January 26, 2008
Product Review: Stock NeuroMaster
About a week ago I stumbled across a neat little stock neural net program called, interestingly enough, Stock NeuroMaster. I downloaded it and played around with it during its 14 day trail evaluation and was surprised by how easy it is to use.
Although I’m a heavy duty RapidMiner modeler, and I build intricate custom neural net/genetic algorithm models, Stock NeuroMaster appeals to me for its simplicity and ease of use.
If you don’t care about how neural nets work, don’t want to deal with the hassle of finding data and making predictions sets, and just want a system to generate BUY and SELL signals with a click of a button, this is the system for you.
The system, from what I’ve deduced, uses a standard backpropogation algorithm. It downloads OHLC closing data for your selected stock symbol, then applies a zigzag line to the peaks and valleys of the stock price (this is done to find profits and losses), generates training BUY/SELL signals, then trains the data for maximum profit, and then builds a model.
Once the model is built, you can update the data by downloading daily/weekly data and then create a prediction with a simple click of a button. The prediction will then automatically generate a BUY/SELL or HOLD signal for the next trading day open. The model will also keep a log of all generated BUY/SELL signals and give you some system statistics on how well the system is performing.
Pros
- Easy to use, no knowledge of neural net algorithms needed
- Clean GUI, easy to read clean charts
- System Performance, Profit and Loss statement, and Buy/Sell Signal summary
- Easy step by step model building and prediction sets
- Yahoo, Metastock, and CSV (format) data compatibility and updates
- Stop loss and trading strategy settings
Cons
- Only a few technical indicators
- No real time data feed or update, only end of the day data downloads
- No “R” position sizing
Bottom line: If you want a neural net program that will generate BUY and SELL signals for stocks, without knowing how the inside guts work, then this is the product for you. Download the Stock NeuroMaster free 14 day trial (zip file) and let me know what you think by leaving me a comment.
January 26th, 2008 at 6:43 pm
But are the signals any good?
January 26th, 2008 at 7:29 pm
The signals seem to be pretty good but with any high volatile and frantic markets, like we’ve been experiencing lately, expect some “off signals”.
Download the product and judge for yourself. The position/performance summary is pretty slick.
January 27th, 2008 at 12:15 am
I’m a Mac user.
January 27th, 2008 at 7:18 am
Then I suggest you download RapidMiner, its java based, and build everything from scratch. :)
January 28th, 2008 at 11:29 am
Why do you care about the indicators?. Isn’t all the information in the indicators already in the price/volume data?. Have you tested their significance for the accuracy of the predictions?
January 28th, 2008 at 11:57 am
Luis,
I like to overlay more indicators over the candlesticks and see where the model is issuing its BUY and SELL signals.
The program compiles a detailed list of performance for the BUY and SELL signals so you can judge the significance for yourself. Download it, play with it, and let me know what you think.
February 19th, 2008 at 3:18 pm
Tom:
I have been using Neuro and it is quite interesting. I have already developed several successful models that I have put into practice.
Strangely enough, the most robust setting that I have found is the Fast-Chaotic, 30-30-30, Investment (Buy/Hold), 5 look back, no filtration. Obviously, use one year of data or more, Associative.
Although the program says the Investment setting is for “Long-tem”, It generates signals every week or so (If you use LONG/SHORT), with relatively good success on volatile ETFs and stocks. I tested dozens of instruments with all time settings, and the longest one see to generate the most profit on average (It caught big moves, cut small losses).
What setting (on average) have you found to work the best for volatile instruments? I know that you are busy now, but would love to read a follow up post eventually as to the modeling settings that you use.
February 21st, 2008 at 6:04 am
RS: I plan on writing a follow up post maybe this week or next, I’m just terribly busy right now (til Friday).
Yes you are right, it manages to find the big moves and cuts the small losses fast. It’s a neat little program and it helps me time my positions better.
February 25th, 2008 at 8:00 pm
Tom:
I am very concerned by Neuro.
I have used it extensively now and I have noticed that the models change frequently.
For instance, I created a model for AAPL that flashed a sell for today’s open. However, when I updated the model after today’s close, the signal was changed from SELL to neutral. What happened to the sell? I have no idea. I did not re-learn the model or anything. It simply changed the existing signal after the update.
This is not the first time that this has happened. Are you aware of this flaw? It certainly has soured me on the program and the model. I don’t think that I would hand under my hard earned cash for such a program without an explanation from the developer as to why the signals are changing without the model being altered.
February 25th, 2008 at 8:34 pm
RS: This hasn’t happened to me on my models but what you say is of grave concern. The neural net shouldn’t be switching signals like that. Where are you getting your quote data from?
