Jiva Dragon | Page 37 | 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

  • #721
  • Nov 2, 2010 1:08am Nov 2, 2010 1:08am
Quoting kauaibobby
Jiva thank you
You explained what the 50/50 is, I wasn’t sure what you meant by that.
It is starting to make sense all the hard work you are doing, I am slowly figuring it out.
I know brokers vary a lot, but I was wondering what we should experiment with our broker.
I know the offset is different for brokers. What about changing other setting what should we try first.
How much difference between FXDD and Alpari do you find you need to change settings.
I hope I am not asking to much of you to help me figure this stuff out, but I want to help and...

>>>I know the offset is different for brokers. What about changing other setting what should we try first.<<<
jm2110 has also requested I post the sequence I am currently using forvoptimizing. I will post that shortly.

>>>How much difference between FXDD and Alpari do you find you need to change settings.<<<

A lot. I start with the FXDD set file and see if it can work at all. Most of the time it doesn't work very well on Alpari. Usually it seems it is the channel that needs to be set differently. But I will post the sequence.
  • #722
  • Nov 2, 2010 2:00am Nov 2, 2010 2:00am
Quoting jm2110
[color=black][font=Verdana]Hi...
>>>As you know from my past comments I am a little dubious on rolling optimizations as I see them as just rolling curve fitting but I am happy to agree to disagree on that. I still fancy trying to optimize this EA over much longer data sets and seeing what the "average" performance is like. <<<

Yes, it would be wonderful to be able to optimize over a longer period of time. However, by the nature of the strategy itself it is not possible to do that .
As I posted earlier in response to this issue, because the main indicator is a moving average channel the value of that channel will change as the market changes. Just as in trying to find the moving average most appropriate for finding pullbacks changes for different trend moves. Sometime a 55 sma will catch the pullbacks, sometimes it is a 34sma. It will vary for a particular trend. So that is why rolling optimizations is really the only realistic and practical method for the Dragon.
If the strategy were based on an an event that was always the same, with no variable involved, then optimization over long periods of time would work nicely. Like traders who use candlesticks and bollinger bands, for instance. Hypothetically, when a bullish harami occurs and the candle closes inside the bollinger band, etc. That is always a case with no variable, it's a fixed occurrence. But the channel is moving with the market and does change its value. Maybe you do not like it, but do you see the issue? How a variable is part of the main strategy. The changing moving average channel.
I really see no practical way to get around this. The Dragon is confirming a strong trend with use of the multi time frame moving averages so it will not enter a trade without that confirmation, it nicely avoids consolidations that way.
So I see no problem in updating on a regular basis. Except, from a personal perspective, it would be nice not to have to do that.
  • #723
  • Nov 2, 2010 2:25am Nov 2, 2010 2:25am
Quoting jm2110
[color=black][font=Verdana]Hi...
>>>Given the speed at which you are pumping out these .set files you have obviously managed to work out a series of variables you optimize in steps rather than all at once as otherwise the time taken to do each optimization would take weeks. Therefore could you please explain what process you are going through i.e. which variables you target first and with what steps and then 2nd 3rd etc. This would help me greatly when doing much longer runs over larger data sets after which I can return with my findings.<<<

http://encefalus.com/wp-content/uplo...roadrunner.gif
You noticed.
Yes, other traders have also been asking for the sequence I am currently using.
I would like to say this is cut in stone, although there is a basic system to it, I allow for some flexibility.

No need to reinvent the wheel I start with settings that were already calculated for v11. Open the setting in v14.1.
1. First thing I do is remove the trailing stops, and then I set the 1st profit, 2nd profit and stop to the same value. I start with the smallest stop value possible. Usually I will try 20 for all three. I run it on control point as a rough draft technique.

If 20 returned a profit and the win rate was high, say over 65%. That's the value I use for the rest of the optimizations. I leave the 1st profit, the 2nd profit and the stop all set to 20 with no trailing stops.

