Tuesday, December 07, 2010
Momentum Still Lagging
The graph shows the total return to a strategy long positive momentum stocks, short negative momentum stocks (it's from Kenneth French's excellent data repository). Historically, stocks that have outperformed over the past 3-18 months tend to outperform over the next 3-18 months, with an optimal point being around 12 months. The above graph is the total return for going long the past winners, short the losers. Historically this strategy has generated a puzzle, because the annualized return of around 8% it is pretty high, and it seems independent of the standard value and growth risk proxies. Sometimes it is added as a risk factor to studies, but mainly because we know it works and don't want to confuse this finding as something else. There is not a good 'risk' explanation for it.
Anyway, it seemed like easy money until March 2009, when it got clobbered, just as many prophets of doom were on CNBC telling Maria Bartiromo another leg down was in the cards. Actually, they would say another big decline 'could' happen (thus if it happened they would take credit for predicting the decline, and if it didn't they would note all they said was 'could'--predicting is so easy on TV once you know these tricks!).
Of course that was basically the start of the US market recovery and all those stocks that had been pounded like banks roared back (the momentum portfolio would have been short those). From March 2009 through September 2009 the momentum strategy lost about 50% and has gone up only about 5% since. Since 1932 (when it lost 77%), its worst draw-down was consistently about 30%, so the standard financial fat tail event happened. If you are lucky when such strategies stop working they flat-line, but this one was not so fortunate. As they say about these statistical patterns: they work until they don't.
A lot of funds were adopting momentum previous to this, especially because it came out the tumultuous 2007-8 period relatively unscathed. Unless momentum starts to really outperform it will go the way of 'low price' stock funds, because a big -50% really hurts all those lifetime annual return statistics.
I remember back in 2003 I convinced my boss to try momentum because it seemed have positive alpha. So, we put it on in the fourth quarter, and it started out ok, so ramped up in December. Of course, December 2003 was really bad for this strategy. Without a long track record, I looked like an idiot. I felt horrible and remember writing Ken French an email, and asking, what can I tell my boss about momentum's lousing December 2003? He was nice enough to reply, though all he said was, 'tell him there's risk'. Alas, that was not a good answer. In real time when you lose money it isn't seen as risk that pulled a bad number from the urn of chance, but a bone-headed mistake, especially when you don't have a benchmark. That wasn't one of my better Christmases.
Subscribe to: Post Comments (Atom)
But the low beta effect is guaranteed to work out of sample?
I like this blog, but what the author is snarky about vs thinks is a good idea seems highly arbitrary.
And a log scale would be nice.
Oh, and those who like momentum generally like it as a combo with the value strategy, which it is negatively correlated with, and the combo has still worked fine out of sample.
Here's the real problem: momentum is not mean reverting. The losses are permemant. Momentum is much riskier (for a long term investor) than a mean reverting strategy with the same volatility.
That is a fair point (mean reversion). However, mean reversion isn't the only good property to have. Another good one is plain old "mean".
An "optimal" portfolio (monthly horizon, not crediting for mean reversion, just ex post optimizing Sharpe) of French's long-short value and momentum factors from his website, 1927-present, is 55% value 45% momentum. And that understates momentum as that's in dollars, momentum is about 40% more volatile (not a bad property, you just have to adjust for it, something that works using rolling out of sample vol estimates).
If you take them both since the "momentum effect" was discovered (starting say 1992 or 1993), you get about the same optimal portfolio. Again, in vol space you'd actually have more momentum than value. Again, this is really out of sample stuff.
I would agree that I'd put more in value given the above results AND the extra fact that value is mean reverting, but given the large negative correlation between value and momentum, if you believe the above (including the out of sample 1993-present which includes the 2009 momentum disaster), I'd guess you'd still have a very significant momentum exposure.
Other pro-value vs. momentum arguments I credit include 1) the obvious that value is lower turnover so the above gross results are closer to net, and 2) momentum isn't just more volatile than value it's more negatively skewed (certainly driven by this 2009 disaster). But, again, I don't think you can push either of these to close to bad enough so that a value investor should not (if if if you believe the 1927-present of 1993-present results) add a significant momentum exposure.
If mean reversion is gonna kick in, it's not very visible at the 5 year horizon.
The worst case rolling 5 years for French's value spread factor, 1926-2010, is -44% vs. -49% for momentum. But the worst 5 years for 50/50 value/momentum is -17%. Seems like at 5 years you still want a ton of momentum.
At 10 years the worst cases for 50/50 are still much better than pure value.
It gets sillier and sillier (fewer data points) at longer horizons, but it doesn't seem mean reversion is close to knocking out momentum (alone or more importantly as a portfolio with value).
More specifically it doesn't knock it out empirically, you can certainly still argue for it on first principles.
There's also no reason one has to follow the strategy blindly. There are numerous permutations of this source of alpha besides simple N period holding and many, many ways of reducing risk / drawdown.
A failure of may quant/academics is the lack of creativity IMO.
Creativity often equals data mining.
That is the fear with the entire momentum strategy. It's a fear I think reduced by 20 years (since about 1990) of out of sample evidence (Eric's bad experience not withstanding, one bad December or 2009 does not eliminate the net evidence) and out of sample evidence around the world (and asset classes). But still these fears are legitimate. Once you start tinkering "creatively" you open up the whole strategy to that again. Not a proof you shouldn't do it, just a warning.
Those advocating that they can reduce or eliminate the downside of a strategy are often selling snake oil...
Why short? It is a bad strategy. If one wants to hedge just hold more cash.
Uh... For money. Just like everything else we talk about here...
Post a Comment