The total return to Momentum was impressive for many decades. It's a simple strategy, basically going long past winners and short the losers, hoping they continue to win and lose. Interestingly, the past returns should go only up to the prior month, because there's slight mean-reversion at the one-month horizon, so most people use the returns from months t-12 through t-1. This highlights the non-fractal nature of stock returns, in that there's momentum in the data from 3-18 months, but mean-reversion at the shorter and longer frequencies.
Even after discovered by Jegadeesh and Titman in 1992, it seemed to work for another 8 years. Since 2000, however, it hasn't worked (see a decent paper on that here). Using Ken French's replication of this strategy, we see the total return pattern above. Note that while from 1932 through 1943 it stagnated, it seemed Madoff-like in its ascendance from the end of WW2 through 2000.
The big drawdowns in the momentum strategy occurred in the big stock rebounds of July-Aug 1932, and March-Sep 2009. These moves would generate losses of over 50%, which since it generated an 8% annual return, this would probably eliminate any particular portfolio manager--such losses are usually lethal.
Now, these long-term simulations tend to have a bunch of survivorship problem issues, and data prior to 1964 is to be taken with a grain of salt (the database was created then, so its much harder to correct errors when you don't remember how you collected data in real-time). Interestingly, while momentum is considered a real factor by some (eg, the Carhart 4-factor model is the Fama-French 3-factor model plus momentum), Fama has been conspicuously avoided treating momentum as a risk factor, nor trying to explain in theoretically in any way, and just looked at it quizzically. That was rather refreshing, in that it's tempting when you have the status he does to explain everything in your field, but instead he just shrugged.
Above are the December returns. These actually made sense because there was a real story here. The idea was that winners had taxable gains, and so not selling them until January would push off a liability; losers had losses that selling prior to January would lower one's taxes. Thus winners have this absence of selling, losers a greater amount of selling. Alas, since 2003, this pattern too has disappeared. In fact, I actually put the trade in December 2003 based on looking at this data, and got crushed. It was the worst return by far relative to my sample of 25 years that used a different universe than French but was basically the same pattern. I actually emailed Ken French at that time to ask if he had any insight, and he merely emailed back: 'it's risky.' My boss didn't think that was a good answer.
The big drawdowns in the momentum strategy occurred in the big stock rebounds of July-Aug 1932, and March-Sep 2009. These moves would generate losses of over 50%, which since it generated an 8% annual return, this would probably eliminate any particular portfolio manager--such losses are usually lethal.
Now, these long-term simulations tend to have a bunch of survivorship problem issues, and data prior to 1964 is to be taken with a grain of salt (the database was created then, so its much harder to correct errors when you don't remember how you collected data in real-time). Interestingly, while momentum is considered a real factor by some (eg, the Carhart 4-factor model is the Fama-French 3-factor model plus momentum), Fama has been conspicuously avoided treating momentum as a risk factor, nor trying to explain in theoretically in any way, and just looked at it quizzically. That was rather refreshing, in that it's tempting when you have the status he does to explain everything in your field, but instead he just shrugged.
Above are the December returns. These actually made sense because there was a real story here. The idea was that winners had taxable gains, and so not selling them until January would push off a liability; losers had losses that selling prior to January would lower one's taxes. Thus winners have this absence of selling, losers a greater amount of selling. Alas, since 2003, this pattern too has disappeared. In fact, I actually put the trade in December 2003 based on looking at this data, and got crushed. It was the worst return by far relative to my sample of 25 years that used a different universe than French but was basically the same pattern. I actually emailed Ken French at that time to ask if he had any insight, and he merely emailed back: 'it's risky.' My boss didn't think that was a good answer.
I think this highlights an important point. At any point in time your strategy is susceptible to a draw down that could cost you your client base. You can't just say 'hey, that's risk!' Investors see it as a failure, not a bad draw from the urn of chance. Returns over time are treated very differently than cross-sectional returns because cross sectional returns have covariances and volatilities amenable to statistical optimization; time-series returns are looked at more like datum in a broader strategy of eliminating all the 'losers' at any point in time. If you are in the bottom 10% at any time for any reason, your portfolio probably has hit an 'absorbing barrier.'
No comments:
Post a Comment