Tracking the 52 Week High: Does Trend Following Work for Stocks?
In Brief
An investing screen based on buying stocks that are close to their 52 week high (and/or selling stocks that are close to their 52 week lows), particularly in industries whose stock prices are close to their 52 week highs (or lows). Similar to other forms of momentum investing, this appears to work because investors tend to under-react to positive (or negative) information about those kinds of stocks.
Background
Every day, financials newspapers like the Wall Street Journal, the Financial Times, and the South China Morning Post publish lists of stocks whose prices have attained new 52-week highs and lows. Many investors approach such a list with trepidation, understandably assuming that this may mean that stocks prices are approaching “nosebleed” territory. However, recent research suggests that this may be exactly the wrong conclusion, perhaps precisely because everyone else is thinking the same thing. As has been discussed elsewhere, academic researchers have found that stock prices look to have "momentum". More recently, work by researchers George and Hwang published in the Journal of Finance has shown that the closer a stock's current price is to its 52-week high, the stronger that stock's performance in the subsequent period. They conclude that “price levels are more important determinants of momentum effects than are past price changes”. In subsequent work this year, they also found an industry level 52 week high effect which was even more pronounced.
Why Does it Work?
George and Hwang surmise that investors use the 52- week high as an “anchor” against which they value stocks, thus they tend to be reluctant to buy a stock as it nears this point regardless of new positive information. As a result, investors underreact when stock prices approach the 52-week high, and consequently, contrary to most investors' expectations, stocks near their 52-week highs tend to be systematically undervalued. Finally, when information prevails and the 52 week high is broken, the market “wakes up” and prices see excess gains. Similarly, when bad news pushes a stock’s price far from its 52-Week High, traders are initially unwilling to sell the stock at prices that are as low as the information implies.
They conclude:
“Traders’ reluctance to revise their priors is price-level dependent. The greatest reluctance is at pricelevels nearest and farthest from the stock’s 52-week high. At prices that are neither near nor far from the 52-week high, priors adjust more quickly and there is no pronounced predictability when information arrives”.
Screen Criteria
The underlying idea is that long positions are taken in stocks whose current price is close to the 52-week high, so some indicative criteria might be:
- Volume above bottom 20% of the database
- Market Capitalisation above bottom 20% of the database
- Proximity of stock to 52W high is within the top 30% of the database (an "individual stock 52-week high" strategy).
Or alternatively:
- Proximity of stock’s industry to its 52W high is within the top 30% of the database (an "industry 52-week high" strategy).
Stocks are assumed to be held for 6 months and then recycled.
How well does it work?
The original research found that, between 1963 – 2001, the average monthly gain to this strategy assuming a 6 month hold was 0.45% - “about twice as large as those associated with other momentum strategies”. Returns were also very persistent. Work by JP Morgan came to similar conclusions about the effectiveness of the strategy.
George and Hwang's latest paper found that that:
"a strategy that buys stocks in industries in which stock prices are close to 52-week highs and shorts stocks in industries in which stock prices are far from 52-week highs generates a monthly return of 0.60% from 1963 to 2009, roughly 50% higher than the profit from the individual 52-week high strategy in the same period".
This was then boosted to 0.78% monthly if stocks were held for three months only.
The academic research was done using a combined long/short strategy. However, the latest paper confirms that investor underreaction to positive news accounted for more of the profits than investor underreaction to negative news, suggesting it may be reasonable to implement the strategy on a long-only basis.
Geek Stuff
Firstly, as is the case with other momentum strategies, the researchers found that the return to the "individual stock 52-week high" strategy is actually negative in January (apparently because loser stocks tend to rebound in January). Interestingly, though, the "industry level 52 week high" strategy did not have this characteristic. Secondly, the more recent industry level work also indicates that the 52-week high effect is mainly driven by investor underreaction to industry, instead of firm-specific, information. Even the "individual 52-week high strategy" seems to work best among stocks with high R-squares and high industry betas (i.e. stocks whose values are more affected by industry factors).
How can I run this Screen?
On Stockopedia Premium, of course! Sign up now for access to our exclusive Beta!
Sources:
It’s worth reading the original October 2004 Journal of Finance paper here, The 52-Week High and Momentum Investing, as well as the subsequent 2011 paper, Industry Information and the 52-Week High Effect. Thanks is due to Mark Carter for alerting us to this work.
Other Sources:
Disclaimer
This is not financial advice. Our content is intended to be used and must be used for information and education purposes only. Please read our disclaimer and terms and conditions to understand our obligations.
