Reversion to the Mean

HoundDogOne · Sep 20, 2006

Reversion to historical mean...
Is a classic form of quantitative analysis.
Mixing it up with "technical analysis"... just dilutes it's power.

It's also a short term strategy... hours to maybe 2-3 weeks max...
Because if you are still hoping for reversion to mean after many months...
You will be regularily be caught in a pricing "paradigm shift" and crushed...
Like those guys at LTCM.

man · Sep 20, 2006

mean reversion is a well sounding term. yet it is in my opinion much more difficult to exploit it than trend following.

first the odds are more difficult, since you will miss the big fat tail in your favour. as a mean reverter you have a quite limited trade potential compared to a trend follower. so if you got hit badly by an anomaly it takes you longer to recover.

on the other hand you can expect higher hit ratios than within trend following.

what is most likely in your favour is slippage. since you are betting against the current trend you will most likely get the price you like.

i think mean reversion requires either a high trading frequency or a big portfolio of similar bets, like trading 50 paired stocks at any one point in time. you need the sheer mass of trades to get the odds right.

minmike · Sep 20, 2006

Pairs trading is an example of mean reversion. If Ford and GM move together most of the time and suddenly move apart you would be playing that they would revert to the mean which is coming back together. I like trading that way. but the Fat tails do hurt.

rdg · Sep 20, 2006

Quote from ImamicPH:

Usually when "reversion to the mean" is used it means returning to the average. For example you're looking at a trading strategy that has a historical win percentage of seventy percent. If you observe ten straight wins, then you can expect to lose in the coming setups.

What a perfect example of the Gambler's Fallacy. A coin flip is 50-50. Given that you have flipped 10 heads in a row, what are the odds that the next one will come up tails?

panzerman · Sep 20, 2006

If you want a times series series that is highly reverting, look at volatility (implied and historical.) In fact using R/S analysis, volatility is often more random than a gaussian time series, that is the exponent is less than 0.5.

Basically, it means that volatility is not fractal, and that if it is up during one time period, it will more likely be down in the next time period rather than continuing in the same direction (and vice versa.)

5Pillars · Sep 20, 2006

You can use Linear Regression channels for a "micro" form of MR measurement to look for potential trade set-ups..........

http://www.charthub.com/images/2006/09/20/Untitled_1.png

When price exceeds a flat or mildly sloping LR channel it can be a good visual reference point to look for your systems trade set-ups.

giggollo · Sep 20, 2006

Quote from man:

i think mean reversion requires either a high trading frequency or a big portfolio of similar bets, like trading 50 paired stocks at any one point in time. you need the sheer mass of trades to get the odds right.

Isn't that also required with trend following techniques?

stevegee58 · Sep 20, 2006

Quote from panzerman:

If you want a times series series that is highly reverting, look at volatility (implied and historical.) In fact using R/S analysis, volatility is often more random than a gaussian time series, that is the exponent is less than 0.5.

Basically, it means that volatility is not fractal, and that if it is up during one time period, it will more likely be down in the next time period rather than continuing in the same direction (and vice versa.)

Hmmm. That's one of the core principles in options trading isn't it? Each market has an "average" or "typical" volatility. When volatility gets out of whack with the average, you look to capitalize on it be either being net long or net short implied volatility.

segv · Sep 20, 2006

I find all of this pseudo-science extremely nauseating. Do you suppose that you are the first person on earth to ever consider the possibility of mean-reversion in asset returns?

http://www.google.com/search?hl=en&lr=&safe=off&q=variance+ratio&btnG=Search

Fama, E., and K. French, 1988, "Permanent and Temporary Components of Stock Prices", Journal of Political Economy 96
http://ideas.repec.org/a/ucp/jpolec/v96y1988i2p246-73.html

Jorion, P., and W. Goetzmann, 1999,"Global Stock Markets in the Twentieth Century", Journal of Finance 54
http://ideas.repec.org/a/bla/jfinan/v54y1999i3p953-980.html

Jorion, P., 2003, "The Long-Term Risks of Global Stock Markets", University of California Irvine
http://www.gsm.uci.edu/~jorion/papers/risk.pdf

... to get you started. The literature on this subject is overwhelming.

-segv

jem · Sep 20, 2006

Quote from HoundDogOne:

Reversion to historical mean...
Is a classic form of quantitative analysis.
Mixing it up with "technical analysis"... just dilutes it's power.

It's also a short term strategy... hours to maybe 2-3 weeks max...
Because if you are still hoping for reversion to mean after many months...
You will be regularily be caught in a pricing "paradigm shift" and crushed...
Like those guys at LTCM.

This is interesting.

Whenever I have had conversations with quants - it seems to me they are doing techinical analysis without charts.

How do you see it being so different.

One guys model seems to be another guys four chartings screens of inter and intra market relationships

Once guys datamines for the reliations of advance decliners and tiki to price, the other guy looks over months of charts.

One guys mean reversion is another guys bollinger or atr bands.

Am I wrong?