Automatic Trading System Creation | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Manual Trading vs Automatic Trading - Suggestions? 58 replies

Regarding the article "Trading As A Business" (Or:Are successful systems automatic?) 6 replies

Automatic Trading 5 replies

Trading a (semi) automatic approach...but want more 11 replies

VTTrader: Automatic trading acting screwy 16 replies

Attachments: Automatic Trading System Creation
Exit Attachments

Automatic Trading System Creation

  • #1
  • First Post: Edited 1:57pm Dec 17, 2015 12:48pm | Edited 1:57pm
Hello fellow traders,

After some good years in the algorithmic trading industry it has become very clear to me that the future of this business is in the automatic creation of algo trading strategies. There is simply no way that a regular person can compete with people who can generate millions of strategies every day when it takes that person weeks/months to manually generate/tune/test an algorithmic trading system. A person manually designing something like an expert advisor might spend lots of time going through charts to design the logic, then backtesting, optimizing, etc. When someone with the adequate software can create what took that person months in just a few hours.

If you know me from my machine learning thread, you may already know that I trade thousands of systems at the same time. All of which are automatically generated using many of the principles described on that thread but there is an entire universe outside of machine learning that can also be considered automatic trading system creation.

I am surprised that there is no thread in ForexFactory that actually discusses automatic system creation and I am starting one to address this need. There are many software solutions already available to the average trader so I think we can share our experience with them as well as their pros/cons. A resource like this is desperately needed since these tools in the hands of novice traders are like circular saws in the hands of toddlers: if you don't know your statistics you'll get burned by automatic creation engines.

I am very busy with the algo trading I need to manage ( I only come to FF occasionally) but just ask questions, discuss things and I'll do my best to answer when I come back,

Cheers!

Jo
  • #2
  • Edited 1:14pm Dec 17, 2015 1:03pm | Edited 1:14pm
Do any of you guys have any experience with automatic algo trading system generation software? Which ones do you use? What pros/cons have they had for you?
  • #3
  • Dec 17, 2015 3:55pm Dec 17, 2015 3:55pm
Subscribed
  • #4
  • Dec 17, 2015 5:03pm Dec 17, 2015 5:03pm
Quoting algoTraderJo
Do any of you guys have any experience with automatic algo trading system generation software? Which ones do you use? What pros/cons have they had for you?
I use softwares from Ward-Mql4 and I occasionally use R software through freelance R-MQL4 coders.I make prediction first then do classification using NN.
I used kernel lab package fromR.There are additional packages that I find interesting from Ward like fuzzy pattern recognation,clustering and their adaptive turboprop packages.Major problem with algo trading is many people are finding difficult find good inputs.
  • #5
  • Dec 17, 2015 5:37pm Dec 17, 2015 5:37pm
Quoting olsen-yersen
{quote} I use softwares from Ward-Mql4 and I occasionally use R software through freelance R-MQL4 coders.I make prediction first then do classification using NN. I used kernel lab package fromR.There are additional packages that I find interesting from Ward like fuzzy pattern recognation,clustering and their adaptive turboprop packages.Major problem with algo trading is many people are finding difficult find good inputs.
But to me this process is not automated because you still need to choose your inputs. Have you tried methods for automatic input selection? Searching for algos using different inputs? One thing is to use an ML algorithm and another is to automatically produce the ML algos you will be using for live trading. Thoughts?
  • #6
  • Dec 17, 2015 5:40pm Dec 17, 2015 5:40pm
Well Subscribed ...
ANCORA IMPARÒ
  • #7
  • Edited 6:08pm Dec 17, 2015 5:49pm | Edited 6:08pm
I want to summarize some of the offers currently available in the market to see if anyone has any opinions or experiences with them. These are the packages that I am currently aware of that can do automatic system generation (I am only mentioning them, I have no link or commercial ties to any of them):

  1. Adaptrade (1495 USD)
  2. Strategyquant (1490 USD)
  3. PriceActionLab (1500 USD)
  4. pKantu (291 USD)
  5. OpenKantu (Free)
  6. DATFRA (199 EUR)

Most are worth a good amount of dough. I have used all of the above and although they all have advantages and disadvantages some are very much more worth their salt than others. Have you had experience with any of them? I am very interested in hearing what other traders might have experienced using them. All of them make it very easy to produce worthless systems if you don't know your statistics. Have you been able to make any of the above programs work for you? Am I missing any hidden gems? Share!

