Systematic trading | Page 22 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Trade another day through Systematic Trading 203 replies

Nonparametric Association Measures in Systematic Trading 57 replies

Anyone use Systematic from Alpari? 1 reply

A Systematic Approach to Markets 68 replies

Systematic Trading vs. Discretionary Trading 14 replies

Attachments: Systematic trading
Exit Attachments

Systematic trading

  • #421
  • Oct 14, 2009 5:36pm Oct 14, 2009 5:36pm
Can you elaborate on what you mean by 'trading correlations'?
The breaking of a wave cannot explain the whole sea.
  • #422
  • Oct 14, 2009 5:41pm Oct 14, 2009 5:41pm
Quoting Craig
Can you elaborate on what you mean by 'trading correlations'?
calculating coefficiencies [and other "correlations"] and trading based on them.. something I have been pondering to do for a long time.

Not just pairs/equities/futures correlations but interest rates, unemployment numbers (and some more advanced stuff like text mining - it should be quite possible, I could go to this if someone is interested, in basically the same way as spam filters work) and so on.
  • #423
  • Oct 14, 2009 5:55pm Oct 14, 2009 5:55pm
Quoting mikkom
calculating coefficiencies [and other "correlations"] and trading based on them.. something I have been pondering to do for a long time.
...so like pairs trading? e.g. Looking for correlations to revert?
The breaking of a wave cannot explain the whole sea.
  • #424
  • Oct 14, 2009 10:33pm Oct 14, 2009 10:33pm
Hi mikkom

are you now working on 4 Major pairs and CAD AUD?

Do you have any comment on adding CAD and AUD.




Quoting mikkom
If anyone is interested, here is some eyecandy (1% risk per trade). It's quite interesting to see how it looks for all the 6 pairs combined - this is the first time I see this so this is quite interesting for me as well
  • #425
  • Oct 15, 2009 5:21am Oct 15, 2009 5:21am
Quoting mikkom
calculating coefficiencies [and other "correlations"] and trading based on them.. something I have been pondering to do for a long time.

Not just pairs/equities/futures correlations but interest rates, unemployment numbers (and some more advanced stuff like text mining - it should be quite possible, I could go to this if someone is interested, in basically the same way as spam filters work) and so on.

I worked for quite a while on a system that traded "lead/lag" relationships/correlations between 2 assets in high frequency. Unfortunately it turned out to be non-tradeable in a realistic setup and generally it was a lot more difficult than I initially thought.

I might have a look into it again because there is definitely "something to it" and it is probably also worthwhile to look at longer term relationships than the ultra-short term I initially looked at.

You are welcome to comment further on your ideas.
  • #426
  • Edited 5:39am Oct 15, 2009 5:27am | Edited 5:39am
Quoting BlackMage
You are welcome to comment further on your ideas.
This is not exactly about correlations but here is something

One of the thoughts I have had for a long time is to use bayesian filtering to predict movements based on past textual and economic number data. The real problem with this is that I want to backtest it for a long time and I just can't find a reliable source for long term number data (finding articles is easier because of google).

Bayesian filtering (and derivatives of it) don't know what it's looking at so it can be used both for numbers and events like "unemployment went up".

I'm quite sure this would work (and it's quite backtestable with a good dataset) but it's not exactly the simplest thing to implement (not hardest either but timeconsuming) because of the amount of data that needs to be processed.

ps. if someone implements this, please contact me - I really want to know if it works. I think few hedge funds use some systems like this based on some interviews I have read.
  • #427
  • Oct 15, 2009 5:58am Oct 15, 2009 5:58am
I read a couple of years ago a very relevant research in Uni of Surrey, UK. I think at that time the papers were public, though at this moment some of them seem to be in password protected area, however I believe you can get a hold of them if you email the authors cause this kind of research is usually public.

http://www.computing.surrey.ac.uk/ai/gida/papers.html
  • #428
  • Oct 18, 2009 4:17am Oct 18, 2009 4:17am
Quoting challenger78
I read a couple of years ago a very relevant research in Uni of Surrey, UK. I think at that time the papers were public, though at this moment some of them seem to be in password protected area, however I believe you can get a hold of them if you email the authors cause this kind of research is usually public.

http://www.computing.surrey.ac.uk/ai/gida/papers.html
Interesting stuff, I have to read this when I have time.

Now as we have a few methods, all we need is a big database of economic data and news to be able to test them - if anyone knows any good sources in machine readable format please put a link here (I don't mind if they cost unless they are hugely expensive)
  • #429
  • Edited 4:53am Oct 18, 2009 4:37am | Edited 4:53am
Does anyone have a good method for calculating past "congestion zones"? I'm thinking about checking momentum for certain periods and calculating deviation or high/low of period, any comments?

I would like to test a few methods that I have tested manually and that seem to work but to be honest, I don't trust manual testing at all. Too much cognitive biases.
  • #430
  • Oct 18, 2009 6:02am Oct 18, 2009 6:02am
Quoting mikkom
Now as we have a few methods, all we need is a big database of economic data and news to be able to test them - if anyone knows any good sources in machine readable format please put a link here (I don't mind if they cost unless they are hugely expensive)

I know Reuters is into machine readable news so they might be worth contacting (but they might also be "hugely expensive").

I have an SQL Server database of US reports going approx. from 1998 to beginning of 2008. (I will check up on the exact time period once I get into office). It contains actual numbers, forecasts and revisions. We made it by writing a crawler for a news site (which name I can't remember exactly either - way to much Vodka last night...) so its public info.

I consider it incomplete myself because it only covers US reports and its not up to date, but it's probably OK to test out some ideas quickly. Let me know if you want it.
  • #431
  • Oct 18, 2009 7:52am Oct 18, 2009 7:52am
Mikkom,

What is the progress of your your ATS on real account.

