A comparative study of static and iterative models of ARIMA and SVR to predict stock indices prices in developed and emerging economies

Abstract

Predicting the stock market is a complex and strenuous task. Moreover, the stock market time series is nonlinear, volatile, dynamic, and chaotic. The efficient market hypothesis (EMH) and random walk hypothesis (RWH) state that it is futile to predict the stock market. Auto-regressive integrated moving average (ARIMA) and support vector regression (SVR) are popular methods in time series forecasting. This study empirically compares static and iterative models of ARIMA and SVR's ability to predict stock market indices in developed and emerging economies. Five global stock indices, two from emerging and three from developing economies, are predicted. In the long-term, in contrast to EMH and RWH, the results show that the SVR has predictable power. Further, the SVR has better predictability in emerging economies than in developed ones in long-term forecasting. The market shows efficient behaviour in daily prediction, and the naïve model is the best performer. Additionally, the ARIMA model is equivalent to the naïve model in daily and long-term prediction.