If it was not acceptable I set all three to maybe 30. If that doesn't work 40. I don't want go over 50. If none of that worked then the channel needs to be optimized.

set everything back to 20. run control point for the channel. Start 5, step 5, end 65.

So like that. first thing is get the channel right and 50/50 to show high win rate. Other wise forget that pair. But they can all be done. I have not had to skip any pair yet.

Once the 50/50 is acceptable, and the channel is optimized.

Next is
2. dss (2 settings together)
3. macd (3 settings together) you can experiment with macd if you want, but I have found it is most useful set to the 1min chart using the macd cross
4 t3 (sometimes t3 is not good to use, sometimes it is).
5. then the moving averages start with higher and work down to lower
6 then back to profit 1
7. trail for profit one = 0 does not need to be set to a different value.
8. profit 2
9. trailing stop profit 2
10. If the results are good, leave the stop as it is. We want that to be as small as possible. If we can decrease a few losses by increasing the stop a little bit then we can optimize. But we want the take profit 1 to be more than the stop loss by as much as possible.

The initial run is control point. After each function has been optimized run the new setting on each tick to see the graph. Sometimes choosing the biggest profit is not the best choice. It may have introduced some losses too. So before going on to the next function make sure what has been optimized so far is acceptable.

I think that just about covers it.







  • #724
  • Nov 2, 2010 3:12am Nov 2, 2010 3:12am
I can understand how different sma settings better fit the market conditions.

However I would like to be concentrating on WHY that is the case. Is it because the market gets more volatile i.e. ATR increases, etc. If we can find some rough link then we can program that into the EA allowing it to automatically predict that the sma should now be looking at a longer or shorter window given how the market is evolving.

What continues to concern me is that if we can understand no link to why this variable moves around then we are just once again curve fitting and just because it happens to fit doesn't mean the very next trade is any more likely to follow this same pattern. The fact in my previous post that you can so clearly see the point on the p&l chart that the optimization took place shows that there clearly was no relationship to this set file in the month leading up to the period you optimized for.....this is worrying.

Even more worrying is when you said that you find your .set files have different SMA settings depending on which broker is used.... I am not disputing at all that this is the case but if you think about it it really is madness.

The EUR is EUR is EUR no matter where it is traded. I am yet to see anyone pitch a strategy where we can successfully arbitrage one mt4 broker against another. We know that if we have multiple mt4 platforms running their prices are within a fraction of a basis point of each other....funnily enough the fx market is very efficient.

However as you have discovered when optimizing you are getting wildly different variables depending on which brokers data you are using. In my opinion this has NOTHING to do with how the market is going to function in a live forward test it just shows that the brokers historical databases have holes in them and your modelling quality is not high enough...sure 90% sounds great but when you finally forward test you are getting 100% quality...that 10% can make massive differences. Sure bid/offer can make a difference in scalping strategies but in trades where we are trying to take 20-100 pips out of each trade the difference between 1 or 2 pips in bid offer is insignificant and DEFINATELY shouldn't be effecting the SMA periods we are using.

At this time the closest you can get to true 100% back testing is by following Birts instructions on achieving 99% through Dukascopy http://eareview.net/tick-data .....it takes a long time to prepare but once you have the data you can be fairly sure that there are no assumptions being made in your testing.

Saying we have a .set file for Alpari doesn't mean it is accurate to Alpari's live feed. All it says is that this is the .set file that gives us the prettiest picture using Alpari's old data which we know we're are only getting 90% modelling quality on.

I always choose to optimise and backtest using the highest quality data I can find - Dukascopy 99% and then take the resulting .set file and run it over Alpari etc...I know it will never look identical but I just double check that it isn't a disaster as I know with 90% quality it is never going to be "right".

If you really stop to think about the fact you are finding you need different sma's to show optimal results on different brokers in the same currencies (assuming the GMT offset has been adjusted) then I think it is time to upgrade to 99% quality because it makes no sense that something as fundamental as the sma would change ...yes you might find the take profit in one broker in 28 pips and another 26 because on a piece of data the price spiked 2 pips higher quickly but 99.9% of the time those broker feeds are going to be virtually identical and definitely not different enough to justify taking different sma snapshots.

