In this piece, I’ll explore the momentum factor in some detail, mainly because of the excellent article by @jesse_livermore on the dangers of back-testing and the recent deterioration in the momentum factor. As Jesse did, I’ll use Fama French data because its publicly available, even though I think that their definition of momentum (total return from one year ago to the end of last month, re-balanced monthly) can be improved upon (total return between 3 and 9 months, combined with a measure of volatility). I show results for both equally weighted and market cap weighted momentum deciles published by Fama/French. The equally weighted data is more interesting for investors building a benchmark-agnostic active portfolio.

Here is the cumulative excess return of the best decile of momentum stocks minus the worst decile since 1927, presented in an obfuscating log chart. I’ve circled a few periods on the chart which look tame or benign when viewed this way.

**Cumulative log-scale growth charts have done more harm than good, because they make backtested strategies look so easy.** They lull investors into what looks like a sure thing. To correct that impression, here is the same spread for rolling one year periods (shown twice, for equal weighted first then market cap weighted. Market cap weighted results tend to be reigned in).

The periods highlighted are momentum crashes, when high momentum stocks get obliterated by low momentum stocks. Coming out of the bottom in 1932, low momentum beat high momentum by 471%…**in one year**. Coming out of the 2009 bottom, high momentum stocks under-performed low momentum by 237%.

It is also useful to see how high momentum did versus the overall market (rather than in a long/short context). Coming out of the 1932 and 2009 bottoms, high momentum under-performed the equal-weighted market by 213% and 52% respectively. It is very difficult to recover from that sort of crash.

These momentum crashes show up in the long term data even as you extend the holding periods to 15-years. Here is the rolling 15-year excess returns of high minus low momentum. Only the middle part of the sample is unaffected by the major factor crashes in the 1930’s and 2000’s.

Despite the crashes, the momentum strategy has been quite consistent over longer holding periods. Here are the base rates/batting averages of the best and worst momentum versus the market.

A 99% win rate (for equal weighted vs. market) in 15-year periods sounds impressive: until you realize that that 1% of the time just happened, and this is all before any sort of transaction costs, which are significant for a strategy that re-balances monthly. Here is the cumulative growth of high momentum vs. the market since 2000. The equal weighted version has done poorly, almost entirely because of the momentum factor crash in 2008/09.

One final interesting comparison is the rolling 15-year excess return for both the best and worst momentum vs. the market. While the spread between high momentum and the market has collapsed, it is interesting to note that the spread between worst momentum decile and the market has been negative 8-9% per year over the past fifteen years (2000-2015).

Momentum has been useful two ways: buying high momentum, and avoiding (or shorting) poor momentum. **Buying high momentum has had a rough 15-years relative to longer-term history, but avoiding poor momentum has continued to work quite well. **As I’ve written before, combining value and momentum has produced very strong results (with considerably lower turnover than a monthly momentum strategy). Like many quality factors, momentum has been very effective when used to avoid certain stocks rather than to select certain stocks.

*********************************

So here is the question for prospective momentum investors must ask: is the deterioration of momentum since 2000 a product of 1) more widespread awareness/adoption of the strategy or 2) the result of the natural inversion (crash) of momentum that happens after severe bear market bottoms, and our bad luck to have had 2 such crashes over the past 15 years.

The answer may be a combination of 1) and 2), but there is no question that the momentum crashes in the 1930’s and 2000’s are a key driver of momentum’s weaker results around those periods–especially in an equally weighted, benchmark-agnostic context.

Momentum crashes cannot be ignored. They are extremely hard to survive, for both professional money managers (career risk atom bomb) and individuals (a test of one’s emotions and discipline). Many factor-investing, quant researchers are trying to solve this issue with momentum, because if we can somehow avoid future momentum crashes, the momentum strategy could continue to be extremely potent (and may be potent even despite future crashes).

The bottom line: you want your factors to be as unpopular as possible (value has been heading that direction), and perhaps momentum crashes serve as purging fires of sorts, washing out those investors who cannot stick with the factor over the very long-term.

**An extra word of caution: this is all data on one factor in isolation. The best factor-based strategies combine different factors into a single portfolio (value, momentum, quality, and so on). I do not advocate a pure, momentum only approach to factor investing. **

What questions do you have? Let me know in the comments below

Great article Patrick!

All of this is being long high momentum versus being short low momentum or the market. That’s not the way most people invest. I’ve read that the terrible performance of momentum crash periods is all due to the short side. Wouldn’t it be better to look at long-only momentum? I’d really like to see how that has done during those crash periods and over the past 15 years versus other 15 year periods. Long/short is irrelevant and misleading for most of us.

Any idea how it holds up internationally ?

An astute observer would note that many of the worst momentum crash periods correspond to bear markets (i.e the market is below a long-term moving average). Any studies on a MA filter for momentum (implement MOM if market > moving average, 10-yr treasuries if below)?

one adjustment to the long/short results would have been to beta-neutralize the ratio. that is dollars long times long momentum beta must equal dollars short times short momentum beta. this will take out a lot of market influence and reveal the “pure”momentum factor. also running a sector-neutral momentum long/short strategy works better as well

Patrick, your “combined with a measure of volatility” comment at the top is intriguing. I have found that taking a measure of volatility into account as a guide to dictating your momentum look back period (as short as 1 month at volatility extremes) has huge advantages in minimizing drawdowns/ preserving capital over a static look back period (by definition 12 months as you pointed out). I’m actually surprised to see very little written about this.