Absolute Momentum

6th Nov '15 by Tom Firth

14 comments 12777 reads

Things, once set in motion, continue until influenced otherwise. We intuitively know this physical law and, thanks to research such as that by Jegadesh and Titman, are beginning to understand it better in the context of asset prices. Less frequently discussed, however, is the difference between two methods of measuring price momentum. The first, and most commonly used, is relative momentum – the momentum of a price as compared to other prices. The second is absolute momentum – the momentum of a price in relation to itself only. We rarely consider this in the physical world where we are naturally concerned with absolute momentum – after all, why would I care that one bullet is slower than another if it’s still going to kill me? When it comes to investing however, the two concepts can lead to surprisingly different results…

The problem with relative momentum

It’s the Olympics and you’ve won tickets to watch the long jump event. The competition is strong this time around and all competitors are performing well. So well, in fact, that you decided to spice things up a little and make a bet with your friend that before the event is up, a new world record will be set.

Above all others, one athlete stands out and is consistently putting in jumps at a level seemingly beyond the rest of the field. He’s the one, you think, as he steps up to the starting point for his final jump. Head down, he sprints for all he’s got, picking up more momentum than you’ve seen all day – this is it! He sails through the air with the best jump of the competition to seize the gold…but not the world record. As the athlete celebrates his victory, you hand over your hard earned cash and hang your head to reflect on where you went wrong.

The answer, you realise, was that you relied too much on relative momentum. In order to break the world record, the athlete didn’t just have to be better than everyone else on that day, but had to overcome an absolute level of performance.

In the context of investing, this distinction may be key. In the midst of a bear market, the share with the best momentum may still exhibit negative momentum! By sticking to a relative momentum strategy we are prone to buy stocks with poor momentum in times of market duress, with unpleasant implications for…

Unlock with Email

Unlock with Google

Already have an account?
Login here

pka

7th Nov '15

Tom,

Thanks for the very interesting article, in which you wrote: "Perhaps this can be remedied (or even improved) with other criteria though, i.e., finding all stocks that meet absolute momentum criteria, and then filtering for value, quality etc. Certainly strategies such as The Naked Trader’s that place some reliance on absolute momentum (by stating that 1 year price change be positive in addition to requiring positive earnings momentum) have performed well."

As I guess you know, Stockopedia has a "Naked Trader-esque" Screen here:

http://www.stockopedia.com/screens/naked-trader-esque-screen-164/

it is the 6th best-performing screen out of 66 Guru Screens on Stockopedia based on its annualised performance over the last 4 years:

http://www.stockopedia.com/screens/?sort=annualised

One of that screen's rules is: "% Price Chg 1y > 0", which is the absolute momentum criterion. I wonder how much of that screen's excellent performance over the past four years is due to that particular rule.

I'm thinking about adding a "% Price Chg 1y > 0" rule to the other rules that I currently use for my high StockRank 'farmed' portfolio. I think the advantage of a "% Price Chg 1y > 0" rule is that it should be very easy to apply each time that I rebalance my portfolio, and hopefully it would be effective in helping to preserve the capital value of my portfolio if there is a severe bear market. I think it is a bit like the 'Stop-Loss' rule which some people use to protect the downside of stocks in their portfolios. A disadvantage of using this rule is that it would increase the number of trades I do, so it would increase my transaction costs.

Blissgull

Surely those shares with the highest relative strength must also be those shares with the highest absolute strength?

This looks like a strategy for timing the market. In effect it says buy the strongest shares in a bull market, sell the weakest shares in a bear market.

Tom Firth

8th Nov '15

Hi pka, thanks for the feedback - I have indeed seen the screen on Stockopedia and note its excellent performance too! There are a few momentum related criteria in there, a mix of earnings (EPS, DPS and Sales) and price (1y change and vs 52 week low).

Essentially, the screen finds companies with positive absolute momentum and makes a couple of adjustments for size, leverage and price. You'll also notice the requirement that the spread be < 400bps.

It is hard, as you say, to identify exactly how much is due to which rule though but I think it is fair to say that momentum is the driving force overall! I reckon that spread requirement is a life saver, especially as the screen chooses to focus on smaller companies.

Not sure how Robbie Burns performed in the last big bear market to be honest, I'd be interested to see, I might have a look on his blog in a minute. A lot of momentum strategies suffer from extreme drawdowns in bear markets but perhaps his positive absolute momentum criteria help as is suggested by the study.

Tom

Hi Blissgul,

Most of the time, your comment is absolutely spot on. In bull markets the set of absolute strength shares contains the set of relative strength shares. In the paper by Gulen and Petkova they find that the absolute criteria explain all of the performance of the relative criteria but not vice versa...interesting in itself.

It is not just in bear markets that the strategy is short though, most of the time it will be both long and short and it is only in the more extreme circumstances that it cuts off one direction. I wouldn't describe it as a market timing strategy, although I can see why it may appear that way. It simply says, if enough companies have returned above(below) x% in the last 12 months, then buy(sell) them all.

Check out this graph from the study (figure 3 on page 38) which shows how many firms are held in the portfolio in each direction:

You can see that most of the time there are a large number of stocks held in each direction, at one point there are about 2500 stocks in the long direction!!