  • #8
  • Dec 17, 2015 6:33pm Dec 17, 2015 6:33pm
Quoting algoTraderJo
{quote} But to me this process is not automated because you still need to choose your inputs. Have you tried methods for automatic input selection? Searching for algos using different inputs? One thing is to use an ML algorithm and another is to automatically produce the ML algos you will be using for live trading. Thoughts?
I use R software and Neuroshell to create algos for live trading.I did not understand when you said ''you still need to choose your inputs''.I search for algos using different inputs.
  • #9
  • Dec 17, 2015 6:39pm Dec 17, 2015 6:39pm
Quoting olsen-yersen
{quote} I use R software and Neuroshell to create algos for live trading.I did not understand when you said ''you still need to choose your inputs''.I search for algos using different inputs.
I see, I was misunderstanding your process. How do you control the data mining bias of your system search process?
  • #10
  • Dec 17, 2015 7:05pm Dec 17, 2015 7:05pm
I use Neuroshell's walk forward analysis to create algos .Addition to your list of softwares TRAIDE software looks interesting: https://www.inovancetech.com/
They are pretty new.
  • #11
  • Dec 18, 2015 12:23am Dec 18, 2015 12:23am
I use R also ... there are so many packages to use which people have written, its fast, free, and you are able to code in almost anything you can imagine to test. The downside is its a steep learning curve, but once you are up to speed on it, its great for testing out new ideas, or performing statistical analysis.

EZ
  • #12
  • Dec 18, 2015 4:20am Dec 18, 2015 4:20am
Quoting olsen-yersen
I use Neuroshell's walk forward analysis to create algos .Addition to your list of softwares TRAIDE software looks interesting: https://www.inovancetech.com/ They are pretty new.
Using a walk forward process does not eliminate data mining bias. You must measure mining bias using something alike white's reality check. Do you use any statistical hypothesis test to measure or account for data mining bias? Also thanks for the recommendation I will check it out!
  • #13
  • Dec 18, 2015 4:22am Dec 18, 2015 4:22am
Quoting ezcurrency
I use R also ... there are so many packages to use which people have written, its fast, free, and you are able to code in almost anything you can imagine to test. The downside is its a steep learning curve, but once you are up to speed on it, its great for testing out new ideas, or performing statistical analysis. EZ
I also used to use R, the problem for me was live trading. The R package is really not meant for this, it's mostly single-core and does not have adequate functionality for handling live trading. Executing more than a few systems using R was not possible for me. Since I needed to execute hundreds of systems as fast as possible R was not an option. Have you run into any such issues?
  • #14
  • Dec 18, 2015 11:18am Dec 18, 2015 11:18am
Quoting algoTraderJo
{quote} Using a walk forward process does not eliminate data mining bias. You must measure mining bias using something alike white's reality check. Do you use any statistical hypothesis test to measure or account for data mining bias? Also thanks for the recommendation I will check it out!
You are right walk forward process is not enough to eliminate data mining bias. Unfortunately white's reality check is not available in Neuroshell. Zorro Trader http://www.zorro-trader.com/faq.php has white's reality check available http://manual.zorro-project.com/new.htm Their developers are in touch with David Aronson to integrate Aronson's TSSB software http://www.tssbsoftware.com/ to Zorro Trader platform.
I always learn something new everyday.But I have a long way to go.Applying statistical tests to measure data mining bias on different platforms requires multiplatform coding skills.it is a steep learning curve.
  • #15
  • Dec 18, 2015 11:24am Dec 18, 2015 11:24am
Quoting algoTraderJo
{quote} I also used to use R, the problem for me was live trading. The R package is really not meant for this, it's mostly single-core and does not have adequate functionality for handling live trading. Executing more than a few systems using R was not possible for me. Since I needed to execute hundreds of systems as fast as possible R was not an option. Have you run into any such issues?
I just use it for doing analysis and quick backtesting. If an idea works, then I'll code it into MT4 and live test on demo for a while. If profitable then try on live act with smaller lot size. If its still matching prior results, then increase lot size.

I also are of the mindset of using multiple systems on multiple pairs for trading (for diversification among systems & pairs). Currently I've developed a trend model and range model (and variations of each), and testing out.

