A measure of uncertainty for stock performance

The prediction of the future returns of individual stocks can be based on the growth rates of several fundamental factors, such as revenues, earnings per share, capital investment, debt, and market share, among others. A stock will be expected to perform well if the growth rates are large, and it will be expected to perform poorly if the growth rates are small or negative. Usually, the growth rates vary, however, and when the growth rates are mixed, that is, some of the rates are large and some are small, the appropriate prediction of stock performance is uncertain. A measure of the extent to which the growth rates are mixed provides a measure of uncertainty in the expected performance of the stock. The paper describes a method to measure the extent to which the growth rates are mixed, based on a fuzzy logic model for combination of evidence. The model was originally developed for a signal-detection task designed to identify situations in which a detection decision cannot be made with reasonable certainty. Identifying when a decision cannot be made is useful in applications such as medical diagnosis, where additional testing can be considered. Standard statistical detection techniques, such as Bayesian maximum likelihood detectors, determine the most probable choice between signal-present and signal-absent for every observed waveform, but they cannot conclude that a decision cannot be made. A signal detector that explicitly characterizes uncertainty can be constructed using fuzzy logic, with membership functions based on the observed values of the waveform features being used for detection