Momentum's Dirty Secret: It Crashes

Momentum has the best long-run track record of any equity factor. It works in stocks, bonds, currencies and commodities. It has worked for over a century. Its risk-adjusted returns embarrass value.

And then, every so often, it detonates.

In a handful of months in 1932, and again in a handful of months in 2009, a standard momentum strategy lost a staggering fraction of its value, wiping out years of accumulated gains. Kent Daniel and Tobias Moskowitz wrote the paper that explains exactly what is happening in those moments, and, crucially, showed that the danger is partly forecastable.

The problem: a strategy with a beautiful average and a horrifying tail

Look at momentum's monthly returns and you see something that should worry you. Most months are modestly good. The distribution is not symmetric. It has a long, ugly left tail: rare months of catastrophic loss.

This is the profile of a strategy that quietly earns small profits and occasionally hands back everything at once. If you judge it by average return or by Sharpe ratio, it looks magnificent. If you actually have to live through 2009 while running it with leverage, you may not have a fund afterwards.

The key idea via analogy: you are short a lottery ticket on the market's rebound

Here is the mechanism, and once you see it the crashes stop being mysterious.

A momentum strategy is long the past winners and short the past losers. Now think about what those two baskets contain after a market crash.

The market has just fallen 50%. What are the "past losers" that momentum is short? They are the companies that got destroyed: the banks, the leveraged, the nearly bankrupt, the ones trading at two dollars with debt they may not survive. And here is the key point about a company in that condition. Its equity has become, in effect, a call option on survival. If the firm goes under, the shares go to zero, and they were nearly there anyway, so the downside is exhausted. But if the firm survives, the stock can go up 400% in weeks.

Meanwhile, what are the "past winners" that momentum is long? Defensive, stable, boring names that held up through the crash. Consumer staples. Utilities. Things that did not fall.

So after a crash, the momentum portfolio has a very particular shape: short a basket of highly levered, option-like, coiled springs, and long a basket of sleepy defensives.

Now the market rebounds. What happens? The beaten-up junk, the short leg, rockets. It triples. The defensive long leg goes up a bit. The short leg annihilates you.

That is a momentum crash. It is not bad luck. It is the structural consequence of what the portfolio holds in the aftermath of a bear market. Daniel and Moskowitz describe the losers' payoff as option-like, and momentum is short that option, so it gets crushed precisely when the option pays.

The forecastable part

This is what elevates the paper from diagnosis to something useful.

The crashes are not random. They cluster in identifiable conditions, what the authors call "panic" states:

• The market has recently fallen sharply.
• Market volatility is high.
• And the crash itself occurs when the market rebounds from that state.

All three of those things are observable in advance, or at least the first two are. You do not need to predict the rebound. You just need to notice that you are in the condition where a rebound would kill you.

So a momentum investor is not helpless. The natural response is to scale down momentum exposure when the market has recently crashed and volatility is high. The authors show that a dynamically managed version of momentum, one that adjusts its exposure based on these conditions, substantially improves the strategy's performance and tames the left tail. Momentum's biggest weakness turns out to be partly manageable, which is a genuinely valuable finding for anyone actually running the trade.

Why it mattered

• It explained the two worst months in momentum's history with a single coherent mechanism, rather than treating them as freak accidents.
• It made risk management part of the factor, not an afterthought. After this paper, running raw, unmanaged momentum looks naive. Volatility scaling and dynamic exposure management became standard practice in momentum implementations.
• It reframed what momentum's premium is for. If momentum is short a rebound option, then part of what you earn in normal times is the premium for writing that option, an insurance-selling business. That is a much more coherent economic story than "trends persist," and it explains why the returns are not a free lunch: you are being paid to take a specific, ugly, rare risk.
• It is a general lesson about backtests. A strategy whose returns come partly from selling disaster insurance will look wonderful in any sample that does not happen to contain a disaster. This is the archetype of that danger.

The honest limitations

• The crashes are rare, so the evidence is thin. The whole phenomenon rests on a small number of episodes, most prominently 1932 and 2009. Building a confident model of a tail from a handful of observations is statistically uncomfortable, and the authors are honest about it.
• Dynamic scaling is fitted after the fact. The dynamic strategy that would have avoided the crashes was designed with full knowledge of when the crashes happened. Whether the same rule protects against the next crash, which may have a different shape, is unknown.
• Being cautious has a cost. Cutting momentum exposure whenever volatility is high means sitting out some of the periods when momentum performs well. The insurance is not free.
• It does not remove the risk, it manages it. Even the dynamically managed version remains a strategy with negative skew. The tail is thinner, not absent.

The one-line takeaway

Momentum crashes because after a market collapse it ends up short a basket of near-bankrupt companies whose shares behave like call options on survival, so when the market rebounds those shorts explode; the saving grace is that the dangerous conditions, a recent crash plus high volatility, are visible in advance.