The article says that men default more than women, controlling variables such as payment history and income, so models should be allowed to account for that (with a gender dummy variable?). But would the same argument be made if current models overpredicted men's defaults compared to women's?