In this issue:
I. Expensive clothes and high tuition – a BIG sale!
II. Hedge fund trader gains edge with neural nets
III. Bug in NeuroShell Trader releases 4.0 and 4.1
IV. TASC September traders’ tips
V. Make sure you continue to get the news
*******************************************************
I. Expensive clothes and high tuition – a BIG sale!
It is that time of year again and our employee’s kids, like all middle class kids these days, are demanding more expensive clothes than ever as they go back to school. Not only that, but many of these kids are in college, and tuition bills this fall are sky high. However, because of the bad economy, our sales have been off somewhat this year, and we haven’t given any bonuses to our embattled employees. You can help by quickly buying more software you’ll need anyway this fall, and we can help by making your purchase easier with a big sale. It’s a win-win situation.
You aren’t going to see a sale like this one from us again, but you’ll have to move quickly! We are offering a whopping 30% off but only for 1 week so they’ll have time for shopping, then 20% off for the following week. The sale discounts apply to any software purchase $200 or over. Absolutely no exceptions, no rainchecks, no waivers, no applying the sale to previous orders, no fooling – so call or email right away. The 30% discount ends after Friday, August 22, and the 20% discount ends after Friday August 29. This sale can’t be combined with any other offers.
*******************************************************
II. Hedge fund trader gains edge with neural nets
Troy Buckner from NuWave Investment Corp. has fine tuned the non-conventional pattern recognition capabilities of the NeuroShell Trader to create a Pattern Recognition Program with positive returns totaling 82.6% since 1996 (including both live and simulated data; live trading began in June of 2001). The Pattern Recognition Program is one of three trading methods used in the company’s Combined Portfolio investment plan.
Buckner, a principal at NuWave, says their Pattern Recognition Program uses the NeuroShell Trader along with the General Regression Neural Nets (GRNN) or prediction nets from the Adaptive Net Indicator add-on. Buckner uses a majority decision from three different nets to cover 35 different international stocks, bonds, currency, and commodities markets.
Buckner has been researching the use of neural nets in trading since 1989. The Pattern Recognition Program is based upon the assumption that markets exhibit a degree of repetitive price action that can be identified throughout history. He believes factors responsible for such repetition include fundamental factors such as supply, demand, economic cycles, interest rates, weather, seasonality, etc., and human factors such as fear, greed and other emotions.
GRNN nets are much more adaptive than typical neural nets, because they further “evolve” or “learn” with the pattern in each new bar they see (i.e., they retrain themselves with each new bar). GRNN nets learn the relationship between a set of inputs and a predicted value such as the momentum of the close by composing a formula of this relationship based on historical patterns.
Buckner said the inputs to his models include summaries of price information that a trader would see by looking at a monthly, weekly, and daily charts such as long term trends, the slope (rate of change), the volatility of bars, and the range of bars. Buckner describes his inputs as what a trader would see overall by looking at directional and volatility price data.
Buckner extends his analysis by looking at similar variables in different time frames. His GRNN models have a 24 day average holding periods for trades.
For the most part it takes only minutes to run models because Buckner has created a batch processor of sorts for the NeuroShell Trader by using a macro program that records key strokes.
“I think the structure of the NeuroShell Trader is superior in that it offers you options that don’t – overtrain a model, unlike the old backpropagation networks,” according to Buckner. “There you had to worry about selecting training and test sets. The Adaptive Net Indicators make it easy to automate because they automatically retrain every bar according to a moving window of history . I find GRNN to be more effective than backprop nets because GRNN works with in-sample data in a more consistent fashion, that is it doesn’t depend on how your select training and test sets or how many hidden neurons you train with. All I have to be concerned with is whether my historical training window is long enough to contain up, down, and sideways price activity.”
Buckner said there are four things that are essential to a systematic trader:
1. have a consistent approach
2. don’t overfit
3. create a model structure that applies across all markets.
4. allow the history of each market to guide trading in that contract (i.e. use S&P 500 history to generate S&P 500 trading signals, T-note history to guide trading of T-note signals etc.)
He likes the GRNN adaptive nets because there are few moving parts. , “It identifies the personality of each market while maintaining consistency with the modeling approach. You can use the same structure for each market, but allow only the history of a given market to affect trading signals in that market. In other words, each market can be modeled independently of the others but with the same structure. Because I use the same inputs and a constant smoothing factor for all markets, the length of history becomes the only real variable for the model created to trade each market.” Experience has taught him that the length of the training history doesn’t matter as much as diversifying across many trading markets.
“Overall, I see that the Pattern Recognition Program gives us a consistent edge coupled with solid risk control as part of our Combined Portfolio,” according to Buckner.
The Nuwave website is www.nuwavecorp.com
*******************************************************
III. Bug in NeuroShell Trader releases 4.0 and 4.1
For any of you who thought that releases 4.0 and 4.1 weren’t giving as good of results as 3.8, you may be right in some cases. We don’t know how this bug got through alpha and beta testing unnoticed, but two alert users caught the error this week. We are sorry for the inconvenience this has no doubt caused.
The problem occurs with either predictions or trading strategies when you are using the “return on account” objective (which unfortunately is the default). A 100% profit could look like a 10% profit, especially if you are:
1. using “buy as many shares as possible with current account balance” AND your account balance grew greatly or
2. if you are using a fixed number of shares AND your stock price has been rising greatly over the course of the optimization.
The two wrong statistics are “return on account” and “required account size”.
Until we get 4.2 up next week, we suggest that you change your objective function if you have been using “return on account”. Use net profit or something else instead, but don’t look at the return on account statistic. Call us if you need help.
*******************************************************
IV. TASC September Traders’ Tips
The September 2003 issue of Technical Analysis of Stocks and Commodities magazine includes two Traders’ Tips that can be implemented in the NeuroShell Trader. Markos Katsanos’ article on “Detecting Breakouts in Intraday Charts” describes an indicator called the Finite Volume Elements Indicator (FVE). This is a version of a money flow indicator that considers both intra- and interday price action as well as minimal price changes. Viktor Likhovidov’s article on the “Index of Chart Sentiment” describes an indicator for intraday traders that can identify the dominant direction of future market movements and simultaneously detect possible turning points. The charts including these indicators for both of these articles can be downloaded from www.ward.net after we get them up next week, along with charts for previous articles published in TASC. You’ll have to get the text of the articles from the magazine, however, because we aren’t allowed to reprint it.
*******************************************************
V. Make sure you continue to get the news
We were surprised when, after we sent out our June newsletter announcing release 4.0, several of our users called or emailed to ask when release 4.0 would be available. Seems they all changed their email addresses and never notified us of that. So now they’re on the end of our 4.0/4.1 distribution list. There are many others, no doubt, who still don’t know about release 4.0.
We also frequently get calls from NeuroShell 2 users who don’t know about NeuroShell 2 release 4.0, and who don’t know that we have newer, better, faster, more accurate products. So they’re stuck trying to impress their bosses with older technology.
*******************************************************