Hi Michal,
I now understand. Actually after posting my reply and reading about the various systems I now realize there is a substantial difference. Lot's of studying to do, but very interesting none-the-less.
Thanks.
important links
Moderator: moderators
Please add www.kreslik.com to your ad blocker white list.
Thank you for your support.
Thank you for your support.
- michal.kreslik
- rank: 1000+ posts
- Posts: 1048
- Joined: Sat May 13, 2006 2:40 am
- Reputation: 36
- Location: Monte Carlo, Monaco
- Real name: Michal Kreslik
- Gender:
eudamonia wrote:Michal,
I hope this doesn't drive things too off topic, but do you have any links for resources regarding Simulated Annealing? The ones at SmartQuant are offline.
Edward
Ed,
unfortunately, I don't have any cool links regarding simulated annealing method. This technique is on my "to do list" queue for some time already though
I also remember Jeffrey Katz writes in his book The encyclopedia of trading strategies about simulated annealing. This method has been able to solve significant combinatorial problems, including the notorious "traveling salesman problem".
Regarding the links, what about trying this:
[google]simulated annealing[/google]
- eudamonia
- rank: 500+ posts
- Posts: 536
- Joined: Thu Jun 15, 2006 9:50 pm
- Reputation: 0
- Location: Rocklin, CA
- Real name: Edward Heming
- Gender:
- Contact:
Michal,
Thanks. I've already tried Google. I just thought I'd ask in case you had some harder to find links (like the ones for GA you have above) or references.
Edward
Thanks. I've already tried Google. I just thought I'd ask in case you had some harder to find links (like the ones for GA you have above) or references.
Edward
Eudaimonia (pron.: you-die-moan-e-a) (Greek: εὐδαιμονία) is a classical Greek word commonly translated as 'happiness'. The less subjective "human flourishing" is often preferred as a translation.
michal.kreslik wrote:Forest,
you are talking about the neural networks solutions here.
Michal
Michal,
Actually, I think the Neuroshell Trader Professional and Neuroshell Daytrader Professional are genetically and walk-forward capable. I am seriously considering purchasing a 30 day trial to try them out since it doesn't appear to require much programming to insert parameters. On top of the fact it now autotrades with IB, it also interfaces with www.prophet.net for an inexpensive historic data feed.
Anyway, the dialog here in these forums helps to make a decision at which point to jump in.
Forest
Please add www.kreslik.com to your ad blocker white list.
Thank you for your support.
Thank you for your support.
If I do take the plunge, I will indeed let you know, Edward.
If I can insert the trading model I have been working on and optimize it with a reasonably straight equity curve, I am sure you and the others on this forum would find it child's play.
The price difference between Neuroshell Daytrader and a Neoticker solution is indeed significant, but all other things being equal and considering how long it would likely take me to learn to program in Neoticker and apply it's Grid Optimizer, it is really not a comparison. Besides, if something I do actually works well, I am sure somebody here might be tempted to convert it.
I have been playing with a DeMarker indicator that evolved out of the tunnel method and a few other trade management solutions on the 1 hour FOREX and it seems to be very reliable when traded manually. Even though Neuroshell didn't have this particular indicator out of the 800+ that comes with the software, they said they would make it for me which was quite impressive. I will hopefully find out whether a mechanical test agrees enough for me to feel comfortable sharing it with you all.
If I can insert the trading model I have been working on and optimize it with a reasonably straight equity curve, I am sure you and the others on this forum would find it child's play.
The price difference between Neuroshell Daytrader and a Neoticker solution is indeed significant, but all other things being equal and considering how long it would likely take me to learn to program in Neoticker and apply it's Grid Optimizer, it is really not a comparison. Besides, if something I do actually works well, I am sure somebody here might be tempted to convert it.
I have been playing with a DeMarker indicator that evolved out of the tunnel method and a few other trade management solutions on the 1 hour FOREX and it seems to be very reliable when traded manually. Even though Neuroshell didn't have this particular indicator out of the 800+ that comes with the software, they said they would make it for me which was quite impressive. I will hopefully find out whether a mechanical test agrees enough for me to feel comfortable sharing it with you all.
eudamonia wrote:Forest,
Your right I hadn't noticed that Neuroshell Trader Pro had GA capabilities. Please keep us posted on how you like Neuroshell, since I've considered looking into them.
Edward
Okay Edward, I went ahead and purchased Neuroshell Daytrader. I really had reservations since it does not give you too much information (not much graphics) on the site as to it's capabilities and it isn't exactly cheap. However something about the presentation gave me a good feeling.
I received the CD yesterday and just finished the tutorial videos. I am quite impressed. Even though you smart guys can write strategies in C++, Pascal and Visual Basic, you can also just use simple logic A>BorC strings while clicking on indicators to formulate strategies. Just my cup of tea!
It has GO abilities with a simple method of setting range parameters and you can quickly combine different indicators for testing out of their 800+ base of included indicators.
It interfaces directly with Interactive Brokers or will send emails for autotrading your strategies and has three different solutions for historical data with www.prophet.net costing less than $30/mo.
Just my initial reaction but I feel pretty darn good about it so far. Will keep you updated.
