Machine learning in the service of policy targeting: the case of public credit guarantees

Public credit guarantees should be provided to firms that are both creditworthy and credit constrained. We use Machine Learning (ML) predictive tools to propose a targeting rule that includes both objectives. The study elaborates on the case of Italy's Guarantee Fund and demonstrates, by means of ex-post evaluation methods, that the program effectiveness can be increased by ML targeting. We discuss some of the problems in using algorithms for the implementation of public policies, such as transparency and manipulation.