February 2000 Newsletter

I. NeuroShell Trader Professional User Places Second in National Contest

The letter and press release below speak for themselves. Those of you who read our “Real Traders” interviews on www.neuroshell.com will recognize Shashi’s name.

January 31, 2000

Mr. Steve Ward, President
Ward Systems Group, Inc.
Executive Park West
5 Hillcrest Dr.
Frederick, MD 21703

Dear Steve;

Last year I participated in American Skandia’s 2nd annual Internet investment game – MarketMa$ter 2, a national investment contest that involved dynamic asset allocation. There were about 4200 investment professionals competing. I came in second using NeuroShell Trader Professional, even though I started about a month and a half after the other participants! I started off with $10,000.00 at the beginning of the investment period and finished with $23013.31.

I made about 130% in less than 6 months. Thanks to your product I have had somewhat similar returns in my personal account.

Best of all I am receiving a check for $13,013.31 as my award from this contest. Attached is the press release that was sent to me by American Skandia.

Sincerely,
Shashi Mehrotra

FOR IMMEDIATE RELEASE

LOCAL FINANCIAL PLANNER CAPTURES TOP PRIZE IN AMERICAN SKANDIA WEB INVESTMENT GAME

SHELTON, Connecticut – Local Financial Planner – Shashi Mehrotra of Legend Advisors Corp has achieved top ranking in American Skandia’s 2nd annual Internet investment game – MarketMa$ter 2.

The game, which tested the investing skills of Financial Professionals, gave each participant a fictitious $10,000 to invest. From July 1, 1999 – December 31, 1999, players were allowed to buy, hold and sell any of the 46 sub-accounts found in the ASAPsmII variable annuities over the six-month period. This year’s game drew more than 4,200 participants with more than 734 of them earning in excess of 50% return on their mock portfolios.

“This MarketMa$ter game is about keeping your eye on the markets and positioning your money in the most advantageous investment options. It was designed to parallel what all good financial planners do on a regular basis – manage money, ” said Chief Operating Officer Wade Dokken. “We are pleased to recognize Shashi for achieving 2nd place in American Skandia’s latest offering of this challenging game.”

The top five winners succeeded in earning over 100% return in their respective portfolios by strategically investing in a variety of leading investment options found in our flagship variable annuity. “This game is about heightened awareness of our growing investment options as well as fun. Next year, we look forward to delivering a new enhanced MarketMa$ter 2000 as well as possibly a collegiate MarketMa$ter,” said Christian Thwaites, senior vice president of marketing.

MarketMa$ter is sponsored by ASIST American Skandia Information Services and Technology Corporation.

II. Doctorial Thesis Compares NeuroShell to Logistic Regression

Dr. Gordon S. Doig has produced an outstanding and comprehensive study comparing a very popular statistical method and our neural networks. His study completed in April 1999, Severity of Illness Scoring in the Intensive Care Unit: A Comparison of Logistic Regression and Artificial Neural Networks, was Dr. Doig’s doctoral thesis at the University of Western Ontario, London, Ontario, Canada. Dr. Doig used genetic adaptive and backprop nets to make ICU outcome predictions. Rarely have we seen such a complete work. Dr. Doig is now at the London Health Sciences Centre, and we have reproduced the abstract of his work below.

Abstract

Purpose: To compare the predictive performance of a series of logistic regression models (LMs) to a corresponding series of back-propagation artificial neural networks (ANNs).

Location: A 30 bed adult general intensive care unit (ICU) that serves a 600-bed tertiary care teaching hospital.

Patients: Consecutive patients with a duration of ICU stay greater than 72 hours.

Outcome: ICU-based mortality.

Methods: Data were collected on day one and day three of stay using a modified APACHE III methodology. A randomly generated 811 patient developmental database was used to build models using day one data (LM1 and ANN1), day three data (LM2 and ANN2) and a combination of day one and day three data (LMOT and ANNOT). Primary comparisons were based on area under the receiver operating curves (aROC) as measured on a 338 patient validation database. Outcome predictions were also obtained from experienced ICU clinicians on a subset of patients.

Results: Of the 3,728 patients admitted to the ICU during the period from March 1, 1994 through February 28, 1996, 1,181 qualified for entry into the study. There was no significant difference between LM and ANN models developed using day one data. The ANN developed using day three data performed significantly better than the corresponding LM (aROC LM2 0.7158 vs. ANN2 0.7845, p=0.0355). The time dependent ANN model also performed significantly better than the corresponding LM (aROC LMOT 0.7342 vs. ANNOT 0.8095, p=0.0140).

