Yesterday I published some ridiculous and bogus numbers from what I thought was a valid backtest. I mentioned at the time that I still had some validation, and I probably should not have posted them until I had done some more homework, but… well… I did.
In any case, they are not accurate. Once I perform backtests I then validate results by performing walk-forward trade-by-trade testing. After that I then trade them live in small sizes. In my walk-forward testing I discovered something called “look-ahead bias”. It was only in one small parameter, but it was enough to radically skew the results. Look-ahead bias simply means that you are using data from the future… that which can’t be known at the time of the trade.
I was using a volatility factor that was reading the data on the day of the trade. Since the trade is done at the open, I can’t know any of the data for that day until after the day is over, and therefore after the trade is already on. This is an easy thing to discover during validation, and it is now corrected.
The corrected results are still quite good, but not quite as ridiculous as before.
As you can see, there is still a lot of green, which means relatively consistent profits. I like that.
Just to be clear, testing and validation is still not complete. I just felt like I needed to say something since I published the wrong results prior. More to come on this.
Good Trading…
No comments:
Post a Comment