Skip to content

Mo Data: Using Mean-Reversion in the Momentum Factor to Time Momentum

September 12, 2019

In the last post we used the data available for the momentum factor using an ETF (ticker: MOM) which seeks to replicate The Dow Jones Thematic Market Neutral Momentum Index to time when to be in or out of high momentum stocks. Alpha Architect recently did some interesting analysis of the distribution of returns for the same momentum index in this post. One of the challenges was the lack of data available for testing. Ideally we would have much more data. To address this issue I found that the Kenneth French Data Library has daily momentum factor returns. To find a tradeable long-only momentum strategy, I used PDP which is the Invesco DWA Momentum ETF and is based on the Dorsey Wright® Technical Leaders Index (DWA Technical Leaders Index). The strategy is to go long PDP when the momentum factor is oversold (<.5) and this is compared to a strategy that goes long PDP when the momentum factor is overbought (>.5). For more details please read the previous post. The results are in the chart below:

Having data prior to 2013 is valuable because we get to see how the strategy performed during the credit crisis in 2008 and also in the beginning of the explosive bull rally from 2009 to early 2010. Clearly mean-reversion in the momentum factor has worked well in timing high momentum stocks since 2007. While not shown, I tested a wide variety of different parameters and found very similar results. To determine whether this is a pervasive effect across history would require extensive testing with synthetic momentum strategies, however my guess is that this has been effective for at least the last 25 years. Ultimately if it works back to 2007, it is clearly worth using and at least watching as part of your trading in either high momentum stocks or their proxy ETFs or mutual funds. However it is worthwhile doing some additional analysis to determine why this strategy works.

17 Comments leave one →
  1. Alex permalink
    September 13, 2019 12:42 am

    Thanks for the great post. Could you please confirm which indicator for overbought/oversold you used? was it an RSI (Relative Strength Index)?

    • david varadi permalink*
      September 13, 2019 12:04 pm

      Thanks Alex, I used the percentrank of the smoothed 10-day return (using a 5sma of the return).
      best regards,

  2. David Faulkner permalink
    September 13, 2019 8:33 am


    Great articles and something that I have been trying to figure out for a number of years (momentum vs mean reversion).

    Quick question – your entry (buy) signals are clear (MF0.5) but what is your exit signal – do you let momentum run with a different momentum exit or do you exit at MF>0.5? If you are testing for MF>0.5 it seems to me that you must be using a different measure to exit. Maybe negative momentum (in PDP for example)?

    David F

    • david varadi permalink*
      September 13, 2019 12:06 pm

      Thanks David, the entry is .5, although I am sure there are better entry/exit methodologies that are dynamic with different thresholds and also parameters. I was just trying to demonstrate a basic effect.
      best regards,

      • David Faulkner permalink
        September 13, 2019 12:47 pm


        I’m still not quite clear – for your entry when MF>0.5 (orange line) what is your exit trigger?


  3. david varadi permalink*
    September 13, 2019 11:03 pm

    David , the exit trigger is <.5.
    best regards,

  4. September 17, 2019 11:13 am

    Hello David, enjoyed your article as always!

    When do you trigger the entries? On close or open? I ask because I suppose the momentum factor returns are calculated after the close. Does the close vs. next open make a difference? Thanks!

    • david varadi permalink*
      September 17, 2019 12:02 pm

      Thanks Michael, the entries are using adjusted closing data where the entry is long .5. I didn’t test using opening data but that is worth taking a look at (next day open). best regards,

  5. September 25, 2019 8:42 am

    Interesting post.

    Regarding data, have you tried using Barclays Indices? Their momentum indices start from 2002/5 for EU, UK, JP, US stocks. They have both excess-return and total-return. I find it a good resource for back-testing.

    I put together an R notebook using your idea. Only change I made was to use a rolling window to calculate the median. Appreciate your thoughts:

    • david varadi permalink*
      September 26, 2019 3:40 pm

      Shyam, thank you for sending this link and sharing. I assume L1 means below median? Surprised to see that the Hursts on average demonstrate trending behavior but that performance seems to be better when buying below the median generally speaking. Very interesting analysis!
      best regards,

  6. September 26, 2019 10:36 pm

    Yes, L1 means below median. I was surprised as well – wrote it up on my blog:

    I suspect that it has to do something with the windows in use. 5-day L1 returns > 10-day L1 returns. But Hurst is calculated over 5 years so maybe in the very short-term it mean-reverts but trends over all?

    • david varadi permalink*
      November 18, 2019 3:21 pm

      Shyam, very nice analysis! I will present these results in a future blog when I update performance using the indicator. It seems like short-term mean-reversion is the strongest effect.
      best regards,

  7. Diego permalink
    November 11, 2019 10:06 am

    It works for a basket of diversified assets based on the momentum or trend following, for example, applying this mean reversion to a global asset allocation strategy based on momentum such as “Composite dual momentum by Gary Antonaci” or “GTAA Agg. 6 of Meb Faber”. Does the mean reversion applied to these strategies work? Thank you very much for your excellent work.

    • david varadi permalink*
      November 18, 2019 3:24 pm

      Thanks Diego, I would imagine that it does work though I haven’t tested it yet. Ultimately it depends how much the strategy depends on correlated asset classes. A momentum strategy that goes long stocks or bonds versus say sectors,countries or industries will be less affected by this factor since it has to do with systematic risk more than the momentum factor itself.
      best regards,

  8. Diego permalink
    November 11, 2019 11:46 am

    I can use mean reversion to your David Varadi’s Percentile Channels strategy? Thanks

  9. Carl permalink
    January 1, 2020 5:35 pm

    Always use the MOM signal, or you can do it with the same asset that you are going to trade (ex. MTUM, QMOM, GMOM)?
    Ex.: MOM OB/OS Oscillator is 38, but MTUM OB/OS Oscillator is 65. What signal should I take to trade MTUM?
    Thanks for your excellent work!!

    • david varadi permalink*
      January 30, 2020 1:15 pm

      Hi Carl, thank you for the kind words. the asset that you are trading isn’t the same because they don’t include the short side of the momentum trade. Arbitrage forces use both sides (they are the big players), and effectively what we see is that there are limits to stat arb that forces them to periodically unwind and put back on their trades. You can’t clearly notice this by looking at the long side only.
      best regards,

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: