Forecasting the Loss-Given-Default ratio is a key step in financial reporting according to IFRS9. According to regulations, forecasting models are subject to periodic validation. We demonstrate that the standard Student’s t-tests for these models’ predictive ability, as recommended for banks under the Internal-Rating-Based approach, are often not fully adequate to the underlying business assumptions when applied formally to low-default portfolios. Using a value-weighted approach, we develop a statistical test that alleviates this problem. Additional empirical evidence is provided with simulated data.