Regards,
EZ
  • #16
  • Edited 9:47pm Dec 18, 2015 8:21pm | Edited 9:47pm
Quoting algoTraderJo
I want to summarize some of the offers currently available in the market to see if anyone has any opinions or experiences with them. These are the packages that I am currently aware of that can do automatic system generation (I am only mentioning them, I have no link or commercial ties to any of them): Adaptrade (1495 USD) Strategyquant (1490 USD) PriceActionLab (1500 USD) pKantu (291 USD)...
Hi Jo

That post of yours really caught my eye. I am a bit of a novice in this area but have been particularly interested in Adaptrade following a podcast I was listening to with Michael Bryant.

If I was going to ask Santa for a christmas present, are you generally happy with this bit of software (the user guide is attached)? I am not familiar with strategyquant or priceaction quant.

What I like with Adaptrade is that the code is generated for a variety of platforms and was thinking that working in a team with a couple of coders to tweak things a bit and add a bit of further functionality to the code that was produced, you could really punch the EA's out if you are looking to trade a portfolio of systems.

I was wondering how the code generation fares over long range data to further assist in reducing data mining bias. Obviously this is a major feature that needs to be considered with EA generation methods like this.

I would be interested to hear more of your thoughts about Adaptrade before I hit the demo button for a trial.

Regards

C
Attached File(s)
File Type: pdf AdaptradeBuilderUG.pdf   2.4 MB | 789 downloads
  • #17
  • Edited 6:23pm Dec 20, 2015 6:08pm | Edited 6:23pm
Quoting Copernicus
{quote} Hi Jo That post of yours really caught my eye. I am a bit of a novice in this area but have been particularly interested in Adaptrade following a podcast I was listening to with Michael Bryant. If I was going to ask Santa for a christmas present, are you generally happy with this bit of software (the user guide is attached)? I am not familiar with strategyquant or priceaction quant. What I like with Adaptrade is that the code is generated for a variety of platforms and was thinking...
IMHO there is a very very very very big problem when you use genetic programming for the building of algo trading systems. In software like adaptrade you have genetic searching that introduces a bus load of bias into the process that is terribly hard to account for. A substantial weakness of most software I have found online is that they lack ways to evaluate the bias of the mining process. In something like Adaptrade in my experience systems tend to fail because the bias in their creation is humongous. This was my personal experience with it, just my personal experience.

I have started another thread that will be more about purely discussing the theory of all of this (http://www.forexfactory.com/showthre...85#post8655285), we can keep this one to discuss the commercial implementations (since this threads has been moved to the commercial section due to the mentioning and linking to the commercial setups -- I guess). I will use the other one for developing the ideas I want to follow here in FF on this topic.
  • #18
  • Dec 20, 2015 6:56pm Dec 20, 2015 6:56pm
Quoting algoTraderJo
{quote} IMHO there is a very very very very big problem when you use genetic programming for the building of algo trading systems. In software like adaptrade you have genetic searching that introduces a bus load of bias into the process that is terribly hard to account for. A substantial weakness of most software I have found online is that they lack ways to evaluate the bias of the mining process. In something like Adaptrade in my experience systems tend to fail because the bias in their creation is humongous. This was my personal experience with...
Thanks for the response Jo. The curve fitting obviously appears to be an almost insurmountable problems with approaches such as these. I will subscribe to your other thread and am looking forward to the discussions.

Thanks mate.

C
  • #19
  • Dec 21, 2015 6:04am Dec 21, 2015 6:04am
Quoting algoTraderJo
I want to summarize some of the offers currently available in the market to see if anyone has any opinions or experiences with them. These are the packages that I am currently aware of that can do automatic system generation (I am only mentioning them, I have no link or commercial ties to any of them): Adaptrade (1495 USD) Strategyquant (1490 USD) PriceActionLab (1500 USD) pKantu (291 USD)...
Worth A look.
Quantpedia
QuantConnect
Quantopian
Ernie Chan’s blog
Alvarez Quant Trading
Jonathan Kinlay
Quants Portal
Quantifiable Edges
WealthLab
RightEdge prefered by Andreas Clenow http://www.followingthetrend.com/ , the Author of the well Known Following the Trend and the Latest ​Stocks on the Move .​

I use trading blox among others because i'm not a programmer.

Best,
B.R.
Master Your Setup, Master Your self. (NQoos)
  • #20
  • Dec 21, 2015 6:50am Dec 21, 2015 6:50am
hihi, good stuff!

Any recommendation on specialisation on multy-currency and multy-timeframe functionality?

thank you
0 traders viewing now