear market, crash: the thresholds are arbitrary — and that matters

The thresholds, as actually defined

Start with what you came for. Two of the three common labels have published definitions from bodies that regulate or educate US investors.

The SEC's investor education site defines a bear market as “a time when stock prices are declining and market sentiment is pessimistic”, adding that “generally, a bear market occurs when a broad market index falls by 20% or more over at least a two-month period” (Investor.gov glossary). Note the second clause: the official definition carries a duration requirement most headlines drop entirely.

FINRA, in its glossary of stressed-market vocabulary, defines a correction as “when stocks, bonds, commodities or indices reverse course by at least 10 percent before resuming their previous upward or downward trend” (FINRA, Key Terms for Tough Times). Note what is not there: an upper bound. The familiar “10–20%” band that caps a correction where a bear market begins is a convention layered on top of the definition, not part of it.

And the pullback? We went looking for an authoritative 5–10% definition and did not find one. Neither glossary defines “pullback” with a number. It is commentary shorthand — which tells you something about how much structural weight this ladder of terms was built to carry.

Why the thresholds exist at all

A headline needs a moment. “The index is down a bit more than it was” is not a story; “the index has entered a bear market” is. The threshold converts a continuous quantity into a binary event, and binary events are what news is made of. That is not a criticism — shared vocabulary is useful. The problem is not that the labels exist. It is what people do with them once they do.

The 19.8% / 20.2% problem

Picture an index 19.8% below its high. Now picture it 20.2% below. Everything observable is essentially identical between those two states: the same companies earning the same money, the same holders at the same prices, the same volatility, breadth and leadership. One of them gets a name. The other doesn't.

Consider what happens to a rule keyed to that name. Suppose the rule is “go to cash in a bear market.” The index touches −20.1%, you sell. It recovers to −19.4%, the label lifts, you buy. It slips to −20.6%, you sell again. That is a machine for converting an arbitrary line into transaction costs, and the more consequential the action attached, the more expensive the oscillation.

The deeper issue is that the threshold is measured against a prior high — one historical price, possibly set by a euphoric afternoon with no bearing on anything since. Two markets that look identical today can sit 12% and 25% below their respective highs, and draw different labels, for reasons located entirely in the past.

What actually changes at a threshold: nothing mechanical

Worth being blunt about. When an index crosses 20% down, no mechanism engages. No exchange rule fires. No fund is forced to sell. No margin requirement changes. A number crosses a line in a data feed, and a certain kind of headline becomes available for writing.

The one real effect is behavioural, and it runs the wrong way. The label arrives after the decline it names. By construction, you learn you are in a bear market only once the index is already 20% below its high — once a good deal of the damage is done. A signal that can only confirm what already happened is a poor basis for an action that has to happen in the future.

We tested a version of this on ourselves, and it failed

We are not neutral here, so we would rather show the measurement than assert the principle.

Coil's scanner briefly shipped an index-level veto: if the broad index was weak, the engine took no new positions in anything. An intuitive rule, and a close cousin of “stop buying in a correction.” We measured it cold on SPY daily data from January 2016 to July 2026 — 2,649 sessions, including the 2022 bear market — and it was worse than useless.

The days that rule blocked returned +0.48% over the following five sessions, against +0.24% for the days it allowed. It was vetoing the better half of the tape. Swapping the single-day trigger for a structural definition of weakness did not rescue it: inside 2022 — a genuine bear market — structurally weak days ran −0.00% forward-five while structurally healthy days ran −2.05%. Anti-predictive for return in every regime we tested.

What weakness did predict, reliably, was pain along the way: worst intra-window drawdown of −2.63% in a weak tape versus −1.40% in a healthy one. Same measurement, opposite conclusion. Market weakness carries real information about risk and no usable information about return.

The conclusion we drew, and shipped: weakness belongs in the risk channel — how much you commit, how much leverage you allow — and never in the permission channel. It sizes you down; it does not tell you to stop. A down day inside an intact uptrend is often the dip a patient buyer exists to take, the case we make in buy pullbacks, not breakouts.

