Wednesday, September 09, 2009

Market Timing Using a Moving Average

Mebane (me-bany? mee-ban?) Faber has a well-read paper on the SSRN concerning market timing (A Quantitative Approach to Tactical Asset Allocation). I think this because the article is pretty simple, and it seems to work. It's incredibly easy to understand, accessible to anyone who uses Excel. He took the 10 month moving average rule proposed by Robert Siegel in his 2008 edition of Stocks for the Long Run, and found it worked in a variety of markets. I've seen a lot of market timing rules and find most stink, so I looked at the data myself, expecting it not to work.

I used the monthly data from Ken French's website. I went long the market if the index was above its 10 month moving average, went to T-bills otherwise, where 'the market' is the value-weighted US composite. The arithmetic return was slightly higher simply always being long (10.7% vs. 10.0%), but given the volatility of the long-only rule was 50% higher, its geometric return was about the same (9.3% vs. 9.2%). The Sharpe ratio using this index data suggests the market timing rule significantly helps one's investment, taking it from 0.30 to 0.46 in this 1926-2008 period.

US 1927-2008
Long vs. Market Timing Based on 10-Month Moving Avg.

Regular Timing
GeoRet 9.3%9.2%
ArithRet10.7% 10.0%

The result is pretty robust, in that it does not drop off dramatically using a 6-month moving average, or an 18-month moving average.

You can see that the main periods of outperformance were from the 4 big bear markets: 29-33, 73-75, 2000-02, and 2007-08. Perhaps in a simple moving average rule is a wise investment strategy. I would want to look at international data to become more certain, but it's interesting.


vJD said...

Have a look at

and subsequent posts 2 and 3...

alittleknowledge said...

However simple, the strategy may still suffer from data snooping.

Anonymous said...

@ vJD

If nothing else, the site sets a new highwater mark for % of homepage occupied by Flash ads

Anonymous said...

I read your book and I liked it.

In my experience there are an incredilbe number of choices that one has to make in a moving average system/momentum system. The incredible number of choices and the fact that performance will frequently be bad mean that few investors will ever follow one model for a long period of time.

Half the time performance will be good, half the time it will be bad. When it is bad then the psychological need to reset the play clock means the model will be tweaked and the history of one's investment process will be a pastiche of models. Most clients will not know of the model changes.

It is hard to believe that the backtested results are not sensitive to the length of the moving average you selected. Whenever I have looked at moving average approaches the length of the moving average period has always been significant. Same thing with momentum. Momentum passes through at least three stages: reversal, persistence and long-term mean reversion (using data subject to survivor bias)

There is another challenge: universe selection. The results of a momentum strategy or a moving average strategy will depend upon the universe of assets you select.

Another challenge is how frequently one should trade (even assuming a frictionless/ no transaction cost world). For instance, should one expect a 12 month moving average system implemented daily have the same return as a 12 month moving average system implemented daily.

Anonymous said...

Yeah the 10 month SMA is a natural first place to look...why not 11 months? 14.6 months? EMA instead? This has data mining written all over it. Can you seriously get 2% fee + 20% of profits from an investor by advertising a strategy with market returns at 0.46 vs 0.30 Sharpe?

Anonymous said...

Eric, a tip is to use GIF instead of JPG as the image format for graphs like this (limited colors, linear graphics rather than pictures). Or PNG.

Anonymous said...

now adjust for taxes

DT said...

For those that think Mebane's (pronounced Meb-in)system has "data-mining" written all over it you might want to look at the 10-Month SMA a bit more. The 10-Month SMA is a logical place for any technical analyst to start as it is the monthly equivalent of the 200-day and 40-week moving averages. The 200-day and 50-day moving averages are hands down the two most used moving averages on any data platform or anywhere else with a stock chart.

JimJinNJ said...

I follow and use Meb's method employing five asset classes per his book Ivy Portfolio. There is comprehensive research on the 10 month done by Nelson Freeburg.
Contrary to what one expert here suggests, a system being profitable in a range of MA lengths is quite the opposite of evidence of over-optimizing. The Faber article is free. read it before you get to cocky. look at the drawdown of this system vs SP buy and hold. on a rish adjusted basis this simple approach wins hands down. the reason: it is designed to keep you out of deep bear moves.
also the entry is month-end, no the 200 day MA on a daily basis. in effect that means you have a momentum demonstration before you enter.
Everybody has a system--most just don't know it. otherwise you would trade randomly.
also remember this system has built in risk manager. followed exactly you would usually never chase a loser down to the 56% we've just seen in the SP.

Izzahu Ionspot said...


"And because I get this question daily, Mebane is Southern (US), derived from Scottish (McBain or MacBane), and rhymes with “web-in”."

Anonymous said...

It's Jeremy.