Effectiveness of a Soft LTV Limit

This paper evaluates the effectiveness of a soft Loan-to-Value (LTV) limit as a borrower-based macroprudential tool. Our analysis is based on detailed loan-level data from a major Belgian bank, covering all new mortgage originations between 2016 and 2021. Using the 2020 Belgian LTV recommendations as a quasi-natural experiment, we analyze how a non-binding framework that allows tolerance margins affects mortgage lending behavior. We develop a hybrid approach combining machine-learning (ML) predictions of counterfactual treatment status with a difference-in-differences (DiD) and triple-differences (DDD) design. The ML model, trained on pre-policy data, identifies borrowers likely to exceed the 90% LTV threshold in absence of the reform, allowing consistent treatment classification across periods. Our results show that the introduction of a soft LTV limit leads to a significant decline in average LTV ratios and in the share of high-LTV loans (higher than 90%), with stronger effects among constrained borrowers. The adjustment occurs gradually, reflecting banks’ progressive adaptation to supervisory expectations rather than abrupt credit rationing. The reduction in leverage is primarily achieved through higher down payments, which leads to lower monthly repayments and thus lower credit risk. When focusing on the exceptions we find convincing evidence that banks especially favor first-time-buyers, since they remain significantly more present in the above 90% mortgage segment compared to non-FTB. Differences in age, savings, and income also lead to differentiated treatment effects, indicating that banks apply the soft limits in a risk-sensitive and targeted manner. These findings demonstrate that soft borrower-based measures can achieve prudential objectives similar to hard caps without exacerbating credit exclusion. From a policy perspective, the Belgian experience highlights that supervisory guidance can effectively curb excessive leverage while maintaining mortgage accessibility.