Q. The data that I downloaded from ESignal is incorrect, missing or corrupt data. How do I fix this problem?
A. In order to correct these problems you will need to manually delete files that the NeuroShell Trader downloads and stores in the ServersData directory. To do this unload any charts that you believe have incorrect, missing or corrupt data, use the Windows Explorer to delete the ticker file associated with the bad data (for Yahoo ‘ 1 minute data the filename is YHOO.1 min.wsg), then reload the chart.
Q. Why can’t NeuroShell Trader find my AIQ data?
A. In order for NeuroShell Trader to map AIQ data you must use the Data Sources Dialog to add the directory that contains the master.ndx and master.adf files. If you are having difficulty mapping or using your AIQ data you may consider using the “Rebuild Master List” functionality in your AIQ product. Please note that NeuroShell Trader is unable to read data from AIQ’s data CD.
Q. How do I import my data files?
A. First, you must make sure that your data files are in the proper format. The NeuroShell Trader reads text files with the following extensions: *.txt, *.csv, *.prn, and *.asc. It also reads MetaStock, and AIQ. Once you have verified that your data is in the proper format, open the NeuroShell Trader, select Data Sources from the Tools menu, and select the Directories/Files tab. Then click on the Add Directory button. Browse to locate the correct directory and hit OK. Note that the NeuroShell Trader automatically assigns a category but you can change this category if you wish by clicking on the Change Category button after highlighting the directory name you wish to change.
Q. Does the NeuroShell Trader perform optimization of inputs?
A. No, it does not. However, the NeuroShell Trader Professional and NeuroShell DayTrader Professional have a genetic optimizer that assists in the selection of inputs and indicator parameters. On the other hand, after you have trained a net, you can see how important each input was to the net. Just select Detailed Analysis on the final Training Wizard screen (before you hit the Finish button). Then choose the Input Contributions tab.
Q. Can I build my own custom indicators in the NeuroShell Trader?
A. Yes. The NeuroShell Trader comes with built-in functions that you can use to create your own custom indicator. These functions are found in the Indicator Wizard under the following categories: Arithmetic, Boolean, Rules, Trigonometric, and Relational. You can create your own custom indicator by combining any of these functions together. You can even combine them with the other standard indicators that are already included.
Q. Why can’t I add a stock page to my index chart using Edit, Add/Remove Chart Pages?
A. You cannot create stock pages in an index chart or vice versa. This was designed like this because someone might try to add an index page to a stock chart that has indicators on it that are only applicable to stocks. All of the pages in your NeuroShell Trader chart must be constructed from data streams in the same category.
Q. How is the profit calculated in the Train by Profit mode of training?
A. The profit is calculated in the form of a percent return for each trade. For the long side (sell side works the same way), the NeuroShell Trader will buy when the prediction is greater than your buy long threshold and hold the position until the prediction drops below your exit long threshold. Percent return for that trade is then calculated by dividing the point total made in the trade by the security’s value at trade entry. For example, if you bought IBM at 120, stayed long for 10 days, and then exited at 130 your profit would be:
10 ( points made) / 120 (initial security value) = 8.33%
When Training by Profit, the NeuroShell Trader will try to maximize the sum of the percent returns for each trade. The profit figures that are listed on the Training Results screen are annualized figures.
Q. Can I change the periodicity of my chart (from daily to weekly or monthly or (DayTrader Only) from 5 minute bars to 10 minute bars)?
A. Yes, you can do this when you first create the chart or after the chart is displayed. The Chart Wizard will ask you if you want the chart to be in daily, weekly, monthly (DayTrader Only) or intraday format. Once you have created a chart, you can change it to a different periodicity by using the frequency drop down box located at the top of the chart. Note that any predictions and trading strategies on the chart will need to be retrained and rebacktested every time you change the charts periodicity.
Q. How do I actually view the predictions?
A. You can get the predicted value by viewing the data. To do this please see Displaying Data.
Q. Is there a way to evaluate the influence each indicator has on the prediction?
A. Yes, after you have trained a net, you can see how important each input was to the net. Just select Prediction Analysis on the final Prediction Wizard screen (before you hit the Finish button). Then choose the Input Contributions tab. Each input will have a number: the higher the number the more important the net found that input to be.
