Jiva Dragon | Page 39 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Enter The Dragon - GBP/JPY 2 replies

Info on the Borihei-Dragon Scalping System? 15 replies

Jiva's 5,5 Trade Journal 5 replies

The Dragon 441 replies

***Dragon II*** 15 replies

Attachments: Jiva Dragon
Exit Attachments

Jiva Dragon

  • #761
  • Nov 4, 2010 1:46am Nov 4, 2010 1:46am
Quoting jiva34
Beta testing for v14.2

Feedback welcome...

template and .set file in zip..
50-50 edge test 78.77% win rate

106 trades/12 mos.
stop loss 30....
Hi there , I sign up as beta-tester I made my first one

Pair eur/usd , with the set file uploaded by JIva34 in the post I am quoting.

TF H1
1 Year
here is the graph and the report, in the report is amazing to see the max drawdown 2.85% (I see a little difference with results on this quoted post, maybe different time lapse, mine were done in a 1 Year basis)


EDIT : I noticed my backtest was made with 10,000.00 USD now I made it with 1,000.00 and max drawdown was 14.54%, still very acceptable in my humble opinion, profit is obviously the same as I didnt change anything else in this test.
_________-
Sequoia
Attached Image(s) (click to enlarge)
Click to Enlarge Name: Report EU Jiva V14 presets uploaded.jpg Size: 133 KB
Click to Enlarge Name: TesterGraph.gif Size: 17 KB
  • #762
  • Nov 4, 2010 1:56am Nov 4, 2010 1:56am
Quoting Sequoia
Hi there , I sign up as beta-tester I made my first one

Pair eur/usd , with the set file uploaded by JIva34 in the post I am quoting.

TF H1
1 Year
here is the graph and the report, in the report is amazing to see the max drawdown 2.85% (I see a little difference with results on this quoted post, maybe different time lapse, mine were done in a 1 Year basis)


EDIT : I noticed my backtest was made with 10,000.00 USD now I made it with 1,000.00 and max drawdown was 14.54%, still very acceptable...
Great. We're definitely on our way!!
What broker?
  • #763
  • Edited 3:13pm Nov 4, 2010 2:29am | Edited 3:13pm
Very nice looking short entry EURAUD.
Smart, very smart...expert adviser...
This is the old series v11. Huge stop loss 250 pips..

edit first take profit reached, 2nd take profit on this old v11 series is 890 pips.
trailing stop is 77 pips, so either EURAUD makes a big move or the end position stops out at the trailing stop, falls back to break even for 0 pips, closes above the channel hi which will act like a trailing stop or reaches 890 pips...
Attached Image(s) (click to enlarge)
Click to Enlarge Name: euraud.gif Size: 63 KB
Click to Enlarge Name: euraud2.gif Size: 54 KB
  • #764
  • Nov 4, 2010 2:32am Nov 4, 2010 2:32am
Quoting jiva34
Great. We're definitely on our way!!
What broker?
Instaforex
I downloaded all data available from History center, that is the way we should prepare before running backtesting, right?

_________-
Sequoia
  • #765
  • Nov 4, 2010 3:30am Nov 4, 2010 3:30am
Quoting Sequoia
Instaforex
I downloaded all data available from History center, that is the way we should prepare before running backtesting, right?

_________-
Sequoia
yes. so there is 90% modeling..

Backtesting does not guarantee future performance. So we are trying to see how protect against the unforseable with these rolling and walking forward tests, minimal stop losses, and maximum win rates..

If you can do it, it would be a big help, we need to find the set values for the other pairs for v14.2.
Look over the procedure described in post #762 and see if you can try with a pair other than EURUSD and actually discover a usable set. Like maybe work on EURGBP.
Start with the old set values for EURGBP v11 and follow the steps in #762.
http://www.forexfactory.com/showpost...&postcount=762
In your spare time of course!!
  • #766
  • Nov 4, 2010 12:41pm Nov 4, 2010 12:41pm
v11 entered long EURCHF

I find it interesting how the targets so very often coincide perfectly with support and resistance, specifically the Wood pivots, although they were not calculated with any specific reference to the pivots.

