Characterization and Predictive Analysis of Volatile Financial Markets Using Detrended Fluctuation Analysis, Wavelet Decomposition, and Machine Learning

1. Introduction

Comprehending the key characteristics and conducting predictive analytics of financial markets are extremely arduous due to their nonparametric, nonlinear and chaotic nature of temporal movements (Zhang et al. 2015; Ghosh et al. 2017; Jammazi et al. 2017; Tiwari et al. 2018a, b; Jana et al. 2019). Researchers critically examine the behavioral aspects and develop forecasting frameworks for various practical implications (Cipollini et al. 2015, Sharif et al. 2017). Comprehending the properties of temporal evolutionary pattern beforehand is arduous for subsequent predictive modeling tasks (Ghosh et al. 2019, Ghosh et al. 2020) as the said exercise assists in identifying explanatory features.

Plenteous studies report the use of traditional econometric and statistical approaches for forecasting future movements of the financial time series (Granger 1992, Liu and Pan 2019). However, the shortcomings of these techniques become prominent when the time series demonstrates nonparametric, nonstationary and nonlinear patterns. Machine learning, deep learning and computational intelligence algorithms prevail and produce superior forecasts for such time series (Wang et al. 2011, Kao et al. 2013, Hsu et al. 2016, Zhao et al. 2017). One set of machine learning and related models work in a univariate framework using immediate past historical price information of stock markets as input features, while the other models work in a multivariate framework. The majority of the second category models either employ a set of technical indicators or conduct a fundamental analysis to estimate future movements (Zhang Dan et al. 2016, Ni Ping et al. 2011). Fundamental analysis considers several financial ratios estimated annually, biannually, and quarterly for assessing financial health of organizations and employ them to determine the share price at same time intervals. Fundamental analysis cannot be utilized for predicting daily stock prices directly. Technical indicators or univariate forecasting structures are applied for daily price forecasting.