I’ve been starting to use Excel to back test my system/indicator ideas. Tested a few ideas and realized I need to start using PSAR filter to systems and equity curve “surfing”, as Dave Evans would call it, to improve performance. I had my doubts about PSAR though… So I checked some stats about how it performs from one hit stop to the next in the direction it tells you to be in.


 
This is a table that takes DV’s PTO and splits it into long only and short only trend lengths. I then split each short/long trend into profitable and unprofitable stop-t0-stop trends’ ranges in each of the 10 percentiles of the PTO.

PSAR at best struggles in whatever parameter you set it to and fails to capture most of the trends in the long term using the straight forward approach of using one static parameter for the entire back test.

 

So my idea is to build a smarter PSAR that “adapts” to the trends according to

  1. The expected trend lengths and
  2. The most fitting Acceleration Factorparameter according to the best probability among the available 10 parameters (0.1 to 1.0).

 

Still a work in progress.