How Momentum Traders Can Avoid Vicious Drawdowns

27th Nov '12 by Edward Croft

5 comments 12233 reads

Over recent months we've been considering the fact that trading momentum - buying winners and selling losers - works extremely well. In fact, the returns from the strategy are much higher than those from value or growth strategies, albeit at the cost of greater volatility. We've seen that momentum strategies can 'crash', as was witnessed in 2002/3 and 2009, and many hedge funds as a result have given up on it as a trading strategy.

However a groundbreaking research paper by David Blitz, Joop Huij and Martin Martens published in the Journal of Empirical Finance in January 2011 introduced a novel way of investing in momentum that appears to hold some hope for those investors that can't help but chase winning stocks. In it they promoted the idea that stripping the market's influence out of each stock's measured return can seriously improve the profits from a momentum strategy.

To illustrate these ideas, it may help if one thinks of the stock market as a sea of fish. Historically these fish have all been very independent minded and while they've always been affected by ocean currents, they have swum very independently. But since the turn of the millennium the fish have begun to behave more and more like a shoal, more and more like synchronised swimmers, ebbing and flowing in unison.

Before 2000, the average stock's daily move was perhaps 20% correlated with the overall market, but this correlation has risen year after year to reach 50% by the beginning of the year (see image below). Given the growing influence of ETF and index funds and the dominance of macro-politics on market swings, this is perhaps unsurprising, but it's making it a much harder market for traditional stock pickers to outperform in.

Andrew Lapthorne, global head of quant research at Societe Generale, states that when "macro and market events and not company fundamentals drive stock prices then using momentum as a trading signal becomes fraught with danger".

The solution, as described by Blitz, Huji and Martens, is to focus on the momentum of the stock not explained by the impact of the overall stock market. By stripping out the market sensitivity from each stock's share price returns they can then be ranked by what they call the stock's "residual momentum".

Soc Gen, when testing this strategy, confirm that globally annualised returns are on…

Unlock with Email

Unlock with Google

Already have an account?
Login here

Mark Simpson

30th Nov '12

Hi Ed,

Been trying to get my head round this article and see how the principles can be applied to create a screen which is possible to implement without complicated regression analysis. Here's a bit of thinking out loud...

What they seem to be saying is that they are taking residual momentum instead of absolute momentum, i.e. the momentum that remains when you correct for other factors which influence stock returns. They do this by running the regression analysis in a formation period using the so called fama and french factors which are beta, value & size.

Note this isn't far off a subject you must be fed up of me talking about, expected return factor models, except here they are simply regressing to remove the influence of some factors to build a performance model of 1 factor - momentum - rather than including that itself in their regression.

Now I've never been a huge fan of the fama and french model. Particularly when the factors of value and size are used as risk factors to conclude that there is no risk adjusted value or size anomaly!! Also the fact that Haugen clearly shows that beta is negatively correlated to return when you apply a correct methodology kind of invalidates the whole EMH in that area.

However those concerns aside it does make intuitive sense that if you are going to follow a momentum strategy you want to base it on companies that are generally outperforming because the company itself is outperforming, forming a habit of beating market expectations and broker upgrades which are the fundamental factors which drive momentum returns. What you don't wont to be confused by are the other factors that drive stock returns - value, companies that are out-performing because they are cheap, small or just going up because the market is going up and they have a high sensitivity (beta) to market movements.

Now to generate the returns to value and size they are probably just running regressions against those factors which can change during market cycles. But of course the past does not always predict the future so what measurement period you pick really does matter. But what if we simply take very long term data then many published studies show a difference of about 5% the highest and lowest quintile p/e ratio pa and about the same between the largest and smallest companies. We may not capture the changing fashions of value and growth or small cap big cap investing but then we won't suffer from the change periods when fashions change.

So if we have a 1 year RS model our corrective factor would be say Quintile 1 (cheapest) take off 2.5% through to Quintile 5 (most expensive) add 2.5% and Quintile 1 (smallest) take off 2.5% through to Quintile 5 (largest) add 2.5%.

Then we need to correct for beta - but this is easy since the beta's for individual stocks are already in the database. So if we are using or 1 year RS and e.g. our beta is 2 and the market has gone up 10% in the last year then we need to take 10% off the RS to account for the higher than market average beta (the RS has already taken off 1xmarket movement to get the RS)

So we have:

True RS 1Y = RS 1Y - (1 - beta) x Last year's market movement - (Value Quintile - 2) x 2.5% - (Size Quintile - 2) x 2.5%

If I'm right, this gives you a simple estimate of the true increase in the share price without the influence of the market, how sensitive a stock typically is to market movements, its size or cheapness.

Do you think I'm on the right lines here? Any chance this can be replicated in the screener either with the existing setup or with a simple mod?

