Suggestions for improving a Walkforward test?

Discussion in 'Strategy Development' started by BrooksRimes, Mar 7, 2019.

  1. Given a 10 year walkforward test that is generally positive with a fairly smooth equity curve, but 1 recent year with a small loss, what are some suggestions from experienced walkforward testers to make all years positive? Some possibilities are changing the length of the IS and OOS periods, changing the trading system itself, changing the symbol(s) in the test or changing some other test setting. K-Ratio is being used as the optimization target and I've found it to provide the smoothest equity curve.

    Ideas appreciated. Thanks!
     
  2. Lately I've been trying a few different symbols. For example if my system works well on SPY, then I may try it on a few very large, very liquid and very price-discovered stocks. Like DJ 30 members that have been around a very long time. It may seem like an effort to curve-fit a system to a stock, but if 2 year, 5 year, 7 year, 10 year and possibly 12 year (catching the 08-09 meltdown) results are good then there is obviously persistence there and should be worth a look.

    I know that the stock is probably going to be a significant component of SPY, but still results could easily be better than just trading SPY itself.
     
  3. tommcginnis

    tommcginnis

    That smells an awful lot like a bull market. "...changing the trading system itself" might be modified to "...examine the base assumptions of the actual system, and identifying triggers that would exit one system, and incorporate (or, "trade into") a different one.

    My own philosophy (backed by *insufficient* data) is that entry signals need to be much louder than exit signals {which themselves can be a whisper}. Why? I'd rather lose 50% of a gain once or twice, than earn 100% of a loss, again and again and again: "Let your winners run, cut your losses early."
     
    murray t turtle likes this.
  4. By doing this you are destroying the benefits of a walk forward test; eg that the system is fitted using only historic data.

    GAT
     
    minmike likes this.
  5. I would say that depends on the approach. I made decisions before doing the test on inputs such as the list of tickers and the length of the IS and OOS periods. I could have made other decisions with different results. I hoped others may have been down this path and save me some time. I may just have to solve this the hard way.
     
  6. Metamega

    Metamega

    What’s the value of trying to erase this one bad year? Perhaps trying to modify this sample will just lead to a drawdown in a future year and your model will underestimate possible risk...
     
    fan27 likes this.
  7. fan27

    fan27

    Yeah...what @Metamega said. For my own trading, I have performance requirements for strategies that could still be met with a down year as I am more interested in the performance of the portfolio of strategies that I trade. In fact, I spend very little time on any particular strategy as I have an automated method for discovering it, testing it and then discarding it when it no longer performs.
     
    tommcginnis likes this.
  8. ironchef

    ironchef

    To be positive every year, invest in short term treasuries.

    I am very happy with 2 out of 10 years with losses, significant losses by the way because anything else is not realistic, like chasing rainbows.
     
    murray t turtle likes this.
  9. %%
    That could work;
    may not beat inflation.Insured mini bonds could work better. NOT a prediction; all that may lose value.:cool::cool: