Momentum Crash

A momentum crash is a sudden, severe loss in a momentum strategy, which buys assets that have recently risen and sells short those that have recently fallen. Crashes occur over weeks or a few months and can erase years of accumulated gains. They are the main risk of momentum investing and the reason its long-run record, though positive on average, has a markedly lopsided distribution.

The phenomenon was analysed in detail by Kent Daniel and Tobias Moskowitz in a paper titled Momentum Crashes, published in 2016. They showed that the crashes are not random accidents. They cluster in a particular setting: a sharp market rebound following a deep decline, when volatility is high. The two worst episodes for US equity momentum on record came in 1932 and 2009.

The explanation lies in what a momentum portfolio holds after a bear market. During a long fall, the shares that drop most are those most sensitive to the market, with high beta: indebted companies, cyclical businesses and, in 2008, banks. The shares that hold up best are defensive ones with low beta. A cross-sectional momentum strategy, formed by ranking on past returns, is therefore long low-beta shares and short high-beta shares. Its net exposure to the market is strongly negative, even though it was not designed to make a market bet.

When the market turns and rallies, the high-beta losers rebound hardest. The strategy loses on its short positions far more than it gains on its long positions. The beta of a winners-minus-losers portfolio can be written as:

• $\beta_{\text{WML}}$ is the market beta of the long-short momentum portfolio.
• $\beta_{W}$ is the beta of the past winners held long.
• $\beta_{L}$ is the beta of the past losers held short.

Suppose that after a market collapse the winners have a beta of 0.7 and the losers 1.6. The portfolio's beta is $0.7 - 1.6 = -0.9$. If the market then rises 20 percent in a month, the market exposure alone costs the strategy about $0.9 \times 20 = 18$ percent, before any other effect.

Daniel and Moskowitz found that crash losses come almost entirely from the short side. In a rebound, the losers rise by more than their beta would predict. They behave like call options on the market: companies that have fallen close to distress have little further to fall and a great deal to gain if conditions improve. A strategy that is short such shares has, in effect, sold those options, and it collects small premiums in calm times and pays out heavily in a recovery. In the spring of 2009, as markets recovered from the 2008 financial crisis, the most beaten-down US shares more than doubled in three months while past winners rose only modestly.

Forced buying adds to the move. Investors who are short the losers must buy them back as prices rise, producing a short squeeze, and funds running similar strategies with leverage cut positions at the same time.

Momentum returns show strong negative skewness and high kurtosis: many moderate gains and a few very large losses. A measure such as the Sharpe ratio, which uses only the mean and standard deviation, understates this risk. The pattern resembles that of strategies that sell insurance, and it is one argument that momentum profits are, at least in part, compensation for bearing crash risk. The maximum drawdown of a simple momentum strategy over a long history is far larger than its ordinary volatility would suggest.

• Volatility scaling. Crashes tend to follow periods when the strategy's own realized volatility is high. Reducing the position when recent volatility is high and increasing it when low, a form of volatility targeting, has been shown in historical tests to cut the worst losses substantially.
• Dynamic weighting. Daniel and Moskowitz proposed adjusting exposure using forecasts of both the return and the variance of momentum, with less exposure after bear markets when volatility is elevated.
• Beta hedging. The portfolio's market exposure can be estimated and offset with index futures. This works imperfectly, because betas change quickly around turning points.
• Diversification across factors. Value strategies have often done well when momentum crashed, so the two are combined, as discussed under value and momentum strategy.
• Construction choices. Ranking shares on the part of their return not explained by the market, or within industries, reduces the built-in beta bet.

These methods are based on past data, and none removes the risk entirely.

Momentum crashes are not confined to equities. Trend-following strategies in currencies, commodities and bonds have suffered sharp losses at sudden reversals. Crashes are also distinct from the broader stock market crash: the market falls in one, while momentum typically loses most when the market rises. Rapid losses across quantitative equity strategies in August 2007, when many funds unwound similar positions at once, illustrated a related danger, that of crowding in widely used factor investing strategies.