80 pip stop, the old v11 set...10/17


edit several hours later, meanwhile back at the ranchhttp://cdn.forexfactory.com/images/s...nailbiting.gif
Up eventually will it hit the stop before it goes?


edit several more hours later
Stopped out, this one had 80 pip stop loss.
But still, now we see why we are so excited about v14.2. 20 and 30 pip stop losses possible with high success rates..

Attached Thumbnails http://cdn.forexfactory.com/attachme...1&d=1288802451 http://cdn.forexfactory.com/attachme...1&d=1288856868
Attached Image (click to enlarge)
Click to Enlarge Name: eurchf2.gif Size: 74 KB
  • #767
  • Edited 7:12pm Nov 4, 2010 4:58pm | Edited 7:12pm
Working on the walking forward test. Lot of data to sort through and keep filed properly.
What I am looking for is:
Will the set file that has been optimized for the last 12 months in back test be profitable for the next month forward?
Is 12 months the best length of time to optimize?
Would 6 months be better?
Will keep traders posted.
Using v14.2, superior trading options. Relying more on macd cross than dss.
So far results are very positive. Which means this would be a reliable technique to use for keeping the set files updated...
Will look into what is involved with Duskacopy data feeds. 99% modeling..
For now we are slogging along MetaQuotes data at 90%..
And have v11 set up on the pairs in demos...


edit:
interesting that after spending many hours doing the ground work for walking forward testing the results are not at all what I thought they would be.
What I see is that a set file that has been optimized for the previous 6 months and brings in very good profit, when it is applied to the next forward month (for which it has not been optimized) it acts in very unpredictable ways. For 6 such walks forward the total for the walks was a small loss, but when the optimization is looked at for the same period taken as a back test, it shows fabulous profit.
Now what does that say about losing money???
We will get around these things.
Thankfully we are finding the bug a boos..
One thing I notice is that the ea is built around a channel which is a variable. The more constants there are the better. But the entire strategy is built around a variable. It can be a strong point if the channel value can be regularly monitored.
The only constant in the entire system is achieved by using only the macd default setting. Time of day for trading can act a constant also. So we have a rather large window for trading but it is constant and it is in relation to the most active market times.
I will look at walk tests for 12 months for the same period that I tested for 6 months, maybe 6 months is too short a time. I will look for a significant change in increased profit.
And be thinking what constants can be applied to the system? So that when we program in "when this,
then that" the "when this" part is a constant...
  • #768
  • Nov 4, 2010 6:11pm Nov 4, 2010 6:11pm
Hi Jiva, that's me again . I found, that in new setting file for v14.2 (eurchf pair) there is difference between MACD setting used in .set file and MACD displayed from .tpl file. Which one is valid for this pair? Thank you.
  • #769
  • Nov 4, 2010 6:47pm Nov 4, 2010 6:47pm
Quoting kapybara
Hi Jiva, that's me again . I found, that in new setting file for v14.2 (eurchf pair) there is difference between MACD setting used in .set file and MACD displayed from .tpl file. Which one is valid for this pair? Thank you.
Posted the correct files in the zip for EURCHF v14.2
We are using default macd.(12,26,9).
Thank you.
So many details. I think I'm going mad..
Thanks for staying with this. Lots of twists and turns... What works and what doesn't. It's really important otherwise we will only lose our money... There's no getting around that truth... A lesson hard learned by those who are actually successful traders.
  • #770
  • Nov 4, 2010 9:55pm Nov 4, 2010 9:55pm
Quoting jiva34
(Where's jm2110? I think he will have something to say).

But we are scientist, we want to test this. I think it is a valid approach. More so than trying to zoom out to cover data for several years with one channel setting.