What to measure instead

If the label is the wrong input, what is the right one? Two questions that are continuous rather than binary, and both answerable on any given day without knowing what to call the market.

• Is it above its own long average? Not “is the market down,” but is this index or stock trading above its own 50-day and 200-day moving averages. A structural question about the shape of the trend, not a headline question about a percentage from a peak.
• How far below its own high is it? Distance from a trailing high as a live number on a sliding scale, not a bucket. Twelve percent below differs meaningfully from twenty-two, and a scale preserves that difference where a label destroys it.

Both readings degrade gracefully. As the market weakens they move continuously, so anything downstream moves continuously too: position sizes taper rather than snapping to zero, and there is no line to oscillate across because there is no line. If some of that vocabulary is new, our trading glossary defines the terms plainly.

There is a third answer that gets undersold: sometimes the right size is none. Holding cash is a position with its own costs, taken deliberately — not a threshold tripping. That argument is laid out in full in cash is a position.

What a rules-based system does when it cannot know which one this is

Here is the honest constraint at the centre of this. In the middle of a decline, nobody knows whether it is a 9% wobble, a 14% correction or the first third of something much worse. That is not a failure of analysis. The information required to tell them apart does not exist yet.

A system that cannot know has two coherent options: guess, and take a large action on the guess; or build a response that is acceptable across all three outcomes. The second is not glamorous, but it is what survives. It means keeping the entry standard fixed regardless of what the tape is called, so weakness never buys a lower bar. Letting observed weakness reduce size and remove leverage, continuously, so a decline that keeps going finds you carrying less. Keeping exits keyed to each position's own structure rather than to an index label. And accepting that you will neither dodge the whole decline nor sit out the recovery, because doing the first reliably would require prediction.

None of that requires knowing whether this is a correction or a bear market. That is the point. A process whose behaviour does not depend on the label cannot be whipsawed by the label.

How our engine handles it

Full disclosure of our bias: we sell software built on this conclusion. Coil's scanner reads market structure — an index against its own 50- and 200-day averages, and its distance below its own trailing 120-day high — and feeds that into size and leverage only. A weak market halves new-entry size and switches leverage off. It never closes the door on an entry that qualifies on its own merits, because when we measured the version that did, it cost us the better half of the tape.

The takeaway

Learn the definitions. They are real and published, and shared vocabulary is worth having. A correction is at least 10%. A bear market is 20% or more over at least two months. A crash has no definition at all.

Then decline to let any of them tell you what to do. The tape does not know what it is called. A process that does not either has one fewer way to be wrong.

FAQ

What is the difference between a correction and a bear market?

FINRA describes a correction as a reversal of at least 10 percent before the previous trend resumes. The SEC's Investor.gov glossary describes a bear market as a fall of 20% or more in a broad market index over at least a two-month period. Ordinary usage does track real published definitions. What those definitions do not carry is any instruction — they describe how far an index has fallen from a prior high, and say nothing about what happens next.

Is there an official definition of a pullback?

Not one we could find. The 5–10% “pullback” band is common in commentary, but neither FINRA's stressed-markets glossary nor the Investor.gov glossary defines the term with a number. It is shorthand rather than a standard — a useful clue about how much weight the ladder of labels can carry.

How long do bear markets last?

It depends heavily on who is counting. One published tabulation — Ned Davis Research data presented by Hartford Funds, current as of December 2024 — counts 27 bear markets in the S&P 500 since 1928 and puts the average length at 289 days, about 9.6 months. Other providers publish different counts and averages, because the answer depends on where you place each start and end date. An average built on a convention inherits the convention's arbitrariness, and no average tells you the length of any future decline.

Should I change my strategy when the market enters a bear market?

Nothing mechanical changes when an index crosses 20%. Nobody is forced to sell, no rule engages, and the tape at 19.8% down is the same tape as at 20.2% down. A rule that changes behaviour at a fixed line will fire and unfire as the index oscillates across it. If weakness belongs anywhere in a process, our research suggests it belongs in position sizing rather than in permission to act. This is not investment advice, and your circumstances may call for something different.