基于强化学习的联保贷款模式优化博弈分析

Optimization Game Analysis of Joint Guarantee Loan Model Based on Reinforcement Learning

联保贷款是解决中小企业及农户融资难问题的有效解决方法之一.但在联保贷款业务模式中,由于参与各方均为理性的,会选择“搭便车”策略,发生违约行为.因此,在建立联保贷款传统业务模型的基础上,引入信用等级制度.针对联保贷款参与各方通过与环境的交互学习选择的策略调整,采用基于强化学习的演化博弈理论进行讨论,演化结果显示联保参与各方与环境的交互影响过程符合马尔科夫决策过程特征.最后,仿真实验结果表明,优化后的博弈模型提高了博弈各方选择偿还贷款策略的次数,即信用等级制度的引入提升了联保贷款业务模式的效率,有效降低了违约概率.

Co-guarantee loan has proved to be an effective approach to address the financing issue faced by several small and medium-sized enterprises,as well as farmers.However,in the traditional business model of co-guarantee loans,all the parties tend to choose the“free riding”strategy,which directly leads to the decreasing enthusiasm of the party who keeps faith all the time.In this paper,a game-theoretical framework has been put forward,as an endeavor to look into the mechanism whereby a more cooperative business model can be generated.By means of reinforcement learning,the traditional business model was first studied in this framework.Correspondingly,a new business model was then developed and investigated.In such a new business model,all the parties will learn and adjust their policies through interactions with the environment,based on evolutionary game theories.The evolution results show that the interaction between the participants and the environment conforms to the characteristics of Markov decision-making process.The case study was carried out following this new business model,and the simulation results thereof demonstrated that,owing to the newly added regulation,it has been much more likely for all the parties to adopt the cooperative policy.That is to say,the introduction of“credit rating system”improves the efficiency of joint insurance loan business model and effectively reduces the probability of default.