stevehopwoodforex.com
https://www.stevehopwoodforex.com/phpBB3/
Print view

Recovery system with a different twist
https://www.stevehopwoodforex.com/phpBB3/viewtopic.php?t=193		 Page 6 of 6
Author:  gaheitman [ Sat Oct 06, 2012 5:23 pm ]
Post subject:  Re: Recovery system with a different twist
garyfritz wrote:Thanks, George. So it looks like they invented their own measure and gave it an already-well-established name.

I'm not sure I'm interpreting their explanation right. I understand their R to be "the number of streaks of wins + number of streaks of losses." So for the series of blue +/- right after the Z-score formula, I think R is 7: 1 (9 wins), 2 (6 losses), 3 (1 win), 4 (3 losses), etc.

But if I use that definition of R, I get a Z-score of -1.21 on my test system, indicating about a 75% chance of a win following a loss and vice versa. That seems unlikely, since my "streakiniess" calculation on the same series returns 0.52, nearly random.

I think I prefer my measure. :-)
Actually, a streak has to be 2 or more of the same sign in a row, so the R in the example given would be 5. Not sure how that changes things in your test system....

So, are you going to name yours the G-Score or the F-Score? :)
Author:  garyfritz [ Sat Oct 06, 2012 5:32 pm ]
Post subject:  Re: Recovery system with a different twist
OK, if I use that streak definition, my Z is -18. That doesn't help. :D

The Z definition has R in an (R-0.5) term, which makes me suspect maybe R is supposed to be scaled by N, i.e. R = #streaks/N. But that doubles Z to -35.

Without spending more effort on it than it deserves, I don't understand the logic behind their Z score formula, and I can't get sensible results out of it. Unless you want to figure it out, George, I'm not going to bother with it.
Author:  gaheitman [ Sat Oct 06, 2012 6:37 pm ]
Post subject:  Re: Recovery system with a different twist
garyfritz wrote:OK, if I use that streak definition, my Z is -18. That doesn't help. :D

The Z definition has R in an (R-0.5) term, which makes me suspect maybe R is supposed to be scaled by N, i.e. R = #streaks/N. But that doubles Z to -35.

Without spending more effort on it than it deserves, I don't understand the logic behind their Z score formula, and I can't get sensible results out of it. Unless you want to figure it out, George, I'm not going to bother with it.
OK, I've officially spent more time on it than it deserves, but this is what I found...

I think I need to retract my former comment about not counting the streaks of length 1. Testing out the scenarios in excel, it's obvious that if you don't count them, the formula can't tell the difference between these two sequences:

Code: Select all

++++++++++++++++++++--------------------
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-
They should have opposite signs, but they have the same sign (negative). So, switching back to counting all of the series gives a more reasonable result. The first sequence above is -7 and the second is +7. A random sequence also calculates to lower values on either side of zero, as you would expect.

Attached is the ugly spreadsheet I used...

George
All times are UTC Page 6 of 6

Powered by phpBB® Forum Software © phpBB Limited