Dual Momentum on Individual Stocks. Wow.

Dual Momentum on Individual Stocks. Wow.

February 11, 2016 Momentum Investing Research, $mtum
Print Friendly
(Last Updated On: January 18, 2017)

Hot off the press and haven’t had time to reverse engineer and verify, but this is pretty interesting stuff at first glance.

The Enduring Effect of Time-Series Momentum on Stock Returns Over Nearly 100-Years

This study documents the significant profitability of “time-series momentum” strategies in individual stocks in the US markets from 1927 to 2014 and in international markets since 1975. Unlike cross-sectional momentum, time-series stock momentum performs well following both up- and down-market states, and it does not suffer from January losses and market crashes. An easily formed dual-momentum strategy, combining time-series and cross-sectional momentum, generates striking returns of 1.88% per month. We test both risk based and behavioral models for the existence and durability of time-series momentum and suggest the latter offers unique insights into its continuing factor dominance.

A picture is worth a 1,000 words:

dual momentum on stocks
The results are hypothetical results and are NOT an indicator of future results and do NOT represent returns that any investor actually attained. Indexes are unmanaged, do not reflect management or trading fees, and one cannot invest directly in an index. Additional information regarding the construction of these results is available upon request.

 

h.t., A. Miller @ http://www.miller-financial.com/ for sending our way!


Note: This site provides NO information on our value investing ETFs or our momentum investing ETFs. Please refer to this site.


Join thousands of other readers and subscribe to our blog.


Please remember that past performance is not an indicator of future results. Please read our full disclosures. The views and opinions expressed herein are those of the author and do not necessarily reflect the views of Alpha Architect, its affiliates or its employees. This material has been provided to you solely for information and educational purposes and does not constitute an offer or solicitation of an offer or any advice or recommendation to purchase any securities or other financial instruments and may not be construed as such. The factual information set forth herein has been obtained or derived from sources believed by the author and Alpha Architect to be reliable but it is not necessarily all-inclusive and is not guaranteed as to its accuracy and is not to be regarded as a representation or warranty, express or implied, as to the information’s accuracy or completeness, nor should the attached information serve as the basis of any investment decision. No part of this material may be reproduced in any form, or referred to in any other publication, without express written permission from Alpha Architect.


Definitions of common statistics used in our analysis are available here (towards the bottom)




About the Author

Wesley R. Gray, Ph.D.

After serving as a Captain in the United States Marine Corps, Dr. Gray earned a PhD, and worked as a finance professor at Drexel University. Dr. Gray’s interest in bridging the research gap between academia and industry led him to found Alpha Architect, an asset management that delivers affordable active exposures for tax-sensitive investors. Dr. Gray has published four books and a number of academic articles. Wes is a regular contributor to multiple industry outlets, to include the following: Wall Street Journal, Forbes, ETF.com, and the CFA Institute. Dr. Gray earned an MBA and a PhD in finance from the University of Chicago and graduated magna cum laude with a BS from The Wharton School of the University of Pennsylvania.


  • Thomas Musselman

    Looks like its been flat for 15 years and down 25% since 2008. Maybe it no longer works.

  • Lucas

    Have you guys ever explored the research on the one month lag to compute TMOM? This is from the paper referenced above, “There is a one-month skip between formation and holding periods (i.e., month t – 1) to avoid the microstructural bias (Jegadeesh, 1990; Lehmann, 1990).”

  • yes, doesn’t matter that much. Personally think it is much ado about nothing and people get wrapped around the axle on that aspect of ‘generic academic momentum’

  • who knows. I think there are behavioral foundations for the phenomenon and I think it can be painful and not work for a long time. One needs people to give up faith on a strategy in order for it to be sustainable over time. Momentum reminds me a lot of value investing…

  • It doesn’t work anymore because positive autocorrelation disappeared since introduction of computers. Let;s say nothing changed, but risking 30-40 years of time without returns is risk too big to handle. You should adapt to existing situation, past proves nothing.

