Measurement model of credit risk for unlisted agricultural enterprises.

Abstract

This paper aims to measure credit risks of unlisted agricultural enterprises by using the KMV model integrating a CNN-BiLSTM neural network. Initially, the expected default frequencies (EDF) for each listed agricultural enterprise are computed using the Black-Scholes option pricing formula within the KMV framework. We apply the neural network model trained by listed agricultural enterprises to the credit risk analysis of unlisted agricultural enterprises. The EDF and financial data of listed agricultural enterprises undergo Z-score standardization and comparison using CNN-BiLSTM neural networks. Model parameters are then experimented with to determine the optimal CNN-BiLSTM model. This selected optimal CNN-BiLSTM model is applied to standardized financial data of unlisted agricultural enterprises to derive corresponding EDF. Based on the EDF of the listed agricultural enterprises, corresponding rating intervals are determined for unlisted agricultural enterprises. We use unlisted companies in China as an example in empirical analysis. The results demonstrate the effective assessment of credit ratings for unlisted agricultural enterprises using this model, generally aligning with institutional rating outcomes. Given differences in rating systems, the model helps identify hidden credit risks that are challenging to detect through conventional rating methods. It highlights the nonlinear relationship between enterprise credit risks and financial indicators, including debt repayment capacity, operational capability, growth potential, profitability, and debt structure.