The Day Momentum Died

The day momentum died, mean-reversion was born. Now they have both merged into what I call momersion.

The other day I read an article about microscopic momentum where it was defined as follows:

“Multiply each return by the sign of the previous return, creating a new return for our ‘simplified momentum strategy’.”

Actually this is the quant way of doing things. The technical analysis way is as follows:

If close of today > close of yesterday then buy at the close
If close of today < close of yesterday then short at the close

This is how a technical analyst would backtest the result of applying a negation operator to the returns of a price series.

This is the backtest in S&P 500 from 1960, no commissions, one share per trade:

SPX_momentum

The day momentum died marked a paradigm shift in the stock market in many respects. No need to mention them, everyone knows. It was April 14, 2000, or around that time anyway.

The new paradigm was mean-reversion. In system form:

If close of today < close of yesterday then buy at the close
If close of today > close of yesterday then short at the close

This is the backtest of the mean-reversion strategy after the day momentum died:

SPX_mean reversion

The strategy worked well even during the 2000s uptrend and the 2008 plunge. Amazing, isn’t it?

But it could not work for long because quants identified it fast. On July 30, 2009 or around that time, mean-reversion died too. Call it an “arbitraged anomaly”, just like momentum.

So what do we have since? This is how mean-reversion has performed:

SPX_momersion

Momentum came back temporarily, offering hopes but then died again, just like a “dead cat bounce” and mean-reversion has reappeared since.

I call this new paradigm momersion: a little momentum followed by a little mean-reversion so that all those hit by at least 10 cognitive biases cannot make any money easily in the longer-term unless they put some real work to identify value and are also a little lucky.

As I have written before, one could see everything from this 1-lag, 250-day autocorrelation chart:

SPX_auto

Self-explanatory if you do not suffer from the 10 cognitive biases I did not specifically mention.

Momersion will persist until everyone who has built a business on hindsight bias is out of the game. This is how the market works. Value seeking always prevails above and over naive schemes to extract alpha.

You can subscribe here to notifications of new posts by email.

Charting program: Amibroker
Disclaimer

Detailed technical and quantitative analysis of Dow-30 stocks and popular ETFs can be found in our Weekly Premium Report.

© 2015 Michael Harris. All Rights Reserved. We grant a revocable permission to create a hyperlink to this blog subject to certain terms and conditions. Any unauthorized copy, reproduction, distribution, publication, display, modification, or transmission of any part of this blog is strictly prohibited without prior written permission. 

This entry was posted in Quantitative trading and tagged , , . Bookmark the permalink.

6 Responses to The Day Momentum Died

  1. Mark Leeds says:

    Hi Michael: It's interesting what you wrote and you expressed everything simply and clearly. But if I understand correctly, you just considered the 1 day autocorrelation. There's no reason to think that other timeframes won't show other behavior. Whether it's week to week or hour to hour, you probably will get something altogether different if you do the same kind of analysis. Still, a neat way and clear way to think about reversion versus momentum. Thanks. I really like your blog.

    • Hello Mark,

      This is a good point. Actually, the 1-lag (1 day) autocorrelation shows whether positive daily returns are most likely to be followed by more positive daily returns and negative by negative, in the time period considered. If you increase the lag to 2, the autocorrelation is mostly destroyed and becomes random. This is because you are expecting returns to have longer memory. As you increase the lag to 5, for a period of 250 returns, the general trend does not change but you get more variation. You will also notice that periods of sustained negative correlation increase after 2000. For example, for 30-lag, the period increases to two years while positive autocorrelation lasts for less than a year. But If you are increasing the lag, then you are not looking at momentum or mean-reversion properties in price series any longer but instead you are curve-fitting the autocorrelation model to the data, essentially introducing hindsight bias in the analysis.

      As you correctly point out one can apply autocorrelation to returns in all timeframes but then one also needs to find a good explanation of what the results mean in the context of the patterns they exhibit. In the case of the 1-lag, 250-day autocorrelaiton of returns, the trends are clear. However, in other timeframes they are not usually.

      Also, please note that the autocorrelation diagram was not used to prove anything but as a form of illustrative corroboration.

      You could possibly ask at this point: If markets were mean-reverting after 2000, why did some trend-following models work well?

      It is wrong in my opinion to associate trend-following with momentum, as it is done relentlessly nowadays by some circles. You can get trends in mean-reverting mode. Timing trends has little to do with momentum trading but it is only a trend-following attempt using some indicators. Trend-following was successful long before the academic community discovered it and wrongly in my opinion classified it as momentum trading, and specifically price series absolute momentum. However, in essence all of what is classified as momentum nowadays is trend-following, whether it is absolute or relative. Trend-following works when it does and momentum has little to do with it. They just invented a different name for it to possibly avoid offering credit to technical traders that have documented it and successfully used it, such as for example the Turtles and many others. I traded currency futures in the 1990s with success using trend-following models and I never had the illusion that it was something related to momentum.

      Michael

  2. Gariki says:

    Awesome post. Very true.

  3. Darell says:

    Fantastic post, Thanks

  4. Mark Leeds says:

    Hi Michael: Yes, I think going further out with a lower frequency is difficult but going
    to a higher frequency ( but non-HFT ) is the area I'm working on.

    As you said, I think there's both ( one can be within other or vice versa ) and I liked your name for it. Very good name. And As far as names, yes, momentum, trend following, it's all kind of the same concept. In fact, one could view mean reversion as a downward sloping trend.

    My take on all of these behaviors is that the key thing is how one "catches-exploits
    them. That's the name of the game because the existence of these various behaviors is pretty well known. Great blog and thanks.

    • Hello Mark,

      You wrote that " In fact, one could view mean reversion as a downward sloping trend.". However, there can be uptrends in mean-reversion mode, such in the 2000s. The difference in my opinion and according to my research is that uptrends due to momentum are more robust and carry less risk.

      Michael

Leave a Reply