Jono
  • #725
  • Nov 2, 2010 4:25am Nov 2, 2010 4:25am
Quoting jm2110
I can understand how different sma settings better fit the market conditions.

[color=black][font=Verdana]However I would like to be concentrating on WHY that is the case....
jm2110 does it again. Pushes me to the brink.....
How did you know I have been avoiding Duskacopy? I took a glance at some discussion of setting it up and said, yikes.
Well it looks like it's time, to take a second look as Duskacopy!!!

>>>What continues to concern me is that if we can understand no link to why this variable moves around then we are just once again curve fitting and just because it happens to fit doesn't mean the very next trade is any more likely to follow this same pattern. The fact in my previous post that you can so clearly see the point on the p&l chart that the optimization took place shows that there clearly was no relationship to this set file in the month leading up to the period you optimized for.....this is worrying.<<<

My reasoning at the present time is that it is not necessary to find out why a change is occurring. To use another indicator, ATR or whatever, The reason being is that the ea is set up to take trades only in the middle of well established trends. The entries are on pullbacks. It will not be able to catch the beginning of a trend, because of the use of mult time frame moving averages, the slower ones do not turn quickly enough to catch an initial breakout, but when there is a pullback an entry will be identified. Macd on the 1min time frame helps identify the entry.

So if a market turns it will take a little while before the ea will find entries. By the time the moving averages are lined up, the channel value will be identified also. So it is only the middle of the trend that is traded. Using the smallest stop possible reduces any losses that might occur should the ea buy a top or sell a bottom which it can do, since the moving averages are still aligned just before a reversal. But a small stop loss is livable. Hence v14.1 and the ability to set up small stop losses.

The ea does well to stay out of consolidations again due to the required moving average line up.

So I do not think you see how the ea itself actually functions otherwise you would not be arguing along the line of reasoning of trying to optimize over long periods of time. I think you miss the point. What you see as a weak point is actually the ea's strength.

So by testing over long periods of time your results are misleading. You are using the wrong line of reasoning to set up your tests. At least this what I am understanding from your arguments.

Please try to understand the strategy behind the expert adviser. I think you may be caught up in rules for the sake of rules and missing the point.

Thank you for throwing Duksacopy at me. I will be looking into it seriously.

Thank you for putting your valuable time into the thread.

  • #726
  • Nov 2, 2010 5:28am Nov 2, 2010 5:28am
Hi Jiva,

I understand the logic of the system and it is because of this that I am concerned.

Yes by waiting for the series of sma's to all line up you will never be getting the start of a trend, always the mid part as you say. But given you are trying to capture the mid part of the trend why do you always need to change the size of the sma window you're are looking at? The reason no doubt will be because as volatility decreases your optimization will show you need to get into the trend earlier so you can still capture you first ~ 20 pip profit so the sma window will be smaller to allow this. Likewise as volatility increases and therefore the swings are larger you can afford (and is safer) to wait a bit longer for trend confirmation so the sma window will widen.

As we have seen from my backtest not only do some previously highly profitable .set files (and thus sma windows) cause the EA not to be as profitable they can actually cause it to regularly lose money so this is extremely important.

The problem with rolling optimization is “what are you going to optimize?”....the last year every week? Well then the current market situation (the current week) is only going to have 1/52 effect on your optimization so if the market has suddenly changed it is going to take months until your .set file is accurately reflecting the new market behaviour. Or do we optimize just the last months data every week...problem then is there will only be 2 or 3 trades and then we are definitely curve fitting.

Hence I find the preferable approach is to either optimize over a longer period to find an "average" sma window that doesn't hit home runs but also doesn't lose money in most situations...OR try to link the sma window size to market volatility so we are immediately modifying the logic within the EA rather than lagging up to months behind.

Hope that kind of explains what I am saying.

