Backtesting with tick data - page 4
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
[...] noticed that the strategy tester just stopped opening any trades after around 2 weeks. [...]
You'll have to post your code to get help with that... I recommend opening a new thread though.
my dream/aim: What I was hoping was that give the strategy tester good historical data (especially if you can get tick data) and you will get a good insight into how tyour strategy would have really traded over that history (slippage, spread variance etc. accepted). But now im begginnning to dout whether this is attainable. if the strategy engine can actually deliver this. IS THIS AIM ACHIEVABLE WITH METATRADER? Someone throw me some hope!
The MT4 Tester is simply NOT designed to backtest any strategy that relies on inner-M1 movements. That's why it's oversimplified - doesn't support real tick data, variable spread, slippage, delays or any other 'real world' phenomena. For strategies that rely on larger timeframes, it is quite adequate (IMHO).
There are methods to overcome some of the limitations - real tick data and variable spread can be achieved via the methods mentioned on Birt's site. Slippage, errors and delays can be simulated via code. Whether or not this is worth the effort is a matter of opinion, but if u are looking for a platform that natively supports these features then MT4 is not the right one...
I have a few strategies and a few didnt do very well in backtest (despite doign well in my limited forward testing so far) and some did do well in backtesting.
I think this method of tick backtesting does work you know. BUT it does take away the time frame from you (so for instance if you want something to happen on a specific day - complicates it - which is something i will run into now as i want it to stop opening any new trades as we get to the back end of the month contract).
Strategy (over 3 months data)
profit factor: 1.55; total trades: 74 (on M1)
profit factor: 1.91; total trades: 67 (on tick)
Do you think such a strategy (with this profit factor) is anything to get excited about? I did a bit of manual tweak of parameters but i dont think i am in a curve fit situation. only a test with more data of course will tell.
I have a few strategies and a few didnt do very well in backtest (despite doign well in my limited forward testing so far) and some did do well in backtesting.
I think this method of tick backtesting does work you know. BUT it does take away the time frame from you (so for instance if you want something to happen on a specific day - complicates it - which is something i will run into now as i want it to stop opening any new trades as we get to the back end of the month contract).
Strategy (over 3 months data)
profit factor: 1.55; total trades: 74 (on M1)
profit factor: 1.91; total trades: 67 (on tick)
Do you think such a strategy (with this profit factor) is anything to get excited about? I did a bit of manual tweak of parameters but i dont think i am in a curve fit situation. only a test with more data of course will tell.
my 2 cent:
your strategy seems to be realy ticksensitive. the decreased amount of trades is a bit strange. still no variable spread includet. might be far away from reality
The strategy opens new trades once preceding ones have taken a TP or SL - so i guess the trade number difference is due to a difference in resolution - with higher resolution the TP and SL situation is different. Yes, it is pretty tick dependent i guess - why i so anxious to use tick data.
I know its early stages and i have a lot of work still to do.
What level of profit factor (on an actual live trading accoutn, over a long period) would people be excited about? Obviously over 1 but how much over this do people think - wow, this is a good system.
67 trades is a low amount of trades to do statistics, but i my memory doesn't fool me i have read somewhere that everything over PF=2 is consideret better than gambling. so you are near to that. ;)
but i find other values have more importance. for example drawdown/max consecutive losses.
Simply because a trader let run a EA as long at it is profitable, not realy depending how profitable. But then when the EA makes losses the EA maybe will be stopped. So these factors are trader dependent, how much are you willing to loose before the EA gets stopped?
//z
So, i am still testing my EA - am unsure how good it is really though. and seek some advice on the results been getting back.
It is for Light sweet crude oil (symbol CL). I have got tick data for this for the past 2.5 years now. I put this data in M1 bar form and used metatrader to test it. I will go on to test it in its actual tick form in the near future (probably using the method that we have thrashed out in this thread).
Results of this test (started with balance of 10,000):
// profit: 109,145
// profit factor: 2.65
// absolute drawdown: 748
// max drawdown: 30,764
// expected payoff: 90
// trades: 1204
You will notice it has a large drawdown at one point. Now, where it happens means it is of no consequence because by that time enough “fat” has been built up over and above the intial 10,000 deposit. So, just takes the 30,000 dollar hit, takes it down to like 50,000 and then carries back on skywards to 100,000. BUT if i was to start with 10,000 at around that point this drawdown would blow the account. So, i think one would need to startt with 100,000 so that if this drawdown happend at begiinnning of things – would be a 30 perecent loss which is agreeable for someone with an aggressive investment profile. So, the 100,000 profit in 2.5 years then becomes a 100% profit in 2.5 years (as oppossed to 1000%).
I am somewhat concerned that I may have a curve fit. I have manually tuned the parameters (but done no computer optimisation). I guess I am hoping that the long time frame of testing makes it less likely that it is a curve fit? the thing would be to have more data to test on – as a fresh test. but i dont have any more data at this time.
So, you guys have a lot of experience in testing EA’s. With the result parameters here – would these be good enough to make you think bloody hell – maybe this is a good EA?
(ps. with preliminary tuning that im doing now i have been able to get max drawdown down to 20,000 with no real loss of profit. can get max drawdown down further but then start losing end profit, fro example with max drawdown of 13000 but then end profit is less: 60.000)
Anyone can comment? xx