1 registered members (AndrewAMD),
678
guests, and 6
spiders. |
Key:
Admin,
Global Mod,
Mod
|
|
|
Re: Z5
[Re: jcl]
#437450
02/17/14 11:27
02/17/14 11:27
|
Joined: Sep 2013
Posts: 73
Mangal
OP
Junior Member
|
OP
Junior Member
Joined: Sep 2013
Posts: 73
|
When doing backtests, some questions pop up regarding the WFO analysis. Checking the same script in different periods of time I get the following results:
2000-2005, WFO 3 cycles, Anual Return = -58%, nr of trades 55 2000-2005, WFO 10 cycles, Anual Return = 37%, nr of trades 137
Is the result with larger number of trades with 10 cycles rather than with 3 cycles more realiable?
Backtesting the same script through other periods of time: 2005-2010, WFO 3 cycles, AR = 93%, nr of trades 93 2008-2013, WFO 3 cycles, AR = 79%, nr of trades 83.
The results are different depending on the period of time considered. Would that be in backtesting standards a reasonable result?
In the three tested periods of time it gives different results.
Do the Zorro estrategias, Z5, Z12, and so on, have been tested in different periods of time than those one can test by default?
|
|
|
Re: Z5
[Re: jcl]
#437527
02/18/14 21:10
02/18/14 21:10
|
Joined: Sep 2013
Posts: 73
Mangal
OP
Junior Member
|
OP
Junior Member
Joined: Sep 2013
Posts: 73
|
When you say test periods have been selected for best significance, do you follow some specific rule to select them, like uptrend, downtrend and sideways, or you just consider a 10% aleatory sample of, say, 10 -20 year period?
In physical sciences the rule of 10% sample works reasonable well, though nobody knows why, but I don't know if same rule applies to the market.
Last edited by Mangal; 02/18/14 21:12.
|
|
|
Re: Z5
[Re: jcl]
#437599
02/20/14 08:36
02/20/14 08:36
|
Joined: Sep 2013
Posts: 73
Mangal
OP
Junior Member
|
OP
Junior Member
Joined: Sep 2013
Posts: 73
|
The fact that the backtests are based on WFO analysis, if one develops an strategy which results are based on the results obtained in recent period backwards, then one has to keep on optimizing it every certain period of time, say every month or couple of weeks, right? Though in Z5 this is not necessary as you were mentioning somewhere.
Would you advice to keep on optimizing a particular script every certain period of time or rather any number of 50-100 trades backwards for degree of freedom?
This brings the question also that better keep changing the script every certain period of time?. Some particular script adapts better in the inmediate 5 years back than in the 5 years old back in the past. For instante, in one script I get with 10 cycles WFO:
2000-2005, AR = 4 % , nr trades 193 for two degrees of freedom 2005-2010, AR = 191 %, nr trades 313 2008-2013, AR = 77 %, nr trades 282
As far as I understand, if a script doesn't consider the trend and it behaves in a similar way in an upward, downward and sideways period, then it gives more confidence. Other reasons, like periods of less volatility may be also behind the different behaviour.
Last edited by Mangal; 02/20/14 08:58.
|
|
|
Re: Z5
[Re: jcl]
#437754
02/25/14 02:07
02/25/14 02:07
|
Joined: Sep 2013
Posts: 73
Mangal
OP
Junior Member
|
OP
Junior Member
Joined: Sep 2013
Posts: 73
|
There is something that escapes to my understanding here. Supose you take the following script:
function run() { BarPeriod = 1440;
vars A1 = series(priceClose(2)); vars A2 = series(priceClose(1)); vars A3 = series(price(0)); //Vlr = 10*PIP
if (A1[0] < A2[0] and A3[0] < A1[0]) enterLong(); if (A1[0] > A2[0] and A3[0] > A1[0]) enterShort();
//Stop = 30*PIP; //TakeProfit = 500*PIP; ExitTime = 1;
}
It gives the following results:
A1,A2,A3 compiling................ Error 055: EUR/USD 2014 price history missing BackTest: A1,A2,A3 EUR/USD 2009..2014 Profit -33$ MI -1$ DD 33$ Capital 32$ Trades 176 Win 0% Avg -2.4p Bars 2 AR -26% PF 0.00 SR -8.92 UI 0.0% Error 64%
As you can see there is Win 0% and this can not be. If it really would close at the Close of present bar or Open of next bar, there should be some % of win. It could be small but not 0%.
So ExitTime = 1, I don't know exactly where it closes the open position. Or the entry point doesn't happen exactly at the moment indicated in the script. Any light about this?
Last edited by Mangal; 02/25/14 02:25.
|
|
|
|