A hybrid statistical approach for stock market forecasting based on Artificial Neural Network and ARIMA time series models

The time series analysis and forecasting is an essential tool which can be widely applied for identifying the meaningful characteristics for making future ad-judgements; especially making decisions in finance under the numerous type of economic policies and reforms have been regarding as the one of the biggest challenge in the modern economy today. High volatile fluctuations in instability patterns are common phenomena in the Colombo Stock Exchange (CSE), Sri Lanka since after the introduced open economy policies in 1978. As a subset of the literature, very few studies have been focused on CSE to find out the new forecasting approaches for forecasting stock price indices under the high volatility. As a result, main purpose of this study is to take an attempt to understand the behavioral patterns as well as seek to develop a new hybrid forecasting approach based on ARIMA-ANN for estimating price indices in CSE. The model selection criterion results in Akaike information criterion and Schwarz criterion suggested that, ARIMA (4, 1, 3) and ARIMA (1, 1, 1) traditional approaches are suitable for predicting ASPI and SL20 price indices respectively. However, model accuracy testing results of the mean absolute percentage error (MAPE) and Mean absolute deviation (MAD), suggested that new proposed ARIMA-ANN hybrid approach is the most suitable for forecasting price indices under the high volatility than traditional time series forecasting methodologies.