Systems can have monte carlo walk forward done on them.

See the two MC WF below. The one with the higher rating has more systems with profitability below zero. These 2 examples were the best and worst out of 233 systems used.






Select a number of systems, then right click and send them to monte carlo WF. If one wf is being done, use multi threaded. If multiple then use single threaded.


These parameters apply.



For example, nearest 100% means WF parameters will be a max of the nearest 50% above and nearest 50% below the parameters in the original system.



Look on the WF MC tab for the output.









On a very good short only NQ setup, 2500 systems were built using 200 indicators (not all are released yet - and some will be poor), and not the normal 2 pass method to reduce indicators that are likely to fail - so it was a 1 pass method.

This was done to introduce a higher degree of failed systems. However even the worst of the systems were still quite good.

We took the top 300 systems, and used the family command to only get unique systems. This gave us 223 systems.


top 70 MCSR vs bottom 70 MCSR using 250 iterations, anchored  walk forward, then take the worst 35 VSS of the bottom 70 systems and compare to the best 35 VSS  systems of the top 70 wf systems





top 70 MC SR vs bottom 70 using 100 iterations. Note this is no where near as good as 250 iterations. This implies 100 iterations is way too low.


setings used

Dates used for WF as in sample

out of sample dates 2018-2-29 to 202402028



Now a test of 1000 random space was done. This test took under 3 hours on a Intel I9 13900k CPU. On a dual Intel 2690v4 the identical test  took 11 hours.

This gave an even greater degree of improvements in metrics. 32% improvement in fitness. The tests in ave C/D are the in sample, which shows that the in sample metrics of the top and bottom 70 mc sr had near identical performance.




Verification groups used



Below, a test of full search space instead of nearest 100%

These settings were used. Results are similar to nearest 100%



Below, 2500 tests were done, search space 100%. This is not as good as the above tests, so in these early tests 1000 random space tests with 100% search space was best.