Doubts about optimisation process

Forums › ProRealTime English forum › ProOrder support › Doubts about optimisation process

  • This topic has 9 replies, 5 voices, and was last updated 5 years ago by avatarCKW.
Viewing 10 posts - 1 through 10 (of 10 total)
  • 03/22/2018 at 8:30 PM #66121
    Juan Salas

    Hi,

    I have a doubt about the optimisation. If I want to optimise two variables (for example the periods of 2 Moving averages, x and y). Lets said that I combine 1 to 80 step 1 for x, and 1 to 80 step 1 for y.

    During the process I can see that the best results are being displayed according to (profit, gaining per position, success, etc). I have noticed that it displays results of 100% rated by best profit, but some of these 100% results disappear during the optimisation process. When I am about, lets say, 35% of the process I have for example 12 combinations with 100%, and at some point later some of these 100% combinations disappear from the table, giving pass to other 100% results. At the very end of the process I come out with maybe 15 (100%) results, but during the entire running of the optimisations I saw in total in display about 30 (100%) combinations.

    Anyone can tell me why I don’t get all the 100% combinations at the very end of the process and on the contrary I can see a lot of 98%, 94%, 87%, etc.

    Thanks,

    Juan

    03/22/2018 at 8:53 PM #66126
    Vonasi

    I have noticed this too and assumed that the results were incomplete and the test with that set of variables still had some time to complete. I assumed I was seeing incomplete test results displayed and just had to be patient. Maybe PRT does not run through each variable combination completely before moving on to the next? I assumed it depended on the exact calculations involved and the most efficient way of getting the results from them. Maybe someone from PRT can confirm this.

    1 user thanked author for this post.
    avatar Juan Salas
    03/22/2018 at 9:12 PM #66127
    Juan Salas

    Hi Vonasi,

    Thanks for your answer. I have realised that when 100% results are displayed, they have finished combining those two variables. When you combine x (1-20) and y (1-20) and it shows that 4 positions of 100% have resulted from x=20 and y=4, these two variables are done combining each other at least in one direction, so I believe that this result is not definitive. Even being definitive, the last one in the ranking of 100% disappears and yield pass to other results.

    In this sense, I have noticed that the final result of the display may be related to final gaining. They removed combinations of 100% with maybe 123,67€ and give pass to combinations with 97,78% with 245,6€, believing that these results may be also interesting and taken into account for the trader.

    But, I am not sure about this. It would be nice if the display could be customised, so it could show only (by demand) the 100% results for example. I believe the display is what it is, isn’t it??

     

    03/22/2018 at 9:13 PM #66128
    Juan Salas

    Sorry, in the third line I meant the result IS DEFINITIVE

    03/22/2018 at 9:17 PM #66129
    GraHal

    I have also wished in some way I could see ALL the results and make my own decision on what I use to go forward. Or better still I decide what criteria I am optimising for, e.g. max profit or min drawdown etc.

    I’ve noticed that generally when I see 100% win to loss ratio it has been when there has been  very few trades and big drawdown relative to many trades for say, 70% win to loss ratio with lower drawdown.

    2 users thanked author for this post.
    avatar jonjon, avatar Juan Salas
    03/22/2018 at 10:16 PM #66131
    Vonasi
    GraHal wrote:

    I have also wished in some way I could see ALL the results and make my own decision

    True – I would love to have the full list of results so that I can drag them into an Excel spreadsheet and create some graphs that truly give me something to analyse – so that I can fully understand the results of each backtest.

    1 user thanked author for this post.
    avatar GraHal
    03/23/2018 at 6:52 PM #66201
    verdi55

    I think the sorting criterion during optimization is absolute gain, not percentage of winning positions. So, if you have 6400 combinations (which is way too much, I think, anyway), in the end you get only the 100 best results in terms of highest total gain. Everything else is being discarded during the process.

    Why don’t you run the optimization with parameter x first (80 combinations), then take the best or the 2-3 best parameter values for x and run the optimiziation again with parameter y (80 combinations again) ? Usually, you get the same results, but much faster. These curve-fitting processes tend to converge at certain optima for both parameters, whether you run the parameters alone or both combined.

    Yes, and I agree that sometimes it may be valuable to get more than 100 results in order to evaluate them in Excel. For example, when you want to determine the average gain of all parameter runs. This is sometimes more useful than having only the 100 best and selected results that may be worthless in reality, because they are over-optimized.

    2 users thanked author for this post.
    avatar Juan Salas, avatar GraHal
    05/22/2018 at 6:29 AM #71042
    CKW

    Hi,

    Asking a question related to this topic.  I run several the optimization with different sets of variables to ensure i have the complete results as best gain is not the same for best WFE. It’s very tedious and very time consuming. E.g. 500 results copied to excel, It takes me few hours to spot the best WFE by compiling the equity curve 1 by 1….:(

    I really hope PRT can allows us to customization parameters to focus. By the way, i think this is not PRT Priority …

    From optimized results, there are only Gain, %Gain, Nbr Trades, % Win rate and Avg Gain. Anyone know the formula to caculate the “Avg Gain of Winning Trades” and “Avg Loss of Losing Trades” (if possible), at least we can quickly filter the excel by acceptable “Gain/Loss Ratio“.

    Why PRT don’t combine the Optimization & WFE tool ?? It means we can have the option to optimize variables for IS period and also display the results of OOS period on the same table. It will saves us a lot of time!!!

    br,

    CKW

     

    05/22/2018 at 8:34 AM #71068
    GraHal
    CKW wrote:

    Why PRT don’t combine the Optimization & WFE tool ?? It means we can have the option to optimize variables for IS period and also display the results of OOS period on the same table. It will saves us a lot of time!!!

    Good idea … be good to put your idea forward to PRT via their on-site Suggestion Form?

    Click the Suggestions Box under the girly pic! 🙂

    https://www.prorealtime.com/en/contact

    05/22/2018 at 9:09 AM #71077
    CKW

    ha, Done :). Thanks

    1 user thanked author for this post.
    avatar jonjon
  • Author
    Posts
Viewing 10 posts - 1 through 10 (of 10 total)

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
cant get optimisation to work
help about stop trading strategy
help about stopp trading after 2 losses
profit optimisation strats based on 1h n htf
question about acumulated positions
very concerned about ig’s data and/or prt execution of data

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