Frequency for updating parametrers of Algos

Forums › ProRealTime English forum › General trading discussions › Frequency for updating parametrers of Algos

Viewing 12 posts - 1 through 12 (of 12 total)
  • 12/18/2021 at 11:23 PM #183579
    Khaled

    Hello everyone,

    Since the market keep changing, some days the EMA20 works fine, others it’s de DEMA8, and so on, same applies for any other indicator Stoch 5, 8, 14, etc., RSI, and the list goes on… I’ve tried to backtest a couple of strategies while keeping the standard characteristics for each indicator (length, close vs typical close, etc.) and changed the duration of the test: 200 bars, 1k, 15k, 50k, etc. (without changing the parameters) and the same Algo can show excellent result on 1 month, below expectations on 1 year, and very poor on 1 week.

    For running Algos (Live or on Demo), I’m wondering if someone has tried to optimize the variables (Length of indicators for example) of their Algos on a weekly or monthly basis? and what did you get as a result?

    I can take the experiment by myself but it will take at least four weeks to get a meaningful result (based on enough data). So, please share your experience.

    Thanks

    1 user thanked author for this post.
    avatar GraHal
    12/19/2021 at 12:47 AM #183580
    phoentzs

    Personally, I actually try to leave my standard parameters as I created the strategy. I am only optimizing TP / SL and trailing stop. But I know that everyone does it differently. There are certainly some who optimize everything or others who optimize over very long periods of time. Would be interesting to know what is going best.

    1 user thanked author for this post.
    avatar Khaled
    12/19/2021 at 10:20 AM #183587
    GraHal

    I have optimised regularly, even daily on Algos that suit.

    I have not said much on here (as it goes against most folks thinking … 1m bars essential etc) but I’m getting better results since I started optimising regularly.  For one thing, I’ve not enough time left in my life (due to slow PRT for optimising) to optimise over > 10k bars! 🙂

    Through lack of time for analysis of results, I cannot show you a study to prove above, sorry.

    It would be most useful if you could do the study you suggest in your first post (you sound analytical) and then we will know for sure?

    1 user thanked author for this post.
    avatar Khaled
    12/19/2021 at 11:19 AM #183590
    Khaled

    Hi Guys, thank you for your feedback.

    To be clear, I’m not suggesting to change all the parameters on a weekly basis. I came to notice that my entry points are OK (see attached Dow 1 hour). What’s failing me is the Exit tactics within the strategy. Sometime, Exit when Close crosses EMA21 works well, the next day is when Close crosses the Pivot Point or S1, S2, R1, R2, others is when EMA[X1,Y1] crosses over (or under) EMA[X2,Y2], Kijun crosses Tenkan, Fibo levels, etc.

    So what I’ll try to do is (it’s not really Curve Fitting per say):

    1. keep the Indicators parameters standards unchanged over time.
    2. keep the Entry Long/Short strategy unchanged over time.
    3. optimize on a weekly basis the Exit Strategy and the SL, Breakeven.

    My objective is to get 1 to 3 trades a day, no more.

    How do you deal with this Exit module moving target?

    I’ll share the results, but like GarHal said rightly, due to slow PRT, this is taking all my free time… and requires consistency.

    01/01/2022 at 7:13 PM #184277
    Khaled

    Hi Traders,

    Happy New Year!

    So since my first post above, markets have been relatively quiet and I thought I’d rather take an Algo published by JanJ BollingerKletner which I run on DJ on 5min TF and which I optimized as follows:

    1. Period April 2019-Dec. 2020 optimize 4 parameters and only 4 , so I have my “optimal base line”.
    2. Month of Dec.2020 to get the optimal parameters to run in Jan. 2021, always the same 4 parameters.
    3. Month of Jan. 2021 to get the optimal parameters to run in Feb. 2021, and so on until Dec. 2021 run with the optimal parameters of Nov. 2021, always the same 4 parameters.

