Statistical learning theory in equity return forecasting

Abstract

We apply Mangasarian and Bennett's multi-surface method to the problem of allocating financial capital to individual stocks. The strategy constructs market neutral portfolios wherein capital exposure to long positions equals exposure to short positions at the beginning of each weekly period. The optimization model generates excess returns above the S&P 500, even in the presence of reasonable transaction costs. The trading strategy generates statistical arbitrage for trading costs below 10 basis points per transaction. © 2005 Springer Science+Business Media, Inc.