SJ Member  Posts: 49 Joined: 9/28/2012 Location: Olathe, KS  User Profile | I was in the ATM Master Class, so I got most of the skinny on this. I think I've tried with and without checking the scaling box. With a single ranker, it should make no difference. Incidentally, The scaling feature looks at the lowest and highest occurrence of the parameter over the lookback period and scales them to between 0 and 100. If you look into the xml file for each method (XYZ.ITS.txt), you can see those min and max values used for the calculation. In my file, here are the max and min values for VTY_PRICE: RangeMaxLong="0.1029571" RangeMinLong="0.0018801" RangeMaxShort="0.1029571" RangeMinShort="0.0018801" And here are the max and min values for RSI2: RangeMaxLong="100" RangeMinLong="0" RangeMaxShort="100" RangeMinShort="0" What is curious here is that I am a little surprised that RSI(2) hit both 0 and 100 in the data set. But here is the more important point. RSI varies between 0 and 100, but on any given signal day, by definition, RSI(2) was below 10 or there would be no signal generated. So to mix scaled VTY_PRICE and scaled RSI equally in a weighting scheme is not going to be effective as trades are more likely to be spread across the VTY_PRICE range, but will be concentrated in the lower 10% of the RSI range. If your goal is to give both an approximate equal contributing value, you need to scale up RSI by a factor of ten (IMHO). Not to say equal weight is ideal. That is where optimization can help solve the puzzle and I would guess it also resolves the imbalance discrepancy I just outlined. Unfortunately, all that is moot if I can't even get the most basic of ranking functions to act consistent and repeatable. It really looks like it sometimes gets stuck and can't properly distinguish which is the correct sort direction. I would appreciate if anyone else could attempt to recreate my scenario(s). | |