  • Interesting hypothesis, but where is the data? Investors have been blaming computers for market problems since the 1980’s…and the gain in computing power has been exponential well before 2003.

  • Also, did the computers prevent autocorrelation in the early 1930’s as well?

  • another idea to consider is the one highlighted by Asness when he examines momentum in Japan. It doesn’t “work” until you consider the correlation structure with portfolio — especially one with value. So momentum strategies don’t have to work on an absolute basis to be considered invalid. One would need to identify 1) that mom systems don’t work as stand-alone systems and 2) don’t provide diversification benefits for a portfolio, to be considered a ‘failure’.

  • OK, my platform data doesn’t go back to 1930, can’t show the plots, but found one paper: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1088861 So, there were no significant positive autocorrelation near great depression and it started to disappear since 1980, e.g. the start of era of computer technology. More intuitively, prices can grow only with influx of new people (money). When HFTs started to exploit this anomaly, game became more efficient.

  • Also would need to explain the Geczy finding:
    http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2292544
    which extends 100+ years before that period.
    And I assume you are talking about intermediate term momentum? Not sure how many hft folks do that because they’d get their face ripped off trying to arbitrage that effect.

  • Thank you, will explore. One thing though, hope backtests are not on adjusted prices.

  • Thomas Musselman

    Agreed. But to be realistic would a real person stick with a formula that failed to beat the S&P after 15 years, or which was down 25% since the Great Recession 7 years ago? That is an awful lot of under-performance for any real world person (let alone a professional money manager). What is most likely to be the most relevant time period, the most recent past or decades ago given changes in economy and stock market? I would think the most recent is most relevant. I have back-tested momentum strategies which outperformed the S&P so I’m not poo-pooing momentum; e.g. if you take the decile of liquid stocks with the highest 200 day momentum and then from them buy the 1/3 with lowest P/Sales ratio compared to their industry, e.g., you get an excellent return in roughly the same recent time period. So I don’t think the presented test is the best usable momentum formula, even if you want to mix it with other non-momentum strategies. The real advantage of the presentation is the multi-decade data.
    BTW love your site.

  • JAK78

    Much of the under performance since 2000 was caused by the “momentum crash” in 2007. This happens only on the short side of cross-sectional momentum. I’d like to see what performance looks like only on the long side, which is how most of us invest. The big advantage of time-series momentum is its ability to hold up well in market crashes, so shorts aren’t so desirable anyway when you use time-series momentum.

  • JAK78

    Much of the under performance since 2000 was caused by the “momentum crash” in 2009 which happens only on the short side of cross-sectional momentum. I’d like to see what performance looks like only on the long side, which is how most of us invest. The big advantage of time-series momentum is its ability to hold up well in market crashes, so shorts aren’t so desirable anyway when you use time-series momentum.

  • Aaron Smith

    Don’t you think the outperformance by your momentum/low P/Sales backtest may be just what Jak was talking about above? That your backtest is long-only and it was long/short momentum which sucked the past 15 years? The short side is what caused the huge drawdown, so long-only theoretically should be fine.

  • Here are some stats on generic momentum decile and generic value decile (ken french data) from 2000 to 2015. Long-only.

    Summary stats — before costs — show val/mom work and the combo still works great.
    invested growth attached as well.

  • STIMPS

    i wish these academic studies would first define an investable universe. You simply can’t short as easily as they assume, especially if the stocks are illiquid and under $5. Remove what can’t actually be invested under their assumed strategy before analyzing the results. You can get very different results. I understand they are trying to “prove” a concept, but if the concept is based on flawed results, what did they really prove?

