Two researchers have developed a methodology for optimizing a portfolio that avoids the current approach of making precise estimates of expected returns for each stock. Rather, the new approach involves predicting which is the best-performing stock, the next best-performing stock, and so on, but without assigning specific returns.
By ordering the stocks, Neil Chriss, a managing director of hedge fund firm SAC Capital Management Inc., New York, and Robert Almgren, associate professor of mathematics and computer science at the University of Toronto, think they can produce better results than with the traditional mean-variance optimizer.
Specifically, their simulations of portfolios ranging between 25 and 500 stocks between Jan. 19, 1990, and Dec. 31, 2002, came up with better information ratios than portfolios generated by the traditional optimizer. An information ratio is a measure of how much in returns a portfolio produces per unit of risk.
Mr. Chriss presented the new model at the Chicago Quantitative Alliance's annual spring conference in Las Vegas last month. The presentation was based on a recent paper he and Mr. Almgren wrote, titled "Optimal Portfolios From Ordering Information." The paper has yet to be published.
Messrs. Chriss and Almgren are also largely credited for pioneering much of the research on algorithmic block trading in a 1998 paper, "Optimal Execution of Portfolio Transactions."