Do your signals change after you update them right after the close or the next day? I usually update my models in the morning before the markets open if I use Yahoo downloaded quotes. I’ve noticed that Yahoo data can change from at the close to the next morning and it tends to be less reliable right after the close.
February 25th, 2008 at 11:42 pm
Tom:
This may be the issue and I will look into it. I typically update well after the close (After 8 PM or so). I will be more careful with my reviews of the models, but please be aware of this and look out for it. If you haven’t noticed this, that puts me at ease to a certain degree. This may simply be due to a mistake on my part.
I will follow this more closely and begin a log of the signals in my trading journal. I have been placing paper trades based upon the signals, but I didn’t jot down all the specifics from Neuro when doing so. Not that I have had this issue, I will make a back up of the trades to try and source this potential issue. I hope that is simply my own error.
February 26th, 2008 at 6:06 am
RS: The Yahoo quote data, since its free, can be funky at times. I discovered this issue when working with data downloaded via the TraderXL Pro program. Yahoo seems to provide unverified free data which tends to be adjusted over the course of the evening and into the next day before trading resumes.
You might get one set of prices and volumes at 8PM today and then “adjusted” ones at 9PM or 8AM the next morning.
February 27th, 2008 at 6:54 pm
OK, well I will try updating in the morning from now on. The models have been all over the place (constantly changing) and I hope that it is just a data feed issue.
February 28th, 2008 at 5:58 am
RS: Let me know how it goes. You can always try downloading the price data in CSV format and then updating the price data yourself by adding another line. This way you have complete control over the input data.
July 5th, 2008 at 9:06 pm
How did you guys make out with the possible yahoo data issue?
July 6th, 2008 at 6:54 am
Shone: I never had that problem, I dont know if RS fixed the issue our not. Let’s hope he pops in and posts an update comment.
July 6th, 2008 at 9:23 pm
Tom: I have come across the same problem as RS with this program. I too have used it extensively and the predictions seem to jump out of sequence every so often. They will go from Long/Sell one day to Long/Hold the next viseversa
July 6th, 2008 at 9:37 pm
Sorry for the above comment…I didn’t get finished…
Tom: I have come across the same problem as RS with this program. I too have used it extensively and the predictions seem to jump out of sequence every so often. They will go from Long/Sell one day to Long/Hold the next and vise versa.
I don’t think the data feed is the problem, here is why…
I can go to Yahoo and download historical data in a csv files. (1) file for Jan 1, 2007 thru Jan 1, 2008 and (2) for the remainder of 2008. I build a model with (1) file. Then I go into file (2) and start adding daily data, update it, and get new predictions. Please note that I do not relearn the model after adding the new data. When doing this, every so often the predictions will change from the previous days indicators.
If this is a bug in the program it’s a shame. The program works great and is quite accurate until you come across one of these inconsistancies.
July 7th, 2008 at 5:15 am
BA: Thanks for sharing your observations. The program shouldn’t flip-flop predictions like that. I plan on contacting the developers of this program and finding out what the problem is when I get back from vacation. In the meantime I’m removing all advertisement for it on this blog.
July 7th, 2008 at 1:42 pm
Thanks Tom. I have been trying it out and have started to compare it to my friends back testing in meta stock. Seems to be fairly consistent so far and it gave similar buy signals for stock we were watching. Will keep at it for a while to see how it fairs and will post my findings. Overall I like the program.
July 14th, 2008 at 5:58 am
Shone: Thanks for letting me know that its working for you. Just like you, I don’t see these flip flops but that doesn’t mean my other readers who have seen this are hallucinating.
BA and RS: would either of you send me screenshots of before and after cases where this occurred and a brief writeup of what you did and the data files? I’d like to review it and forward it on to the developers.
July 15th, 2008 at 9:53 pm
Tom: I have compiled the information as you requested above. Please send me an e-mail and I will send it to you.
July 16th, 2008 at 4:24 am
BA: use my gmail address, its desertroot -at- gmail -dot- com. thanks.
March 9th, 2009 at 12:01 am
Anyone have any new news regarding this software? I have found it to be extremely useful regarding particular ETFs. In some cases it is just out of the water, but, if you pick the right ETF, it nails it. Anything?
March 20th, 2009 at 7:07 pm
This is one of the main google hits for the program, besides places to download it, so just for the record, I spent a good amount of time with the demo, and the concerns about positions changing stated above are correct, I would not recommend the program. AI Solver, Rapidminer and Merchant of Venice are all freeware programs more worthy of one’s time..
March 21st, 2009 at 10:47 pm
Manatrader: I stopped using this product because the positions flip flopped. Rapidminer is by far the best IMHO.