So let us see.
Hi Jiva,

I agree it is silly to discount something until you have tested it and that is why I stopped objecting - I have been observing while you discovered on your own what I have been pushing you towards.

The fact that I can take one of your set files and run it over a longer period and visually pinpoint on the p&l chart the point where the optimization started told me all I needed to know. Incredibly you had managed to curve fit. I say incredibly because usually it would be easy to fit 30 or so trades over a year but with your new 14.2 you are fitting 130+ trades and still coming up with a .set file that is no better that tossing a coin for periods outside the optimized period.....that truly is incredible.

The reason you are able to do this is that you have so many moving variables that you can structure a .set file that is almost like a "key" to that specific years data "lock". As you have rightly pointed out in your more recent posts you can try to hold some of these constant (macd etc) but as soon as you do that you will find that the EA will no longer perform nearly as well as it has been.

Behind the scenes I have been running backtests across much larger periods but am yet to find a combination that will make the EA successful over most market situations. So we come back to where we started.

It is clever that you have developed a key to fit most locks the real trick now is trying to understand WHY those variables are changing. The stoploss and Takeprofit settings have some effect but it is the macd periods etc that truly modify the EA's behaviour. It may be that there is no pattern to the changes that we can ascertain and unfortunately if that is the case it is very difficult to trade off a model that can only tell you what has happened and not what should happen.

But as you say we are scientists so I will continue trying to understand how better to predict the changes in these variables relative to what the recent price action has been.

I think the best place to start would be to do a 50:50 edge test on each 3mths or 6mths data i.e. keeping the SL and TP settings fixed and solving for all other variables and then looking at how those other variables are changing...are they random or can we find some pattern in their movement that we can link to the movement of some other indicator (or combination of indicators) running over the same data and therefore program the EA to move the variables on its own as the market dictates.

Nobody said this was going to be easy!

Jono
  • #771
  • Nov 4, 2010 10:13pm Nov 4, 2010 10:13pm
Jiva,
You may have already done this in your optimization process. Why not optimize for 1st 6mths, then with settings run for next 6mths and see results.
  • #772
  • Edited 11:13pm Nov 4, 2010 10:57pm | Edited 11:13pm
Quoting jm2110
Hi Jiva,

I agree it is silly to discount something until you have tested it and that is why I stopped objecting - I have been observing while you discovered on your own what I have been pushing you towards.

[color=black][font=Verdana]The fact that I can take one of your set files and run it over a longer period and visually pinpoint on the p&l chart the point where the optimization started told me all I needed to know. Incredibly you had managed to curve fit....
Again, thanks ever so much for your in depth analysis.
We will get there...
The main thing is the channel.
A breach of the channel is the signal.
The problem is the channel value is a variable.
We can find a solution.
We are saying the same thing now.
To be able to tie in to some other measure that is not a variable.

I was very surprised at the results of the forward walks. Of course it was a small sampling but still, it was a bit of a shocker.
Traders can be easily fooled by these back tests. Back tests are useful to see if the mechanics of the robot are opening and closing trades properly but aside from that, for predicting profits, hmmm. Well not of much value from what I just saw.
A bit like getting into the car and driving by looking in the rear view mirror. "Well the road has been straight for a mile or so now"....

I have a few ideas I want to check out..
http://t1.gstatic.com/images?q=tbn:x...-view1.jpg&t=1
  • #773
  • Nov 4, 2010 11:16pm Nov 4, 2010 11:16pm
Quoting junglist0000
Jiva,
You may have already done this in your optimization process. Why not optimize for 1st 6mths, then with settings run for next 6mths and see results.
Yes,
That is the basic idea behind the forward walks I have been doing.
I would backtest for 6 months then use that setting for a forward month.
Rear view mirror......
http://www.whoslosing.com/blog/php/Tree-car-crash.jpg
  • #774
  • Nov 4, 2010 11:30pm Nov 4, 2010 11:30pm
Quoting jiva34
Traders can be easily fooled by these back tests. Back tests are useful to see if the mechanics of the robot are opening and closing trades properly but aside from that, for predicting profits, hmmm. Well not of much value from what I just saw.