This makes the strategy highly impractical for all but a but a quant fund (and even then it could be very expensive). For me, what it does show is the importance of ensuring you are exposed to genuine positive momentum if going long (and vice versa if short), and not just positive relative momentum. There are times when there are very few stocks indeed with high positive momentum yet a relative strategy will blindly buy the top decile anyway which is likely a key driving force behind the drawdowns.

Tom

peterg

The study appears to be comparing two very different strategies, and the biggest difference is not the use of absolute or relative momentum, but the use of a long short strategy. If you look at how the absolute and relative strategy returns deviate, what the results effectively show is that a long/short strategy will outperform a long only strategy when markets are falling. It's a worthwhile observation, if hardly new, since it does point out that if you can protect yourself from falling markets successfully you can outperform. However, I'm not at all sure it tells you much about the relative merits of absolute and relative momentum strategies is used in the same, long only, way.

Peter

Tom, You wrote:

"You can see that most of the time there are a large number of stocks held in each direction, at one point there are about 2500 stocks in the long direction! This makes the strategy highly impractical for all but a but a quant fund (and even then it could be very expensive)."

A practical way that the strategy could be implemented by a private investor is to always select, say, 30 stocks from however many there are available in each direction (i.e. long and short), unless there are fewer than 30 stocks available in a particular direction in which case one would not invest in that direction. For example, if there are 100 stocks available in the long direction, one would select 30 from them in some manner. I think selection by random sampling would be likely to give results closest to those of the paper. But one could instead select the 30 with the strongest absolute momentum in the relevant direction, or the 30 with the best (or worst if short) value using whatever criterion one prefers, and/or one could try to avoid buying too many companies in the same stock market sector to ensure some diversification of the portfolio.

9th Nov '15

Hi Peter,

The study is comparing two long/short strategies, one that selects its portfolio based on relative strength and one on absolute strength.

You still raise a valid point however in that the study does not accurately tell you about the relative merits of the long only aspect of either strategy. Check out the blog post I linked in the article, here's the link again:

http://blog.alphaarchitect.com/2015/10/13/absolute...

They perform the test again but split out the performance into the long/short sides and find the following:

You can see that the two long portfolios (abs str winners and rel str winners) both generate very similar CAGRs. However you can see that the abs str losers portfolio did a much better job of identifying stocks that would underperform.

Bearing this in mind, I don't think your claim that the test shows the outperformance of a long/short strategy over a long only is accurate. I think it does provide some evidence that an absolute momentum strategy better captures the impact of the momentum effect than the relative momentum strategy, it's just that on the long side there isn't much it.

It's a good idea but there is still an issue with sampling from a population that large.

I have been lazy and am short on time so have opted for a sample size calculator instead of doing the maths myself:

http://www.raosoft.com/samplesize.html

Play around with the numbers a little but across a population of 2,500, if I say a 5% margin for error and a 95% confidence interval, then assuming a 50/50 distribution my recommended sample size is 334. So to be 95% confident that I would capture the effect exhibited by the population, I would still have to choose 334 stocks at random...which remains an enormous portfolio.

I think the better option is to first gain evidence as to whether the performance is still strong at more stringent cut off points, i.e., the top/bottom 5% of the historical distribution or/and to combine the strategy with other criteria like value/quality etc.

Tom

Tom, you wrote: "Play around with the numbers a little but across a population of 2,500, if I say a 5% margin for error and a 95% confidence interval, then assuming a 50/50 distribution my recommended sample size is 334. So to be 95% confident that I would capture the effect exhibited by the population, I would still have to choose 334 stocks at random...which remains an enormous portfolio."

I think you are being too stringent on the error margin and confidence interval. If one used less stringent criteria, the acceptable sample size would be greatly decreased. The disadvantage of using a random sample of, say, 30 stocks from the qualifying population is that in any particular year the deviation of the sample from the paper's results might be large in either direction. However, I would expect that, over a reasonable number of years, the deviations in the two directions would tend to cancel each other out, and the cumulative results from a randomly selected 30 stock portfolio would be similar to those of the paper.

Given the randomness of individual stock returns, I really wouldn't be comfortable with a higher margin for error - if anything I would want it lower.

Assuming we keep it at 5% then a sample size of 30 will only give us a 41.5% confidence interval. Again, I'm not comfortable with that. True, all this says is that result shown by the study will be within the range suggested by our sample 41.5% of the time and it doesn't say whether the sample result will be higher or lower...but I don't like those odds.

Tom

Thanks Tom,

I'd clearly missed the fact that the relative strength portfolio was also long/short in the detail.

Peter

cig

It's a detail here, but there there's usually a strong clustering effect around sectors or stock types with factor investing, so to get a given level of confidence you need more stocks than a formula assuming uncorrelated observations gives.

This is true though this mixes badly with leverage, including emotional leverage: if an investor does -50% in a given year with a concentrated sample due to sample idiosyncrasies (let's say a typical broad sample of the same strategy did -10% that year), will the thought that statistically they are due(!) +50% worth of randomness over the following few years of narrow sampling be enough for the investor to persist?

Good point - I was just using the sample size calculator to get a rough baseline figure to be honest. I'm not really in favour of the random sample method here and think that the more we drill down into details like that the less happy I'd be to follow it!

Interesting thinking re the emotional aspect of such a technique in post 13 by the way - not only would you be worried about underperforming the index but also about underperforming your own system!

Absolute Momentum

Unlock the rest of this article with a 14 day trial

About the Author

Tom Firth

14 comments

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

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