Sunday, September 12, 2010

The Power of Momentum

9/20 Update:

It looks like the results using Shiller's data are no good as Shiller uses monthly average data for his index (rather than month-end), which apparently morphs the results from positive to negative in this case. Many thanks to Michael from MarketSci blog for the info.

In looking at results using actual month-end data,
it looks like the only way a month-to-month momentum strategy has outperformed is with substantially higher down month triggers (between 5-10% down months).

Initial thoughts... downward momentum trading strategies likely only work in severe down markets. Otherwise, it is rather dangerous to short an asset class over the long term that tends to be noisy (i.e. volatile / mean-reverting) and "should" have a bias to rise over the long term.


Note: Updated figures, which changes the results.


All data used in the below charts created by EconomPic was adapted from historical S&P 500 data (index + dividends) from Irrational Exuberance. I have uploaded the adapted data to Google Docs and all three indices (total return, momentum, and momentum -1.85%) to share and to serve as a check (if there are errors, let me know) for all my readers here.

To the post...


Bank of England’s Andrew Haldane (hat tip Felix Salmon) with some rather jaw dropping analysis showing that a momentum investment strategy consisting simply of buying when the previous month was positive and shorting when negative, significantly outperforms a simple value investing strategy based on the dividend discount model.

The chart below is Andrew's (for more background on the chart, go here):



Felix ponders how the results would look relative to a simple buy and hold strategy:

Still, I would have loved to see a third line, showing the results of a simple buy-and-hold strategy. Sometimes the easiest things to do are also the most profitable of all.
Here it is (going back all the way to 1871)... it turns out that the easiest thing to do (i.e. a buy and hold listed below as 'total return') was NOT the most profitable by a factor of 80 (according to my calculations, a $1 investment in 1871 is worth ~$125,000 in a buy and hold strategy vs. ~$240,000 for the simple momentum strategy described above over that same time frame).



BUT, the reason for the huge discrepancy in return is due almost solely to what happened during the Great Depression, when a buy and hold investor would have lost more than 80% of their investment, while a momentum investor would have tripled their investment over that same period.

To show how huge an impact... if we take the same data starting in 1940 (i.e. post Great Depression), the buy and hold "total return" investor would have actually outperformed by a factor of more than 8 (~$1350 vs ~$150).



BUT.... when one data mines the historical data (always a fun thing to do) and only shorts the market following monthly returns down more than 1.85% (prevents whipsawing I guess), the momentum strategy is a huge winner turning $1 from 1871 into more than $7,000,000 dollars today and significantly outperforming a buy and hold investment both pre and post 1940.



Source: Irrational Exuberance

17 comments:

DIY Investor said...

Interesting analysis. It isn't clear, of course, that one could have done anything close to this because mutual funds weren't around in the 1870s etc. etc. I liked how you looked a bit closer at the data and ran it from 1940.

Anonymous said...

In your Google Doc, I don't think your formula for total return (column G) is right. Take G1680 for example. You start with $8340.49 from the previous month-end. You have price growth of 3.93% [1197.32/1152.05-1], and a dividend yield of 0.16% [21.91/12/1152.05] for the month.

On your base of $8340.49 these are $327.74 price performance and $13.22 dividends. Instead you're showing a price return of $327.74 and a dividend of $1.83.

If you adjust the formula from

(B1680/B1679)*G1679+(C1679/12)

to

((B1680+C1679/12)/B1679)*G1679

then I think you get a final value for your total return index of about $125,000 and about $240,000 for the momentum strategy.

Jake said...

anon- fixed in the charts and doc

Anonymous said...

It would be interesting to see this on different timescales. Daily, weekly, quarterly, etc.

I remember reading a similar study using the top five or ten mutual funds in any given month back in the 90's. Inertia is your friend.

"Cassandra" said...

Transactions costs and taxes are a few obvious flies in the momentum ointment. Betting simply (or is it simply betting??!) against the crowd (for extended periods) carries its attendant risks...

Michael said...

Hi,

this looks good for the S&P, but what about other stock markets like Nikkei or DAX. I doubt that it will look as good.

Anonymous said...

Try plugging in -90% for the threshold into the spreadsheet.

This should be exactly the same as buying and holding, however you get a ratio of 140.94 ... that doesn't seem
right!

Maybe this is an artifact of downloading into excel, but I doubt it!

Jake said...

when i put in -90% it matches as if it was a buy and hold strategy, which makes sense as the market has never dropped 90% in a month. this was also downloaded into Excel, so not sure what the problem is for you...

Michael (MarketSci) said...

EconomPic - FYI...posted a follow up to Haldane's strategy here:

http://marketsci.wordpress.com/2010/09/20/re-the-power-of-momentum/

In a nutshell, it's a dud because Shiller's S&P 500 price represents the AVERAGE price for the month, not the month-end.

That means it's not reproducible (and completely falls apart when applied to actual prices).

Just my small contribution to the discussion.

Thanks,
ms

MARC said...

QUESTION
If one were to implement this. Would you sell only when the previous month end data was negative? In other words as long as each month is positive you just hold? Waiting for a down month to get out.
Appreciate any help here.
Thank you

Jake said...

Marc- in theory that is correct, but take a look at MarketSci's analysis. Looks like using month-end data (i.e. what should be used vs. the average monthly data within Shiller's data) produces a massively different result.

MARC said...

Yipes. Yes I just noticed that. Thanks for the reply/warning.
Joe Marc

MARC said...

might have sent this twice. but,
thanks Jake appreciate the reply/ warning
Joe Marc

Burt said...

Why can't we use Schiller's monthly average data? It should be possible to average the next month's price by buying a little every day.

Am I missing some statistical fluke? Something about spurious autocorrelation in the averages?

MARC said...

Burt,
I thought about that, If one were to buy 1/4 every Monday close it seems you would be getting prtty much the average prise for the month. Wish I had computing knowledge and skills, I would test that out.

MARC said...

correction, In thinking about what I just wrote, (1/4 each wk.) Since you would be holding the first weeks trade for 4 wks and the last trade for 1 week I guess that wouldn't be average. Better to allocate monthly investment into 10 units. Invest 1st weeks with one unit, 2nd week with 2 units, 3rd wk with 3 units and 4th week with 4 units, That seems to me to be a way of getting the "average" price working for you.

Frank said...

Burt, Marc,

the problem with that approach is that you don’t know during the month if the month will be up or down on a month-end basis (you always know with hindsight only), so being a gradual net-buyer during the month means you’re always assuming (before the fact) the month will be up, not down, otherwise you’d have to be a gradual net-seller, not a net-buyer.

That is a prophecy (month up or down) recorded after the fact, a classic example of hindsight bias.

If you'd be interested in Haldane’s approach using the end-of-period prices (not the average price, of course) for different time frames (daily, weekly, quartely, semiannual and annual), take a look at

http://www.tradingtheodds.com/2010/09/the-power-of-momentum/

Best,
Frank
alias TradingTheOdds
http://www.tradingtheodds.com

Share via Twitter

Facebook Share