  • Alexandre Rubesam

    In my 2013 paper “The Disappearance of Momentum”, I used a regime-switching model to investigate the momentum premium over the period from 1927 to 2010. We concluded that the momentum premium disappeared after the late 1990s, most likely due to the increased activity of sophisticated arbitrageurs. This graph confirms our conclusion: momentum has been flat/negative since 1999 (our model estimated Nov 1999 as the point in which momentum ceased to be profitable). We also estimate that at least 50% of the momentum profits during the 1994-2000 period (a period of large momentum profits) were due to the boom (even bubble) in high-tech stocks. That is, any periods with such overperformance in a specific sector will look, ex-post, as a strong period of momentum performance.

    Link to the paper: http://www.tandfonline.com/doi/abs/10.1080/1351847X.2013.865654

    On SSRN: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=968176

  • Also, whether strategy is still working can be estimated using Monte Carlo. If it works it’s unlikely it will fall below lowest percentiles.

  • Alexandre Rubesam

    I did something similar in my paper although I used bootstrapping instead of Monte Carlo to estimate the probability of the momentum premium observed in the 2000-2010 period, given the probability distribution in previous period. Monte Carlo seems like a nice idea, but you have to assume a dgp.

  • Mark

    Hi Wes,

    I tested a long only dual momentum strategy on stocks from 1/1/2007 to 24/5/2016, while there is a huge drawdown in 2008 (~65%) the strategy still returns 6.5% annually over the period with a Sharpe of .37 on par with the market. Testing exclusively during the latest bull market from 1/1/2010 to 24/5/2016 returns an attractive 13.43% with a Sharpe of 0.75. Adding in RAA seems to add quite a bit of value, testing over the same periods yields 13.5% and 16% annually with Sharpes of .79 and .78 respectively.

    However I was not able to observe the value of the Frog In the Pan effect in my testing, as described in the paper. If I apply a double sort first using long only momentum then filter on FIP I do observer better returns when compared to simply selecting the initial long only momentum portfolio, however the double sorted portfolio has fewer names because of the extra filter. If I compare a long only momentum portfolio of similar size to the double sorted FIP portfolio the risk adjusted returns are similar. I don’t see any real value to adding FIP. I’ve seen the same phenomenon with other acceleration type tweaks to momentum strategies. In general the double sort looks like it adds value but not when compared to a long only momentum portfolio of equal size. I know you guys tested FIP and observed improved performance, when you do the comparison are you comparing the double sorted portfolio to the original long only momentum portfolio or to a long only momentum portfolio of equal size to the double sorted portfolio?

    Regards,
    Mark

  • Hey Mark,
    Thanks for sharing your analysis and work — awesome.
    The FIP effect is a marginal element relative to the big muscle movement of buying strong relative strength. Akin to value investing — buying cheap is the big muscle movement, adding quality is a marginal element. With a short horizon dataset it is probably difficult to detect much difference.

    I am tracking on your point about portfolio construction and needing to control for this difference in portfolio size.

    We’re aware of that and we explored that issue when we did extensive testing to confirm that the FIP was likely robust and not just noise.

    What size portfolios are you looking at ? And what is your initial universe? Are you looking at the decile? And how many firms?

  • Mark

    Hi Wes,

    Thanks for getting back to me. My universe is stocks trading on either the NASDAQ or NYSE with market cap > $2billion. For the momentum portfolio I select the top 100 stocks, for FIP I select the top 20 from the initial 100 based on the FIP metric. If I do that my portfolio size is 20, and compared to the initial portfolio of 100 the strategy outperforms. However if I simply select the top 20 momentum stocks without bothering to do the extra filtering on FIP I get similar results. I’ve tried a couple of different parameters with the same result. For example if I take the top 100 momentum stocks then filter again on FIP and take the top 50, I get similar results to simply taking the top 50 momentum stocks.

    regards,
    Mark

  • Mark,
    just ran 2007 to 2014 since i had data handy.
    hq mom is mom+good FIP, generic mom is 2-12, and lq mom is mom+bad FIP
    So over this period FIP does relatively worse than generic. That is probably why you are seeing those stats on your end. The net implies net of 1.80%/yr in costs and all returns are total returns

  • Mark

    Thanks a lot Wes, that’s really interesting.