Dynamic Forecasting for Systemic Risk in China’s Commercial Banking Industry Based on Sequence Decomposition and Reconstruction

Abstract

The Pressure Index of commercial banks is an effective measure of the systemic risk in the sector. This helps decision makers and market participants assess the potential levels of stress that commercial banks may face when confronted with impending risks. This study proposes a method for forecasting future trends in a Pressure Index for systemic risk prediction. The banking stress index is specifically constructed through an extreme value approach, followed by a non-stationary time series decomposition using variational mode decomposition (VMD). The number of decompositions was determined using the fuzzy entropy (FE) rule. These models were then used to construct autoregressive integrated moving average (ARIMA), artificial neural network (ANN), backpropagation neural network (BP), recurrent neural network (RNN), and long short-term memory (LSTM) models for independent prediction. The empirical results demonstrate the significant advantages of the VMD technique for forecasting non-linear and non-stationary complex time series. These findings highlight the substantial benefits of using VMD in forecasting intricate temporal patterns, especially in cases where traditional methods may face challenges in effectively capturing underlying dynamics. The VMD-ARIMA model showed superior prediction accuracy compared with the other models. Our study aims to model and forecast the data of the banking stress index, which is of utmost importance for the central bank in formulating macroeconomic policies and for commercial banks in managing credit risk.

Keywords

• China's commercial banking industry
• systemic risk
• forecasting
• variational mode decomposition