The predictions obtained from ICU consultants (aROC 0.8210) discriminated significantly better than LMOT (aROC 0.6814, p=0.0015) but there was no difference between the consultants and ANNOT (aROC 0.8094, p=0.7684).

Conclusion: Although the 1,181 patients who became eligible for entry into this study represented only 32 percent of all ICU admissions, they accounted for 80 percent of the resources (costs) expended. ANNs demonstrated significantly better predictive performance in this clinically important group of patients. Four potential reasons are discussed: 1) ANNs are insensitive to problems associated with multicollinearity; 2) ANNs place importance on novel predictors; 3) ANNs automatically model nonlinear relationships and ; 4) ANNs implicitly detect all possible interaction terms.

Keywords: intensive care, critical care, severity-of-illness, logistic regression, artificial neural networks, genetic algorithms, back-propagation, receiver operating characteristic, predictive model building.

III. NeuroShell Classifier in the Hospital Labor Ward

Dr. Ken Giuffre, Director of Anesthesia Research, and his research team at Hackensack University Medical Center of UMDNJ, have constructed a model that predicts caesarean sections on a busy labor ward. Their paper was presented at the 1998 meeting of the American Society of Anesthesiologists, and full publication is pending their use addition of statistics (ANOVA) and a training set of 1000 patients. Following is the summary of their paper:

A Neural Net/Genetic Algorithm Model to Predict Caesarean Section in a Busy Labor Ward.
Giuffre KA, et.al.
Department of Anesthesiology
Hackensack University Medical Center,
30 Prospect Ave.
Hackensack NJ 07601

Using computer-based artificial neural networks (ANN), we sought to determine if epidural analgesia influences cesarean section (c/s) rates while creating a general predictive model of c/s in a diverse obstetric population.

We used 46 variables taken from 162 charts of consecutive patients admitted to HUMC for labor and delivery in March 1997 for model training. Two additional and separate chart groups of 20 and 38 patients were used to test the model’s predictive value respectively before and after removing patients who had undergone prior c/s. Backpropagation neural networks (BNN, NeuroShell 2, Ward Systems Group, Frederick MD) alone had no predictive value. A separate classification model (Kohonen net, NeuroShell 2, Ward Systems Group) revealed patients with a prior c/s as comprising a different group with a c/s rate of 62% (p<0.001). After removing patients who had undergone prior c/s from the teaching and test sets, and constructing a classification neural net with genetic-based alterations (NeuroShell Classifier, Ward Systems Group). We created a highly predictive model that utilized 9 inputs and correctly classified 32 of 38 (84%) of patients in the test group. Of the 6 missed classifications, 1 was a false negative (specificity 97%) and 5 were false positives (sensitivity 87%). No relationship between epidural analgesia and c/s was seen the final training group from which patients experiencing prior c/s were removed.We conclude that: 1) ANN modeling of patient data is more effective when utilizing a system that employs a genetic algorithm to fine-tune net architecture. 2) It is possible to train an ANN data-model that generalizes well enough to a test group for predicting c/s. 3) Epidural analgesia in patients with no prior c/s history has no effect on c/s rates.IV. The Capital Growth Services Story

Capital Growth Services is a Lake Tahoe Company that manages individual client pension accounts, IRAs, savings, etc. The company manages a large amount of funds through the efforts of the owner and manager, Mark Behrendsen.

Mark’s primary philosophy is to “limit risk first.” This is done by reducing the amount of time clients are invested in equities. The company does quite well considering its conservative mandate, and thanks in large part to the NeuroShell Trader Professional. Mark’s conservative trading strategy is able to capture about 85% of the potential profit in an up market. “I’m only interested in the ‘gravy’, which is why I limit the amount of time invested in equities,” Mark explains.

Mark used a competitor’s neural net system several years ago without much success. He liked the concept of a neural net being able to weight input factors automatically, but the package he previously purchased was “just too laborious to use.” So, Mark was a little soured on artificial intelligence until an acquaintance bragged about building a successful neural net and offered to sell it to him for $250,000!

