Sign up today and take advantage of member-only content — the kind of timely, cutting edge industry insight that only Money Management Executive can deliver.
  • Exclusive Online Only Content
  • Free Daily Email News Alerts
  • Asset Management Blogs

Fund Firm Turns to Artificial Intelligence: Claims Computer Model Exceeds Quantitative Methods

Registered investment advisory firm Legend Advisory Corp. of Palm Beach Gardens, Fla., believes it has found the secret to successful investing through the use of predictive asset allocation. The firm's executives claim to be the first to apply so-called artificial intelligence to broad global asset classes to allow the firm to predict, four months in advance, which of seven classes will show the strongest performance.

Legend, with $1.3 billion in assets under management, is a subsidiary of Waddell & Reed of Overland Park, Kan.

The firm has spent $10 million and more than 15 years developing and refining its system, which is now being applied to five model mutual fund portfolios offered to the firm's clients, many of which are participants in the 403(b) plans of not-for-profit organizations. The computer system responsible for digesting the many thousands of data bits each month is dubbed AANN, the acronym for Asset Allocation Neural Network. Legend said it "can think and learn like humans but at lightening speed and with the ability to see dimensions beyond that of the human brain or traditional computer models."

Global Sector Shifts

AANN is used across seven asset classes: large- and small-cap domestic stocks, international equity, international fixed income, domestic investment grade and high-yield bonds, and cash. AANN's specific sector recommendations are then fed into the firm's separate optimization model, which creates the specific asset class weightings appropriate for each of the model portfolios' risk profile. The portfolios range from ultra conservative to aggressive.

According to Shashi Mehrotra, chief investment officer of Legend, unlike typical quantitative investment models, AANN utilizes both fundamental and technical analysis and processes a slew of complex data, including unit labor costs, productivity figures, global yield spreads, bond market default premiums and the moving averages of gold when compared to equities and currencies. Every month, AANN analyzes 32 customized inputs, each of which is compressed from huge volumes of financial data.

"The dimensions that AANN can see and detect in financial markets are beyond human comprehension," Mehrotra said.

Also unlike traditional rules-based quant models that digest data according to a set of pre-established rules and set parameters for each possible scenario encountered, AANN is a free-range computer that is allowed to draw its own conclusions by analyzing many variables, although the firm does impose certain risk boundaries.

"We limited it early on and made mistakes," Mehrotra admitted. "Now we follow what she tells us to do as far as allocations."

Legend also back-tested the computer model over the past 10 years, feeding it data it would have scoured, and found that it would have delivered a 7.73% average annual return.

Neural networks, which sift through voluminous data in record time and use interconnected processes to recognize subtle patterns that might otherwise be imperceptible to mere mortals, are winding their way further into the investment management industry. The goal? To help investment firms predict which asset classes are ready to outperform or which stocks are perfectly ripe for picking or selling.

These neural networks parallel the human brain in that their interconnected processes transmit data over pathways similar to how a cluster of human neurons digests information and then "learns" how to anticipate future events. Hence, these computer-modeling networks are often referred to as artificial intelligence.

"Artificial intelligence looks at patterns and notices that interest rates aren't working in tandem this time, or that the impact of inflation is different," Mehrotra explained. "This is how humans learn."

Experience has shown that AANN will typically decide it is time to reallocate approximately 2-1/2 times per year, Mehrotra said. "We let her decide when to move. There's no calendar or seasonal factors. That definitely distinguishes us," he added. "I think we've created a new frontier in the investment management industry." Legend is now working to leverage AANN's predictive reasoning within other specific sectors such as emerging markets.

Hot on Newton'

Although Legend believes it is the first firm to apply artificial intelligence to broad global asset classes, it isn't the first investment management firm to use artificial intelligence. American Century Investments of Kansas City, Mo., utilizes a proprietarily developed neural network in its almost four-year-old Newton Fund. The diminutive small-cap fund, with only $6 million in assets, applies artificial intelligence modeling to help capture and interpret investor behavior. This allows for the portfolio manager to identify companies whose price patterns suggest investors will continue purchasing them, signaling candidates for purchase, and those whose price patterns suggest a decline, signaling those companies to sell. The artificial intelligence process uses technical analysis and is designed to quickly respond to market changes.

The idea emerged when James Stowers III, the son of American Century's founder, was interested in coming up with a way to use technical analysis only to manage a fund, not just to pick a single stock or two, said John Small, vice president and portfolio manager of the Newton Fund. In a nutshell, tens of thousands of data points are input and thousands of companies are ranked on a scale from negative five to positive five, with higher-ranking stocks considered for investment, he explained.

Over the last 18 months, the firm has layered on its own complementary optimization modeling techniques that take the next step after identifying which stocks to buy. The optimization model determines the best possible price at which each portfolio nominee stock should be purchased, Small said.

(c) 2007 Money Management Executive and SourceMedia, Inc. All Rights Reserved.