If you have any questions re Dukascopy capture and conversion etc feel free to pm me.

Jono
  • #727
  • Edited 10:00am Nov 2, 2010 9:42am | Edited 10:00am
Quoting jm2110
Hi Jiva,

I understand the logic of the system and it is because of this that I am concerned.

[color=black][font=Verdana]Yes by waiting for the series of sma's to all line up you will never be getting the start of a trend, always the mid part as you say. But given you are trying to capture the mid part of the trend why do you always need to change the size of the sma window you're are looking at? The reason no doubt will be because as volatility decreases your...
>>>But given you are trying to capture the mid part of the trend why do you always need to change the size of the sma window you're are looking at? The reason no doubt will be because as volatility decreases your optimization will show you need to get into the trend earlier so you can still capture you first ~ 20 pip profit so the sma window will be smaller to allow this. Likewise as volatility increases and therefore the swings are larger you can afford (and is safer) to wait a bit longer for trend confirmation so the sma window will widen.<<<

I believe the volatility issue is dealt with sufficiently by the hi/lo channel. Once the channel extremes are breached by a close outside of either the hi or lo in relation to the respective trend direction it is a signal that the trend is resuming. Long a breach above channel hi, low a breach below. That is:a breach being- the candle closes inside then a candle closes outside.
For seeing into the future what the channel setting will be I do not think is practical. If traders could see into the future we would all be zillionaires by now. I do not see how ATR would be a reliable indicator for this. I have not worked with ATR per se, although I have looked at it a bit. It does not impress me enough to incorporate it into the coding...
  • #728
  • Edited 1:57pm Nov 2, 2010 11:33am | Edited 1:57pm
Quoting jm2110
Hi Jiva,

I understand the logic of the system and it is because of this that I am concerned.

[color=black][font=Verdana]Yes by waiting for the series of sma's to all line up you will never be getting the start of a trend, always the mid part as you say. But given you are trying to capture the mid part of the trend why do you always need to change the size of the sma window you're are looking at? The reason no doubt will be because as volatility decreases your...
Thanks to jm2110 for his persistence..
There is much very good insight and lots to be learned from experienced traders like jm..
I am changing the optimizing approach a bit. One thing I have decided is that slower moving average hi/lo channels are preferable to faster ones for this strategy. I have been able to optimize the 50/50 test back to 3 years so far and will see how far back we can go using the slower channels with fairly consistent results, at least on EURUSD in v14.1.

I hope traders see why we urge not to use real money yet...The Dragon is under development....

The graph is a rough draft. 15 min setting 36 months previous 20 pips stop 20 pips limit, 70 ema channel. Pretty much bare bones 50/50 test.
Attached Image (click to enlarge)
Click to Enlarge Name: eurusdv1411-250-503yrscontrol.gif Size: 9 KB
  • #729
  • Nov 2, 2010 2:52pm Nov 2, 2010 2:52pm
Hi Jiva,

today early morning one sell trade occured, which was taken by version11 (version11.jpg) -> both positions on EURUSD hit SLs (2x110 pips). Here is screen shot from version14.1 chart (currently updated) -> I browsed for the critic time (version14.1_M1.jpg). Now I'm wondering, if it (v14.1) would take that trade or not. In what position should be MACD indi, to confirm trading conditions for that trade? Can you explain me this new setting condition (MACD crossing), please?

Thank you.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: version11.jpg Size: 190 KB
Click to Enlarge Name: version14.1_M1.jpg Size: 157 KB
  • #730
  • Nov 2, 2010 3:10pm Nov 2, 2010 3:10pm
Quoting kapybara
Hi Jiva,

today early morning one sell trade occured, which was taken by version11 (version11.jpg) -> both positions on EURUSD hit SLs (2x110 pips). Here is screen shot from version14.1 chart (currently updated) -> I browsed for the critic time (version14.1_M1.jpg). Now I'm wondering, if it (v14.1) would take that trade or not. In what position should be MACD indi, to confirm trading conditions for that trade? Can you explain me this new setting condition (MACD crossing), please?

