Fractal Inspection and Machine Learning-Based Predictive Modelling Framework for Financial Markets

Abstract

This paper presents a novel framework to conduct empirical investigation and carry out predictive modelling on daily index prices of Bombay stock exchange, Dow Jones Industrial Average, Hang Seng Index, NIFTY 50, NASDAQ and NIKKEI, representing developed and emerging economies. We examine the equity markets in detail to check whether they follow pure random walk models or not. Mandelbrot’s rescaled range analysis-based Hurst exponent and fractal dimensional index have been estimated to assess the dynamics of the daily prices of the chosen stock indices. Four advanced machine learning algorithms for predictive modelling—adaptive neuro-fuzzy inference system, dynamic evolving neuro-fuzzy inference system, Jordan neural network, support vector regression and random forest—have been used to build forecasting frameworks to predict the future index prices. Fractal inspection strongly rejects random walk hypothesis and suggests that the considered stock indices exhibit significant persistent trend. The results of the predictive modelling exercises show that prices can be effectively forecasted. The research framework and the overall findings can be useful for the investors and traders to a large extent.