Walk forward optimization software?

Discussion in 'Trading Software' started by yoyo2000, Jul 10, 2005.

  1. yoyo2000


    It's hard to find the software for walk forward optimization,including seeking the most robust variable combination and other analysis,some suggestion please?

    PS: bolter tried to develop in WLD,but the project suspended because of the instability of WLD's debug circumstance.It's realy a pity!!
  2. I am interested to know what do you expect to do with walk forward data.

    Are you looking at your optimized results, then filter them by the walk forward results?
  3. yoyo2000


    My purpose is to conform whether the result of a system on some contracts are robust,the theory of walk forward optimization ( WFO for short),is to test a system on a period of data,and verify it on another period.

    For example,I test my 20/50 cross SMA system on the period of 1991.1.1-1998.12.31 of a basket of stocks,I got a 2.87 MAR(not bad),so I want to see how good it is on the period from 1999.1.1 to 1999.12.31.

    It's not a big problem to preform the WFO on a single test,but it's a disaster for system developer to perform it on several date intervals for many parameters combinations without a smart software.
  4. QuantDeveloper, www.smartquant.com/quantdeveloper.php can do it out of the box. It shows you in sample and out of sample performance on the same plot after strategy optimization, so that you can see if your perfect in-sample performance is nothing but a direct result of overfitting.

    There is strategy optimization demo video at the bottom of QuantDeveloper page.

  5. yoyo2000


    hmmm....It seem that's a powerful software,but it doesn't say how to pick the best ( = superior and the most stable ) parameters among the parameters combinations from the optimization,especially when parameters are more than 3.

    The WFO should be repeated several times,for example,the whole data range is from 1991.1.1 to 2005.7.11,the first in-sample test is from 1991.1.1 to 1998.12.31,then the software should pick the best ( = superior and the most stable ) parameters combination on this period and perform it on the first out of sample test (from 1999.1.1-1999.12.31),then record the result;then test on the second in sample period(from 1992.1.1 to 1999.12.31) with all of the parameters combinations,and picks the best ( = superior and the most stable ) parameters combination,and perform it on the period from 2000.1.1 to 2000.12.31,and record the result,...

    The most problem is how to pick the best ( = superior and the most stable ) parameters combination.
  6. How about VisualStudios .NET ?

    I'd say C# is good.
  7. yoyo2000


    TSGannGalt,you mean......program it line by line?

    It's my first opinion,but......I withdrawed later because of my poor program ability,then my friend told me that a member named Bolter in WLD's forum decided to make a WFO progress,some of their thoughts and concepts are great ,address is here:


    But some time later,his project suspended because of the instability of WLD's debug circumstance,and no more progress till now,how disappointed it is!

    So I post here,hoping a good solution......

  8. www.linnsoft.com.
  9. saxon


  10. yoyo2000


    Thanks,I'll learn and study it.It seem very complex.
    #10     Jul 13, 2005