Deadline June still stands
I figure the curve-fitting comes from using 2 filters instead of just the one.
I've have an arsenal of code for about half a dozen basic measures of market action, for instance ATR, DMI, CCI and a few I cobbled together like one I call 'CloseZone' which is just the smaller of the distance from the high to the close or the low to the close, doubled.
I try them as simple on/off filters to prevent / allow trading, they're either high or low (relative to a longer MA of the same), up or down (higher or lower MA compared to yesterday), and either rising or falling (compared to the longer MA again).
So I've got about 36 options to chose from on the filter front, and I guess combining 2 filters just curve fits the whole algorithm.
I was naively hoping
to have a system to trade for the Bund by this weekend but that little project target has been blown out of the water.
I curve fitted my way through all my historical data for the Bund, well the last 8 years of it, and I don't have an algorithm.
I've got a simple choice here I think:
(A) chuck out my algorithm as I can't find a profitable system unless I curve fit, and try to find some other style of algorithm to work with
(B) make the algorithm simpler by disallowing more than one filter and deleting any other extraneous code switches in there which don't really look good. This will mean I can't find decent optimisations for more and more of the optimisation windows I look at, leading to more abandoned optimisation windows.
i.e. either choice = more work, gobbling up days to my deadline.
Plus I think I should move onto a different market and not run my next optimisations over the same data.