Pathfinder swing TS

Forums › ProRealTime English forum › ProOrder support › Pathfinder swing TS

Viewing 15 posts - 1,546 through 1,560 (of 2,005 total)
1 2 3 4 5 … 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 … 130 131 132 133 134
  • 09/21/2017 at 3:23 PM #46830
    Despair

    Very simple, you must NOT use all data to optimize.! If you have for example (like the backtest above for PA) data beginning 2009. Only optimize with data from 2009-2014 and then test the result on 2015-2017. This is what i would try. Real WFA would be nice but I don’t see how you want to do it with a seasonal strategy like pathfinder.

    If you use all data for optimization you are just creating fantasy backtests like the one above. 90% winners and gain/loss ratio 60?? This is more than ridiculous. If it was possible for anybody to just install PRT and then code a strategy that has this stats live everybody would be a millionaire.

    Look at professional CTAs, the best are usually around 35% profit per year (and this are high educated full-time professionals). Every year one sticks out and does maybe 60-70% but then he had a lucky year and next year is back to normal. Don’t get me wrong, it is possible to make more than 35% a year but mostly through diversification and luck.

    09/21/2017 at 3:49 PM #46839
    wp01

    Thanks Despair.

    I think that nobody thought to be a millionaire by just install PRT. If that was the situation, the code would not be here for everyone to download.

    I understand what you are saying but also the way you describe how to test it remains data of the past. I still doesn’t say anything about 2018 and so on.

    I probably don’t understand it but if from all the 30+ algo’s (diversification) a couple of percentages remain of those fantasy backtests it could be very profitable.

    The past already showed that. And the creater said: “I’ve been profitable for many years….” Probably this is not his only project 🙂

    Could you be so kind to create the BT’s for example from the PA from MAT_CH here above and show us with pictures and comment how you see it from your point of view.

    Unfortunately i’m not the high educated full-time professional so i hope you can enlighten me.

    Thanks.

     

    09/21/2017 at 5:28 PM #46852
    Despair

    Nobody knows what will be in 2018 and I don’t say (or think) that pathfinder is a bad system but I would never ever invest a single euro in a system based on the BT you guys produce here.

    I’m not familiar with the pathfinder code and I’m not motivated to dive that deep into it but do as I said. Optimize one of your algos (best one you have a lot of data on) in the following way: Take one of the futures that has history beginning 1970. Then you optimize your parameters with the data 1970-2009 but don’t have a look at newer data at all.

    When you are done with your optimization you run it with the parameters you found on all data and then you look at the equity curve. Does it after 2009 look the same as before? This is what you want. If your algo works then on the new data it wasn’t optimized on similar to how it performed on the old data THEN you have a good result! And can take the conclusion “Hey this worked even on data I did not optimize on. This can work in the future too!”. If you got this confirmation you can optimize also using the last 7 years (maybe skipping the early years instead) and get “fresh” parameters and then you could put it live.

    And I say again, I really think pathfinder is a good system (maybe) but I never saw a test that said anything of it and I’m very sorry (and I will probably not make friends with this comment) to say, IMO all you guys optimize (and test) wrongly.

    I also addressed Reiner several times here in the forum concerning this but he so far always ignores my questions.

    1 user thanked author for this post.
    avatar Pfeiler
    09/21/2017 at 7:27 PM #46860
    wp01

    Thanks Despair for your input. I encouraged you to give your opinion and you did. That has nothing to do with making friends or not.

    Looking critical to an issue with multiple eyes always works the best.

    The problem on the forum of Prorealcode is in my point of view that i think 80% of the visitors know nothing or almost nothing.

    They give their comment without support, copy/paste what they want and come to the conclusion that it didn’t work. After the first loss they

    blame even the uploader of the code. The other 20% knows more than enough or is capable to create something. It is clear that most of them

    do not share anything at all or share a little bit. I can recall a couple of names. What i do not understand that people do not take effort to make

    something better. Are they waiting long enough until it is posted and that they can copy it? It discourage me i can tell you.

    I belong to the 80% with one difference from the rest. I stepped forward to try to make it better. To be honest: i know my limits and i already reached it.

