Indian Journal of Science and Technology

Abstract

Accurately predicting stock prices has been long one of the desires of investors in the financial markets. As such, various methods and rationales have been proposed for predicting corporate stock prices. One of the most widely used analytical methods to predict time series is technical analysis in which analysts assess indicators and models to find appropriate strategies and gain profits. Currently, a large number of technical indicators are available but it is not possible to use all of them. Therefore, selecting appropriate indicators for predicting requires great skills. One of such indicators is the simple moving average that can be defined for different time periods. The aim of this study is to find out the most appropriate moving average by assessing the simple moving averages for periods of 1 to 200 days using a feature selection algorithm called Correlation-Based Filter. To this end, data from petrochemical companies listin Tehran Stock Exchange for the period 2009-2015 are analyzed. The results show that five 5, 25, 48, 50 and 89 day simple moving averages are more appropriate to predict stock prices in the Tehran Stock Exchange. To test this hypothesis, we utilize Multilayer Perceptron Artificial Neural Network as a predictor method, and on the one hand, stock prices are predicted using 5 selected indicators, and on the other hand, stock prices are predicted using all 200 indicators. Our results also suggest that using the five selective simple moving averages are more accurate for predicting stock prices.
Keywords: Artificial Neural Network, Correlation-Based Filter, Simple Moving Averages, Technical Analysis