7 comments
The x axis is just price/(256-day high). Stockopedia's ratio "% vs 52w high" is the same thing but expressed as a percent difference from the high. So e.g. 0.70 is the same as -30%, 0.90 is -10% and 1.00 is 0%.
The y axis is called the "likelihood ratio" and is a bit tricky to explain without going into Bayes Theorem and how it used to combine different evidence to give an overall probability of some event. But basically values greater than one mean that "success" is more likely for the particular value on the x-axis and values less than one mean it is less likely. I have defined "success" here as "the price after 6 months is greater than the original price" which is similar to the metric Stockopedia use in assessing stockrank performance (although they typically use a 1-year period.)
If you want to use this info in a Stockopedia screen, I would suggest that you set a rule which requires the "% vs 52w high" ratio to be greater than a point where the relmax256 variable is above 1 on the graph. So about 0.82 or -18%.
However the effect is not that great on its own (the likelihood swings from 0.8 to 1.2 roughly) so it needs to be combined with other rules to improve the predictive power (which is pretty much also the Stockopedia philosophy I think.) It would be nice if we could access graphs like this for each Stockopedia ratio so that we had some idea how to set the ratio.
But lots of caveats apply. Cig has mentioned "curve fitting" where you optimise too hard for what has happened in the past. There are also problems with defining a useful definition of 'success'. A range of interesting (and possibly misleading) things happen if you ask for prices to be at least 20% higher within 6 months for example. And of course all of these predictions are based on probability not certainty.
Nick, Thanks for that explanation. The graphs make a bit more sense to me now. Out of interest, how many years of data did you use to construct your graphs?
My FTSE100 stock data goes back to 2006-01-01 but the other stocks only go back to 2010-01-01.
It turns out that my original data was faulty. Cig mentions the danger of curve fitting, and in fact I had accidentally dropped most of the stocks from the data and hence was seeing a pattern from a very small sample.
When I corrected this to look at the whole FTSE350 the 52-week and 26-week versions work in a similar way as you can see in the attached graphs below. (In practice I have used 128-day and 256-day periods rather than 26-week and 52-week periods.) Likelihood ratios above 1.00 (grey line) correspond to situations where the performance is better than the 'prevalence' ( = what you would expect without using this criterion to select) and below 1.00 where it is worse.
The main reason that I was using only the FTSE350 is that when you start to consider stocks which are smaller than this we see much more outlier behaviour which can also cause curve-fitting behaviour. However, for completeness here is a similar 256-day plot using 660 stocks in total which clearly has a similar shape to the FTSE350 one.
Nick, I'm afraid that I don't understand your graphs, and I expect the same is true of many other people. Could you provide some more explanation of what they mean?
I have tried back-testing this with my own historical data and I found that (using the FTSE 350) the results are much more stable using a 26-week high than a 52-week high.
I computed a likelihood ratio by dividing the probability of "success" for a given value of the 52-week high / current price by the probability of success "a priori" (based on the entire population.) I then computed it again for a 26-week high. "Success" in this case is a price after 6 months which is higher than the original price.
With a 26-week high, using samples from a range of periods and stock groups, the point at which the likelihood ratio moves though 1 is somewhere near -20% and this is quite stable if I vary the meaning of "success" and/or the periods which I sample. However with a 52-week high there are often periods where stocks which have fallen in value by -30% or even -40% are favoured. But it isn't stable. The curves change shape as you make small changes to the definition of success or sampling period. So it is much less useful for building a strategy around.
Now maybe my data or method is at fault but for now the suspicion is that "Mr Market" has quite a short memory when it comes to momentum and six months is more than enough to establish a momentum trend. Also, over this period, there may be something human behavioural about -20%. Perhaps a lot of traders use a trailing 20% stop-loss for example.
Anyway, it would be very interesting for me if you were to provide a 26-week high (and low) as well as the 52-week high (and low). I would also be interested to hear what results you get from a back-test with your own data.
There was a curious paper (on US data if I remember right) that found that 6 months lagged by 6 months (e.g. look at the momentum from 52 to 26 weeks before the trade time, ignoring the 6 months immediately before it -- yes, it's counter-intuitive) worked better than all timeframes to date. You could try that on your dataset. I forgot the guy's theory, but I thought upside on the left-hand side of a 1 year chart might make them likeable to TA fans.
On methodology, 350 stocks is small, so beware of curve fitting.