[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: luck -- system back testing



PureBytes Links

Trading Reference Links

This discussion on backtesting is exactly why one should use a walk forward
analysis to test a system on out of sample data.  For example, if you have
15 years of data -- develop the system using 1 to 5 years of data then walk
forward test using years 5-6,  6 -7, 7-8, 8-9 etc.  After that you can test
using 5.5 - 6.5, 6.5 - 7.5 etc.  you may tile the outsample data as many
ways as you wish.  From there it is possible to determine system efficiency
using this form of analysis produces sufficient tests to determine expected
value variance and max DD with statistically significant sample size.
Further rigor can be applied by creating synthetic data from the outsample
set and performing another walk forward on the new data sample for example.
Using outsample data walk forward by 1 year increments once you have burned
data 5 - 15 years create another synthetic data set and perform the walk
forward analysis again.  If you use the original outsample data and 2 sets
of synthetic data you now have 30 tests of you system with a 12 mo trading
horizon.  Ideally the above process should be performed until 200 tests of
the system are completed. --

Good Hunting
Glenn