Using neural networks to predict abnormal returns of quarterly earnings

Artificial neural networks are used in conjunction with the Sharpe-Linter form of the capital asset pricing method to predict when the returns on US stocks will be greater than financial risk models would predict. The advantage of using a nonlinear approach is to model the financial system more accurately than linear techniques. The Sharpe-Lintner form is used to control risk and determine abnormal returns of stocks. Inputs include ratios of recent to past stock price averages over pre-event time periods, similarly, stock volume ratios, and previous quarter standardized unexpected earnings. The earnings data is quarterly and runs from the first quarter of 1993 to the second quarter of 1998. Event periods that had the smallest width around the earnings report tended to be easier to predict abnormal returns. In addition, event periods that were closest to the event (the earnings report) were more accurate at predicting the abnormal returns of stocks.