A Deep Reinforcement Learning-driven Vine Copula Method for Correlation Structure Analysis of Mortgage


Controlling risk is the key to playing a core role in financial services and effectively serving the high-quality development of the real economy. And the correlation analysis between the characteristic variables is the foundation of risk tracing and prevention and control. As the reform and innovation of rural finance, the mortgage loan of agricultural land management right is important to reduce risk effectively and promote financial innovation to a larger scope and solve the problem of control diffculty for peasant loans by studying its risk characteristic variables associated structure. However, there are multiple variable risk factors affecting the mortgage loan debt default of agricultural land management right, and the combined correlation structure of its characteristic variables has high-dimensional complexity. Therefore, an effective modeling method is needed. To this end, this paper proposes a deep reinforcement learning-driven algorithm framework based on the C-vine copula function in vine copula. In our model, the C-vine copula function uses a binary function combination to conveniently and intuitively describe the structural correlation between variables. Deep reinforcement learning, with outstanding nonlinear fitting and high-dimensional space representation capabilities, automatically learns in exploration, and plays a key role in the modeling of complex high-dimensional variable structure correlations. According to the distribution of data, the variables and copula function types at each level are selected to effectively improve the total log-likelihood of the model. The results show that in the correlation structure of the variables influencing the default of the mortgage loan debt of the agricultural land management right, the generation order of the variables is loan amount, interest rate, mortgaged farmland area, household’s expenditure, output value of major crops, age, household’s income and the distance from village halls to the nearest farmland trading center. We also found that paying attention to the dependency relationships among the tails of variables is crucial for comprehensive risk analysis and effective prevention. This paper sheds light on an intelligent modeling method of reinforcement learning-driven vine copula to mine the correlation variable structure of farmland mortgage loans, which provides support and has important significance for effectively controlling the rural financial debt default risk.

China Journal of Econometrics