In this issue:
I. We still need your vote in the Stocks & Commodities Readers Choice Awards
II. Editorial Commentary from Steve Ward, CEO
III. Pairs trading
IV. A better Quote.com Server Selector
V. NeuroShell Trader Companion Products
VI. One way to stop this newsletter
*******************************************************
I. We still need your vote in the Stocks & Commodities Readers Choice Awards
The November issue of Technical Analysis of Stocks and Commodities magazine was the issue that opened the voting for the Readers Choice Awards. We’re asking for your vote and would like to say that it’s for your benefit as well as ours. Prospective buyers use these ratings to make their decisions. When they purchase NeuroShell Trader because of your vote, that new revenue means FREE tech support for you and updated software. So please vote now. Here’s how:
Get your copy of Stocks & Commodities magazine and visit www.Traders.com. Look in the upper portion of the web site for the 2006 Readers’ Choice Awards and click on the Vote Now! link.
On the page that is displayed enter your subscriber ID from your magazine address label (look at their example to see which numbers on the label are the subscriber ID). After you are logged in, look for the Artificial Intelligence Software (Expert, Neural) category on page 3. Be sure to mark your ballot for NeuroShell Trader Professional (Ward Systems Group Inc.). While you are there you can also vote for our GeneHunter product as your second choice (after all GeneHunter is the optimizer in NeuroShell Trader Pro.)
Now please go to page 3 and look for the Standalone Analytical Software, in the $1000 and more category. Look for NeuroShell Trader there and vote.
Vote carefully. Once you’ve voted in a category and then return to that page, you’re not allowed to change your vote.
You have until December 31, 2005 to vote, but please do it now so you don’t forget. You can also win a prize. Contest rules state that 10 random voters will win “The Professional Traders’ Starter Kit”, a $495 value. Voting results will appear in the Bonus Issue, which is mailed in February 2006.
If you have any trouble voting, our technical support department will even help you with that!
*******************************************************
II. Editorial Commentary from Steve Ward, CEO
Many of the recipients of this newsletter are not NeuroShell Trader users, having purchased one our general purpose neural network or genetic algorithm packages in the past. Those users no doubt are investing their nest eggs in mutual funds, long term stock portfolios, or real estate. A few of our NeuroShell Trader users are also more interested in long term investment than trading. This editorial is for both groups, and it is about long term investing.
This weekend I read a very good article in the Baltimore Sun by Will Deener, who writes for the Dallas Morning News. Pamela Yip also contributed to the article. The title caught my eye right away: Forget promises of 10%; Experts advise investors to lower expectations on stock returns. Now this wasn’t one of those doom and gloom forecasts that anyone could make and not be proven right or wrong for many years. It had some pretty convincing “proof”, if you will, that long term investors are likely to see no more than 6 to 7% over the next 10, 20, 30 years. I’m not going to bore you with the details because I’m going to give you the link so you can read it yourself, but here’s one of the final assertions:
“If investors believe they are going to get 10 percent returns, they have to believe that we are going to have another bubble, … and if history is any guide it may be another 70 years before we see another bubble.”
Their conclusion is that unless you have some can’t miss strategy to beat the market, that you had better SAVE more for retirement. Now I’m biased of course, but I have been selling the benefits of trading over investing for years now, and of course we sell tools to help you build strategies to beat the market (I won’t go so far as to insinuate your strategies will be can’t miss).
Of course that bubble the final assertion is talking about is a stock market bubble. I personally got caught in a real estate bubble a number of years ago, and I can tell you that wasn’t pretty either. I started buying single family properties in Dallas – Ft. Worth around 1979. The prices were low and the market was getting hot. After a few years, something quite unexpected happened – the market got over built and the savings and loan crisis hit, and hit Texas especially hard. I still don’t know where things went wrong, but my losses are too painful to try to remember. If you are too invested in real estate, remember there can be both local and national bubbles.
Anyway, here’s the Baltimore Sun link. Read it soon because I don’t know how long the paper keeps these articles around (they may make you register, but the wisdom there is worth it):
http://www.baltimoresun.com/business/investing/bal-ym.stockreturns13nov13,1,1834296.story?coll=bal-investing-headlines&ctrack=1&cset=true
*******************************************************
III. How to Build Pairs Hedging Trading Strategies
Sometimes our users ask us how to build a chart wherein one stock B is sold short when another stock A is purchased, and vice versa. The NeuroShell Trader is designed so that each subchart of a chart deals with buying and selling the “primary” stock in that subchart. Of course, you can have other stocks in that subchart (as “Other instrument data”), but buying and selling only takes place on the primary issue.
A simple method
Therefore, the easiest way to do this hedging is semi-manual – just put one stock (A) in the chart. When you get a buy signal for A, tell your broker to sell B at the same time. When you get a sell signal for A, you know to buy B too. How hard can that be? You can even optimize the chart for A and your scheme still works.