A bit like getting into the car and driving
It is true it is easy to be fooled. However LONG dated backtests are valid as if you have a model that can function over the past 5+ years (and ideally as in Birts tests back 10yrs) then it becomes statistically impossible that you have curve fitted over that many changing market environments.

You have done pretty well to fit over the past year with SL and TP at similar widths but as we have already seen even 18 months fitting is challenging. So anything that backtests over a long period and still looks consistently profitable with minimal drawdowns should be considered a valid guide for predicting profits.

As you say we are both on the same page now and you see that in order to do this we need to understand the relationship between price action and channel size.....let's hope we can find it.

Jono
  • #775
  • Nov 5, 2010 12:27am Nov 5, 2010 12:27am
There are many advices but I want to see what they tried and how the result is.....
I hope they aren't just a critic....
  • #776
  • Nov 5, 2010 12:49am Nov 5, 2010 12:49am
Quoting Taiyakixz
There are many advices but I want to see what they tried and how the result is.....
I hope they aren't just a critic....
I am just a critic.
You are the ones doing all the work.
I know you are getting tests with good results optomized for the one year.
It would seem you need to test random periods of time. like the last 3 years, total, maybe june 2009 to august 2009 and march 2010 to now...................
Things keep changing all the time.
Thankyou for spending so much time and effort to figure out what works.
Practice, practice, practice makes perfect. The more you put in the more you get out i hope.
Iam trying to back test and forward, gets frustrating and complicated
I always thought things should be simple, somehow, at least i have read that alot before.
  • #777
  • Nov 5, 2010 12:57am Nov 5, 2010 12:57am
Quoting Taiyakixz
There are many advices but I want to see what they tried and how the result is.....
I hope they aren't just a critic....
I think jm2110 has good points about the curve fitting for back testing and how misleading that can be going forward.
We are doing some more forward walks using different time spans...
Let us see...
We will post results...
It is very time consuming. So it will take a little time to compile.
  • #778
  • Nov 5, 2010 1:04am Nov 5, 2010 1:04am
Quoting kauaibobby
I am just a critic.
You are the ones doing all the work.
I know you are getting tests with...
Hey thanks for doing some testing.
If we keep an open mind the solutions are out there...
We'll find them.
The amount of details involved is mind boggling.
A big problem is just labeling and filing everything so I know what it is, so I can access later on.
  • #779
  • Nov 5, 2010 1:20am Nov 5, 2010 1:20am
Am taking backtest to 5yrs EURUSD and will see about forward walks...
  • #780
  • Nov 5, 2010 1:23am Nov 5, 2010 1:23am
Quoting Taiyakixz
There are many advices but I want to see what they tried and how the result is.....
I hope they aren't just a critic....
The idea of posting idea's on these forums is you get to share from everyone's experiences and hopefully create something better.

Over a week ago I pointed out that despite the initial results from Jiva looked good there was a flaw in the logic due to curve fitting and rolling optimizations were in my experience not the way to go. I was trying to fast forward Jiva and direct attention to where things are headed now.

If I had all the answers and knew how to make your EA perfect I would but this is a learning process for all. What I did know immediately when I ran the set files over much higher quality data from longer periods and could spot the optimized period was that this was a curve fit. Jiva now has discovered this the long way around by manually doing the forward walks but next time it won't take this extra week to make that conclusion.

I understand what you say about criticism for criticism sake but I think the process Jiva has been through and the fact that everyone is now coming to the same conclusion shows that I am not wasting either of your time.

I have the same goal as all of us here and that is to make a consistently profitable EA, let's hope that is possible.
1 trader viewing now