Thank you.
Hi, Yes this is exactly why v14.1 has been created.
to be able to use as small as possible stop losses.......
No v14.1 would not have taken the trade. I had all three EURUSD settings up. v11 1 hr, v11 15 min and v14.1
see this post for what happened.
http://www.forexfactory.com/showpost...&postcount=715
In v14.1 macd is set for a confirmation of a crossing on the 1min. It seems to be very effective for preventing bad entries.

We have been a little preoccupied experimenting with some of the ideas jm2110 has been presenting. Seeing how to maximize the optimizing:nerd:
  • #731
  • Nov 2, 2010 3:14pm Nov 2, 2010 3:14pm
The chart shows what the trades from v14.1 look like. v14.1 is very good catching the middle of a trend.
there were 2 losses (x 2 positions) on the chart, stopped out for 20 pips per position.
EURUSD backtest 12 months FXDD 1h..
Obviously we do not know what the future will be and fitting settings to a back test is not going to tell us what the future will be. My thinking is to do regularly 50/50 tests, and optimizing to keep up with market changes.
There is argument against this, that it will not be effective. Let us test going forward a bit and see. We want to get this right before committing any hard earned money...
I think the ea is sound enough to be able to avoid consolidations and able to capture middle of the trend trades as long as we keep up with the regular optimization of the settings.
So let us make an experiment and see. v14.1 gives us what we need: very small stop losses and laser focused entries with macd on the 1min.
I will be putting more time into getting the settings for the other pairs ready for testing on v14.1
Attached Image (click to enlarge)
Click to Enlarge Name: eurusd4.gif Size: 48 KB
  • #732
  • Nov 2, 2010 3:57pm Nov 2, 2010 3:57pm
We will need to get used to getting some stopped out trades with v14.1. But the stops will be small and the overall strategy is a lot safer than v11.
It is unrealistic to think there will not be bad trades and losses with any strategy or system. The idea is to keep the losses small and keep the gains big.

We will be working on getting .set files for more pairs fro v14.1 over the next few days. Traders who would like to do some of the set files are welcome. Rough guidelines are posted here:
http://www.forexfactory.com/showpost...&postcount=725

Questions, comments, discussion always welcome.....
  • #733
  • Edited 3:14am Nov 3, 2010 12:09am | Edited 3:14am
http://t0.gstatic.com/images?q=tbn:A...c29aJ3F6NuCsg=
We might be having some technical difficulties.
The macd setting doesn't seem to be working properly.
macd is set to check for a cross on the 1m
but comments display macd cross as up on 1m
but we can see from the indicator window macd cross is clearly down?

Lots to learn trying to get this right.
v14.1 may have been released too early. sorry for the inconvenience they may have caused traders. We wanted to be set up with v14.1 for the action after Fed Weds. but it looks like we will have to make do with v11 and the big stop losses until we can be sure macd is working properly in v14.1.That may be part of the reason why it has not taken any trades at all since it's release.
Ah well, at least we have v11 for Weds. Big stop losses and all...

We will have to see what Taiyakixz can do to make sure things are really sound and fully technically up and running before we try to release it again.
This happens a lot. Little bugs here and there that need to be ironed out.
v11 has all the bugs taken care of but it took some doing to get there. So no surprise if there a few little buggy critters lurking in v14.1.
Again sorry for the inconvenience.
A lesson we have learned. To be very,very certain the the technical aspect is sound before trying to make the upgrade available.
Attached Image (click to enlarge)
Click to Enlarge Name: eurcad.gif Size: 63 KB
  • #734
  • Nov 3, 2010 1:15am Nov 3, 2010 1:15am
Quoting jiva34
http://t0.gstatic.com/images?q=tbn:A...c29aJ3F6NuCsg=
We might be having some technical difficulties.
The macd setting doesn't seem to be working properly.
macd is set to check for a cross on the 1m
but comments display macd cross as up on 1m
but we can see from the indicator window macd cross is clearly down?