If you think you might forget about B when you get signals for A, there’s even a way to remind yourself. You could build alerts for buying and selling A, and in the text of those alerts, you remind yourself to sell and buy B!
A more automatic method
However, with a little more work you can build a trading strategy that hedges automatically, and the hedging strategy can even be optimized as a hedge instead of optimizing one stock independently.
To understand our hedging method, you may want to download our four example charts from www.ward.net. Note that these examples have data saved in them.
The first example is called Pairs.cht. The example assumes that you believe that Hormel (HRL) usually moves in a direction opposite of McDonalds (MCD). (That probably isn’t the case, but this is just an example). So in this example, the two stocks to be hedged are each on a different chart page.
The stocks are being hedged based upon the RSI indicator. MCD is purchased when its RSI is less than 40. At such time Hormel is shorted. When the MCD RSI > 60, then the previous MCD position is reversed to short, and so Hormel is reversed to long. At any given time, the trading strategy is always in the market long one stock and short the other. The strategy is based on rules about MCD only, but that doesn’t have to be the case.
The key to hedging two stocks inside a single trading strategy is letting the trading strategy know which stock it is working on. To accomplish this all we have to do is compare the sum of the last 10 closes to the sum of the last 10 MCD closes. If they are the same, we are working on MCD. If they aren’t the same, we are working on HRL. Note that we take advantage of loading MCD as other instrument data.
Optimizing the automatic method
Now let’s talk about optimizing a pairs chart. For this discussion we’ll use pairs optimize.cht, the same chart but after some optimization was accomplished.
The goal is to perform an optimization such that the optimized model maintains the pairs hedge. In other words, the optimizer returns a model where HRL is short when MCD is long, and vice versa. To do this we have to keep the rules exactly opposite each other. That in turn means that:
1. The lookback periods for the long rules must match lookbacks for the short rules
2. The rules for MCD have to be the same rules for HRL
Number 1 above is accomplished by creating an association link called “lookback” that all the lookback parameters are associated with, so that optimization keeps them the same. The RSI thresholds have to be the same too, so we created links called “Micky short” and “Micky long”, letting them vary from 50 to 80 and 20 to 50 respectively.
Number 2 above is accomplished simply by selecting “Optimize across all chart pages” on the Optimization tab. That assures that whatever rules we use for MCD are the same rules we use for HRL.
We actually stopped optimization early because with this data the optimizer tended to make one chart very profitable at the expense of the other. That is not necessarily bad, but more negatively correlated stocks may not exhibit such a tendency under optimization.
Note that the parameter 10 in the Sum indicators is set not to optimize. If you are going to build more than a few pairs charts you may want to save this indicator as a custom indicator with the parameter 10 hidden.
Extending the technique
If you want to use neural nets in pairs trading, you will have to use one of our add-on neural nets in a trading strategy.
Another popular trading strategy involves buying one stock or the other based upon the relative strength between the prices of the two related stocks. When the relative strength indicates that A becomes much lower than usual compared to stock B, it is time to buy A. If B goes lower, we buy B instead. There is no shorting. Pairs relative strength.cht will demonstrate that. It is a chart with Honeywell and IBM. We have created a relative strength indicator of HON to IBM, and decided by inspection that if that strength is < 0.3 we want to buy HON and sell IBM, and if it is >0.5, we want to buy IBM and sell Honeywell.
The optimized version is in Pairs optimize relative strength.cht. As before we used optimization links and optimized across all chart pages. This time to keep the profits strong in both stocks we tried a different objective function.
*******************************************************
IV. A better Quote.com Server Selector
Quote.com maintains a number of servers around the country. The Quote.com software that we have installed in the NeuroShell Trader, Trader Pro, and DayTrader Pro decides which servers will be used by the Trader. Unfortunately, the Quote.com servers occasionally stop communicating for unknown reasons, possibly because they crash, or possibly because Internet links in the path become saturated, we don’t know. However, we do know that the Quote.com software they supplied does not always switch to a working server, making it impossible to connect sometimes.
To alleviate that kind of problem we built a utility called the Quote.com Server Selector, which we install in the NeuroShell Trader folder. However, after a while newer versions of Windows 2000 and XP caused the utility to stop working.
After Hurricane Wilma, the Miami servers had a problem and for the next few days they were missing data for Oct 25. In order to get that data you had to make sure you logged onto a Santa Clara server.
So now we have rebuilt and improved the Quote.com Server Selector. You can download it from www.ward.net and use it but we can’t guarantee it will work with all types of firewalls.
Simply store the file you download into the NeuroShell Trader Folder, replacing the old one. Then run QuoteComServerSelector.exe just before you bring up NeuroShell. You’ll probably want to make a shortcut on your desktop for this program. After the program runs, it will have re-arranged the Quote.com file which specifies server order. It puts the servers that are currently responding the fastest higher up in the server list. It also sets a command line that prevents Quote.com from rearranging the list as it normally does. It also has a feature that allows you to move a specific server(s) to the top of the list. (Qcharts itself has features to tell you where the servers are, because they are not always labeled).