    The result is a bit surprising. The optimal base line parameters determined on the period April 2019 to Dec. 2020 kept for each month of 2021 generated 9.196€ of Net Profit, while the parameters optimized on a monthly basis generated 2.080€ of Net Profit, a factor of 4x.

    Which part is due to the System itself and which part is due to the methodology or the frequency?

    It makes me wonder whether it makes sense to optimize frequently. I’m going to do the same exercise with Q3/2021 vs weekly from Oct. to Dec. 2021 with the same Algo and see if the trend is the same or if it’s the monthly frequency that is the problem. It’s a bit fastidious, but I’ll try to do the same with another Algo and may be a different TF.

    Note 1: that the Algo I tested was originally designed to run under 1h TF.

    Note 2 : I reinvested the gains, so there is a compounding effet.

    Any thoughts?

    01/01/2022 at 10:13 PM #184284
    GraHal
    Khaled wrote:

    The result is a bit surprising.

    Thank you for your hard work and sharing with us all!

    I agree the results are surprising.

    Your 1st column – Period – is identified as month-20 … is this a typo or have you maybe looked up the results for 2020 (should be 2021)?

    Or have I misinterpreted … it is nearly time for my bed! 🙂

    01/01/2022 at 10:13 PM #184285
    Khaled

    Here is the comparison of the Q3/2021 on a weekly basis, parameters updated each week, VS the same period run with optimal parameters of Q2/2021. the quarterly version slightly outperforms the first one and both underperform the buy & hold. The four parameters in play are: Average Period, Donchian Period, Breakeven and SL.

    What do you think?

    01/01/2022 at 10:14 PM #184287
    Khaled

    You’re right its month-2021, typo

    1 user thanked author for this post.
    avatar GraHal
    01/01/2022 at 10:19 PM #184288
    Khaled

    The problem I see from this quick study is that if the Average Period of the Actual Signal is far away from the Algo hard coded Average Period then either the Algo doesn’t take a trade or takes a big loss.

    If I were to take a decision quickly, then I would update my Algos every six months to 1 year, not less. With that said, I have been surprised that I put an Algo on 29/12/2020 in Demo and didn’t updated it at all. It generated >500% in 2021! (with no reinvestment of Gains, just 1 NQ lot)

    Cheers

    01/01/2022 at 10:40 PM #184289
    GraHal

    Using optimised Parameters from Previous Week did produce several very good weeks (far better than all but 1 week of the baseline optimised)

    Also using optimised Parameters from Previous Week produced several bad weeks.  If the previous week had been a downtrend week and the following week is an uptrend week then it is understandable that this will produce bad results for the week.

    Some of the surprising results may be due to the strategy?  If the strategy had a filter which prevented trades being executed when the running week is a different market from the optimised week then results may be better overall?

    Or …

    If uptrend use parameters A,B,C,D.

    If downtrend use parameters W,X,Y,Z.

    I’m off to bed, woke at 05:50 this morning! 🙂

    1 user thanked author for this post.
    avatar Khaled
    01/01/2022 at 10:43 PM #184290
    Khaled

    I’ve attached the file of the strategy in my previous post (it’s from Juan where I added an additional filter with Donchian)

    Good night!

    01/02/2022 at 8:46 PM #184363
    Khaled

    I suspect the problem comes from the period of sampling, not from the frequency of updates. for the Weekly case, I have compared (1) the trades with parameters established over a period of 3 months with (2) trades with parameters established only over the previous week. So even if the System is robust, the sampling period of one week is too short to cover various technical configurations.

    When I’ve time, I’ll make the test with parameters established on sliding 3-month vs those determined the previous quarter and will let you know.

    1 user thanked author for this post.
    avatar GraHal
  • Author
    Posts
Viewing 12 posts - 1 through 12 (of 12 total)

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
algos don’t want to start up in auto trading
high frequency micro trading strategy – aapl
is there a easier way to hide the code on algos?

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