Cheers,

Danger

Edward Croft

Hi DS,

Since writing this article I've actually been experimenting with the Fama & French factors that are published on their website. They provide publicly the monthly returns for each of 4 factors - the market - risk free rate, HML (value or high book/market minus low boo/market), SMB (size or small minus big), and also WML (momentum or winners minus losers). It's very straightforward to run regressions against this data using either excel spreadsheets (downloadable from the web) or common code libraries. Have a look here - http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html#Developed

I think the danger with the approach taken in your comment is that each year the market moves wildly differently - weighted more towards some factors than others - using the average approach might give unexpected results.

Anyway - we are getting into higher brow areas here. The goal I have is to start creating our own library of factor loadings specific for each market (i.e. UK at first) - especially the Haugen set you mention - but also factors based on quality and financial risk - once we've done that we may start trying to build some serious portfolios based on this stuff. But it's probably another 12 months away if I'm realistic - by all accounts you need 36 months history to do it well.

Currently we have been regressing the Fama/French factors against various funds, screens and the market as a whole. It's quite fun to see how little alpha the top fund managers really have - but that's for another article !

Hi Ed,

Good work on starting building he data set for expected return factor models - since you have such a great set of factors available in stockopedia it makes sense to include as many as you can.

I think the danger with the approach taken in your comment is that each year the market moves wildly differently - weighted more towards some factors than others - using the average approach might give unexpected results

Yes you are right the idea of their model is the 'time-varying exposure to the fame & french factors' which does mean that they are trying to asses what the current market returns are due to size and value and correct for those.

However this still doesn't make a lot of sense to me, just becasue the market favoured value stocks in the past year (or whatever period you are using to calculate the return to value) does it mean that it will favour value stocks in the follow year (or whatever your optimal holding period is)? I've never seen any research showing statistically significant positive serial correlation to value or size and intuitevly I would expect it to be negatively serially correlated - i.e. mean reverting. Have you come across any research or run your own regressions on this?

The other thing with this is that they are running a residual momentum strategy so they are using the fama & french factors to remove the influence of value, size and beta from their measurement of momentum and then going long the shares with the greatest momentum. This means that even if you assume that there is positive serial correlation for value and size, when value is performing well and will do in the future they are favouring holding growth stocks and when small cap stocks are going to do well they are favouring holding large caps! Doesn't sound a great strategy to me and actually the opposite of applying expected return factor model.

However, although I can't see the logic of correcting momentum for time-varying value and size efects, correcting a momentum strategy for a share's beta makes a lot of sense to me. It means that you don't just end up holding low or negative beta shares after a significant market decline or high beta shares when the market has been hitting new highs. And as you point out in the original article more and more shares are being influenced by the overall market direction rather than individual performance. Maybe the risk-on, risk-off nature of current trading is having this effect or it's just a general part fo the increaingly correlated returns fo asset prices that we've seen ove rhte recent past. Whatever it is it makessense to correct for it in a momentum strategy.

So while you are right that if you want to correct for the time-varying risk factors of value and size you can run a regression or just look them up from ken french's published data I'm not convinced you actually need them.

A good momentum screen imo opinion would be:

high RS

beta < 1 so we know that the RS is 'real' (and contrary to the EMH low beta is a return factor in its own right)

low EV/EBITDA. (I know you are not a fan of this in momentum strategies but let's just call it an emotional crutch - if I'm going to steel myself to buy something that's gone up a lot in the last year I'm at least i'm not going to stack the deck against me by overpaying for uncertain future growth!!)

Some quaility measure - roce or piotroski

Earnings Surprise 1Y >0 (I'm also a fan of understanding the fundamental reasons behind momentum too so I'd be looking for evidence of earnings surprise to be a driver of the excess return.)

Maybe I'll set this up as a screen to track.

Cheers,

Danger

pka

6th Aug '13

I wonder whether simply dividing a share's Relative Strength by its Beta would be a useful way to devise a Beta-corrected Relative Strength measure. It would have the merits of being both easy to calculate and easy to understand.

Mark Carter

28th Dec '14

I am far from understanding this paper. It's quite a long one. Equation (1) seems to be key to whole thing.

I think figure 1 on page 56 is particularly striking; and perhaps proves the opposite of what the authors intended. Total return momentum was actually superior in the period 1940 to 2000. So a lot has to do with the starting point! Note also that the TTM (total return momentum) had a big drawdown between 1930-5, only to see it massively outperform RM (Residual Momentum) subsequently.

The chart also shows stagnancy, and another massive drawdown after 2005 (2008, perhaps?) after a period of stagnation in 2000. So you could argue that TTR is now permanently stagnant, or else due for a massive revival. The choice is yours.

How momentum traders can avoid vicious drawdowns

Unlock the rest of this article with a 14 day trial

About the Author

Edward Croft

5 comments

Unlock the metrics for all these stocks and more...

Unlock the metrics for all these stocks and more...