A Proposal of Stock Price Predictor Using Associated Memory

The novel method [1] we propose for predicting stock prices is a case-based reasoning predictor based on associative stock price data memory using Self-Organizing and IncrementalNeural Networks (SOINN) [2]. When a user inputs stock price data, the predictor outputs the most likely prediction based on statistically summarizing similar stock price pattern. It also outputs all cases included in the prediction. Our method has following advantages: (a) our predictor gives users grounds by giving all cases consisting of the prediction using associative memory. Users thereby recognize and are ready for prediction risk. (b) Our predictor avoids large prediction failures because it modifies itself through online learning and continues to learn without its learning parameters being reassigned. This makes it much safer where investment loss may be large. (c) Our predictor is as profitable as previous work while realizing unique, useful functions, as shown by experimental results using actual stock price data from the US and Japan markets between 1998 and 2005.