A granular machine learning framework for forecasting high-frequency financial market variables during the recent black swan event

【Author】 Ghosh, Indranil; Jana, Rabin K.

【Source】TECHNOLOGICAL FORECASTING AND SOCIAL CHANGE

【影响因子】10.884

【Abstract】This paper analyses highly voluminous 1-minute intraday movements of the closing prices of Bitcoin, crude oil, the Dow Jones Industrial Average (DJIA) and the euro-U.S. dollar exchange rate in various non-overlapping regimes spanning across the COVID-19 pandemic, a recent black swan event. The empirical characteristics of the intraday dynamics of chosen assets are gauged using nonlinear modelling tools and tests. The proposed methodological framework utilises maximal overlap discrete wavelet transformation, Bayesian structural time series forecasting and random walk with drift to evaluate the quantum of predictability of select variables in different phases across the pandemic horizon. The framework survives a series of performance and statistical checks to justify the efficacy of drawing highly accurate predictions from big data setups. The findings suggest that, despite chaotic traces, all variables can be precisely forecast across different sub-regimes, eliminating the adverse effects of the first and second waves of the COVID-19 pandemic. High-frequency movements of Bitcoin and Euro-U.S. dollar exchange rates are relatively better predictable, signifying resilience during the pandemic. Finally, the outcome of this research will help mitigate financial risk during volatile periods.

【Keywords】Big data analytics; High-frequency financial market forecasting; Nonlinear dynamics; Maximal overlap discrete wavelet; transformation; Bayesian structural time series

【发表时间】2023 SEP

【收录时间】2023-08-18

【文献类型】

【主题类别】

--