Time Series and Neural Network Analysis

This chapter discusses and compares the performances of the traditional time-series models and the neural network (NN) model to see which one does a better job of predicting changes in stock prices and to identify critical predictors in forecasting stock prices in order to increase forecasting accuracy for professionals in the market. Time-series analysis is somewhat parallel to technical analysis, but it differs from the latter by using different statistical methods and models to analyze historical stock prices and predict the future prices. Neural network approaches can make important contributions since they can incorporate very large number of variables and observations into their models. In this study, the authors apply the traditional time-series decomposition (TSD), Holt/Winters (H/W) models, Box–Jenkins (B/J) methodology, and neural network (NN) model to 50 randomly selected stocks from September 1, 1998 to December 31, 2010 with a total of 3105 observations for each company’s close stock price. This sample period covers high tech boom and bust, the historical 9/11 event, housing boom and bust, and the recent serious recession and current slow recovery. During this exceptionally uncertain period of global economic and financial crises, it is expected that stock prices are extremely difficult to predict.