    So if you say that Pathfinder probably is a good system, than i do not understand that you are not interested in making it better. If i had the knowledge

    to make it better i would stay up all night and day to make it better. Maybe you can consider again and try to find some motivation.

    Thanks.

    1 user thanked author for this post.
    avatar Nicolas
    09/21/2017 at 7:39 PM #46863
    Despair

    I said I do not know if pathfinder is a robust system because I never saw a test that was useful IMO.

    I also understand perfectly what you write. And I also wrote myself before on the forum that I’m not really motivated to share my work here because most users don’t contribute and (even worse) don’t even take the time to learn to code and to understand the code and just copy/paste. I have no problem with helping people or sharing an indicator or code snippet for this or that but I refuse to share complete strategies because I don’t want to support the laziness of most people (I put a lot of time and work in my programming).

    So what I did, I saw pathfinder and took the idea of creating a seasonal system for commodities but didn’t even look at Reiners code. Instead I wrote my own seasonal strategy that i run on several securities.

    The pathfinder code when I scrolled over it looked pretty long and there where so many parameters that one could optimize (I don’t think this is a good thing and promotes curve fitting) that I thought programming my own is less work.

    But sure I can have look at it again. When you optimize your versions, which parameters do you optimize?

    09/21/2017 at 7:54 PM #46864
    wp01

    If we optimize we do as described on page 33 of the other forum:

    https://www.prorealcode.com/topic/pathfinder-trading-system/page/33/#post-19123

    Thanks for taking the time.

    09/22/2017 at 8:25 AM #46885
    Mat_CH

    Hi despair, hi all,

    i have follow your advice and i have do an optimization with date 2009-2014 and the result is this.

    I think is a good result anyway.

    What is your opinion?

    Maybe a good idea would be to make an average of optimization between 2009-2014 and 2009-2017.

     

     

    1 user thanked author for this post.
    avatar ozz87
    09/22/2017 at 10:11 AM #46898
    Despair

    That is definitely an improvement over your first test! Then I think the history on PA is a little short for such a kind of strategy. Develop on only 5 years often gave me misleading results for my seasonal strategy. Have a look at another security. For example cotton. There you have data beginning 19070. Perform an optimization from 1970-2009 and then run the result s test on 2010-2017. I would be curious to see that test.

    09/22/2017 at 10:28 AM #46900
    wp01

    What i was wondering Despair, that there is also a difference in my opiniun in BT’s where the variables are for all the years and where the variables only apply for half the months multiplied by x- years.

    Like for example the post of the DAX from me a couple of days ago. In that code there are for 10 months 2 times different variables for just the long site.

    I think the deviation is then less than when you have for example 6 variables for about 10 years.

     

    09/22/2017 at 10:43 AM #46904
    Despair

    Fitting the the also on every 2 week period might cause curve fitting.

    09/22/2017 at 1:36 PM #46930
    O-jay8

    Hello Guys,

    Here my algos for Oct1

    AEX, ASX, Bund, Cac

    3 users thanked author for this post.
    avatar wp01, avatar dajvop, avatar Andyswede
    09/22/2017 at 1:38 PM #46939
    O-jay8

    Dow Short/long
    Dax
    Euro Stoxx

    2 users thanked author for this post.
    avatar wp01, avatar dajvop
    09/22/2017 at 1:39 PM #46945
    O-jay8

    FTSE, Hangseng, IBEX short,  Nasdaq

    2 users thanked author for this post.
    avatar wp01, avatar dajvop
    09/22/2017 at 1:41 PM #46950
    O-jay8

    Nikkei Short, SMI, US 500

    2 users thanked author for this post.
    avatar wp01, avatar dajvop
    09/22/2017 at 3:53 PM #46972
    O-jay8

    Commodities
    Corn long/ short
    Cacoa short
    Coffee

    2 users thanked author for this post.
    avatar wp01, avatar dajvop
  • Author
    Posts
Viewing 15 posts - 1,546 through 1,560 (of 2,005 total)
1 2 3 4 5 … 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 … 130 131 132 133 134

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
noob help please sl and ts
us100 mini – overnight swing strategy – ai

Create your free account now and post your request to benefit from the help of the community
Register or Login