I agree. The equity curve "looks" unusually smooth. Also, on the use of a Sharpe ratio, even if you're not pitching this strategy to anyone, the measurement and similar ones can be useful to show whether you are trading noise.
You are absolutely spot on, I am using discretion now and yes it improves. Im taking every signal but im getting a feel for when there are sentiment driven days and when things are technical. So yes i agree with the discretion part you mentioned completely. Im just not sure i can trust my emotions and like a plan to follow so maybe ill have to build this into my system once ive been live trading for a good period.
Hi, please carry on as im interested in what you are saying. I firmly believe ive found the latter but im open to being wrong. I really want to dismiss weather ive curve fitted and would like to discuss this further. I do agree with you all that the profit factor is too low and needs improving because although i say this is scalable, i just realised that even if i was trading 10 x higher volume, id still be left with a low profit factor as i demostrated in posting the 250 share version. So i will look at ways to improve, its hard though because spy, es or main market instruments are not very trendable and i find i can't run targets so much. Maybe if i take the system to a basket of stocks ill do better as i will be diversified and also non correlated a little with weak and strong stocks. This is a starting point as ive always wanted to take this to stocks. Also i should be able to narrow down entry critria with stocks as there is so much choice meaning the profit factor will be higher, but ill have to search for a filter that makes that difference. The steps i took with the backtest was this: In the beginning i found many many settings that gave excellent returns for a few years, but seem to not hold true over 10 years of data. So not sure if this was correct, but I then made it a critria for me to only look at systems that were year on year showing good profits over a 10 year period with 1000+ tardes. So i was now dismissing any system i found that didn't meet the 10 year consec profitability critria, again not sure if this was a correct way to test or not. My reason for the 10 year critria was i wanted something that was not luck or coincedent and felt if intially a system showed consistentcy this way then i could move on to step 2 etc........Ive thrown out countless systems all with profit factors above 2 , but again i may be wrong but i felt consistency was the key, and if a system showed 10 years of profitability then my chances were higher that the results would continue. All along in my mind the whole process ive been very concious not to curve fit and read most articles on the subject. Also common sense tells me that how is it possible to curve fit over 6000 trades and 10 years of data?, with just the same set of parameters?.............Am i wrong to dismiss the higher profit factor systems id found even though they were too inconsistent with not enough trades?...........Maybe i was, maybe i should focus on getting more frequency for the higher profit factor systems.....hmmmm.
Im really interested in what you are saying here. Im sure i have not curve fitted but i am in a biased postion so we need to investigate further. Let me know what you need to count curve fitting out. I have WFO tests and things like that if you want me to post. please read the steps i took with the backtest. What does 'look ahead' mean? sorry dont understand this term. Also looking ahead hole, what would indicate this?, thanks
Well the sharpe ratio is 0.74 which i know i will recieve flack for hence i didn't want to post it. Yes it is low but like i say its not a stat i personally use. The idea is i have enough money for one year's survival, so even if i had a crap year i wouldn't go broke. Not affecting the initial bank size to start this up so to speak. As for equity curve being unusually smooth, then this is great news for me, it means you like it. Now as long as everything is done ok, which is why im interested in this 'look ahead hole', if someone could clarify. Regards,
The curve being smooth implies the look ahead hole, and the Sharpe ratio being low puts it towards noise, which completes the picture for me. It fits, although not at the same time So .. I'd work on those two things. Not much margin for error, as someone else suggested.
Look ahead occurs in a backtest when the system logic references any price level that would be otherwise unknown during actual trading. For example the system If the high of tomorrow is higher than the high of today then buy at the open of tomorrow involves look ahead. But this is a trivial example. There are cases when bugs in the platform used impose a look ahead without you knowing that. Are you using TS? I would not trust any high level language unless I backtested manually myself.
Im quite sure my rules are nothing like that, but how can i be sure i have not used look ahead?....What would be the one thing i could do to rule this out?......yes tradestation is not ideal but thats way ive had things programmed into the entries and exits which are not standard in the normal platform, to make my testing realistic.
So i have fitted 10 years of data to suit and over 6000 trades to suit and fit?....Is this possible over such a large sample?, with the same set of rules, i guess it is but highly unlikely surely? So there is no edge?, is it just coincedent then it makes money year on year? sorry im obviously missing something but it will become clearer once i can digest what your saying. Let me do some research on this 'look ahead'. Thanks for your comments regards,