Tax credit management system (TCMS) implemented by the State Administration of Taxation (SAT) of China in 2014 is an innovative mechanism of tax collection and management. This paper attempts to incorporate the TCMS into the model of Allingham and Sandmo [Allingham, M. G. and Sandmo, A., Income tax evasion: A theoretical analysis, J. Public Econ. 1(3–4) (1972) 323–338], and then proposes an agent-based tax compliance model combined with EWA model. In this model, enterprises are divided into three categories: enterprises paying taxes for better credit, paying taxes to avoid a punishment, and not paying taxes. Enterprises can borrow from banks based on their own tax credit rating. The experimental results based on the model show that: (i) When the probability of spot check is low, the actual tax behavior of some enterprises is inconsistent with the results of tax credit publicized by tax authorities. (ii) In order to reduce the proportion of tax evasion, we try to change the parameter variables of integral reward, audit probability, punishment, financial subsidy and loan interest rate. The results show that the punishment of fixed high loan interest rate on tax evasion enterprises can effectively restrain tax evasion and increase the proportion of enterprises with high credit rating.

中文翻译：

---

EWA学习模型中征税行为动态博弈与信用评价的演化分析

国家税务总局（SAT）于2014年实施的税收信用管理制度（TCMS）是税收征管的创新机制。本文试图将 TCMS 纳入 Allingham 和 Sandmo 的模型 [Allingham, MG 和 Sandmo, A.，所得税逃税：理论分析，Ĵ.公共经济。 1(3-4) (1972) 323-338]，然后结合 EWA 模型提出了基于代理的税收遵从模型。在这个模型中，企业被分为三类：为更好的信用而纳税、为逃避惩罚而纳税和不纳税。企业可以根据自己的税收信用等级向银行借款。基于模型的实验结果表明：(i)在抽查概率较低时，部分企业的实际纳税行为与税务机关公布的税收抵免结果不一致。(ii) 为了降低偷税漏税的比例，我们尝试改变积分奖励、审计概率、惩罚、财政补贴和贷款利率等参数变量。