Great EA in backtest! - page 77

 
xxDavidxSxx:
I found this removed/blocked in the code. I unblocked it and am running exact same back test on $jpy to see if there is a difference. Dave

all the %'s were exactly the same and it took the same # of trades. But the $winnings was 1k more. Probly nothing to do with the change. 1k diff. out of 35k isn't much. I'll try running again w/o change and test on euro$ too.

I am running CT on 2 pairs. I renamed CT a different name for each pair used on.

Dave

 
xxDavidxSxx:
I found this removed/blocked in the code. I unblocked it and am running exact same back test on $jpy to see if there is a difference. Dave

hi Dave,

double BuyPossibilityQualityMid; is a variable declaration only. i've checked and BuyPossibilityQualityMid is not used in any calculations (v1.85).

 
kalamari:
1.85g is the same as 1.85f, trailing stop corrected only. so i added magicnumber autocalculation to v1.85g and renamed to v1.85g2, because we already have 1.85h. version 1.85g2 attached

Changing to g give me 4 losing trades in a row, whereas I had 3 winning trades in a row with 1.85f. I'm going back to 1.85f for now.

Thanks,

Fx Diva

 
fxdiva:
Changing to g give me 4 losing trades in a row, whereas I had 3 winning trades in a row with 1.85f. I'm going back to 1.85f for now.

it is because of trailing stop. try to disable it.

 

ive noticed something in CalculatePossibility (int shift) code

DecisionValue = iClose(Symbol(), 0, ValuePeriod * shift) - iOpen(Symbol(), 0, ValuePeriod * shift);

PrevDecisionValue = iClose(Symbol(), 0, ValuePeriod * (shift + 1)) - iOpen(Symbol(), 0, ValuePeriod * (shift + 1));

PrevDecisionValue is calculated with use of ValuePeriod. since there is ValuePeriodPrev shouldnt we use it to calculate PrevDecisionValue?

edit: or simply save DecisionValue to PrevDecisionValue after calculation?

 

Im just testing this out but i been doing backtesting since early MT3 days so i have an idea how it works:) There is a lot of stuff posted elsewhere in Russian about this system. In essence this is still only a frame work and not at all really an up and running system. The paid pro version is for that cant cant vouch thats its much better then this freebie.

Anyway lots of talk about Nueral nets and learning. Thats something to be patched in later and might even be losely in place on the paid pro version. I personally am working on Genetic Programming not NN based system. This script does not learn it dont even look at last trades other then to see how much money you got left. but simply looks back over the last 30 bars to get a feel of the market, spreads, volatiltiy etc and stores these parameters in about 90 seconds. The rest of the calcs are all probabilty based. That said there are mathmatical grounds for slightly increased performance in the slightly larger time frames ie 30 to 60 mins. At 4H and above techs can play a part in discovering postions becuase your over the noise floor but this script is not really made to trade longer time frames nor do i think its suited to do so. If you like 4 H trading then use something else instead. Also you will die of boredom i think waiting for a week to collect 15 pips.:)

Some things to note about this the scalper mode 1 min must never be activated at the same time as longer time frame trading ie +5 min bars. Make sure BlockPipsator = true. Most brokers will not allow this feature as it puts a lot of strain on the server by asking what their price is every tick and then try to place an order 1 pip each side of it to collect 4 pips. In fact some will ring or email mail you to tell you your about to be placed on phone trade only!! In addition and this is VERY important..MT only allows backtesting to 1 min. During that 1 min they are often lots of bars which the backtester will never see particualry in around mild news time session open etc. So in effect your days of backtesting to see the effect of this feature will never be realised unless you run live. Becuase of this the backtester will mistakingly believe that stops like 5 9 or 11 pips etc are good values when in fact you should double this in realtime. The author of this script suggest 18 pips if not higher to extract 5 pips realtime. i agree with this based on my own scalper based systems. While on backtesting ALWAYS download the 1 min ticks from Alpari so you get 90% modeliing. This is far from perfect but close enough to get a feeling but performance in realtime will always be least 50% lower then MT for scalpers. On large time frames its pip perfect. I have EA running for a year and the forward and backtest results always conform to exact stop and tp points. Becuase Alpari provide this tick data it only really works correctly on their setup. For bench mark testing and comparing results use Alpari only. Traders using others will be faced with different time zones and complelty different qaulity data so you will never get the same results even when time shift is applied becuase your broker data will overwrite last 8 weeks or so of ticks etc even when downloaded from Alpari.

SO due to this you should be looking towards longer time frames i suggest 15 mins will give about 2 trades a day. 30 mins and 1 Hour may only give 3 or 4 trades a week. Also on larger time frames you need bigger stops. There is no such thing as a free lunch. You must give in order to recieve. If you want 20 pips be prepared to give 40. In fact the technical information suggest that you dont use any stops ie set stops 0 this will envoke dynamic auto stops which generally run out to as far as 70/100 pips on 15 mins bars. There is a significant gain increase if you do this but really you need to stick around and keep an eye on the system. You might be right to think this is terrible risk but the odds of a stop being hit is reduced about 20 fold. This allows perhaps as much 50 consecutive winners at 25 pips each before you take a 70 stop hit.

There is no need for intelligence becuase basically this is running on luck. The reason is under 30 mins is deemed chaos noise so if your trading noise and chaos its all down to money man and probabilty that price will return beyond where is came from. The author goes on to say this is NOT a full time auto-machine. Its up to you to make sure you stop trading before major news. This script works on normal market noise not impussilve news burst will kill it everytime. Becuase of this manaul interaction its actually not possible to back test it properly. So its a descretional tool. It will cope with euro land news etc but definite off for NFP!! I think its interesting and worthy of a closer look but it has its place in the tool chest.

Oh i forgot to mention becuase we are trading chaos there is no advantage to setting trailing stops. Thats used for break outs on larger time frames and even then a Tp and hard stop works much better. In fact my tests show trailing stops is a bad thing becuase often in realtime you get bang bang up down 25 pips in 5 seconds which will clear out trailers but wont even be seen on 1 min test bars.

 

Nice post Bolt. very informative.

what are your suggestions on 1 hr time frame. You spoke alot about 30 min or less. But I belive 1hr time frame is the way to go. It gives the best risk/reward ratio. even with no s/l so that the auto s/l takes over, risk reward is close to 1/1. When looking at the losses it takes compaired to the wins.

Dave

 

I re-ran my $/jpy back test with s/l set to "0". Allowing the EA to set its own automatic s/l.

Results were horrible. The win% was up slightly from 63% to 67%. But the profits were down to 24k from 35k, and max draw down was 73%.

73% is way to much of a draw down.

Dave

 

I will be out of town for a few days....I will check in as occasion permits from the road with our laptop. I think I'll leave my Ea's running on low lot settings. I'm not sure if I will restore the allow both long and short trades or not. Judging from just todays' performance I'm not sure it helped to limit them.

 
xxDavidxSxx:
I re-ran my $/jpy back test with s/l set to "0". Allowing the EA to set its own automatic s/l.

Results were horrible. The win% was up slightly from 63% to 67%. But the profits were down to 24k from 35k, and max draw down was 73%.

73% is way to much of a draw down.

Dave

73% is suicide