You can run QuoteComServerSelector.exe any time you want while NeuroShell is up. However, you will have to bring down NeuroShell and restart it again before any new changes take effect. The reason for this is that the Quote.com software reads its server list only when it is first loaded by NeuroShell.
In order for the Quote.Com Server Selector to work, there must be a copy of the ContinuumClient.ini file in your NeuroShell Trader Folder. This file will only be there if you have previously run Quote.com at least one time in the past.
Go to www.ward.net, Tips and Techniques, to download the new Server Selector. Put it into your NeuroShell Trader folder. Then make a shortcut on your desktop that refers to the exe. Run the exe before you bring up the NeuroShell DayTrader (or Trader Pro if you use Quote.com for end of day data).
Anytime you want to go back to letting Quote.com arrange the servers, just erase the file continuumclient.ini in the NeuroShell folder before you bring up NeuroShell.
This utility will be included in all shipments of release 4.9 and above.
*******************************************************
V. NeuroShell Trader Companion Products
Companion products are different than add-ons in that companion products are stand alone software systems that do not function totally within NeuroShell Trader Series. However, the results can be utilized within NeuroShell Trader Series.
NeuroShell Predictor and Classifier
NeuroShell Predictor and NeuroShell Classifier are stand alone general purpose neural network programs that offer more “hands on” control of the networks and data compared to the Trader’s Prediction Wizard. For modeling experts and other advanced users who want more interactivity with neural nets and training data than the Prediction Wizard supplies, these companions may be just the answer.
The NeuroShell Predictor and Classifier train on data in text file format that the NeuroShell Trader series and other programs can export. The data can be manipulated in a spreadsheet like EXCEL* in a variety of ways before a neural network is trained. Once trained, that network can be linked into NeuroShell Trader series so that it “fires” (executes) and produces a signal on every new bar, even intraday in the NeuroShell DayTrader Professional. That signal is like any other indicator in NeuroShell Trader in that it can be fed into other indicators, used in trading strategies, and even the Prediction Wizard. At that point the completed neural net is completely integrated into NeuroShell Trader and no further manual intervention is required until if or when you want to retrain the network.
Here are some manipulations of training data you might want to accomplish in your own way:
Removing outliers from the training set
Creating your own buy/sell points by visual inspection
Creating your own inputs/outputs or gathering them from other programs
Preprocessing of inputs/outputs that you are more comfortable handling in EXCEL than in the Indicator Wizard
Randomization or other “shuffling” on the training data
Clustering or de-clustering of the training set
Introduction of noise in the training set
Balancing unbalanced training data
There are other reasons for using these companion neural network programs:
NeuroShell Classifier allows categorizations like “Buy”, “Sell”, and “Hold” rather than price forecasting
One training method has a “one hold out” or “leave one out” during training to reduce overfitting
One training method functions like a “nearest neighbor” classifier or predictor
One training method allows real time tightening or relaxing of nets after training
One Classifier training method can be weighted to reduce false positives or false negatives
NeuroShell Classifier provides a probability (confidence) that its classifications are correct
Predictor and Classifier nets can be fired from EXCEL or computer programs
Predictor and Classifier nets can be fired from TradeStation* with a special interface program
One training method weighs inputs by genetic algorithm
Other factors to consider:
NeuroShell Trader Professional versions are required to build the interface that fires nets
Companion products require more manual effort to train nets than does the Prediction Wizard
The NeuroShell Trader Professional optimizer cannot be utilized during training
The number of inputs is limited to 150
The number of training patterns is limited to 16,000
One training method trains and fires quickly, but the other can be very slow with a large training set
NeuroShell Predictor and Classifier rely on EXCEL or NeuroShell Trader for data manipulation
NeuroShell RunTime Server is required to fire networks
The TradeStation Interface requires use and knowledge of Easy Language*
Pricing
There is special bundle pricing for those owners of NeuroShell Trader Professional or NeuroShell DayTrader Professional who have been owners for over 30 days (add shipping costs for all bundles, or subtract $75 for Internet download. TradeStation interface is ONLY available by download, so do not subtract $75 from that):
NeuroShell Predictor + NeuroShell RunTime Server: $545
NeuroShell Classifier + NeuroShell RunTime Server: $545
NeuroShell Predictor + NeuroShell Classifier + NeuroShell RunTime Server: $845
TradeStation Interface for NeuroShell Predictor and NeuroShell Classifier: $249
TradeStation Interface if you do not own NeuroShell Trader: $299
Visit these sites for more information on NeuroShell Predictor and NeuroShell Classifier
* TradeStation and Easy Language are trademarks of TradeStation Securities. EXCEL is a trademark of Microsoft Corp.
*******************************************************
VI. One way to stop this newsletter from coming
It is really easy. Just change your email address and don’t tell us.
*******************************************************