Banks are considered more opaque than non-financial firms due to the type of assets they own. Different measures of opacity have been used in the literature, but most of them are noisy proxies for the ‘true opacity.’ Searching for a better proxy for the ‘true opacity’ of banks, I propose a new approach based on the MIMIC model of Joreskog and Goldberger (1975). The model assumes that bank opacity is unobservable and latent, but there are several observable causes and indicators of opacity. The MIMIC model assumes that the latent opacity is caused by the type of bank assets (i.e., types of loans, trading assets, etc.) and banks’ information environment (i.e., number of analysts covering banks and number of 8-K filings). In addition, market microstructure variables are used as proxies for indicators of opacity. Using the latent opacity computed from the MIMIC model and parametric and nonparametric regression discontinuity design (‘RDD’), I study the impact of stress tests on the opacity of banks. I find that the opacity of mid-size banks ($10B<Assets<$50B) performing bank-run stress tests increased significantly for the period they were not required to disclose the results to the public. Large banks (Assets>$50B) that are required to release the results to the public had no significant difference in opacity before and after stress tests. The findings suggest that stress tests generate valuable information about the banks, and non-disclosure of the results to the public increased bank opacity. Therefore, the lack of public disclosure of the stress test results increased adverse selection problems in banks.