Abstract:

Stock prediction is an important issue in finance. LAMSTAR is a system of networks for storage, recognition, comparison and decision. This paper attem pts to explore the LAMb＇TAR network application in short-term stock market prediction. For each prediction, the stock features extracted from the histo  orical data are fed to the LAMSTAR network, in which the multi-dimensional non-linear connections between the features are detected and encoded in link weights. If the stock price is predicted to go up in the following trading day, LAMSTAR will send out a buy signal to initiate a transaction. Three expe rimental results with exciting returns of different companies are presented to validate the efficiency of this approach.

Key words: neural network, LAMSTAR, stock prediction;short term prediction