Q. Can the prediction results be printed or recalled to compare with other predictions?
A. Yes, the prediction results can be viewed by selecting the prediction and choosing Modify Selected Prediction from the Edit menu (or from the right-click). At this point you can print the predictions results from final Prediction Wizard screen by selecting the print button.
Q. After I train a prediction and plot it on my chart, how can I view the training results again?
A. You can access the Training Results screen again by selecting the prediction on your chart (using the right mouse button) and selecting the Modify Selected Prediction option.
Q. How can I delete indicators from the chart or from memory?
A. You can delete indicators from your chart by highlighting them and pressing the delete key. Remember that the NeuroShell Trader stores indicators that are being used in predictions and these can be deleted from your chart (which will clear some memory), but will not be completely deleted from memory until the prediction has been removed. Also, an indicator that is based on other indicators does not require those indicators to be displayed on the chart (in fact it takes up more memory to have them displayed on the chart). Additionally, if any unnecessary indicators hidden from view (check the Hide/Unhide Data screen), you may unhide them and delete them as well.
Q. I cannot see the entire indicator stream because it goes out of the edit window. How can I fix this?
A. You can expand any window to see more of the indicator stream. You do this by placing your mouse in the lower right hand corner of the form screen and wait for the double arrow to appear, then resize the window.
Q. Why do I get ‘System Out of Memory’ when I try to open a chart of a data file I have on my hard drive?
A. You are probably trying to open the chart by using the Open Chart menu command from the File menu. The Open Chart command lets you browse through folders to find previously saved NeuroShell Trader charts. You cannot directly open a MetaStock, CSI, AIQ, ASCII, or any other type file using the Open Chart command. You first need to add the data directory under Tools – Data Sources and then create a new chart with the Chart Wizard.
Q. What is the difference between the neural nets in NeuroShell 2 and those used in the NeuroShell Trader?
A. TurboProp 2, which is the neural net algorithm used in the NeuroShell Trader, is uniquely different than any of the algorithms in NeuroShell 2. First of all, TurboProp 2 prevents over fitting without using a test set. This means that you can train on all of your historical data without having to hold any out. Second, TurboProp 2 is the result of several years of research and is, therefore, newer technology than the older neural nets in NeuroShell 2. Third, TurboProp 2 trains much, much faster than many of the neural nets in NeuroShell 2. This means the time it takes you to develop a profitable model will be that much shorter. Finally, Turboprop 2 is much more accurate than any net that has ever built to date (that we know of).
Q. What is an acceptable range for the Training and Evaluation Average Error figures?
A. There is no particular range. This number will vary depending on what you are trying to predict. The real question is, can you make money on the predictions? As our profit based training will illustrate, you don’t have to get good predictions at all as long is the direction is mostly right.
Q. When a normalization technique is used in the prediction, such as the percent change in price, is this technique applied to all the instruments and indicators in the study?
A. No, this is only applied to the value you are trying to predict. This normalization technique does not apply to indicators or inputs that will be used to make that prediction. If you want to use inter-market analysis and would like to include the price of another market as an input to your prediction, then it is a good idea to add it’s percent change rather than its raw data value. The algorithm does scale all of your inputs and does some normalizing for you, but you will probably have a better model if you use technical indicators to pre-process your raw data.
Q. When a network has been trained, can the resulting analysis be applied to new data without re-training? How is this done?
A. Yes, after you have trained a net, it will automatically be applied to any new data that is added to your chart. For example, if you download your data every day in MetaStock format and train a net on one of those files, then new predictions will be made whenever the data for that chart is updated.
Q. Is the NeuroShell Trader capable of using a prediction to generate buy/sell signals on new data?
A. Yes, you can create buy/sell signals by applying an indicator to your prediction. For example, your indictor could be something like ‘If prediction > .5 then 1 else 0’. This would result in a binary 1 or 0 buy/sell indicator being plotted. Or you could decide to train the net to produce a 1 or a 0 (or 1 , -1).
Q. Why is it that when I compared the predictions from the NeuroShell Trader with an older back propagation neural net and found that accuracy of the nets was about the same?
