Discovery Process Overview
In the previous posts on ranking system design I described the screen backtest setup, 107 stock factors and provided an EXCEL spreadsheet for recording results. I received some feedback regarding the factor listed in the spreadsheet that will be incorporated into a future version.
I have decided to perform a shortened version of the factor tests and include the actual results. I feel this will be more beneficial to the reader. In the shortened version I test both the "Even Stock ID" universe and the "Odd Stock ID" universe independently and then determine which stock factors are strong predictors for both universes. Those factors will be included in the (preliminary) ranking system.
For my own personal ports and R2G models I test each stock sector independently and then analyze the results. The stock factors that prove to be strong predictors for multiple sectors are then included in the preliminary ranking system.
Since I am not performing the long version of the test, I don't mind sharing the results with readers.
Even Stock ID Test Results
Below is a screenshot of the Even Stock-ID results. As can be seen, there are eight highly ranked factors with a relatively low correlation to each other. The correlation is calculated using the average stock price of top 20% of stocks for each factor.
Odd Stock ID Test Results
Then the entire 107 stock factors were retested, this time using the Odd Stock-ID universe. A different set of stock factors emerged as shown below.
I will stop here since I have decided to use the shortened process and only test the 107 factors against the even and odd stock universes. If I had used the long process then I would now test each of the 107 factors against each sector within the stock universe. That would obviously take a long time since I don't have any automation tools.
I am finished with the stock discovery process. Now I will decide what stock factors to use for the ranking system. But that is the subject of the next post.