Tracking Mean Reversion After Bad Months

Does your investment approach have you in the dark? Feel like having an advisor or portfolio manager is like the blind leading the blind? Maybe it is time to take control of your portfolio with a systematic approach.

I have talked a lot about various strategies for systematic trend-following and mean reversion on World Beta. In "Time to Put Money to Work" we examined what happens after really bad months in asset classes.

What were the key take-aways?

- It does not pay to buy an asset class after a really bad month for the following 1 month.

- 12 months later the return is not much different than average.

- 3 and 6 month returns, however, are stronger. You pick up on average about 3-4% abnormal returns buying after a terrible month. That annualizes to about 10% per annum.

A simple strategy would be:

After an asset class has a terrible month (i.e. MSCI EAFE in January), wait a month then take a 2 month position. i.e. buy March 1 with a two month hold. Investors could simply overweight a position or use options.

That post performed nicely. Buying March 1 with a two month hold (using the iShares (EFA)), I show a return of 5.9%, and foreign emerging (EEM) would have done about 5%, and foreign REITs (RWX), would have done about 4.9%.

June was pretty awful in US Stocks, Foreign Stocks, and REITs in particular (returns at the end of the post). A simple strategy would be to wait a month, and then buy August 1st for a two month hold. Corresponding ETFs include (SPY), (IWM), EFA, EEM, (IYR), and RWX (and there are plenty of substitutions). I consider the trigger to be -10% in equity-like asset classes, and -5% in US bonds. One could also drill down this model into smaller asset classes, but I have not tested the short term mean reversion there. Some truly awful performances for June include:

Financials (XLF)
Netherlands (EWN)
Sweden (EWD)
India (INP)
Infrastructure (MGU)
Private Equity (PSP)
Foreign Private Equity (PFP)

Comments

[billb:]

Interesting. Have you tried the inverse? In other words, take the best performing asset classes for a given month, waited a month and gone long? What tools did you use to test this theory?

2008 Jul 02 07:40 AM

[SmartETF:]

Our research concludes that you are on the right track but your focus is too linear. Mean reversion over the long-term is an academic boon for getting a Nobel Laureate designation but it does not translate into a workable application in the real world. For example, MVO demonstrates domestic equities have returned 10% over 80 years. Therefore, you should get a 10% return on average. In the real world, the domestic equity market is down over the past 1, 3, and 10 years; yes 10 years. Granted it worked in the 80’s & 90’s, but not the 60’s & 70’s, and definitely not this era. It’s like a broken clock that is right twice a day; it is devoid of market cycles.

Short-term MVO is very interesting and much more meaningful. The question is, and will always be, what time frame is best for analyzing the time series of data (aka, time parameter estimation). I think you are off track when you try to curve fit your data by selecting a particular number of months. Markets don’t move in a linear pattern like monthly. You will have much greater success by rebalancing when markets move by a defined level of volatility or price (or both). Take volatility as an example, last February the market hit an extreme level of volatility (and price drop); buying at the level would have been very profitable. It is these extreme moves (up & down) that create the fat-tails of distributions and are reflected in the extreme technical patterns like Relative Strength. A more scientific approach is to go with a Noble winning approach from 2002 (in effect tossing the MPT model from 1959) and incorporate Generalized Auto-Regressive Conditional Heteroskedasticity (GARCH) which examines the clustering of data; basically, a scientific approach to short-term mean-variance. The analogy is MVO works like the Farmer’s Almanac for predicting weather; whereas GARCH acts like the Doppler Radar. Alternatively, you can use price and volatility movement to create a poor man’s GARCH model to track short term mean-variance. Cheers -

2008 Jul 02 01:47 PM

[WorldBeta:]

Thanks Smart ETF - your comments are some of the few that are worthwhile and intelligent.

I agree with you for the most part, and I examined a similar vol clustering in an old post on World Beta. Shoot me an email when you get the chance.

2008 Jul 02 02:21 PM

[WorldBeta:]

worldbeta.blogspot.com...

2008 Jul 02 03:06 PM