CXO Advisory had an interesting post (click here for the post) on momentum earlier this week. In the post, CXO goes over Robert Novy-Marx’s November 2009 paper, “Is Momentum Really Momentum?” I read this paper when it came out, but it was one of those things that got put on the shelf for future research (and ultimately forgotten about!). When I saw the post on CXO it piqued my interest again, and I think it was very timely given the current state of the market.
The premise of Novy-Marx’s paper is that maybe the momentum effect isn’t really momentum, but more of an echo. It is common for researchers to use a trailing 12-month price return when constructing momentum models. The paper broke the 12-month ranking period into two subperiods: months 12-6 and months 6-0. (The actual paper skips the most recent month as does most momentum literature. I am not doing that. I am simply running everything as of the current month end. ) So what’s more important? Is it the current momentum, or the momentum from 6 months ago?
Novy-Marx’s data indicates the current momentum (most recent 6 months) is not as important as the previous momentum (the earliest 6 months). You can view his paper or the CXO blog entry for the data. I found that conclusion very thought-provoking considering the current state of the market. It seems everyone we speak with expects relative strength models to be undergoing major changes because of the market changes this week. I think the overall perception is that the faster you get on a trend, the better your performance will be. Or maybe you just feel better because at least you are doing something! The data in the paper indicates the most recent momentum data isn’t as important as the long-term data. I think that’s exactly the opposite of what most people think.
I ran Novy-Marx’s factor on our database to see how it looked on our data. There are a couple of differences. Our universes are different. I am using a universe similar to the S&P 500 + S&P 400. Novy-Marx used the top 20% of market cap out of the CRSP database. I used the top decile instead of the top 20% of ranks. This doesn’t make as big of a difference as you would think. When I ran the data using the top 20%, the cumulative numbers were very similar. And finally, I am not skipping a month between the ranking date and portfolio formation. I don’t believe any of these will have a material impact on results. I also think it is good to have a slightly different rule set and universe– it just helps reduce the data snooping bias that can crop up in these types of studies.
(click to enlarge)
The table above shows the results we generated using Novy-Marx’s factor. I have also included returns from a 12-month, 6-month, and 3-month price return factor. These returns were generated by taking the top decile of ranks each month end, holding the portfolio for 1 month, then reconstituting and rebalancing the whole portfolio at the next month end.
The 12-Month Echo factor does significantly better than either the 12-month or 6-month factor. And it blows a 3-month return factor out of the water. The table indicates the data at the back-end of an intermediate momentum factor is more important to returns than the near-term data. So as this market continues to gyrate wildly, keep in mind where the best long-term returns come from. It’s not the most recent data that everyone can’t stop overreacting to!