Macroeconomic factors, industrial enterprises, and debt default prediction: Based on the VAR-GRU model

Abstract

This study uses a dynamic factor model to construct predictive factors and applies a machine learning-based vector autoregressive model to predict the possibility of corporate bond defaults. The vector autoregressive (VAR) model mainly examines the dynamic interaction relationships among multiple variables, so as to explain the dynamic impacts of various economic shocks on economic variables. It mainly studies the relationships among endogenous variables. Endogenous variables are those variables that are involved in the model and determined within the model system. Exogenous variables, on the other hand, are variables determined by factors outside the model. The Gated Recurrent Unit (GRU), which is a type of Recurrent Neural Network (RNN), can address issues such as the inability of RNNs to have long-term memory and the gradients in backpropagation. It is relatively easy to train. According to data from March 2014 to November 2021, the relevant findings are twofold. 1) A regulatory-based stress test is a crucial tool for measuring the financial sector's resilience in response to challenging macroeconomic conditions. 2) Macroeconomic conditions that may seem unrealistic during economic booms are now often used by regulators as benchmarks for evaluating the losses and capital requirements for market and credit portfolios.