Monday, October 15, 2007

Correlation Ideation

Let’s say, for the sake of argument, that you are intrigued by the 71% gains that MOS has logged in the past eight weeks in Portfolio A1, but for whatever reason do not want to own that particular stock. Perhaps you have an opinion that the fertilizer stocks are overbought or that a supercycle is just beginning in this sector. Which stocks should you be looking at? I recommend visits to three free web sites that can help you answer this and other related questions: Market Topology; Sector SPDR Correlation Tracker; and DeepMarket.com’s correlation tool. Each of these sites has some particular strengths that I discuss below.

My first stop to evaluate correlation data is usually at MarketTopology.com. Once there, you need to click on the Equities Markets: USA link to arrive at their “i-work” page. From here, just enter the ticker and either click on the ‘Calculate’ button to return data in a table (usually the better choice) or try ‘Map’ to see a graphical representation of the securities with the highest correlation. There are several other boxes you can use to filter the results; these should be self-explanatory and ripe for experimentation. In the case of MOS, the four highest correlations returned are POT, CF, AGU, and TRA – all companies in the fertilizer sector. The next two most correlated securities are both materials ETFs: VAW, the Vanguard Materials ETF; and IYM, the iShares Dow Jones Basic Materials Sector Index Fund. It is these types of discoveries that make tangential company and sector research more fun and interesting. Note also that the table also has a column for ‘Average Daily Volatility’ for those interested in identifying highly correlated stocks or ETF that are significantly more or less volatile than the baseline security.

Among the three sites discussed here, the ease of use award would probably go to the Sector SPDR Correlation Tracker, which simply asks for a ticker and generates three lists: highest correlation sector SPDRs; highest correlation stocks/ETFs; and lowest correlation stocks/ETFs. As an added bonus, you can generate java comparison charts for any four securities on these lists for additional analysis. Let’s say you are interested in the FXI, but prefer to take a position in an individual stock instead of the ETF. Using the Sector SPDR correlation tracker, you would be pointed in the direction of CHL, CEO, LFC, and BIDU.

At the bottom of the list is DeepMarket.com, which scores high for content, but low for aesthetics. Their correlation tracking tool lists the top 5 highest positive correlations and (lowest) negative correlations for the past 10, 30, 100, and 200 day trading periods. The site provides the correlation coefficient and a rudimentary line chart, but little else. What I do like is the ability to slice and dice correlations over four different time periods (the longer time periods probably provide the most value,) but apart from that feature, the other two sites are to be preferred.

I should mention that while I have focused on positive correlations here, each site provides a list of the most extreme negative correlations as well. While these generally are not as strong correlations as the positive correlations, they do provide and excellent jumping off point for someone looking to add securities to a portfolio that may be inversely correlated to some of the portfolio’s riskier holdings. This type of approach is admittedly more art than science at the individual security level, but for those unable to evaluate portfolio level correlation data, it is a substitute worth exploring.

8 comments:

  1. Unfortunately correlations only measure consistency in directions of two graphs (scatterplots) over a certain period- unspecified here. Therefore if two stocks have gone up (or down) consistently over a period their correlations will be high, although they may be totally unrelated substantially, and there may be major inconsistencies over time. I tried out Sector Spider Correlation tracker with DRYS, and the results can be seen at http://stockcharts.com/charts/performance/perf.html?drys,flr,exm,cam,fti,hhd,jec,lfc,rimm,gxc

    DRYS and EXM are in the same sector, but the remainder is not. Performance is also highly variable, as you can see on the perfchart.

    ReplyDelete
  2. All good points, anon.

    For the record, I ran DRYS through MarketTopology.com and had considerably better results.

    I am a fan of that particular site and suggest that you play around with that one if you are looking for correlation data that is easier to filter and manipulate. Of course, you'd need to run through a bunch of data points to come up with some scientific basis for a top choice, but my personal experience has been best with MarketTopology. (Unfortunately, I don't see any indication of the time period they use for the correlation data, but that doesn't water down my recommendation.)

    Cheers,

    -Bill

    ReplyDelete
  3. I have a feeling there is a huge autocorrelation bias ?

    ReplyDelete
  4. Regarding autocorrelation bias, while I'm far from expert on the subject, it would not surprise me.

    Thanks for the insight.

    Cheers,

    -Bill

    ReplyDelete
  5. Bill, maybe it was just not enough tests of the data, but I was unimpressed with the select SPDR correlation tracker. It seems to defy economic and market intuition. I tried a couple of large cap duopolies as a test: KO & PEP, and CL & PG.

    Market topology "got it right" while PEP and PG where nowhere to be found on the SPDR list.

    Also unimpressed that big losers like BZH are (surprise, surprise) uncorrelated with big winners like ISRG. That's not correlation trading, that's just good stock picking. Maybe I don't understand the point of the exercise.

    ReplyDelete
  6. At MarketTopology, time series are de-trended, i.e. correlations are computed on daily fluctuations (1 year) and not absolute price.

    The result is that the measure is not influenced by the overall trends, only short term cooperative movements are highlighted.

    Hoping this is helpful information.

    Xavier

    ReplyDelete
  7. Thanks for the clarifications, Xavier. These are very helpful.

    Cheers,

    -Bill

    ReplyDelete
  8. It's really a great and useful piece of info. I am happy that you simply shared this helpful info with us. Please keep us informed like this. Thank you for sharing.
    Take a look at my weblog diets that work for women

    ReplyDelete