Tuesday, July 28, 2009

Know Your Mean Reverters

The Mean Reversion Swing crowd has been taking it on the chin during the last two weeks and they are feeling quite mean about it -- I'm certain. Since I recently added a nice testable measure of mean reversion performance to the ETF Rewind in the form of David Varadi's (CSS Analytics) Relative Close "DV-2" indicator (see here for a nice explanation by MarketSci), I thought it would be a good time to review just how nasty it has really been.

The chart below plots the results of fading the unbounded DV-2 statistic for the SPY over the last 250 trading days versus the VIX. The hypothetical equity curve represents frictionless, additive returns. Although the tested results yielded nearly a 120% return, I wish to emphasize that this is not a trading system in and of itself. Although with raw returns like that -- it may certainly be the basis of one!

As shown, last year ended up being terrific for mean reversion methods, particularly when volatility was near its peak. By the same token, even if it feels otherwise, the recent 9% draw-down from peak equity is not without precedent in either magnitude or duration (see yellow bands).

While clearly the "trend has been your friend" of late and cycle periodicities have arguably shifted, it's far too early to pronounce mean reversion dead! Also, I would suggest that the aforementioned losses could have been materially reduced by ignoring or even reversing signals during extreme market environments as measured by a variety of metrics.

Sector Performance

This is a high level post, but I thought I'd take it one level deeper by searching for any particular sector source of the SPY's raw performance trading the DV. Looking across twelve major sectors below, my only conclusion is that the breadth of the S&P500 is what give it its edge as rotation among sectors keeps the the index more or less centered through time.

In fact, only Real Estate (IYR) significantly outperformed the SPY, while Consumer Discretionary stocks (XLY) showed no measurable edge on a raw basis whatsoever (though I imagine the savvy mechanical trader may have ideas to improve on that).

Here are several added take aways that I wish to emphasize from this post:
  1. Know what environment you are trading in (reverting/ trending/ break-out);
  2. Trading methods themselves revert (nice addition by David V.);
  3. Track your methodology's equity curve like it was a stock (even when you aren't trading it); and,
  4. Make sure you are trading the correct vehicle(s) for your method (see dispersion above).
Each night the ETF Rewind report provides just such broad market environment measures, as well as the DV indicator and over 65 additional key statistics for nearly 180 ETFs representing every major asset class.

PS - Welcome to the Blogosphere CSS Analytics!


Woodshedder said...


Jeff Pietsch CFA said...

Ha! Maybe I'll change my name to 'Slap Daddy' and apply for a job at IBC. I could use the CNBC exposure! -- Thanks Wood...

Tom said...

Hi Jeff, what are you fading exactly with the unbounded DV2 - change in the indicator value vs the previous day, or just long if negative, short if positive? Thanks, you do awesome work here!

Jeff Pietsch CFA said...

Hi Tom, merely the latter... by your question alone I imagine that you can imagine easy improvements... Regards, Jeff

michael168 said...

Nice piece. We like where you are vs IBC. Keep it up and CNBC will come knocking. I am not sure that would be a good thing as I tend to fade cnbc. :-)

Regards, mrice

Woodshedder said...

Jeff, just getting back to this post as I'm finally getting around to messing with DV2.

Slap Daddy it is...lol...

If nothing else, Jeremy, who does the graphics, will make you a nice header.

By the way, my blog got 10s of hits of exposure after the Kneale mention. In other words, either he has no viewers, everyone already knows my blog, so they didn't visit, or nobody cares, or all of the above.