A. Neural nets are the best modeling tools available, and the new Turboprop2 is far superior for most problems than backpropagation. However, no modeling tool can perform any better if it is given poor inputs. The old saying, “Garbage In, Garbage Out” rings true. Predicting the stock market with technical indicators is a very difficult process (if it weren’t, there would be thousands of rich neural net users). The reason the stock market is so difficult is that there are literally millions of possible variables that can be used for each of the millions of combinations of output types. We have one customer who earned a 300 percent return last year in actual daily trading by predicting things like the S&P, but he didn’t get there by just feeding open, high, low, close, and volume into his nets. He did a lot of work experimenting and using his own savvy to arrive at appropriate things (LEADING indicators) to feed his models. Neural nets are not magic; they are simply great modeling tools. You have to feed them things that are predictive. To see that this is true, just feed one of your nets the output as an input and you will suddenly see it nail the answer. If you have the right inputs, you will get good outputs. If you don’t, no modeling tool can make it happen. It takes time.
Q. How do I create a network that will learn to generate trading signals under specific conditions?
A. You can do this by first creating an indicator that outputs a 1 for a buy or a -1 for a sell and choose it as the value you are trying to predict. Using IF THEN rules, you can build the indicator to sense the specific conditions in which you are interested.
Q. When simulating a market order, does the trading strategy take into account whether the opening price is at the bid or the ask price?
A. When simulating a market order, a trading strategy only uses the next bars opening price and does not take into consideration whether the opening price was at the bid or ask. If you set slippage to the expected average bid/ask spread, the entry and exit prices will be adjusted as if the opening price was at the opposite side of the bid ask spread as the entry or exit would occur.
Q. When I export data to a file, only the date and the actual price values are showing up. What am I doing wrong?
A. When you go to export data items to a file, it will ask you to add the items you want to export. Make sure you select ‘Out of sample: Predicted ‘ ‘ or whatever value you would like to export to file.
Q. How do I use the neural net prediction once I have developed it?
A. The nets are automatically applied to your chart when you add more data. All you have to do is add data to the files (new dates). Then when you load the chart, you just scroll over to the right side and view the new predictions. You can view the data by Displaying Data. Additionally, you can build a trading strategy based upon your prediction by selecting Trading Strategy from the Insert menu. This will allow you to place trading points on your graph using rules similar to the following:
If prediction > x then buy long
If prediction < -x then sell long
Q. When I add new data to my chart, do I have to re-train the neural net?
A. No. Once you have trained a neural net prediction and applied it to your chart, it will generate new predictions every time your chart is updated with new data. The prediction becomes just like another indicator, only a leading one rather than a lagging one. At some point, you may want to re-train your neural net. We have some users that re-train every six months, other retrain every two weeks. It all depends on the robustness of your model and the market you are trying to predict.
Q. When I load a text file that I created, why does the NeuroShell Trader only plot one of my data streams?
A. You can insert other variables from your data file by choosing Existing Data/Calculations from the Insert Menu. This will allow you to insert other variables from your data file. It is important to remember that every column in your data file must have a label row, which will enable you to plot the right value.
Q. How can I use logarithms to normalize data?
A. You can do this by applying the Log indicator to your data before you add it as an input to your prediction. The Log indicator is found in the Arithmetic category.
Q. Why do I get an error when I try to open MetaStock, CSI, AIQ or ASCII files?
A. You are probably trying to open the file by using the Open Chart command from the File menu. The Open Chart command lets you browse through folders to fine previously saved NeuroShell Trader charts. You cannot directly open a MetaStock, CSI, AIQ, or ASCII, or any other type file using the Open Chart command. You first need to add the data directory under Tools, Data Sources and then create a new chart with the Chart Wizard.
Q. When I Auto Scan for data files the program either stops responding or crashes. What do I do?
A. The Auto Scan feature is not always able to locate all of your data files. If you have older or corrupted data files then this could be what is causing the program to stall. To work around the problem, if you know which directories contain your data files, then skip the Auto Scan feature and just add the directories one at a time. If one of those directories is still causing the program to crash, then look in the directory to verify that all of the files are data files and everything looks normal. If you still are having difficulties, then email some or all of the files in that directory to Ward Systems Group Technical Support so the problem can be corrected. Your directory may contain a file that we never ran across in our product testing, so by emailing it to us we can correct the bug that is causing the program to crash.
