N=All

Reading a great book (shown on left) about "Big Data".  It talks about how companies such as Google, Fair Isaac, Target, Amazon etc can use the messy but copious data they collect to predict things like whether you will pay your bills, what book you might want to buy, your income level or whether you might be pregnant.  It does seem a bit "Big Brotherish", but some of what is done makes things better for consumers.

N=All is the reversal of standard statistics where you take samples (think about polling by Gallup) and instead you use all the data as the ability/cost to collect it, store it and analyze it is decreasing daily.  One area that requires a leap of faith (certainly for a statistician like me) is to be just satisfied with correlation and not to worry so much about cause and effect.  High correlation means that you can measure when one variable goes up the dependent variable tends to go up as well.  The classic example of correlation versus cause and effect is that there is a high correlation between the number or priests in a city and the gallons of beer drunk in the city annually.  But just because there is a high correlation does not mean you have cause and effect (the priests (we hope) are not the ones drinking all the beer).

Since this is a blog about Magic Formula Investing, there is certainly a "big data" element to MFI.  But we are far away from "N=All".  Greenblatt surely did not pull in all descriptive variables when coming up with earnings yield and return on capital as indicators of stocks that will out-perform.  Arguably, you could build a model using additional variables that increase predictive power.  You could say I have done (informally) a bit of that.  I have done studies that look at correlation (though I have not technically calculated the coefficient of correlation) between certain variables and returns.  Some seem to be cause and effect related (at least intuitively, such as dividend yield), while others seem more spurious, such as number of letters in ticker symbol or first letter.

When I retire and have loads of free time, it might be interesting to try and build a more robust predictive model.  I am sure many have tried, but what the heck!  Maybe I will see something new.

Pop Goes The Theory

I have hypothesized that my dividend portfolio is underperforming as the Fed signals the beginning of tapering, dividend stocks may become less attractive.  But looking at my MFI tracking portfolio and then comparing the performance of dividend stocks against all the stocks, it appears dividend stocks (at least in that universe) are still doing well.  For portfolios 12/31/2012 and forward, the average dividend/mfi stock is up an extremely satisfactory 23.4%.  The average non dividend/mfi stock is up 15.3%, still good (Russell 3K is 7.7%) but not as good as the divvy subset.

Now you know!