The Legend Group is a 35-year-old, generally conservative company that invests retirement funds and manages about $850 million in assets using asset allocation mainly in equity, debt, and cash. Legend is a progressive company, always open to new ideas. As early as five years ago, Legend Advisory recognized the potential of neural networks and was willing to make the investment of time and money to follow their instincts that this could be a wave of the future. The Legend Group is now using Shashi’s neural nets to make asset allocation decisions to manage these assets. 

Glenn Ferris, Shashi’s supervisor, is a dynamic, progressive individual. The pattern recognition potential of neural networks appealed to Glenn and he foresaw their usefulness in the investment business. 

Shashi started with the Ward Systems Group’s NeuroShell 2 to begin his work. The first six months proved to a difficult learning experience, but Shashi remained committed to the concept. During this time he began to develop his own technical indicators, his preference being the combination of the Advance/Decline Line (ADL) and New/High New/Low Oscillators. His unique formula for combining these standard technical indicators would later prove very valuable. He built a number of other good indicators using ADL and volume, including many powerful indicators that are very simple products and ratios of indicators. 

The first successful model was an asset allocation model. The Legend Group invests in equities in Europe, Large Cap, and Small Cap US stocks, as well as smaller amounts in emerging markets. Basically, Shashi has built a net for each sector to predict the performance of that sector.
 
 

Year to Date performances on their Pioneer funds Strategic Asset Management (SAM) product for all the five models (in order of aggressiveness) are listed below:
 

Model	YTD Return  (As of 5/31/98)	Long-Term  Equity – Debt Split
Pioneer SAM Conservative Model	9.54%	40% (Equity), 60% (Debt)
Pioneer SAM Moderate Model	14.36%	60% (Equity), 40% (Debt)
Pioneer SAM Aggressive Model	16.62%	80% (Equity), 20% (Debt)
Pioneer SAM Select Global Capital Appreciation	14.92%	90% (Equity), 10% (Debt)
Pioneer SAM Select Global Capital Appreciation	16.51%	100% (Equity), 0% (Debt)
S&P 500	12.41%
Russell 2000	4.48%
DJIA	12.54%
DJ Global – US	11.82%

These are impressive performance records, especially since the company is conservative and does not invest in commodities, options, or futures. Shashi says his betas are from .5 to .85 on the products.

Shashi uses 13 inputs, which he selects from an initial 30 inputs by using the contribution factors he gets from the networks, as well as some correlation studies he uses to avoid closely correlated inputs. He often finds that, once he has a good indicator, the moving average of that indicator does better than the indicator itself. His output is the predicted direction of the market, not the magnitude.

Shashi recently switched to the NeuroShell Trader after attending one of Ward Systems Group’s NeuroShell Trader seminars. He likes the Trader especially because of its ability to automatically and easily perform walk forward testing - something he considers very important. After walk forward testing, Shashi goes into production with the NeuroShell Trader’s "current network", which has been trained with the most recent data. He uses weekly data files and trains on 4 to 5 years of history for his asset allocation models.

"This stuff is pretty exciting – it’s a very powerful tool," Shashi insists. "Easy to learn in the beginning, but it will be a long time before I can use all the power that’s there. Every time I look at it I learn something new."

In addition to the asset allocation models, Shashi is now building market-timing models for several individual stocks and the general market (S&P 500). Shashi has already back-tested a prototype for the S&P 500, which did 22.59% in 1996, 34.03% in 1997, and for the first quarter of 1998 it matched the S&P at 12.41%. Over the time frame of 1/1/87 to 5/31/98 the average annual return is about 19% with his model as opposed to 14% with the buy and hold strategy (Both figures exclude dividends). 

We asked Shashi if he learned any of his techniques from our class. "The class opened a new horizon in my mind; it got me working on new planes. Inputs from instructors and other students were very helpful, since the students came from different backgrounds."

Shashi is now a vice president of the company!

UPDATE  December 10, 1999

Legend Group has now started managing stock purchases for individuals, getting their timing signals from the NeuroShell Trader Professional. Shashi has made models for four stocks and is actively trading two of those. Since April 1, 1999, one of those has produced a return of 95.1%. During the same period, buy and hold was 28.8%. Shashi says he has changed from predicting percent of change in close to percent of change in open, and it has improved his trading significantly.

UPDATE  January 13, 2005

Shashi's company now has over $2 Billion under management, and is still using our neural nets to do asset allocation.

UPDATE  May 27, 2005

Shashi has taken NeuroShell a step farther than the usual market predictions to extracting economic predictions from sector data. For 2004 the customer correctly made 11 out of 12 economic, market and sector predictions. So far in 2005, his new predictions are accurate as well.

He created a synthetic index drawn from different sectors as the basis for the economic forecasts. He modified eight to twelve standard technical indicators for each sector. He used a neural net prediction to predict a four month gain or loss for each sector. He’s been creating the sector models for two and a half years and has used those models to manage money for the Legend Group for one and a half years.

Mehrotra initially presented the results of his predictions at the National Advisors’ Conference in San Antonio in October 2004. At the meeting nine of the predictions were judged correct at the time, but figures from the end of the year validated two additional predictions.

UPDATE  December 4, 2007

Shashi was interviewed for an article that appeared on InvestmentNews.com at http://www.investmentnews.com/apps/pbcs.dll/article?AID=/20071203/FREE/712030324/1017/TOC on December 3, 2007.

The writer described how artificial intelligence "...has been described by some as the second generation of quantitative investing because it has the flexibility to get smarter through the expansion of input data", but that "...when computers manage the portfolio, the first challenge is often preventing the human brain from getting in the way." Shashi has found this to be true, as he is quoted saying, "I used to second-guess her, but I was wrong nine out of 10 times."