Q. How can I use the NeuroShell Trader to make predictions on a daily chart several months into the future?
A. You can make predictions 6 weeks, 3 months, and even 6 months in advance on a daily chart. You would just change your Days Ahead parameter to be the appropriate number of days. For example, a 6-week prediction on a daily chart is the same as a 30-day prediction (assuming 5 days in a week). A 3-month prediction is a 60-day prediction (assuming 20 days in a month), etc.
Q. I would like to create a prediction and then plug the results in a database. How do I do this?
A. After training, you can export the prediction values for use by another program or database. You can export your trained predictions, indicators, and closing prices to a text file by choosing Export Chart/Data from the Tools menu.
Q. Since I track stocks by groups, is there a way to separate batches into groups for stocks, currencies, and indexes?
A. Yes, you can create a chart for stocks, one for currencies, and one for indexes. Remember that you can add another instrument to a chart by selecting Add/Remove Chart Pages from the Edit menu.
Q. It looks like the program is retraining each time I open the chart or switch to a different page in the chart. Why is this?
A. The software does not retrain the neural net each time you load the chart page. It is just loading the price data into memory and calculating the values for all of the indicators and predictions that are currently on your chart.
Q. How would I build a portfolio of health care stocks and generate new predictions for them each day?
A. You can create a new chart called health care stocks and add multiple stocks to the chart by highlighting them in the selection box. If you already have a chart to which you would like to add additional stocks, then you could just choose Add/Remove Chart Pages from the Edit Menu. Once you have a chart with different pages in it for each health care stock, you would just build a prediction for the one stock and it would automatically be applied to all of the other stocks in the chart. All you will have to do now is open the chart each day after your data has been updated and new predictions will be made for each of the stocks. You can also add buy/sell indicators to make it easier to determine which stocks are predicted to rise or fall.
Q. What is the difference between the ‘Prediction‘ and the ‘Prediction Signal’?
A. The ‘Prediction’ values are displayed on the chart for the bar that they are predicted, where as the ‘Prediction Signal’ is displayed on the chart for the bar that the prediction was made on. This means that the ‘Prediction’ will lag the ‘Prediction Signal’ by the number of periods into the future that you are predicting. Be sure that all inputs to a different Prediction or Trading Strategy are created using the ‘Signal’ version of the prediction, other wise your prediction or trading strategy will use values that were predict for today, not for the future.
Q. Why does my prediction just seem to mirror the target price, only it is offset by the number of days in the prediction?
A. Neural nets are great modeling tools, but no model is any better than the inputs you are using to build the model. It sounds like you are trying to predict something like the close and are using the open, high, low, and close as inputs to your prediction. The neural net is unable to find any patterns in the inputs you are giving it, so it is forced to just go with whatever will get it the closest to the prediction. In this case, it is probably the most recent close. If you changed your prediction to include things like fundamental indicators, sentiment indicators, or technical indicators like velocity, acceleration, or Bollinger Band % b you would probably start to get better predictions.
Q. How can I prevent all of the data directories I add from being put in the Miscellaneous Category?
A. When you add a directory into the NeuroShell Trader it is automatically added into the Miscellaneous Category. You can change the category for any directory by choosing Data Sources from the Tools menu, selecting the Directories/Files tab, highlighting the category, and selecting the Change Category button. This brings up a list of all the possible category names that you can use.
Q. Why are there gaps in my prediction or custom indicator where nothing is plotted?
A. Gaps or blank spaces in your prediction or indicator are caused by the NeuroShell Trader not getting all of the inputs it needs to calculate the prediction or indicator. For example, if you have a neural net that makes a prediction on the S&P and uses the Treasury Bonds as an input, a prediction will not be generated on a day when there is no value for the Treasury Bonds. The same is true for an indicator – if your indicator formula uses an input that either returns a null value or is blank for that day, the NeuroShell Trader will plot a blank space on your chart.
