Portfolio Design

Advanced concepts in stock investment portfolio design.  Fundamentals, technical analysis and many other related topics are discussed.

Stock Factor Selection Process


In the last post I presented the results of the Even and Odd universe tests. Now I will decide which factors to go forward with.

After looking at the stock factor scores (rating and correlations) for the two (odd and even) universes, I realized that if I filter out the lower rated factors and highly correlated factors as presented in the spreadsheet that I would be left with very few factors.

So I have decided to ignore the correlation and rating filters and include all stock factors that are in both odd and even top 25 lists.  That will give me a preliminary ranking system that I can optimize using the ranking system optimization process. Then I will walk through the optimization process using a more advanced process than I previously presented.

Factor Analysis

I processed the top 25 ranked factors and summarized them in the table shown below.  The two columns on the right labeled Even and Odd are the ranking position (1...25) for each factor, one (1) being the best.


Preliminary Ranking System

Then I assembled a ranking system consisting of all the top factors common to both universes.  The preliminary ranking system is publicly accessible and can be found here:



Ranking System Performance

For the benefit of Portfolio123 users who are new to the website, I am including the steps for generating the ranking system performance.  The procedure uses the stock universe which is publicly available here:  https://www.portfolio123.com/app/screen/summary/83611?st=1

Performance Using 3 Month Rebalance

Go to the Stock Ranking System Factors page (the first page you see).  Then click on the Performance link at the left side of the page as shown below.


Configure the settings for maximum time period and three month rebalance.  The appropriate benchmark is the S&P 1500 Composite, so select this benchmark (if it is important to you).  Select the stock universe Enter a minimum stock price of $1,50.  This was the minimum price used in the factor selection process. Then click on Run.


The performance chart is in the form of a bar chart with the benchmark performance on the very left side, followed by 20 stock buckets. Each bucket represents the performance over time of a 5 percentile band rebalanced every three months. The left-most buck (following the benchmark) represents the performance of stocks with a ranking of 0-5.  The next bucket consists of stocks with ranking 5-10 next, ...  The right most bucket is the bar of most interest, consisting of the performance for the top ranked stocks 95-100. The performance chart is shown below.


For this preliminary ranking system, the top ranked 5% of stocks achieved almost 19% annualized return.  

Performance Using Weekly Rebalance

Now lets determine the performance with weekly rebalance.  Go back to the setup screen by clicking on Change Settings.


Only the rebalance period needs to be changed to Weekly.  All other settings remain the same. Click on Run.


The performance chart is then displayed with the results for weekly rebalance.  Note that the top 5% of stocks have higher performance with weekly rebalance (23% versus 18.5%).  That is due to the higher turnover rate with weekly rebalance.



A process has been described that discovers stock factors and assembles them into a non-optimized ranking system that "works" for three month rebalance as well as weekly rebalance.  At this point I will say that this ranking system is "competent" but not exceptional.  The performance you get at this stage is totally dependent on how much work you want to put into the discovery phase.  For example, you can screen using different rules such as Rank>95 instead of Rank>80.  Or you may have discovered superior stock factors that I do not have in my list of 107.  You are only limited by your imagination and the effort you want to expend.

In the next post I will continue on with ranking system optimization.  This will be more advanced than I previously described and will eliminate factors with high correlation, and adjust the factor weightings.

by Steve Auger