Optimising – Curse or a Blessing??

Forums › ProRealTime English forum › General trading discussions › Optimising – Curse or a Blessing??

Viewing 15 posts - 31 through 45 (of 77 total)
1 2 3 4 5 6
  • 04/13/2018 at 8:47 AM #68246
    jebus89

    id like to share how i make my systems.. Lets say im using 1h on wall st, i get (premium) about 10-11 years of backtest.

    1. i create my thesis/system on entire data set, no optimizing what so ever. If i get a PROFITABLE (not a fantastic but good looking EQ curve) i go to step 2.
    2. then I use ONLY data between say 2012-2016 january. So entire 2016+2017+2018 is “out of sample”, and i start optimizing.
    3. After i have optimized EVERYTHING that i need to optimize, i go back to checking ALL the data. If its curve-fitted then 2016-2018 should basicly look like shit. If its not curve-fitted then 2016-2018 should be “ish” as good as backtest.
    4. If everything looks good, i optimize on the entire data-set with vartiables CLOSE to the ones that has already been chosen. Lets use RSI and macd as an example here. If the “closed dataset” says RSI = 55 but entire dataset wants me to use rsi = 57, then ill use 57. Thats so close to 55 that it shouldnt be enough to curvefit my entire thing.
    5. after the last optimzation, unless its REALLY connected to that 1 market, i should be able to see good-ish results in similar markets. So if i make an algo in DAX, i should see some good results in Wall st/tech 100/spain/nikkei/US 500. (obviously u might want to use bigger stop and targets just because of the markets being different size (3000 pips vs 25.000 pips from us 500 -> wall st…)
    6. If it looks ROBUST (aka profitable in out of sample data as in other markets, the other years etc) then i run it in demo for 1 month. Then live trade.

    (i also check monte carlo simulations, walk forward etc. The more TESTING done, the more confidence i can have when i run it live.)

     

    Edit: Dont use optimizng as an excuse or as a tool to help you make profitable systems. Use optimizing to find the BEST values for your already profitable system!! 🙂  It should WORK without being optimized. if your just taking 3 random indicators and optimizng everything from 0-100 then your gonna data-mine and curvefit your system.

    If you however create a system that works in 5/5 similar markets WITHOUT optimizing, you should feel safe when asking for a better value from your base value +/-10 points..

    If you use RSI = 70 as a base, and that seems good in OOS and other markets. Then try asking for values between 65-75 and see if you can optimize it. But never ask for RSI 0-100.. thats curve-fitting waiting to happen

    3 users thanked author for this post.
    avatar buffster76, avatar Leo, avatar GraHal
    04/13/2018 at 9:28 AM #68250
    Despair

    Your procedure looks sound but I think running just 1-2 months on demo is not sufficient to judge the quality (if it is not a strategy trading very often, like several trades per day). I think it is more like 4-6 months evaluation that you need.

    04/13/2018 at 9:38 AM #68251
    jebus89

    @Despair I can of course see why you would want more demo-time! More new data = more confidence if it goes the correct way!

    At the same time! (and i only speak for myself with my own experience of course) im pretty confident in that my systems PROBABLY wont work forever. So im thinking if it does what its suppose to do for 1-2 months in demo (usually about 10-30 trades) and i can see from backtesting horrible parts of the markets (bear markets and big turns down for longer periods) that the system will not lose a lot of money, then that makes me confident enough. Also if my systems stop working i 6 months i would like to trade 4-5 out of those 6 months.

    But yea i hear you, i also feel like the systems i have for now wont just stop after 6 months, they’re just too basic and (hopefully) not curve-fitted so i feel confident in running them 12 months straight now.

     

    I would never run a system that could leave me broke if the market tourned sour for 6 months..

    04/13/2018 at 9:52 AM #68252
    GraHal

    Got any Systems you might share jebus89 … all sounds good to me!

    I watch closely any / all Systems I am running … all part of the fun!?? 🙂 🙂

    04/13/2018 at 10:01 AM #68253
    Despair

    @Jebus: I have seen a study regarding just this topic. It examined how many strategies that where profitable for a test period of X months continued to be profitable afterwards and this study showed that it takes 6 months to sort out the good ones with reasonable probability. After 1,2 or 3 months there were still too many that turned sour after the test period. I was also amazed and didn’t thought it would take so long but this was the result of the study. Wished I had the link to the study…

    Then you you are talking about up to 30 trades within 1 month. This is really a lot. This means the strategy trades more often then once per day on average. With so many trades one could maybe get along with 3 months evaluation only.

    04/13/2018 at 10:42 AM #68255
    jebus89

    Oh that sounds interesting. If u find the link let me know, would love to read.

     

    Im still pretty new in this game and havnt been algo-trading for even 12 months so who am i to say whats what! 30 trades is more like 2 months of demo 🙂

     

    So what the study showed, if a strategy is profitable for over 6 months then theres a big chance of it being profitable for.. how long? 🙂

     

    i have a couple of strats that takes 1,x trades pr day. but majority is more like 0,5~ pr day.

    04/13/2018 at 11:39 AM #68259
    Despair

    I will post the link, if I find it but it was like I said. It needs 6 months of test period to be reasonable sure that it proceeds in the same way.

    The study didn’t say anything about for how long after this test period the strategy will work but from my own experience, a good strategy can work for many years (with eventually reoptimizing parameters). Of course a strategy can stop working through regime shift from bull to bear market but that mustn’t be the case so quickly. If I read what you write, it sounds like you expect a strategy to only work a few months. Such a sudden death of a good strategy I have yet to experience.

