BACK-TESTING EP Y-6 S&P 500 Y-6

It's going to take some time for me to get my portfolios in order, but I feel good about this decision.  It's going to be more work, but these small portfolios are giving inaccurate results.

I thought it would be useful to remind myself why I'm doing this in the first place.  This started because I wanted to get an idea of whether my EP model was predictive.  I did a quick and dirty test against a somewhat current S&P 500 portfolio I had in my list.  We have to start somewhere right?  Seemed like a good choice.

The results were very encouraging.  Encouraging enough for me to go further with the testing.  Specifically to refine it, remove biases and isolate other factors that could be responsible for the results.

I purchased a list of S&P constituent members that went back years and years.  It was very well done, it shows all the ins and outs of members each quarter.  They're called Siblis Research, for anyone who's interested.  I used this to build S&P lists for each year.  I included every stock that was a member at some point in the year.  I then had to scrub the data including removing financials.  

This model doesn't work on financials.  I've been working on a version that will.  I've build it, but not tested it.  So much testing to do.

The results run from April 1st to March 31st.  I think that range is appropriate, given that 231/300 stocks in this group have a December 31st year end.  We're working on an enhancement to the software that will let the user sort by year end but this is good enough for now.  

As an aside, I'm a believer in good enough.  This will probably sound dreadful coming from an accountant and a financial analyst, but for me, making an estimate based on "close enough" or the best one can do is good enough.  Especially when there is already a lack of precision, which is always the case if you are trying to see into the future, which is what a model is trying to do.

Here's a couple of examples of models or calculations that make me scratch me head with their pointless attempt at being overly precise: the Sharpe Ratio and the standard deviation model.  In the former the numerator is supposed to have the risk free rate subtracted from it.  I think that is a big "who cares".  Does that step really add anything?  It feels arbitrary to me.  Also the standard deviation calculation - in one version you divide by N and another N-1.  I just don't think the "-1" is important.  Just leave it at N always, don't try to pretend that reducing the denominator by one makes it better.  It's elitist.  Keeps the masses away.

Ah, feels so good to admit that to the world.

Anyway, without further ado, here are the results for the EP model back-tested on Y-6 run on a portfolio of S&P constituents (x-financials) from Y-6.

One Year Return

Two Year Return Three Year Return Four Year Return Five Year Return Six Year Return Return for Year Two Return for Year Three Return for Year Four Return for Year Five Return for Year Six Observations: 

I think these are pretty dandy results. Not perfect, but geez, I hope no one reading this is actually expecting that.  Nobody bats 1,000 but one can improve.  My goal is to figure out the rules of this model.  For example, how long do the results persist?  Also, are there "size rules" - does it work better on large cap stocks than small cap stocks?  How does it perform during market stress?  Lots of questions.

I learned that if I work with too small a portfolio, outliers can take over the results and skew them.  So, I'm changing my strategy and moving to larger portfolios, but I am still categorizing by sector.  Perhaps that will fail too, and I might have to change my testing to categorize by market cap.  I'll probably have to do that anyway to determine if size plays a roll.

Tonnes of testing to do, kind of excited.  I feel like I'm chasing a true discovery.  

My goal is to get the testing done by year end.