March 30th, 2009 at 2:01 am
Dear Traders.
I’m the author of Stock NeuroMaster, and i got several complaints about the problem that you described above. After that, we started our testing, because, for sure, it should not happen, that the signals are changed after receiving new data when updating the model. Actually, it would not be possible, because the software is designed to work correctly, without looking at the data in the past. It was difficult to determine the reason of such a behaviour, because signal changes were not too often, not with every model, and not every day/update.
As you perhaps know, signals changing problem can occur because of delayed data on Yahoo. When you update the model at the moment when the latest data is not recorded into Yahoo database yet, you see non-closed bar. That bar is stored in the model, but it is not final, and the final bar has different prices (close price at least). So, when you get the forecasting results, using that non-closed bar, and
after tomorrow update, that bar is replaced with final one, you may have different signal/recommendation on previous bar. I think that’s the most possible reason of such behavior.
Data downloading is performed by 2 steps. First step is downloading historical data from yahoo’s database, which is marked ‘historical quotes’, and then, immediately after that, current realtime quote is downloaded and added into the model for displaying on the chart. So that last bar may reflect not final data, but latest realtime (20
minutes delayed) data.
So, we were tracking all the data flows, and model changes, during our testing and debugging, during the last 2 weeks. And we found the problem and solved it. The reason was in incorrect model saving function, which moved the point of modelling period several bars backward, this way stored model was reopened, but the system was trying to build new signals on previous several bars as they were marked as ‘new’.
You could notice, that if you didnt save the model after updating, close the model and reopen it again, the model looks updated, with latest bars on the chart, AND signals never changes this way. After saving and reopening, signal were changed sometimes. This tells us that the problem is exactly in model saving procedure.
And, another important thing, is that we made several very useful enhancements in forecasting core, during our deep testing, and, the results of forecasting are better now, even though they were good in previous versions.
So, everyone who is still interested in our software, please download the latest version from our website:
http://stockneuromaster.com/snmaster.zip
and i will be glad to get any kind of feedback on it.
If you need extended trial period to estimate this new version (this is ver. 1.33), please contact me directly.
Sorry for the confusion once again, i hope that you will find it good enough now, with all these fixes and enhancements.
Thanks again for your interest in my software, guys!
Best regards,
Konstantin Grek.
April 2nd, 2009 at 4:10 pm
Dear Admin.
I posted the message a week ago, about the problem described above, and that the problem is solved already, so users can download the latest fixed version of our software to estimate it. But my message still tells “Your comment is awaiting moderation”.
Can you please make it visible?
Thank you.
April 3rd, 2009 at 5:44 am
Konstantin,
Thanks for fixing the issue but I have version 1.31 and would need to upgrade to 1.33. Will you give me a free upgrade or do I have to pay? I would hope you extend free upgrades to users of 1.31.
April 3rd, 2009 at 6:10 am
Sure, everyone who has ver. 1.31 registered, will have new version with no extra charge. Please download it from our website, and if you have any problem with activating it-please contact me directly, i’ll try to help.
Thanks.
Best regards,
Konstantin Grek.
April 3rd, 2009 at 9:38 am
Thanks!
May 13th, 2009 at 1:48 am
I just started using this software. I’ve read the manual twice, but I still don’t understand why the model I create is called 50-50-50, or 30,30,30. What do these numbers mean?
If anyone knows the answer, I sure would appreciate your input. Thanks!
May 13th, 2009 at 5:38 am
Mike, those numbers refer to the neural network configuration itself. The 50-50-50 reference to the hidden layers and the number of nerual network nodes. The higher the numbers, the more nodes and the more complex the model. This also slows down the learning time.
May 14th, 2009 at 12:02 am
Tom, thanks for your explanation. I’m a lot closer to understanding this now. However, I still don’t see how you can get a label of 30-30-30 for a manual model that has 3 layers of 30 Neurons each layer. I don’t understand where the first two of the three 30′s or 50″s
(in some models) comes from. Wouldn’t M-3-30 be better? Then, M for manual, 3 for the number of layers and 30 for the number of neurons in each layer exactly describes the choices of the user when creating the model. If my questions and observations are making you get nuts, I apologize.
March 20th, 2010 at 2:36 am
I just found this software on the internet after using some craptastic software called Trading Solutions. They should call it losing solutions because that's what it does and I have yet to see a positive review outside their website. How many of you trade with this software and how much profit do you make on average (or lose if this is the case)?
Thanks!
March 22nd, 2010 at 5:13 am
@Tony: I predominately use Rapidminer for my market timing/trading and have not use StockNeuromaster for trading.