Plutux
12-1 Momentum Factor

The momentum crash: the factor's worst days are all the same day

One rare, violent, structural failure; one slow leak; and the question mark that hangs over every anomaly with a paper trail.

12-1 Momentum Factor — When it fails

Key takeaway

  • At a sharp market bottom the previous losers rally hardest — and the factor is positioned exactly wrong for that, by construction
  • The crash is not bad luck; it is the ranking inverting, and no parameter change removes it
  • Costs and crowding both eat a documented edge from opposite ends, and neither shows up in the published tables

Anatomy of a momentum crash

Walk through what the portfolio holds at the bottom of a bear market. The 12-1 ranking was computed over the decline, so the top of the list is whatever fell least — defensives, cash-rich names, the least bad. The bottom of the list is everything that was crushed. Then the market turns, and the sharpest rallies belong to the crushed names, because that is what a relief rally is.

The factor is short that rebound in spirit even when it is long-only in form: it owns none of the names doing the rallying and all of the names being rotated out of. 2009 is the canonical example — in the months off the March low, the previous losers led by an enormous margin, and momentum portfolios gave back years of accumulated edge in weeks.

The uncomfortable part is that this is not a malfunction. The ranking did what it always does; the market inverted underneath it. Any fix that would dodge the crash — faster rebalancing, a reversal detector — degrades the factor everywhere else, and the literature's conclusion is that the crash risk is part of what the average return is compensation for.

A price path falling steeply, bottoming, and climbing back above its startA cyan line rises, drops sharply to a low marked as the crash, then recovers past its starting level. A dashed horizontal line marks the pre-crash level.level before the fallthe part you live throughthe part you read about
The factor's worst moments cluster at the turn between the two halves of this picture — computed on the left half, run over into the right.

The slow leak: costs against a thin edge

Between crashes, the factor's enemy is arithmetic. High turnover means the strategy pays its costs many times a year, against a documented pre-cost edge that is modest to begin with. An implementation with wide spreads, fat commissions or short-term tax drag can be a faithful copy of the paper and still net out to nothing.

This failure is quieter than the crash and more common: nothing dramatic happens, the screen keeps producing lists, and the account simply underperforms the factor it is supposedly tracking, by the exact amount of its frictions.

The published-anomaly problem

Everything about this factor is public: the paper, the parameters, the decades of replications, the ETFs tracking it. An edge that is in a journal is an edge everyone can crowd into, and crowding has two costs — it can compress the future return, and it makes the exits narrower when the crowd leaves at once, which is one mechanism behind the crashes.

A rising backtest curve that goes flat where the test data endsA line climbs smoothly across the left half of the chart, then a vertical divider marks the end of the tested period and the line moves sideways and slightly down after it.tuned on thismet this laterThe left half is not evidence. It is the data the rules wereshaped to fit.
The gap between a published record and live results is the most dependable finding in strategy research, and a factor with this much fame should be expected to show it.
  • The case that the edge persists: it has survived three decades of being public, and the standing explanation is that it pays for a risk — the crash — that most capital refuses to hold.
  • The case for humility: persistence in the past is the same evidence every decayed anomaly had, right up until it decayed.
  • The practical posture: expect less than the papers report, and size so that either outcome is survivable.

A quirk specific to the threshold version

The shipped implementation replaces the rank with absolute thresholds, and that changes the failure profile. In broad declines it goes empty rather than holding the least-bad decile — which softens the bear-market ride — but around the +10% line it can flicker, admitting and ejecting the same name across consecutive rebalances with costs attached each time. A ranking has no such boundary; a threshold always does.

Common questions

Can a filter avoid the momentum crash?
Proposals exist in the literature — scaling exposure down when market volatility spikes is the best known. They help in backtests, add parameters, and blunt the factor in normal times. None removes the underlying mechanism, which is that the ranking is computed on a regime that has just ended.
Is momentum dead now that everyone knows about it?
It has been declared dead repeatedly since publication and has so far kept showing up in the data. The honest answer is that nobody knows — which argues not for avoiding the factor but for expecting less than the historical tables and treating multi-year droughts as within the range of normal outcomes rather than proof either way.

These are documented methods described for study. Nothing here is investment advice, a recommendation, or a claim about future returns — every system on this page has losing periods, and the pages say where.

Reading about a system is not having one.

Plutux is where you write your own rules down, test them against real data, and keep the record your memory would otherwise rewrite. Join the waitlist for early access.

Plutux는 투자자문업자가 아닙니다. 시장 데이터와 AI가 생성한 분석은 정보 제공 및 교육 목적일 뿐 투자 자문이 아닙니다. 면책 조항

© Plutux Technology Limited 2026
When 12-1 Momentum Fails: Crashes, Turnover Drag and Crowding | Plutux