Discussion in 'Trading Software' started by sluggworth, Nov 21, 2005.
Experience with product, positive or negative would be appreciated.
The first thing you should understand is that Neuroshell will not produce huge returns (at least for me). If you work at it you can get nice returns. I'm happy with between a consistent 10% to 20%. Others may claim more, I don't.
It is not an easy product to learn. You must study it and play with it. I would not expect to get much out of it for 2 to 3 months at a minimum.
It is loaded with indicators. If it has been created, it is in Neuroshell Trader (NST) or you can create it. I suggest you look at some back issues of Stocks and Commodities to see how NST implements things. There are some indicators which are not readily created using the built-in functions. You can use any product that can produce a DLL (Basic, C, C+, C#, etc) to roll your own.
You can create predictions using the indicators to give you buy or sell signals. You can then create strategies to buy or sell based on the predictions. The strategy can be something like - If a position is already open ignore, if it before 10 AM ignore, if indicator 1 AND indicator 2 indicate buy do so if none of the major averages are down for the day unless it is Tuesday.
All of these things can be optimized or not. You can tell the program to pick the best strategy rules, pick the best indicators, pick the best parameters for the indicators, etc. Each of these can be handled as a group or on an individual basis.
You can use outside data to restrict actions, that is, do not sell Wheat (short) if the rainfall distribution in Kansas is normal. You can do all the normal things such as setting up reversal systems, long only if, short only if, etc. NST handles any ASCII data you can feed it and many of the popular proprietary feeds.
Money management is available to a limited extent. You can do $ investing, fixed number of shares/contracts, buy with money available, margin, etc. Commissions and slippage are easily entered.
It can perform portfolio-wide analysis of an idea. I use a 10 stock subset of the S&P 100 to test a system and refine it. I then run the system against the entire 100 to see if it is robust.
A mistake people make is expecting too much. Neural networks and genetic algorithms do not create solutions. They help you quickly find out whether an idea you have about the characteristics of some trading vehicle is valid. It will help verify (or not) your theorized edge.
About a year and a half ago Ward Systems stated that they were going to make a number of significant changes to NST and introduce them in Version 5.0. In his October 2005 newsletter, Steve Ward advised that 5.0 was in final beta testing. One of the problems with version 4.x is its treatment of commodities. Because NST was originally designed as a stock system, it can only handle one contract point value so testing a portfolio of commodities\futures is not valid if the point size is different (say wheat @50 and bean oil @ 600 or the e-mini S&P @50 and the standard S&P @250). Version 5.0 is to fix this problem.
NST does not have a broker interface for trade execution but Ward has stated that there will be an API in 5.0 to solve this problem. He stated that they would provide one working interface as an example and documentation to assist in development for other brokers. Since Steve had mentioned in the private Ward Forum that he uses Interactive Brokers, it's my guess that the demo\working interface will be to IB's API, but that is only a guess.
The charts in NST leave something to be desired. There is no drawing tool. The only thing you get on the charts are what the program produces. If you program Pivots, they will show up, if you program (use the built-in function) for regression lines, you'll see them, but you cannot draw a trendline by hand or manually add\remove a fib, or a channel, etc. You can reveal or hide program drawn indicators. Once again, 5.0 is supposed to add a number of charting features.
Programming (scripting) is done by point and click assembly of your special purpose indicators, predictions, and strategies. I did not like this at first, I'd rather type things out. However, you quickly acclimate, and you never need to go hunting for "variable not found" as you cannot mistype. You can, of course, mess up your logic and get strange results.
Optimizing indicators can be a chore. Let's ay you have a MACD set for 14,26,9 and an RSI at 14. The current program allows you to optimize all the parameters - or not. If you want to optimize everything except the 26 period portion of the MACD you must go into the indicator and reduce the search range from the default 15 to 42 to 26 to 26. LOTs of users requested that a simple method to lock any indicator while allowing the rest to optimize be added. I strongly suspect 5.0 will provide this feature.
Ward help is very good and free. It is not toll free but at today's rates that is inconsequential.
Ward has a moderated-web site for registered users. There are good discussions but there is rarely anything negative on the site. All posts are delayed until a moderator reviews them. Thus, there is not any spontaneity such as occurs here on ET. I think the users suffer but Ward will not change.