Forest
- eudamonia
- rank: 500+ posts
- Posts: 536
- Joined: Thu Jun 15, 2006 9:50 pm
- Reputation: 0
- Location: Rocklin, CA
- Real name: Edward Heming
- Gender:
- Contact:
O.k. I've been hunting around for a good article or two on multi-objective algorithms:
http://citeseer.ist.psu.edu/horn93multiobjective.html
Cite Seer also has lots of great articles on GAs and other technical research papers.
Edward
http://citeseer.ist.psu.edu/horn93multiobjective.html
Cite Seer also has lots of great articles on GAs and other technical research papers.
Edward
Eudaimonia (pron.: you-die-moan-e-a) (Greek: εὐδαιμονία) is a classical Greek word commonly translated as 'happiness'. The less subjective "human flourishing" is often preferred as a translation.
Hello everybody,
I wanted to point you to a couple resources and ask a question.
http://www.swarmintelligence.org/
(or google for:) Particle swarm optimization, a class of algos to search intractable problem spaces. Newer but similar to GA. Best feature is probably the fact that it's extremely easy to code, almost a must for those like me that want to develop their entire simulation platform (I did it just to have full control)
http://www.dtreg.com/
Classification and Regression Trees, may be useful for pattern classification. You may like treeboosting and decision tree forests, apparently much more efficient than NN. Forests are black boxes but trees are not. The link is to one of the commercial packages available just because I found the sw manual very clear in explaining the underlying concepts (can download it there). Open source code for the algos is available on the net if I recall.
Now for the question. Is anybody willing to share how you use your optimization tools? I'm just thinking/trying kind of a brute force approach (mimicking some of what's available in most commercial sw):
-code parametric entry and exit rule
-establish parameters' limits, fitness function
-optimize, walk forward, optimize, walk forward...
-look for positive expectation (changing rules, parameters intervals)
-check if it was just luck (via acrary's edge test, a variant of bootstrapping technique... more stuff to google for)
-add money management and integrate in systems portfolio
Unluckily I'm stuck at the positive expectation thing LOL! Can't find it yet, but hey, I just tested my first set of rules
So how do you select rules (indicators?)? I considered some stat analysis of the time series first (à la Michal), but wouldn't that be looking into the future even before starting to test? Or do you use GAs to select the very rules as well (I read a paper where they were doing that)?
Thanks!
I wanted to point you to a couple resources and ask a question.
http://www.swarmintelligence.org/
(or google for:) Particle swarm optimization, a class of algos to search intractable problem spaces. Newer but similar to GA. Best feature is probably the fact that it's extremely easy to code, almost a must for those like me that want to develop their entire simulation platform (I did it just to have full control)
http://www.dtreg.com/
Classification and Regression Trees, may be useful for pattern classification. You may like treeboosting and decision tree forests, apparently much more efficient than NN. Forests are black boxes but trees are not. The link is to one of the commercial packages available just because I found the sw manual very clear in explaining the underlying concepts (can download it there). Open source code for the algos is available on the net if I recall.
Now for the question. Is anybody willing to share how you use your optimization tools? I'm just thinking/trying kind of a brute force approach (mimicking some of what's available in most commercial sw):
-code parametric entry and exit rule
-establish parameters' limits, fitness function
-optimize, walk forward, optimize, walk forward...
-look for positive expectation (changing rules, parameters intervals)
-check if it was just luck (via acrary's edge test, a variant of bootstrapping technique... more stuff to google for)
-add money management and integrate in systems portfolio
Unluckily I'm stuck at the positive expectation thing LOL! Can't find it yet, but hey, I just tested my first set of rules
So how do you select rules (indicators?)? I considered some stat analysis of the time series first (à la Michal), but wouldn't that be looking into the future even before starting to test? Or do you use GAs to select the very rules as well (I read a paper where they were doing that)?
Thanks!
System development is One Hell Of A Job!
- michal.kreslik
- rank: 1000+ posts
- Posts: 1048
- Joined: Sat May 13, 2006 2:40 am
- Reputation: 36
- Location: Monte Carlo, Monaco
- Real name: Michal Kreslik
- Gender:
Hello, Neuro!
Thanks for the post. I'm now leading an academic project on Time Series Prediction in Financial Business Systems using Alternative Search Methods.
Those alternative search methods include genetical algorithms, particle swarm optimization, ant colony optimization, artificial immune systems, simulated annealing, tabu search and several other metaheuristics.
Just as you say, I decided to write my own multi-functional optimizer in C# to get a total control. It will be a tough but rewarding work.
Discussion on using the metaheuristics in finding new models from basic mathematical buidling blocks (so not just the common-or-garden paremeter search!) is a very broad theme though.
Michal
Thanks for the post. I'm now leading an academic project on Time Series Prediction in Financial Business Systems using Alternative Search Methods.
Those alternative search methods include genetical algorithms, particle swarm optimization, ant colony optimization, artificial immune systems, simulated annealing, tabu search and several other metaheuristics.
Just as you say, I decided to write my own multi-functional optimizer in C# to get a total control. It will be a tough but rewarding work.
Discussion on using the metaheuristics in finding new models from basic mathematical buidling blocks (so not just the common-or-garden paremeter search!) is a very broad theme though.
Michal
Please add www.kreslik.com to your ad blocker white list.
Thank you for your support.
Thank you for your support.