Mark figured if his acquaintance could do that, he could too, so he talked to a Ward Systems Group representative and purchased the NeuroShell Trader Professional.

He has designed a model which combines major aspects of the NeuroShell Trader Pro: first, the ability to build neural network trading models and next, incorporate the nets into the Trading Strategy Wizard. Mark, although guarded, was willing to give us some information about the model he has constructed. The model is a neural network trading strategy and generates about 8 trades a year.

Capital Growth Services trades no-load mutual funds long only. Mark’s “secret”, so to speak, is that he makes his predictions using only the Nasdaq composite, but then makes the trades using growth mutual funds.

“My neural net inputs are exponential moving averages and Wilder’s indicators; just basic stuff,” says Mark. “I also have some price momentum, but I use no volume indicators. I predict the close one or two days out, using a 3 year training set with 2 six month walk forward tests.”

Mark does optimize his inputs using Return on Account as the objective function. Interestingly, he uses the full 80 hidden neurons during optimization which we normally recommend against because it takes too long and increases the danger of over-fitting. Mark has considered that, however, and he takes care of those two problems by severely limiting the range of the variables during optimization.

“You guys should be asking $5,000 for the Trader Professional,” he laughingly told our interviewer. “It is one of the most powerful pieces of trading software ever developed. A person will never be able to try all the features, because they’ll be making a ton of money before they even get close.”

Editors note: This article is now in the Real Traders section of www.neuroshell.com, including a picture of Mark with his Dodge Viper.

V. Zinc Coating Life Prediction

Galvanizing, which produces a zinc coating on steel surface, is one of the most effective methods for corrosion protection of steel. Like other metals and alloys, galvanized zinc coatings corrode at certain rates depending on the environmental conditions. The corrosion rates of zinc coated steels in atmospheric environments have a wide range, about two orders of magnitude. It is important to know the specific corrosion rate in a given application environment in order to effectively use zinc coated steels in outdoor structures.

Traditionally, a common method for estimation of the life of galvanized steels has been the use of a generalized value for the different types of atmospheres known as rural, industrial, urban and marine. However, this non-specific approach is no longer adequate to meet the demands of the market place. Today, the users of galvanized steels are increasingly asking for information on performance certainty, that is, on life prediction. Also, as products are becoming more application-specific, more relevant information on corrosion rates is required, which calls for more accurate prediction methods.

A new prediction method was developed using statistical methods, neural network technology and an extensive database by Dr. X.G. Zhang at Cominco Product Technology Centre. The neural network models were developed using commercial software, NeuroShell 2, by Ward Systems Group, Inc. The models take the basic and readily available weather and air pollution information as inputs and produces corrosion rates as output. The method is now being utilized to develop Internet based prediction software.

VI. Clarification of NeuroShell DayTrader Issues

In the January newsletter, we announced the NeuroShell DayTrader Professional. If you missed that newsletter, you can read the article on either www.neuroshell.com, or if you are a Trader user, on www.ward.net. It may be worth visiting just to see Granny in her beret!

Since the DayTrader announcement we have had several emails from people upset because we did not interface with the TradeStation 2000 Global Server.

The Global Server was a top priority, and still is. However, when we tried to interface with it, we had lots of problems and crashes. We don’t know if the problem was the Global Server or us, but we DID try very hard. We will still be trying, but for obvious reasons, we will make no promises on when or even if.

Rest assured we didn’t take the right turn and introduce the DayTrader sans Omega by design. We did it as a favor, so everyone could get started using the DayTrader right away. The attendees at our last class who saw the DayTrader said paying the $79+exchanges fees for Quote.com was a small price to pay compared to what can be accomplished with the DayTrader and daytrading. They felt they could just cancel the Quote.com subscription if and when we can support whatever feed they currently use. The overwhelming feeling, even among those with TradeStation, was “get it out!”

We also heard from some of our users who were concerned that we didn’t interface THEIR favorite data feed. If you want to wait until or if we can interface with your favorite data feed, you may, but if you really want to use the DayTrader, you can get started easily. If you use the internet already, you are ready to go. The DayTrader takes care of all connections. We use the DayTrader every day, and some of us have just a 56k modem. You shouldn’t have to abandon your current data feed, and there should be no conflict with TS2000. The ability to use NeuroShell in real time should be well worth the relatively minor cost of another data feed!

Join the revolution!

Was this article helpful?

Related Articles