Menu

Options

Bookmark Thread

First Page First Unread Last Page Last Post

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

Quoting charvo
what we don't have, is a robust tester that can test many imaginations. for example, multi-pairs, basket trading, and so on. now only a few pro software/API can do it. but if you develop this, this will be very costly and ultimately cannot be free.
What do you mean by "cannot be free" and "is very costly"? Yes, my tester works with portfolio of strategies and equities so it's probably "pro" software if you want to define it that way (see the past thread).

Anyway I have decided to go on with my plan, the more I think of it, the better the idea feels. I'll post the link here as soon as I have something ready. The system will basically be a web interface to a simple version of my rule based tester.
those kind of 'pro' tester are very difficult to develop. the development has to be professional and teamwork (usually). therefore, it is supposed to be a commercial thing, by a corporation or a company. i never saw one 'pro' tester developed by one person (privately). then as a result, these kind of commercial tester generally cannot be free because the developers invest too much into the development

that's my limited understanding, of coz. i believe someone (including you) can develop a robust tester on your own, but the journey will be difficult. good luck.

i have indeed seen many 'free' testers based on MT4 or other platforms. but they are limited both in functionality as well as multi-pair/market ability.

Quoting mikkom
What do you mean by "cannot be free" and "is very costly"? Yes, my tester works with portfolio of strategies and equities so it's probably "pro" software if you want to define it that way (see the past thread).
Quoting charvo
those kind of 'pro' tester are very difficult to develop. the development has to be professional and teamwork (usually). therefore, it is supposed to be a commercial thing, by a corporation or a company. i never saw one 'pro' tester developed by one person (privately). then as a result, these kind of commercial tester generally cannot be free because the developers invest too much into the development

that's my limited understanding, of coz. i believe someone (including you) can develop a robust tester on your own, but the journey will be...
Have you read this thread at all?

TLDR The tester already exists.
Quoting mikkom
Comments? Is this silly idea?
Hi Mikkom. You can definitely count on my involvement. Do you need betatesters at this stage?

Quoting charvo
i have indeed seen many 'free' testers based on MT4 or other platforms.
Hi Charvo. Would you be able to post some examples?

Cheers

m
  •  rayzeel
  • | Joined Nov 2008 | Status: Automated/DiscretionaryTrader/Coder | 37 Posts
Hi guys,
mikkom... your tester would be a good solution for people not capable to develope their own tool... I personally got a break through in this subject (took me some time to learn "how to").

I have some question if you don't mind... You said that you moved from currency market to stock (if I understand correctly), and my question is whether it was good decision and if it's not a problem please point main adventages and disadventages of that two markets.

I'm curious because currently Im working on currencies and it seems that I cant move any forward for better performance across different crosses. Over - diversification is not a solution I suppose...

Could you provide some advice as You are large amount of steps ahead from me

Question is for all of You guys if you want to share yours experience. I will be greatful for each answer.

Thanks.
Quoting rayzeel
I'm curious because currently Im working on currencies and it seems that I cant move any forward for better performance across different crosses. Over - diversification is not a solution I suppose...
Diversification only works if you have a strategy that works on some principles that are common on multiple markets. If you can find a strategy that works equally well on multiple markets that is an indication that the strategy actually works (as there is more data).
  •  rayzeel
  • | Joined Nov 2008 | Status: Automated/DiscretionaryTrader/Coder | 37 Posts
I do have such strategies. They working on various currency crosses with good results but all crosses seems behave similar and I just wondering if I could find better diversification on equities or other market.

And another question... What do you think about changing parameters on fly ? I mean any kind of walk forward approach. I'm about to implement walk forward testing into my app and maybe you have some experience on this ground ? Currently Im testing about 8 years of data and verifying results on few years out of sample data. It looks good but I dont have a clue when should I change parameters. How long should I keep portfolio running before making new test and preparing new portfolio. Walk forward seems to be an answer but I'm not sure... I think I have to build and check what it will show anyway to get an answers...

If I remamber Craig doing WF tests.

Instead of doing periodicals test Im planning to make optimization after each new trade is over. The fitness function could be an example 5 losses in row. After optimization I then change parameter set for next trade. Is someone tested something like this ?

Im working for proprietary automated trading company and we getting some results doing long term tests but we started to think about improving performance by WF as we think that today market shape is different than few years ago and maybe it will be better to fit our strategies only for current market and not care what was in 2008 an example...
hi guys

I have reqad most of the thread
great work.
I am a coder. & coded thousands of different algos.
not as complicated as yours but fair enough.

one thing I want to share with you.
when I have an algo in my mind after coded & tested if it does not match my criteria I ignore it.
I do NO optimization.

