Machine learning approaches for constructing the national anti-money laundering index.

• The LASSO and random forests are employed to jointly identify the five key factors affecting AML, which have policy implications for regulatory authorities to optimize the allocation of AML resources under the risk-based approach. • The random forests five-factor (RF-FF) model proposed in this paper has high prediction accuracy and good out-of-sample predictive ability for the MER-AML index. • The time-series national AML index constructed based on the RF-FF model contributes to overcoming the limitations of existing methods for measuring AML systems. This paper proposes a methodology for constructing the national anti-money laundering (AML) index based on Mutual Evaluation reports and machine learning models. We employ LASSO and random forests to jointly identify the key factors affecting AML, which have policy implications for regulatory authorities to optimize the allocation of AML resources. The random forests five-factor (RF-FF) model proposed in this paper has high prediction accuracy (86.31%) and good out-of-sample predictive ability for the MER-AML index, which is significantly better than competing models such as OLS and relaxed LASSO. The time-series national AML index constructed based on the RF-FF model contributes to overcoming the limitations of existing methods, providing fresh perspectives on the measurement of AML systems, and facilitating empirical studies related to evaluating the controversial AML regime. [ABSTRACT FROM AUTHOR]