If anything like your alpha factory chart , it would be fair to say you have the "holy grail"

And if I was a big bank I 'd be giving you a blank check
  • #432
  • Oct 18, 2009 7:55am Oct 18, 2009 7:55am
Quoting BlackMage
I know Reuters is into machine readable news so they might be worth contacting (but they might also be "hugely expensive").

I have an SQL Server database of US reports going approx. from 1998 to beginning of 2008. (I will check up on the exact time period once I get into office). It contains actual numbers, forecasts and revisions. We made it by writing a crawler for a news site (which name I can't remember exactly either - way to much Vodka last night...) so its public info.

I consider it incomplete myself because it only covers US reports...
That's what I have been thinking to do as well, I have written a couple of news aggregators in the past so this is nothing new to me.. Reuters is probably too expensive to use for testing. It's usually a bad sign that a company doesn't publish the prices for services they offer..

When you have sobered up, would it be possible to post the news site? I might write a small crawler myself too.
  • #433
  • Edited 9:25am Oct 18, 2009 8:00am | Edited 9:25am
Quoting spekkiefx
Mikkom,

What is the progress of your your ATS on real account.

If anything like your alpha factory chart , it would be fair to say you have the "holy grail"

And if I was a big bank I 'd be giving you a blank check
It's quite impressive but not the same as the chart I posted because it has some correlation checks (so it doesn't use all strategies) so my drawdowns have been bigger. I have only been trading the full set for about 3 months (subset for 6 months), some of the strategies I have been trading longer but not automatically.

I have made some really bad "discretionary" trades on the same account too (I always know where the market goes and when I do, I tend to play too big - too bad the market doesn't agree) so it's not up as much as it would be without them but let's say the profit is quite well anyway

I posted some of the initial curves of the systems back when I started this set here but these are not from the full strategy set I'm trading now. Just a small note, I'm trading this account with a bit too high risk until the size is decent (I use linear risk lowering when the account size grows) my current percentage DDs are higher than they would be on fixed risk account (and on actual euros they will be much lower in the next 10 years because of this).

ps. yes, I consider this set to be a kind of holy grail for me but it's slow (I trade dailies so only once a day) and there are and will be DD's so my holy grail might not be someone elses holy grail.
  • #434
  • Oct 18, 2009 9:25am Oct 18, 2009 9:25am
It was Breifing.com:

http://www.breifing.com/Investor/Pub...icCalendar.htm
  • #435
  • Oct 18, 2009 2:54pm Oct 18, 2009 2:54pm
Mikkom,

Is it right you are using mt4 as your execution framework. You obviously have the java skills so why not build your own? Surely this would be more flexible/stable?
  • #436
  • Edited 4:53pm Oct 18, 2009 4:30pm | Edited 4:53pm
Quoting spekkiefx
Mikkom,

Is it right you are using mt4 as your execution framework. You obviously have the java skills so why not build your own? Surely this would be more flexible/stable?
Well it's more of a vice versa, I was using mt4 before I built my own framework and believe or not but almost all the traiding apis there are on brokers are quite horrible and far from easy to use. With mt4 I'm at least able to switch brokers when I want to and it works.

My framework could work as a execution framework too but at this point I'm using mt4 until I have time to implement few trading implementations - much harder than you would think because of the awful apis, even FIX is far from what you should call standardized..
  • #437
  • Oct 18, 2009 5:29pm Oct 18, 2009 5:29pm
Quoting mikkom
Well it's more of a vice versa, I was using mt4 before I built my own framework and believe or not but almost all the traiding apis there are on brokers are quite horrible and far from easy to use. With mt4 I'm at least able to switch brokers when I want to and it works.

My framework could work as a execution framework too but at this point I'm using mt4 until I have time to implement few trading implementations - much harder than you would think because of the awful apis, even FIX is far from what you should call standardized..

Change of trend here!

I believe a lot of (ECN)brokers are now starting to offer currenex(or hotspotfxi). They have a java api!! If you have the volume you could negotiate a good rate per million ($20/million and lower should be possible). You could build a solid framework going all the way and go on holiday once completed , keep urself informed through messages. I am rather surprised you are not more focussed on this.
  • #438
  • Oct 18, 2009 5:32pm Oct 18, 2009 5:32pm
Quoting spekkiefx
Change of trend here!

I believe a lot of (ECN)brokers are now starting to offer currenex(or hotspotfxi). They have a java api!! If you have the volume you could negotiate a good rate per million ($20/million and lower should be possible). You could build a solid framework going all the way and go on holiday once completed , keep urself informed through messages. I am rather surprised you are not more focussed on this.
Well I already have a working trading implementation ("EA"). Yes I probably pay a bit more on commission at the moment but at this point, I'm happy with what I have and it works. In the future I probably code another trading implementation but I'm in no hurry.
  • #439
  • Oct 18, 2009 5:50pm Oct 18, 2009 5:50pm
Quoting mikkom
Well I already have a working trading implementation ("EA"). Yes I probably pay a bit more on commission at the moment but at this point, I'm happy with what I have and it works. In the future I probably code another trading implementation but I'm in no hurry.
Don't get me wrong no attack here as I have no "real" working strategies on the moment. But just wandering as you created your testing java framework in such a short time span, why not create an execution one?
  • #440
  • Oct 19, 2009 7:36am Oct 19, 2009 7:36am
Quoting spekkiefx
Don't get me wrong no attack here as I have no "real" working strategies on the moment. But just wandering as you created your testing java framework in such a short time span, why not create an execution one?
Because testing is what sucks in metatrader, charts, trade statistics and such are quite okay.
2 traders viewing now