I only use neutral winners.

any optimizition is curve fitting & fails on live.
I call optimization as a "strecth wear on elephent" any move of elephent wipes it out

good luck.
  •  rayzeel
  • | Joined Nov 2008 | Status: Automated/DiscretionaryTrader/Coder | 37 Posts
Yes, you are right Stevewalker. Too fit always causing fail, but where is the border ? 2 or 3 optimized parameters maximum ? Should strategy work over all available data or just on most actual one ? Im about to assume that strategy should be prepared to work on current market shape and be reevaluated after let say ... one month...

For my strategies I optimizing from 2 to 4 parameters and strategies working over various of currency pairs.
Quoting rayzeel
Yes, you are right Stevewalker. Too fit always causing fail, but where is the border ? 2 or 3 optimized parameters maximum ? Should strategy work over all available data or just on most actual one ? Im about to assume that strategy should be prepared to work on current market shape and be reevaluated after let say ... one month...

For my strategies I optimizing from 2 to 4 parameters and strategies working over various of currency pairs.
IMHO the optimization only works if you can compare the results correctly. If you have a very large amount of non-correlated data and you can find a fitness method that works well (without aplifying noise/curve fitting) then optimization might work.

Basically it alll depends on your algo. One example woudl be neural networks [or similar], you can't fine-tune by hand at all so you have to optimize by machine.
Quoting rayzeel
Yes, you are right Stevewalker. Too fit always causing fail, but where is the border ? 2 or 3 optimized parameters maximum ? Should strategy work over all available data or just on most actual one ? Im about to assume that strategy should be prepared to work on current market shape and be reevaluated after let say ... one month...

For my strategies I optimizing from 2 to 4 parameters and strategies working over various of currency pairs.
It is possible to both under and over fit the data. If no optimization is performed then bias may be less, but variance of returns out of sample is still a major concern. If optimization is used on a small amount of data it is very easy to introduce bias to the results. The defense against this is increasing the sample size.

I won't say that optimizing on a small amount of data cannot work in some limited cases. I am 14 weeks into a forward test on a system that was designed using 4 weeks of data with 10 additional weeks of out of sample validation. If you find a period that generalizes well and if your strategy is based on principles that generalize well then it can be done using limited data. It is important to decide whether the objective is to maximize past returns or future returns and to set up the tests appropriately. But this system is probably the exception rather than the rule and may be a lucky survivor. My expectation is that over time the results will degrade significantly, requiring additional optimization, though it continues to surprise me.
  •  rayzeel
  • | Joined Nov 2008 | Status: Automated/DiscretionaryTrader/Coder | 37 Posts
Quote
My expectation is that over time the results will degrade significantly, requiring additional optimization, though it continues to surprise me.

That's sounds familiar to me. I've also looking only for most simple strategies and they are working also in OOS data. But I think that another optimization should be made after some time on market. I'm just gathering information before I do some steps to any WF tests. I see that I have to find out what will works best just doing tests 'coz there is no clear answers how to do that
Long time without posting here, I have been developing a new set of strategies on totally different market (and trading them too, don't ask about the market) mostly using very large datasets and Weka. If you haven't checked out Weka and are interested in data mining, please do. It's quite interesting that most papers that I have read recently use weka as a platform for their learning algorithms so there is a lot of cutting edge stuff available quite easily.

http://www.cs.waikato.ac.nz/ml/weka/
This is worth reading.. Any comments? (BlackMage?)

http://www.businessweek.com/articles...-rise-and-fall
Quoting mikkom
This is worth reading.. Any comments? (BlackMage?) http://www.businessweek.com/articles...-rise-and-fall

finally....cant wait to see how 2013 results will be.
wo-yoy! wo-yoy! wo-yoy! wo-yoi! wo-yoy-yoy-yoy!
The HFT guys were the architects of their own demise, they basically cannibalized each other.
But then lots of firms are doing it tough these days, no volume, no volatility and lots of staff.
However, things seem to be picking up somewhat.
The breaking of a wave cannot explain the whole sea.
One link more

http://www.bbc.co.uk/news/business-23095938
Just found out about this nicely organized algo trading resources wiki page (seems to be rip from another already-took-down page). Not sure if its up-to-date though.
http://gutone.com/mediawiki/index.php?title=Automated_Trading_Resources
This might be interesting to some of you..

http://queue.acm.org/detail.cfm?id=2534976
Quoting mikkom
This might be interesting to some of you.. http://queue.acm.org/detail.cfm?id=2534976
Thanks, mikkom you orient you in HFT ?
0 traders viewing now