Measurement of Systemic Risk Based on the QRDCCNN Model

Abstract

Measuring and preventing systemic risk have always been core issues in finance. To accurately capture systemic risk, this is the first introduction of the Quantile Regression Dilated Causal Convolution Neural Network (QRDCCNN) model for assessing systemic risk. This model focuses on the causal consistency of financial time series and effectively expands the model's receptive field by increasing the dilation rate layer by layer. The study selects the daily closing prices of the S\&P 500 index and 38 US financial institutions as subjects. The QRDCCNN model is employed to measure the VaR of each financial institution and the CoVaR of the financial system when these institutions are in extreme risk conditions. This paper compares the results of the QRDCCNN model with those from the DCC-GARCH, quantile regression, QRNN, and QRCNN models using the Kupiec test. The research results show that the QRDCCNN model has the highest accuracy, followed by QRNN and QRCNN models, while the DCC-GARCH model has the lowest accuracy.