Nueroshell, Tradecision

Hi all
Been trying out NS and have had mixed results. I do have to say its a handful, the attitude at tech support is truly one sided. Like here's where you can find it in the help files. Thinking of giving back so they can help themselves to my return.

Anybody have any thing positive on NS.

Tradecision. Tradecision looks pretty well put together and they have been eager to answer questions unlike Ward systems group. So am thinking of trying it out.

Thanks

 

Oops NS is Neuroshell

 

I use both these software packages. Any of my dealings with NeuroShell were during the purchase process and they were very helpful. Their product is first rate and is generally regarded as the benchmark against which all other neural software is compared. This said it does take a lot of effort to use it to its full potential.

Tradecision is also an excellent neural network package. Plus it has first rate charting and data managing. Their technical support is top notch. Any inquiries are generally answered within a matter of hours and they go out of their way to be helpful.

Having said all this none of these packages will provide you with a silver bullet for trading stocks. To build a successful trading system based on neural nets is an extremely complicated
task requiring a lot of effort and time. However if you are determined to do so
either of these packages would be a good place to start.

Lets know if I can be of any help.

LotsOfLoot

 

Thanks very much Lots for your input. I am diligently working on systems. However as you say it takes a lot of time and work.

One thing I would like to ask you is if you run TD and NS back to back for predictions systems or strategies do they tend to come up with the same outputs or very different results? I really like TD's package design as it seems to more of an intensive charting program.

I am thinking of course of trying that my self. However the paralysis of analysis comes to mind and NST is enough Wahoo at the moment.

I am aware of NST being the bench mark I guess thats why I was so supprised they seem so uneager to help me get up to speed quicky on the product. Inversley if they dont care whether you send it back or not says they have enough customers make enough money and dont have time to be bothered if you dont want to spend endless trial and error hours learning the basics or help you get that first system together they told you "yea yea yea it will do that" to get you to buy it. Anyway enough of that. Just wanted to let others know to pin them down good before gettiting it.

Its funny the first prediction I did for last Wednesday on a 5min bar chart I ran all day and it could do no wrong. I was absolutley astounded. I could not make a mistake just wait for the time in price to meet the indicator and a reversal would occur.
That happended for 16 trades. Thursday it was completley wrong and was giving inverse signals and has sense. I have changed nothing so am scrapping that particular stategy.


Obviously I have some work to do.

thanks again.

 

If you program a simple strategy such as one based on a moving average crossover system to generate buy/sell signals then you will get similiar results with NS and TD.

However, if you develop a system based around neural networks then the correlation between those run on NS and TD are not so predictable. This can be due to the fact buy/sell signals generated by NS and TD will depend on their ability to identify say price patterns. Now as NS and TD will each program their various algoritms differently a pattern identified by one may not be identified by the other. Hence a different set of buy/sell signals.

My best advise would be that rather then try to develop a trading system for a specific stock, develop what you consider a workable system and test it against a wide range of stocks eg the S&P 500. Then trade those stocks that give the best and most consistent results. You can do this by setting up a portfolio of stocks and testing them against your trading system.

Hope this has being some help.

 

Started slowly with Td STD; now using RT version (take a gander at my humble opus: http://www.elitetrader.com/so/?action=view&SR_ProductID=150).

Asked dozen of questions at the beginning – so yeah, support guys really sweated to make sure I’m pleased with TD. I’m not an AI/NN fan, just playing with model builder time to time. So, agree with lotsofloot – NN’s rather a time-consuming task. Regularly run Sim manager & MM editor (real gem) to check my ideas; apply “on-the-fly” auto-trends/EW/patterns to find current market phase. I like their overall "thinking" approach yet there’s some room for improvements. Waiting for the direct trading feature.

 

I have never tried NS, but I saw they added automated trading with IB so that's a big plus to any other NN software that I know right now. I see a boatload of plugins and extensions being available for NS so it seems this is the de facto standard in the NN department.

Tradingsolutions (http://www.tradingsolutions.com) just recently released version 4.0 of their software. They have a 30 day free trial, can't hurt to check it out. It doesn't come with any mind blowing additions from what I saw but it's nice to see these guys are still working on their software. I heard a little bird chirping they will release an automated trading interface (I guess to catch up with NS) in the coming months.

Can't comment on TD. Will check & bookmark their website now.

 

I tried Tradecision earlier this year and sent it back. My chief complaint was that it came up with different backtest results over repeated runs, despite the fact that none of the test parameters had changed. IOW I would do a backtest run and obtain one result, and using the same parameters would run the backtest over again and get a different result. Which result do you believe? The company was helpful in responding to my emails and attributed the phenomenon to the nature of neural networks. At that point I abandoned any further exploration of neural network systems. As stated earlier, I don't see them as being a silver bullet.

 

Using TD model builder I’m constantly getting different results with each NN model re-train but that's ok, I can't worry. Simply look through their findings (manuals & guides) and other AI related stuff to familiarize with the concept of local error minima. BTW building a committee with a sound money management idea is important while uisng NNs.

No wonder (maybe at such “bug-eyed” customer's requests) they just recently added a check box (retrain NN model with the same weights) allowing you to have the same results )

If I use a finished model for a trading system backtesting (strategy builder) I always have the same results/reports/graphs. However, I suggest paying attention to some indicators, which accumulate value in the recalculation process, such as AccDist, RSI, On Balance Vol, a few others. They can lead to dif. results while performing a system becktesting if there’re dif. start dates of the same strategy.

 

Standard feed forward nets, like backprop have these problems. Methods like Radial nets and Kernel regression do not. You run them over and over again you always get the same results. When using standard nets it is recommended you retrain 10 times or more and pick the best net in terms of learning the training set to use for your walk forward system rules.

I have all three methods in my Neural Studio addin for TradersStudio. This addin requires owning TradersStudio.

http://www.tradersstudio.com/Produc...ID/5/psnavcmd/CatalogItemDetails/Default.aspx