AI and Neural Network System

[c1borg]

I have been involved in testing Artificial Inteligence AI and Neural Networks for several years and produced some quite sophisticated results on another forum which worked primarily on bots for online poker, but also had some sub forums dedicated to exploring Forex trading. Several members here have PM’d me expressing an interest so would like to share a newly developed complete system with you to :-

a) Get your thoughts and opinions
b) See if its useful
c) Can it be incorporated in any way into trading as an added tool
d) Be improved in any way

If it proves to be useful all well and good if it’s a load of rubbish I would like Steve to delete the thread and I will abandon it and move on. Btw he has kindly given his permission to present this to the forum. Hopefully I have posted the topic in the most relevent section, if not feel free to move it.

So firstly this is NOT backtesting and doesnt use Craptester, although it does perform tests on past data to come up with a future prediction of the price movement. The system does give a prediction average accuracy of around 75% of the future trend UP or DOWN for the next week when applied to at least 2-3 years of previous data. So to put it simply you will know correctly 7 out of 10 times the direction of the trend for next week, not the price but the trend UP or DOWN from Fridays closing price, if you can make use of this I cannot say but hopefully it will be another useful tool in your armory.

The system is based on a research paper :youknow: which predicted historical volatility for options in the CAX40, DAX and the Greek FTSE/ASE 20 which I have attached (just for info dont study it ). The prediction element has been copied and adapted for Forex prediction but seems to work for stocks and commodities just as well.

Over the next few posts I will attempt to describe the system in detail and hopefully I will not distract you from the brilliant EA’s developing on here, but if you have the time and inclination I hope you follow the posts and try things for yourselves. I am no expert by any means on the software used or Neural Networks (hopefully there are some lurking on here) but will try and answer any questions. Please be patient as I don’t have a huge amount of spare time to post.

[c1borg]

Ok step 1 is to download the software to enable us to run the process, Rapidminer studio is available here https://my.rapidminer.com/nexus/account ... #downloads it is available for free to a single user. If you want to follow along and work on the process you will have to install the software. I would put it on the best available hardware you have to spare, as the process will run quicker. It currently takes around 1 Hr on my laptop which is by no means high spec so dont worry if you dont have mega processing power.

[FXNZ]

Good Stuff. This is the way of the future using Big data and cognitive computing. IBM has lots of Watson API 's
I am looking into as well.

[SteveHopwood]

Just to distance myself from this.

Yes, the OP approached me to post. I gave him my blessing.

There my blessing ends.

One of the first paragragraphs in the pdf in the OP's initial post reads thus:

Analysts have claimed over the last years that the volatility of an asset is caused solely by the random arrival of new information about the future returns from the underlying asset. It is a common belief that volatility is of great importance in finance and it is one of the critical factors determining option prices and consequently driving option-trading strategies. This article discusses an empirical option trading methodology based on efficient volatility direction forecasts. Although in most cases accurate volatility forecasts are hard to obtain, forecasting the direction is significantly easier. Increase in the directional accuracy leads to profitable investment strategies. The net gain is depended on the size of the changes as well; however successful volatility forecasts in terms of directional accuracy was found to be sufficient for positive results. In order to evaluate the proposed methodology weekly data from CAX40, DAX and the Greek FTSE/ASE 20 stock indices were used.

In other words: prices will go up and down; sometimes they will go up further than they go down; others they will go down further than they will go op; others they will just wander up and down.

Gosh. We never knew that.

Looks as though HGB'nT has this stuff sussed though. Some of the Slopey family are not doing so badly as well.

OP, only an idiot is attracted by long words and involved sentences within stupidly long paragraphs. Those of us with the occasional firing synapse are unimpressed by either. I could dismantle that paragraph sentence by sentence with ease. I cannot be bothered.

This thread is locked until I am sure that the OP has noticed. Try a stunt like this again OP, and I will remove your ability to post at all.

Capiche?


----------------------------------

Subsequent edit: I had intended that last paragraph to read: "This thread is locked until I am sure that the OP has noticed, and then I will delete it." I have decided to leave it in place as a warning to others.

I remind members that bollocks is not allowed at SHF.

[SteveHopwood]

This thread is open again and I apologise to c1borg for closing it in the first place. I just received this PM from BobbyT:

Hi Steve,

Hope your doing well and getting ready for some blistering 20C heat coming your way in a couple of months

I just wanted to drop you a line about this thread:
http://www.stevehopwoodforex.com/phpBB3 ... 79#p160579

The thread was locked and the attachments were removed before I had a chance to look them over....but...

The 1st paragraph you quoted from the pdf is a direct copy/paste from a research article. c1borg states that his thread is based on a research article.

I know the language in the quoted paragraph looks like bullshit but it is most likely the abstract of a peer-reviewed research article. How do I know this? I am a researcher myself. Granted, a molecular biologist and not a computational finance researcher, but I can pick academic writing from a hundred yards. And that quoted paragraph is most certainly academic writing, specifically, a paper abstract.

In the interest of free exploration and exchange of ideas I gently ask you to unlock the thread to see where it goes. Machine learning, the initial subject of the thread (in the form of neural networks), is an interesting method of trading and one that we do not explore enough here at SHF. I feel (yes I know,...feelings) that we, as a community, could benefit from allowing moderated discussion around machine learning software, algorithms and techniques/strategies. It is something that I personally dabble with a few times a year (when I have the time).

Feel free to ignore this if you like. Hell, even post it and flame me. Just don't ban me. I love this place too much

Cheers,
BobbyT

BobbyT is not the first one here to find that I can be swayed by rational argument. So carry on folks; I shall not interfere again.

[BobbyT]

Thanks for re-opening this Steve

C1borg, I look forward to seeing more about this. Even just getting first hand experience with the platform is a bonus.

I have played with Zorro a bit but have had trouble getting data to align properly. I'm not sure if its me, the source data or the platform. I'm working (slowly) through it regardless.

Now the white paper is available again I'll have a look at it over the weekend and see if there's anything I can contribute

[c1borg]

Ok step 1 is to download the software to enable us to run the process, Rapidminer studio is available here https://my.rapidminer.com/nexus/account ... #downloads it is available for free to a single user. If you want to follow along and work on the process you will have to install the software. I would put it on the best available hardware you have to spare, as the process will run quicker. It currently takes around 1 Hr on my laptop which is by no means high spec so dont worry if you dont have mega processing power.

Ok Step 1 hasnt changed however my free time has, so replies on this subject may not be as quick. But as soon as you have got to grips with Rapidminer I will release my system file(s) for you to try and see what you think.

[rapple]

Dear C1borg,

I am following with interest.

Kind regards Richard

[c1borg]

Ok guys and girls hopefully you are up to speed with Rapidminer so to step 1 here is my RMP process file and a sample data file in the correct format to run the experiment. Obviously you can download and create your own data files to experiment with when you get to grips with the system.

The data file needs to be loaded in here by double clicking and drilling down into the process.

Capture.PNG

and an output file path for the results here, again double click to add the path

Capture1.png

[rapple]

Hi C1borg

Have run your Processes and it worked perfectly.

Now just need to understand the results.

Cheers Richard