Forecasting the nikkiei average: a comparison between a neural network model and regression analysis.

• Author: Heiat, Abbas

INTRODUCTION

Many investors, including individuals and institutions, are interested in forecasting stock prices in financial markets. Forecasting the stock prices, however, is difficult and complicated. This study examines forecasting methods with two different widely-used forecasting techniques: regression models and neural networks. This research should provide individuals and institutions some insight for how to forecast and analyse stock prices in the Japanese markets. Bahmani-Oskooee and Sohrabian (1992) reported about the effects of the exchange rate on stock prices. Their study showed that S&P 500 index and the effective exchange rate influence each other in the short-term but not in the long-term. Abdalla and Murinde examined the relationship between the exchange rate and stock markets in the developing countries. They found that the exchange rate had a strong effect on stock prices. Their findings were consistent with some earlier research on developed countries. Dheeriya studied causality among international stock markets and found that some markets influenced other markets. The results showed positive correlations among the markets. Dheeriya also discovered that all individual markets responded to other markets significantly. Huang reported on the relationship among the U.S., Japan, and the South China stock markets. Huang found that the U.S. stock markets had more influence on the markets of South China than the Japanese markets do. Mclntyre-Bhatty extended the previous research using neural networks in financial market analysis. Compared to regression analysis, the performance of the neural network system was superior in stable, bearish, and bullish markets. Oh and Han studied the accuracy of the neural networks for interest rate prediction by using change-point detection. They concluded that neural network models are more reliable than other models in forecasting interest rates.

Eakins compared regression models and neural network models by examining the relationship between the institutional ownership percentages in common stock and financial ratios. Eakins found that neural networks were more accurate than the regression model. In contrast to Eakins' work, Below reported that linear regression models are superior to neural network models. He examined the determinants of institutional investment demand and common stock, as well as the relationship between institutional investment decisions and financial ratios.

DATA

The data...