The users are rather protective and if you go to Yahoo Groups you'll find virtually no traffic. The program is expensive and the users do not want to support pirated copies. They restrict themselves to the "crippled" Ward forum.
You can use load the program on as many machines as you wish, but only a single copy can be run at one time.
5.0 will probably be released late this year or early next year. Ward has always allowed upgrading for a $50 or so disk shipping and handling charge. I assume he will do so this time as well. However, if you are thinking of buying I'd recommend talking to sales about 4.8 vs 5.0.
Hope that helps, but I've got to stop now or I'll miss dinner.
very informative post, Jack.
Thank you for the excellent reply.
Jack nailed it with his reply - everything I would have mentioned and more.
I spent 2 years or so playing with NSDTP. It's an awesome package, and has some really interesting indicators - I was especially intrigued by the "clustering" indicators add-on.
Did I ever make a dime from directly using NSDTP? No. But looking back I see it as just one of the steps I had to go through to get where I am today in terms of keeping things simple.
You should be asking yourself - "Why do I need something like NSDT?"
What is the answer to that?
In my case results have been deceiving .
I got hold of one of these "pirated copies" and started fumbling with it.
Did some study. Let me tell you that, should I have purchase the product, the 30 days money-back guarantee would not have alowed me to play with it enough to see its shortcommings.
Neural networks: plainly deceiving. Maybe my fault but widely backed up by other people's opinion in this forum.
Then tried to build strategies with indicators only and the help of the optimizer.
Surprise: I got very profitable buy/sell signals using historical data. Curve fitting ? I don't know.
But then I discovered that after adding new daily data to the data file, every time Neuroshell detected a new buy/sell signal needed to be issued, it asked to rebacktest (otherwise I discovered no signal will ever be issued) and all previous buy sell signals were modified :eek:.
Maybe only my personal experience with a pirated copy but AFAIC I'm not willing to fork all that money and see if Ward Systems honor their money back guarantee. Besides, as I said before, 30 days is too little time, specially taking in account that many of these days would be lost in the round trip from ward to my office and back, to evaluate the product.
The problem quoted above was operator error. Recalculation after a strategy has been backtested is a user selectable option.
Although it was not available to you, Ward has an excellent customer support policy and telephone or email support is free and without a time limit. You might ask them for a 60 day trial. They are a pretty flexible group. However, don't ask right now. They released Version 5.0 last week and there are enough new features and subtle changes that I suspect the help folks will be busy for a while.
Hi and thanks for taking interest in this topic.
In my case, every time after adding new data, if NS detected a nrew buy/sell signal needed to be issued, it asked to rebacktest.
Not a big problem since I got used to hit the "stop optimizing" button and partameters remained unchanged. I could also take note of the parameters and rebuild another "chart page" with the same strategy but "no optimization" so I did not go thru the hassle of hitting the stop optimization button to keep the parameters remain unchanged.
But I want to state clearly that in my expeience unless I rebacktested NO buy/sell signal would be issued ever
The problem (among quite many other shortcommings) was that after adding data for -let's say- Fri march 24, NS issued a buy/sell signal for Fri march 17 and -since a market order was specified- it used Sat march 18's close to buy/sell (but stated in report that "fill date" was march 24). I guess I would have needed a time machine to go back to saturday 18th and trade. But then, if I had such a device, why would I need NS ?
BTW this hapened feeding daily data but using weekly bars.
Other shortcommings/bugs (IMHO):
In v3.8 is that you cannot select on which day of the week you want to trade.
Using a trailing price for "long exit" will sell at the NS calculated trailing price which does not correspond to next weekly close (still on weekly bars example here).
These don't seem minor glitches. They may have been adressed in later releases but at the time I could not believe that such a highly priced product could be so "unfinished".
I have more (sadly, too many) of these "glitches" in store to comment on if someone's interested.
Now I am looking into TradingSolutions which also has a 30 day moneyback guarantee. Since it's available for download, it eliminates the shipping time and cost problem.
I am in no way affiliated to ward (evidently) or tradingsolutions. If I go for the latter I will try to keep you informed.
Separate names with a comma.