Chris, I think the Keystone system is based on the changes Fosback made around 2000 or 2001. I presume Hulbert is continuing to track the older system from 20 years ago. Actually, I think the system described in Market Logic was also slightly different but Fosback updated it in the late '70s - early '80s.
The latest version accounts for the propensity for Mondays to underperform. This smacks a bit of curve-fitting if you include the 1987 period - If I were backtesting this system, I would intentionally exclude '87 from my sample. The latest system also attempts to reduce the number of round trip trades by lumping the end-of-year holiday periods together.
You should read Tim Hayes' book The Research Driven Investor. He outlines a system testing methodology that will help to minimize over-optimization and curve fitting. I continue to notice that most people who back test systems tend to use all the historical data they have and then optimize from there.
Hayes' suggests testing a system against a sample period, optimizing the system, then testing it again against an out-of-sample period. If the system holds up in out-of-sample testing, the next step is to test in in real time (NDR tests indicators in real time, sometimes for years before they deploy an indicator).
Good luck with your test - hope you'll post the results.
