A momentum crash is the defining tail risk of the momentum factor: infrequent but severe and persistent strings of losses in which past losers violently outrun past winners, gutting a cross-sectional winners-minus-losers (WML) strategy. The core tension is that momentum earns one of the most robust premia in finance during normal regimes, yet its return distribution is strongly negatively skewed — it behaves like collecting small steady premiums while implicitly writing a catastrophe option. The crashes are not random: they cluster in "panic states" — bear markets, high volatility, and especially during sharp market rebounds — which makes them partly forecastable and therefore partly manageable.

How the crash forms

The mechanism, formalized by Daniel & Moskowitz (2016), is a time-varying beta asymmetry in the two legs of the WML portfolio:

• After a prolonged bear market, the "loser" decile fills with stocks that have already collapsed — often highly leveraged, near-bankruptcy firms whose equity now behaves like a deep out-of-the-money call on the firm. Their downside is capped (already near zero), their upside is enormous.
• The "winner" decile fills with defensive, low-beta survivors.
• This makes the short (loser) leg high-beta and the long (winner) leg low-beta. The momentum portfolio therefore carries a large, conditional negative market beta exactly when the market is most likely to snap back.
• When the rebound comes, the high-beta losers rip upward, the defensive winners lag, and the WML strategy — short the very stocks that rally hardest — suffers a tail loss. Daniel & Moskowitz describe the loser leg's payoff as analogous to a written call option on the market, with the strike near the crash trough.

Two crashes anchor the literature. In July–August 1932, after the market had fallen ~90% from its 1929 peak, the WML strategy lost roughly -88% in two months — over those two months Daniel & Moskowitz report the past-loser decile up about +232% versus the past-winner decile up only about +32% (figures vary slightly by sample vintage; an earlier draft on the 1927–2010 sample reports +236% / +30%). In March–May 2009, the past-loser decile rose roughly +156% to +163% while the past-winner decile gained only single digits (the 1927–2010 draft reports the winner decile up ~6.5%), producing a momentum drawdown of more than -73% over the quarter (Daniel & Moskowitz). Both crashes occurred near market bottoms, not tops.

How it's used in practice — managing the crash

Because crash risk is conditional and forecastable, the practical literature centers on dynamically scaling exposure rather than abandoning momentum:

• Constant-volatility scaling (Barroso & Santa-Clara, 2015). Scale the WML position inversely to its own recent realized volatility (they use the prior ~6 months of daily returns) to target a constant strategy volatility (their target ~12% annualized). Because momentum volatility is highly persistent and predictable, this nearly eliminates the worst crashes. They report the Sharpe ratio rising from about 0.53 (unmanaged) to about 0.97 (risk-managed), with sharply reduced kurtosis and left-skew.
• Dynamic momentum (Daniel & Moskowitz, 2016). Scale weight by the forecast of momentum's mean and variance — a mean-variance-optimal weight that cuts or even flips exposure in panic states (high volatility + bear-market indicator). They report this roughly doubles the alpha and Sharpe ratio of static momentum and is not explained by standard factors.
• Idiosyncratic / residual momentum. Rank stocks on their factor-model residual (e.g., Fama-French residuals) rather than raw returns, which strips out the beta loadings that drive the crash. Studies (e.g., Blitz, Huij & Martens) find smoother, less crash-prone profiles.

The unifying insight: the crash is a beta problem, and every fix either neutralizes the conditional beta (residual momentum) or shrinks gross exposure when conditional beta and volatility are high (vol scaling, dynamic weighting).

Adoption, debate & evidence

Momentum crashes are a well-established, peer-reviewed phenomenon — Daniel & Moskowitz (JFE 2016) and Barroso & Santa-Clara (JFE 2015) are among the most cited factor papers of the decade, and the negative skew is visible in the long-run Fama-French momentum (UMD/MOM) series itself. The existence of the tail is not seriously contested.

What is debated is the fixes:

• Volatility scaling's out-of-sample robustness is generally well-regarded (the predictability of volatility is a strong, stable empirical regularity), but it is not a free lunch: scaling can increase gross leverage in calm periods (transaction costs, financing, shortability constraints) and can lag a fast regime change.
• The dynamic (mean-forecasting) approach is more contested because it relies on predicting momentum's conditional mean, which is noisier than predicting its variance — some researchers argue most of Barroso/Daniel's gains come from the variance term alone.
• A live concern is factor crowding / decay: post-publication, raw momentum's premium and the efficacy of simple fixes may be partially arbitraged away, though this is an open empirical question rather than a settled result.

Honest framing: the crash is real and large; vol-scaling robustly tames it; mean-timing the rebound is harder and should be treated with more skepticism.

Strengths & limitations

When the framing works: It correctly identifies when momentum is dangerous (after deep drawdowns, when volatility is elevated, when a violent rebound is plausible) and gives a concrete lever — cut exposure — instead of a binary in/out call.

When it fails: The signal is conditional, not a precise timer. A crash can be brief and you can de-risk into the recovery; vol-scaling cuts exposure after volatility has already spiked, so it mitigates but does not preempt the first leg of a crash. Implementation frictions (short borrow on the very loser stocks that crash you, leverage limits) can blunt the academic results.

The #1 misuse: Treating momentum's high unconditional Sharpe as if returns were Gaussian, sizing accordingly, and being wiped out by a single left-tail event. The whole point of the crash literature is that average return badly misrepresents momentum's risk; you must size for the skew.

Sources

• Daniel, K. & Moskowitz, T. (2016). Momentum Crashes. Journal of Financial Economics 122(2): 221–247. NBER WP 20439 — primary source for beta asymmetry, written-call analogy, 1932/2009 magnitudes, dynamic strategy. https://www.nber.org/system/files/working_papers/w20439/w20439.pdf
• Barroso, P. & Santa-Clara, P. (2015). Momentum Has Its Moments. Journal of Financial Economics 116(1): 111–120 — constant-volatility scaling, Sharpe 0.53 → 0.97. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2041429
• Chicago Booth Review, Understanding Momentum Crashes — accessible summary of Daniel & Moskowitz. https://www.chicagobooth.edu/review/understanding-momentum-crashes
• Alpha Architect, Risk of Momentum Crashes: can it be reduced? — empirical comparison of vol-scaling, idiosyncratic, dynamic methods (landscape/robustness view). https://alphaarchitect.com/risk-of-momentum-crashes/
• Blitz, Huij & Martens, Residual Momentum — basis for idiosyncratic-momentum crash mitigation (secondary).

Dispute flagged: the existence/magnitude of crashes is settled; the out-of-sample robustness of the mean-forecasting dynamic fix (vs. variance-scaling alone) and post-publication factor decay are genuinely contested.