Lots to learn trying to get this right.
v14.1 mat have been released too early. sorry for the inconvenience they may have caused traders. We wanted...
Thanks for your hard work Jiva, Luckily it seems like a minor bug, let's hope can be easily solved, I will close temporally my V14 charts, and make some tests in V11 , a good prevention to avoid using the system live yet , but I am sure in future this will ROCK very hard.


__________-
Sequoia
  • #735
  • Edited Nov 4, 2010 12:07am Nov 3, 2010 1:40am | Edited Nov 4, 2010 12:07am
v11
Always check the first page for the most recent .set files
updated optimized EURUSD FXDD 1hr 10/27/10
Attached Thumbnails http://cdn.forexfactory.com/attachme...1&d=1288154675
Attached Files
Attached File(s)
File Type: zip eumtf1hopte.zip   3 KB | 304 downloads
  • #736
  • Edited Nov 4, 2010 12:08am Nov 3, 2010 1:44am | Edited Nov 4, 2010 12:08am
v11
Always check the first page for the most recent .set files
Updated optimized .set file GBPUSD FXDD 10/27/10

Attached Thumbnails
Attached File(s)
File Type: zip eugbmtfoptb.zip   3 KB | 200 downloads
  • #737
  • Edited Nov 4, 2010 12:08am Nov 3, 2010 1:46am | Edited Nov 4, 2010 12:08am
v11
Please always check the first page for most recent downloads

Updated optimizations for GBPUSD 10/28/10
Attached Thumbnails http://cdn.forexfactory.com/attachme...1&d=1288242826
Attached File(s)
File Type: zip gumtfopt10-28.zip   3 KB | 203 downloads
  • #738
  • Edited Nov 4, 2010 12:09am Nov 3, 2010 1:48am | Edited Nov 4, 2010 12:09am
v11
Please always check the first page for latest downloads.

Updated optimizations for EURCHF FXDD.
Attached Thumbnails http://cdn.forexfactory.com/attachme...1&d=1288245949
Attached File(s)
File Type: zip euchmtfopt10-28.zip   3 KB | 258 downloads
  • #739
  • Edited Nov 4, 2010 12:09am Nov 3, 2010 1:54am | Edited Nov 4, 2010 12:09am
v11
Please always check the first page for the most recent updated set files.

Updated AUDUSD 1h optimized FXDD 10/28/10
Attached Thumbnails http://cdn.forexfactory.com/attachme...1&d=1288316152
Attached File(s)
File Type: zip audusdnoloss10-28.zip   3 KB | 239 downloads
  • #740
  • Edited Nov 4, 2010 12:09am Nov 3, 2010 1:57am | Edited Nov 4, 2010 12:09am
v11
Please always refer to the first page for the most recent updated .set files.

EURAUD 1h optimized FXDD 10/30/10

Thanks to jm2110, and his in depth critiques, we are now including a "50/50 statistical edge" test for every new update.
This is a test where all the settings for channel, moving averages, dss, etc remain the same. The settings for stop loss and take profit will be set equal to each other. If there is a statistical edge to the strategy it will show wins more than 50% of the time, preferably at least 65%. So that adjusting with stops and trailing stops is truly optimizing and not merely curve fitting.
The 2nd graph is the 50-50. It tested at 57.14% win rate.

I am seeing how to use this very valuable tool. It may be that the 50-50 is the first optimization to run along with the channel. The Dragon's bare bones. Then we dress him up later... http://cdn.forexfactory.com/images/s...m/big_grin.gif

Traders, please keep your insights, comments, and ideas coming. Thank you, jm. Like I said, you always learn the most from your toughest customers....http://cdn.forexfactory.com/images/s...im/big_hug.gif
Attached Thumbnails http://cdn.forexfactory.com/attachme...1&d=1288476231 http://cdn.forexfactory.com/attachme...1&d=1288476721
Attached File(s)
File Type: zip euraud10-30.zip   3 KB | 219 downloads
2 traders viewing now