Assessment and Comparison of linear and Non-linear Methods for Forecasting Returns on Stock Market Index

Nowadays, Investing in the stock exchange market is an important component of the countries economy. Being able to forecast the stock price variations is of great significance for the stockholders as it will enable them to obtain the highest return on their investment. The stock price index illustrates the general state of the stock market and could help stockholders make forecasts for their investments. This article assesses and forecasts returns based on Tehran Stock Exchange Index (TEPIX) using the daily data between 22/11/2010 and 18/03/2014 and different forecasting models such as ARIMA, GARCH, and Artificial Neural Networks (ANN) and Adaptive Network-based Fuzzy Inference System (ANFIS). Evaluating the accuracy of above mentioned models through their forecasting measures (e.g. RMSE, MAE, MAPE, TIC and CSP) reveals that ANFIS has a better performance of all models. The statistical comparison of the results through Diebold-Mariano Test also demonstrates a significant difference in models accuracy.