Predictability Test of Stock Market Price Index in Iran Investment Market and comparing Linear and Nonlinear models predictability potentials
Subject Areas : Financial Economics
Karim
Emami
1
(Assistant Professor, Research and Science Branch, Islamic Azad University)
Ghodratollah
Emamverdi
2
(Post graduated from Department of Economic, Research and Science Branch, Islamic Azad University)
Keywords: Artificial Neural Network, Predictability, Non-Linearity test, Non-Linear & Linear Time Series Models (additives & multiplicative), Market Price Index,
Abstract :
Since the highly complicated Time Series such as Stock Market Prices are usually stochastic, their changes are assumed to be unpredictable. Some tests which have been used to study the statistical observations related to the economical variables e.g. Stock Market Price, are often go wrong while encountering the chaotic data and recognize them as stochastic ones, though these data are actually generated from the deterministic systems which bear few tribulations. For this reason the predictable and non-linear tests such as HURST, BDS, Runs Test, and Correlation Dimension have been used to study the existence of deterministic chaotic trend and non-linear process in Time Series of Daily Stock Market Price Index of TEHRAN STOCK EXCHANGE from 23 rd October, 2000 to 24 th September, 2002. The result of the above mentioned tests shows the predictability and the existence of a non-linear process in the sample data. After the illustration of predictability and the non-linear process in daily stock index data, then the linear time series models (AR), non-linear (GARCH) and Artificial Neural Network (ANN) have been estimated to present a suitable model for predicting the Stock Price Index. Comparing the potential of predictability of these models by such criteria as: CDC, RMSE, MAE, MAPE and U-THEIL inequality coefficient, it has been revealed that there is the highest potential of predictability in Artificial Neural Network models than the other ones