Q. What are continuous futures (also called continuous contracts)?
A. Continuous futures are data series that consist of individual futures contracts that are “glued” together. The method that they are glued together by depends on the data source that you are using.
A. Return on Account is the percentage return over a period of time on your money invested. Another way of saying this is that it is (net profit)/(account size required). Return on Account is very simple and straightforward when considering only one trade, but it can get more complicated if there are several trades at different price levels, as we will see in some examples that follow.
Return on Trades is the cumulative sum of the percentage returns of individual trades. If you make three trades, during which you made 5%, 7%, and 10%, then your Return on Trades is just 5% + 7% + 10% = 22%.
It can be noted here that the percentages 5%, 7%, and 10% above are actually the percentages you’d get if you computed Return on Account separately for each individual trade. However, the Return on Account for the whole series of trades will not be 22% as you will see.
Return on Account is best for comparison with “buy and hold”, which is just the percent change in price. If you compare with Return on Trades you may erroneously conclude that you aren’t beating buy and hold when you may be. That is because return on Trades looks much worse on a trending stock.
Some examples will make the differences between these figures more clear. For purposes of clarity, we will assume no commissions and no slippage in these examples.
Example 1: You buy 100 shares of a stock at 50 and sell them at 60.
Return on Account:
The required account size is 100 shares * $50 = $5000.
The net profit is = 100 * (60 – 50) = $1000.
Therefore Return on Account = 1000/5000 = .2 or 20%
Return on Trades:
The sum of the trade percentages = 20%
Therefore Return on Trades = 20%
Example 2: You buy 100 shares of a stock at 50, sell them at 80, repurchase them at 80, and then sell them again at 90.
Return on Account:
The required account size is 100 shares * $50 = $5000
The net profit is = 100 * (80 – 50) + 100 * (90 – 80) = 3000 + 1000 = $4000.
Therefore Return on Account = 4000/5000 = .8 or 80%
Return on Trades:
We compute the percentages individually:
Trade 1 percentage is 100 * (80 – 50)/(100*50) = 3000/5000 = .6 or 60%.
Trade 2 percentage is 100 * (90 – 80)/(100*80) = 1000/8000 = .125 or 12.5%. Note that the required account size is larger for trade 2, since it cost more to get into the trade.
Therefore Return on Trades = 60% + 12.5% = 72.5%
Example 3: You buy 100 shares of a stock at 50, sell them at 80, repurchase them at 85, and then sell them again at 90.
Note that this is the same as case 2, except that we repurchased at $85 instead of the $80 at which we last sold the shares. This means that we had to come up with an additional 100*5 = $500 to make the second trade. Therefore, the required account size is now $5000 + $500 = $5500.
Return on Account:
The required account size is $5500
The net profit is = 100 * (80 – 50) + 100 * (90 – 85) = 3000 + 500 = $3500.
Therefore Return on Account = 3500/5500 = .636 or 63.6%
Return on Trades:
We again compute the percentages individually:
Trade 1 percentage is 100 * (80 – 50)/(100*50) = 3000/5000 = .6 or 60% as before.
Trade 2 percentage is 100 * (90 – 85)/(100*85) = 500/8500 = .059 or 5.9%. Note that the required account size is larger for trade 2, since it cost more to get into the trade.
Therefore Return on Trades = 60% + 5.9% = 65.9%
Example 4: You buy 100 shares of a stock at 50, sell them for a loss at 45, repurchase them again after they have risen to 80, and then sell them again at 90.
Note that this is similar to case 3 because we had to come up with an additional $80 – $45 = $35 per share for a total of $3500 to make the second trade. Therefore, the required account size is now $5000 + $3500 = $8500.
Return on Account:
The required account size is $8500
The net profit is = 100 * (45 – 50) + 100 * (90 – 80) = -500 + 1000 = $500.
Therefore Return on Account = 500/8500 = .0588 or 5.88%
Return on Trades:
We again compute the percentages individually:
Trade 1 percentage is 100 * (45 – 50)/(100*50) = -500/5000 = -.1 or -10%.
Trade 2 percentage is 100 * (90 – 80)/(100*80) = 1000/8000 = .125 or 12.5%.
Therefore Return on Trades = -10% + 12.5% = 2.5%