    For example I have strategies that worked very well last year when things were pretty bullish but the same strategies earned good profit when things went south this year. So even if we have a regime shift it needn’t render your strategies worthless (at least if they are not long only or so).

    The trade frequency you mention 0,5-1 trade/day is for example for a strategy on daily bars almost impossible. There you maybe have 1 trade every 1-2 weeks and the strategy can stay in a trade for several weeks. This is at least so for my best strategies.

    04/13/2018 at 2:47 PM #68269
    jebus89

    Youre correct, i dont trade on daily candles, all my algos are 15min – 1h.

     

    How often do you feel the need to re-optimize?

    How can you tell that its time to do so, and not just a “slow period” for your algo?

    Im glad to hear that all your strategies have made it for so long! Thats good news.
    personally ive activated most of my algos in january and its been good so far.

    2 have been going on live for about 6 months now.

     

    if these keep performing for years, i would be happy 😀

    04/13/2018 at 7:50 PM #68313
    Despair

    How often I reoptimize depends on the strategy. For strategies on daily bars i reoptimize maybe ever 6-8 months. If you have a WFA to begin with you can simply take the same intervalls that worked fine in the WFA.

    Well by far not all my strategies have survived. This would be amazing. But I’m in the tripple digits % for the year, so I can hardly complain.

    1 user thanked author for this post.
    avatar Inertia
    04/13/2018 at 9:12 PM #68321
    verdi55
    GraHal wrote:

    You seem / sound disillusioned verdi55 ?? Have you lost a bundle recently or have you now come round to realising “All is Folly” (Ecclesiastes ** ?) ?? Maybe when we get Mutlti-Timeframe and better still Multi-Instrument then success rates may improve? GraHal PS Is Multi-Instrument ever likely to be a possibility??

    Taking illusions away is one of the purposes of this website, as Nicolas has repeatedly stated.

    Multi-Timeframe is just one other possibility for different idealization of the past, called “curve-fitting”.

    Never wondered, why you sign each time, when you start a trading system “Past performance is no indicator of real performance” ?

    Or why all employees of ProRealtime are forbidden to trade themselves by their employment contracts ?

    The reason is to inhibit them to get crazy like most people here.

    04/13/2018 at 9:16 PM #68322
    verdi55
    Vonasi wrote:
    verdi55 wrote:

    It is not knowledge, only experience. For most people, experience of what does not work. Knowledge means something that works and is reproducible.

    The definition of knowledge is – facts, information, and skills acquired through experience or education; the theoretical or practical understanding of a subject. Knowledge is gained from experience of what did not work. By knowing what did not work we have a better likelyhood of now doing what might work. We have a theoretical idea to move forward with – otherwise known as knowledge. Unfortunately my knowledge has not progressed very far on why you keep posting videos from YouTube. I know why you probably shouldn’t so I guess that I am halfway there with that knowledge! 🙂 p.s. Even Greece is now developed enough to no longer have video shops!

    In my eyes, knowledge is : history, chemistry, physics, and the like. Or a software like Prorealtime, where things in most cases work as expected.

    Trading never works as expected, as we all know. Most people here never make any money, I am sure, although they keep dreaming of it. And i do not exclude myself.

    04/13/2018 at 9:56 PM #68324
    verdi55
    GraHal wrote:

    “All is Folly” (Ecclesiastes ** ?)

    I am not well-versed in the bible, although I do know the final fugue of Verdi’s last opera “Falstaff”.

    “burla”, in this case, I think, does not mean folly, but more “fun”, and the piece says that man is constantly led astray by his reason.

    I will neither post the full text nor the video here !

    04/13/2018 at 11:13 PM #68327
    verdi55

    Remember Lot’s wife who became a pillar of salt because she was only lookin’ back

    04/14/2018 at 6:08 AM #68331
    Vonasi
    verdi55 wrote:

    Multi-Timeframe is just one other possibility for different idealization of the past, called “curve-fitting”.

    Glad to see the old Verdi55 back with some good knowledge on show (and no videos!)   🙂

    I agree that multi time frame is very much another tool to help curve fit strategies but I personally can’t wait for it as it will make auto strategies easier to code if for example you want to look at the monthly candles, weekly candles and daily candles and decide that today is a good day to enter but only once the markets have opened on the hourly chart to avoid large spread. At the moment with single time frame if auto trading then you are forced to enter at midnight. The downside will be that I suspect that backtesting will then be limited to the shorter hourly chart history. Maybe Nicolas could confirm whether this is going to be the case or not?

    04/14/2018 at 4:59 PM #68362
    Nicolas
    Vonasi wrote:

    The downside will be that I suspect that backtesting will then be limited to the shorter hourly chart history.

    I confirm that this is the case as the version I am currently testing.

    1 user thanked author for this post.
    avatar Vonasi
  • Author
    Posts
Viewing 15 posts - 31 through 45 (of 77 total)
1 2 3 4 5 6

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
display daily or yearly high or low
10.3 bug alert or what??
15m or 4hrs trading
2 or more screener?
2 or more time windows for orders
alert in prt to enable buy or sell signal
all systems stopped > 10000 orders placed, edited or cancelled in one day
are we going sideways or down?
automated scalping system based on williams fractals or quadratic semaphores
autotrade or buy & hold?

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