GSB Forums - Powered by XMB 1.9.11

Setting for EURGBP

rws - 16-6-2017 at 02:08 PM

I find there is few trading activity and few interesting systems when trying EURGBP on 10 min. Find a picture of a typical performance.

Most systems only have 25% activity in the training period.
Especially when compared to ES there are few high quality systems found.

What settings could I use improve this?

I am very impressed with the ES system and how good it performes OOS.

Would a metric net profit/buy-and-hold ratio make sense like lately added to Adaptrade?

Thanks

admin - 16-6-2017 at 04:04 PM

Hi RWS,
Forex is an area of weakness for me personally. I spoke to someone who ran coarse on SQ and Adaptrade, and he said Adaptrade works, but doesnt work on forex. (Sq I think did, but I've never managed to get a system out of SQ that works - though Im sure it can be done.) I suspect GSB in the current form wont work on fx either. The good news is in the GSB community, there seems to be the expertise to make forex systems. This is going to mean we need the correct secondary filters/ truisms and possibly architecture modifications. My guess is this is a month away.
You could try commission on fitness of 0, and fitness netprofit only. Both of these will increase the number of trades. I doubt it will fix the problem.

moveo - 17-6-2017 at 06:08 AM

well, much depends on settings

rws - 17-6-2017 at 09:52 AM

Oke thanks these are very good values.

I did already test with market on day "false" but that
did not make a difference.
It seems you have another optimizer active which I don't have.

At the left side at Optimizition I can only set training, test and validation data
but you have 3 more options.
You have a setting genetic algorithm which I don't have.
Do I miss something?
Thanks

rws - 17-6-2017 at 10:25 AM

I already found out GUI->advanced->true
Thanks

rws - 17-6-2017 at 10:58 AM

You have set X ax to auto.
If I set the X ax to auto I also have a normal looking graph

If you set the X ax to date then the picture could be much different having long inactivity periods.

I tried to run with the supplied EURGPB.120 but it did not find any trades at all after 30 minutes with exactly the same as your settings.

Could I please ask you, how much history do you use and what is the difference between the two EURGPB.130 files? Did you bootstrap the data in one of them?

I see GSB finds much better with 2 similar but different data files.

Quote: Originally posted by moveo

well, much depends on settings

moveo - 17-6-2017 at 02:51 PM

I use data from now on 6 years back.
Just for fun, gave it a second run, looks also good

rws - 17-6-2017 at 03:01 PM

Yes indeed it looks very good.

What session do you use? 24 hours?
I am trying to reproduce

Could I please ask you what the difference is between your 2 input data of EURGBP130? Or are they the same?

Thanks

Quote: Originally posted by moveo

I use data from now on 6 years back.
Just for fun, gave it a second run, looks also good

admin - 18-6-2017 at 05:45 PM

I find this thread interesting, and Im wondering how good the systems will prove to be.
No one has yet done any WF on the systems. It would be good to publish this. Note 100 population and 100 generations is a bit low for WF, so I would try 200 x 200 or more.
Note also if you use secondary filter of closeMinusCloseD or genetic could be very significant. For ES its much better to use CLoseMinusCloseD on secondary filters,
but forex I suspect not. As GSB gets more advanced on Secondary filters, this will be less of an issue. So please experiment with the secondary filter to see whats best.
Using secondary data with bigger time frames on the same symbol would be worth trading as well.

Secondary Forex Data...

andrew.gibbs777 - 18-6-2017 at 05:56 PM

Suggest Forex Futures data may act as a reasonable secondary filter... (Just an idea).

admin - 19-6-2017 at 12:52 AM

Hi Andrew, I think you mean futures data as a secondary data stream? I had some success with this many years ago.

rws - 19-6-2017 at 12:02 PM

Hi admin,

Will there be something like bootstrapped data?
What is your idea of using this as a secondary data stream?

Thanks

admin - 19-6-2017 at 03:26 PM

Hi RWS. I don't know alot about bootstrapped data. Can you explain how it might be useful in GSB?
For secondary data using bigger time frames is a good idea, ie 60 min, 240, daily and using the same time frame as data1 fx, but the futures of it. Maybe even bigger time frames on the futures data too.

rws - 20-6-2017 at 08:49 AM

I understand from Adaptrade that changing the values of the OHLC data a bit could lead to more robust systems.

admin - 20-6-2017 at 03:52 PM

I think what you mean is add noise to test robustness of the system. Is that correct? Adaptraded needed this as there is no WF.
If this is what you mean, I may add this at some stage, but I have another concept to do a similar thing. There is not much need for this sort of stress testing as we have WF and a better architecture. Still im open to added it, but not for some time.

rws - 20-6-2017 at 04:32 PM

EURGBP data is incorrect in GSB. It has big holes in all timerframes.
That was the reason of the inactivity periods I saw.
I understand Peter will update the data and I hope the new testdata will have more currencies and more history.

Quote: Originally posted by rws

I find there is few trading activity and few interesting systems when trying EURGBP on 10 min. Find a picture of a typical performance.

Most systems only have 25% activity in the training period.
Especially when compared to ES there are few high quality systems found.

What settings could I use improve this?

I am very impressed with the ES system and how good it performes OOS.

Would a metric net profit/buy-and-hold ratio make sense like lately added to Adaptrade?

Thanks

rws - 21-6-2017 at 04:44 AM

Oke thanks.

It is surprising how much better OOS performance is when adding 2 similar but not the same tickers. I also tried it on DAX only vs DAX + a derived one and it was something more than 2 times better OOS on the average system OOS.

Sofar I have seen this is one of the big things in GSB that helps and I guess it had something to do with avoiding overfitting?? Could that be true?

So I thought about random variation in OHLC data to achieve the same.

I wonder if having both WF and randomize would be even better or not
I could try to corrupt OHLC data myself and test.

Quote: Originally posted by admin

I think what you mean is add noise to test robustness of the system. Is that correct? Adaptraded needed this as there is no WF.
If this is what you mean, I may add this at some stage, but I have another concept to do a similar thing. There is not much need for this sort of stress testing as we have WF and a better architecture. Still im open to added it, but not for some time.

admin - 21-6-2017 at 04:53 AM

What do you mean by "vs DAX + a derived one". Can you give more detail on how you used this concept? It sounds worth exploring if its as good as you say.
Related is that the cash trends much better than the futures, so vip to have the cash indices. However there is more to the merits of data2+ than this.
I researched this a lot in the previous generation of GSB, and know alt data streams are vip.

rws - 21-6-2017 at 03:48 PM

with "derived one" I mean nothing special, just following your advice.

like having the dax and daxk as second data.

https://www.investing.com/analysis/the-dax-versus-the-daxk-1...

I was just wondering if it was overfitting when just using 1 data because it finds better systems with 2 data.

Quote: Originally posted by admin

What do you mean by "vs DAX + a derived one". Can you give more detail on how you used this concept? It sounds worth exploring if its as good as you say.
Related is that the cash trends much better than the futures, so vip to have the cash indices. However there is more to the merits of data2+ than this.
I researched this a lot in the previous generation of GSB, and know alt data streams are vip.

admin - 21-6-2017 at 04:51 PM

Its not due to over fitting, its due to the cash trends much better than the futures.