WF Testing Interpretation…
- This topic has 29 replies, 5 voices, and was last updated 3 years ago by
.
-
https://www.prorealcode.com/topic/day-month-year-strategy-robustness-tester/
1 user thanked author for this post.
Is that the correct sequence?
Yes, but some period needs to be not included in the period over you optimise variables … then you have a true OOS period over which to do WF.
Best if the period not included in optimise runs is the most recent, for example, the most recent 10k bars.My Method
A twist on above and not using WF at all, but you get the idea …I optimise over the most recent 10k bars and then backtest over 100k bars and if my Algo does not break the bank / dip loads below zero (over the 90k bars OOS) then I put it on Demo Forward Test for at least 40 to 50 trades.
1 user thanked author for this post.
i dont bother with demo, maybe I shoud?
Not putting a strategy on live demo is like jumping out of a plane with a parachute that you packed yourself having never ever packed a parachute before. It might work but then again it might not. Would you still jump knowing that?
No matter how much we try to not curve fit a strategy and no matter how much we WF and robustness test it there is still a chance that it is a pile of poop. Put it on forward test live and sit back on your hands and don’t be tempted to put it really live until you have plenty of OOS trades to show you how it really performs. Then jump!
1 user thanked author for this post.
Dear people – FWIW, this is what I do; amd saying in advance, this is nothing really new, but Monobrow perhaps can now grasp this better.
First off, the last time I worked on this 1 second system, was March 2020. This is already quite crucial, because I have a LOT of WF to examine.
ehh … WF ? why ? Why not WB(ackward) ? … and that thus is the whole idea (at least mine) …You create a system (algo) on any random period of time – as long as you can load the data.
Look here. I am not interested in Out of Sample data whatsoever; What I am interested in, is a super-easy means to run a back test over numerous periods of time, without pressing buttons. OK, to be honest, 1. 😉
Obviously (?) it is crucial that you run the Backtest with “WF” functionality as much as possible over a period you did not optimize at all. Like my example here, because it was March last year I worked on it, while I run the system today. But, would I develop it “today” and use the last week, for example, then I’d try to backtest it over e.g. last November *and* use the WF means. And as you can see I selected a 97:3 ratio because that apparently gives me a nice spread of the starting and ending times. Of course you got it that for a 1 second system it is super crucial to not always have that nice gain fallen into your backtesting, so while the one period of backtest may incorporate it, the other will (quite explicitly) miss it (and even run into a big loss). Here you can see the importance of that, because it looks quite more red-ish :
Let’s keep in mind that I can do this for November, but also for September etc. (each is a separate run which as many “Repetitions” as you like, and besides this, I could also change the start-date/time myself (shift it a few hours forward) and repeat the lot).
I surely would allow myself to re-optimize after looking at the results by the above described means, and the more red lines get out of the way, the more gain there will be. … If I only keep on running all these varying start and end times.
Of course this implies a kind of over-optimization but in the end you’d do that to net always gain – seen over the longer period of time.Btw, I would always try to make each day profitable, like this one (ran over 3 days only, but with 30 days I’d aim for the same) :
Yes, 3 days is not much, but it is still 121 trades and commission (the biggest pain) is well included.
1 user thanked author for this post.
Sorry I should have added the word ‘demo’. Yes forward test live is put it on demo live.
You should be aware though that even when you do this your forward test could fool you. If your strategy opens a trade and stays on the market for a period of time then during that time you are potentially missing other trades and those trades could all be losers. So your forward demo live test tells you that your strategy is great whereas in reality you just got lucky with the bar that you chose to start it on. So even after a long successful forward test I would recommend further robustness testing (I use my VRT and SEB techniques) on the new OOS data that your forward test was carried out on just to confirm that you didn’t just get lucky.
1 user thanked author for this post.
got a link?
https://www.prorealcode.com/topic/a-strategy-for-every-bar/
1 user thanked author for this post.
Find exclusive trading pro-tools on 
