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

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.

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.

• 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.