High win% system – faster reveal of curvefitting in demo live run?

Forums › ProRealTime English forum › General trading discussions › High win% system – faster reveal of curvefitting in demo live run?

Viewing 9 posts - 1 through 9 (of 9 total)
  • 08/03/2018 at 7:57 PM #77508
    YachtTradaren

    Im trying to develope a scalping strategy. I highly suspect that its over optimized. However, this made realize that a high win% system should likely reveal itself as curve fitted in a earlier stage than f.e a trendfollowing system with less win%, during demo live testing. With “earlier stage” I mean time. Do you agree with this?

    08/04/2018 at 8:04 AM #77521
    Vonasi

    Impossible to confirm because if a strategy is curve fitted to the most recent data then it may carry on working for a while but then when the market changes to a different structure it will most likely fail. We do not know when that change can occur – it might be in one hours time or 1 months time or one years time!

    A low win rate trend following system will need longer forward testing if the quantity of trades is lower due to the fact that when you are onto a winner you will hold it for a long time so you might start a forward test with either a higher than expected win rate or a lower than expected win rate depending on when you first found that first winning trend to follow.

    Personally I gave up on the concept of scalping a long time ago. You are fighting the spread while working with a very short period of very recent very noisy data to base your idea on. Maybe with MTF now available it might be worth looking back into but we still unfortunately have the short amount of data to work with as soon as we start switching down to faster time frame charts.

    08/04/2018 at 11:30 AM #77536
    YachtTradaren

    Okok. The more weeks my system continues to perform well during the forward testing the more probable the system is to be robust, is that correct? Also isnt it better if the correlation of the forward testing and the backtesting is high, versus a lower correlation between them but with increasignly profits during forward period?

     

    08/04/2018 at 12:04 PM #77538
    YachtTradaren

    About scalping, a friend of mine had a quite interesting journey. He would just by purely on feeling on his Iphone :S 😀 When the price according to him “had gone up too much” he would sell and vice versa. The were more or less no use of SL. He was trading with warrants – mini futures (with inbuilt issuer SL). Of course sometime the drawdown periods of the account was crazy, at one point it was ~50% down from the initial starting capital but he recovered. However the holding periods were about 15 min on average I would guess. But it happend a few times that the positions was hold over night when it was in a losing state (happend like 20 times).

    Surprisingly this account had a good streak for like one year with, to me, very impressive returns. At one point it was up 100%. There were like 8 positions on avg per day and target of about 5-15 points. This was in DAX, and during the trend of september, october 2017 he got wiped. I think he woulda survivied with some proper risk management, and thats why im a believer in this type of strategy. There were no routines or strucutre, all decisions: size, entry, target etc etc was based on feelings. I was living with him a lot so I witnessed the most of it.

    08/04/2018 at 12:14 PM #77540
    Vonasi
    YachtTradaren wrote:

    Okok. The more weeks my system continues to perform well during the forward testing the more probable the system is to be robust, is that correct? Also isnt it better if the correlation of the forward testing and the backtesting is high, versus a lower correlation between them but with increasignly profits during forward period?

    Yes but if the back-testing is only based on a small sample of recent data then the likely hood of it still working in a years time is impossible to quantify. For example – you back test on three months data then forward test for three months and it performs the same. Maybe you got lucky and the market this three months happens to be behaving the same as it did in your three month back test period. Then you go live with real money and a market changing event happens the like of which was not seen in your very short six month testing period and your strategy empties your account. It is the problem with fast time frame testing.

    08/04/2018 at 12:21 PM #77542
    Vonasi
    YachtTradaren wrote:

    he got wiped.

    That says all I need to know about his trading style.

    As for trading on feeling – I often have a feeling that a horse is going to win a race because I like the name but I resist the urge to bet on it as it is just gambling. If however I read the form book and the form book told me this horse had never been beaten and that all the other horses in the race only had three legs then I would bet on it because my back tests and experience would tell me that a four legged horse who has never been beaten will always beat a three legged horse. If I bet just based on my feelings how do I not know that there is not a six legged horse in the race?

    I suspect your friend got off on the adrenalin rush of the highs and lows of the winning and the losing.

    08/06/2018 at 12:31 AM #77589
    YachtTradaren

    I suspect your friend got off on the adrenalin rush of the highs and lows of the winning and the losing.

    Yes, hahha actually thats spot on. He told me he loves to gamble and he also refers to himself as one. Its good tho that he realized he couldnt continue trading, so maybe not a true gambler after all.

    What is a decent amount of data in your opinion for timeframes below 60 min?

    Also when I apply a filter that increases my win chance slightly but it also removes a huge amount of trades it doesnt feel good. That is bad right? if so, is there some formula to calculate ratio/relation between amount of trades and win chance, like weighing the win chance against the amount of trades or is this irrelevant?

    Sorry for so many questions and thanks for taking time. If you cba answer more its fine 😀

    08/06/2018 at 8:16 AM #77606
    Vonasi

    It is impossible to put a level on the amount of data that you need – the simple answer is as much as you can get and preferably more. A lot of people seem to go straight to whatever time frame chart that they hope to finally be trading on but in my opinion they would be better to start on a monthly chart and work out a simple filter that would only have them trading at the right times and then study the weekly and work out the same thing – then the daily and then move on to the hourly. The reason for this is that when you are staring at the limited data in a sub 1 hour chart you need to have an idea of the overall market picture. If you are creating a long only strategy for example it is good to know whether you are currently trying to develop it on a market in a downturn or a market that is hitting new highs as otherwise when you trade it you may be in for a shock if the market structure suddenly changes on you. It is best to only let it trade when the market structure matches what you developed it on. It seems obvious but most people just try to develop a five minute strategy on a couple of years of data and then expect it to work forever in any market conditions and then seem surprised when the forward test does not match the back test.

    A filter that wipes out masses of trades for only a tiny gain is a filter that is not worth having. I test every condition by removing them one at a time to see what true effect they are having. If not much difference then they get binned and if a lot less trades and no real gain then they get binned too.

    As for quantity of trades in any test – more is better. Having said that if the strategy is very simple and only cherry picks one or two prime trades a year then you can possibly expect very few trades but very high quality trades. After a lot of staring at charts and indicators you start to get a feel for what is fitted and what is actually just matching your expected probability. In the real live trading world a few very high quality trades is what we want so as to minimise risk exposure but in the back testing world a large quantity of high quality trades is what we want so as to have confidence in our test. No one ever said that trading is easy!

    08/06/2018 at 6:10 PM #77665
    YachtTradaren

    Awesome answer. Im sure this information will help me a lot, it seems very logical, Thanks!

  • Author
    Posts
Viewing 9 posts - 1 through 9 (of 9 total)

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
buy a long every week if the high is higher than the last high
differnce between live & demo
long time to run system code causes rejection?
proorder – changing a parameter/variable live in a running trading system
v12 for prt-ig now live on live

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