question regarding – "Why we employ walk-forward testing to avoid curve-fitting"

Home Forums Logical Invest Forum question regarding – "Why we employ walk-forward testing to avoid curve-fitting"

This topic contains 1 reply, has 2 voices, and was last updated by  Alexander Horn 3 months, 1 week ago.

Viewing 2 posts - 1 through 2 (of 2 total)
  • Author
    Posts
  • #67339

    rikder
    Participant

    In a recent blog post https://logical-invest.com/walk-forward-testing-avoid-curve-fitting-backtesting/ you wrote:

    “For us at Logical-Invest this means, that we have to constantly check the validity of an investment approach and if needed, we must do small changes without changing the original character of the strategy. ”

    How do you decide when changes are needed? What does constantly “checking the validity” involve exactly?

    It would be interesting to go back to your original strategy for something like UIS and see if any of the changes added much to the overall results. I would hope the results of trading the original version of the strategy with no changes would be similar to the backtest of the newest version, thus demonstrating that the “original character” of the strategy is sound. If on the other hand, any of the original versions were “broken” it would be concerning.

    If you could provide some data to alleviate this concern I would appreciate it. It is important you demonstrate that your changes have never been implemented in order to fix a broken strategy. Thank you!

    #68454

    Alexander Horn
    Keymaster

    You wrote: “you demonstrate that your changes have never been implemented in order to fix a broken strategy”.

    Probably not the answer you are expecting, but what Frank is explaining in the article is that indeed we do review our strategies explicitly to ensure the fundamental base is not broken – and if: we do fix it, where possible – or abandon it, where needed.

    Example: The last 20+ years have been marked by ever-decreasing bond yields, especially after 2008/09. This made ultra-long bonds the perfect hedge: Mostly negatively correlated with equities while offering positive return bias. This long-term trend somewhat stopped when the FED started hiking – so we looked for a new hedge and adapted our hedging approach. PDCA: Plan, do, confirm, adjust (if necessary).

    An alternative approach would be to use a backtest with 100 years of historic data – this might indeed reduce the need for changes, but also the opportunity of finding alpha.

    Another example: Our Maximum Yield Strategy is based on harvesting the “fear premium” from shorting volatility. This is a very opportunistic strategy as it relies on two fundamental assumptions:

    – There is a market for such fear-premium harvesting – which has not been available before the GFC 2008/09, so adios long-term backtesting.
    – There is (mostly) contango in the volatility curve, which has been for most time between 2010 – 2018, and then suddenly there wasn’t anymore and in Feb 2018 the strategy stopped working for some months, and then started working again with a new set of parameters reflecting the new market environment.

    Again, the 20+ years backtesting approach would not work here, you would have lost the opportunity to harvest 8 years of very nice returns, which even including the tough years in 2015 and 2018 still surpasses 30+% CAGR.

    So in summary, yes, as good CEO’s of our investment approach we strive to detect and harvest investment opportunities, but also have the flexibility to strategically, i.e. long-term adjust to changing market environments.

    This is a great area of discussion, so hopefully can further expand on this topic.

Viewing 2 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic.