A Robust Adaptive Signal Decomposition Method for Enhanced Mode Extraction in Financial Time Series

Accurately forecasting energy commodity prices and financial time series has long been a challenge for policymakers and energy market participants. Various existing decomposition techniques were introduced in this scenario, like EMD, EEMD, and CEEMDAN but these techniques in the majority of the cases perform poorly in cases when a time series data exhibits irregular fluctuations and has a trend. This study introduces a novel decomposition, the Robust Adaptive Decomposition (RAD) technique, for time-series data forecasting. As its name implies robust means not duly affected by the presence of outliers, here Adaptive means having the ability to change to suit changing conditions and in signal processing decomposition is the process of dividing a signal into its subcomponents. So, this novel technique can decompose complex time-series signals like stock prices or Brent Oil prices. The RAD technique consists of several steps (i) choosing appropriate weights for each value of the dataset (ii) finding the weighted mean of adjacent points (iii) connecting these weighted points with cubic spline (iv) performing sifting as given in methodology (v) stop sifting when stopping criteria reach. The performance of RAD-ARIMA and RAD-LSTM was compared with six hybrid models such as EMD-ARIMA, EMD-LSTM, EEMD-ARIMA, EEMD-LSTM, CEEMDAN-ARIMA, CEEMDAN-LSTM, this novel technique outperformed the other six hybrid models for the simulated scenarios and a real-world data set of daily Brent Spot prices. Also, the RAD technique copes with the problem of model fitting and the labor involved with the classical decomposition techniques.