This paper assesses the sensitivity of solvency stress testing results to the choice of credit risk variable and level of data aggregation at which the stress test is conducted. In practice, both choices are often determined by technical considerations, such as data availability. Using data for the Belgian banking system, we find that the impact of a stress test on banks’ Tier 1 ratios can differ substantially depending on the credit risk variable considered, but much less so on the level of aggregation. If solvency stress tests are going to be used as a supervisory tool or to set regulatory capital requirements, there is a need to further harmonise their execution across institutions and supervisors in order to enhance comparability. This is certainly relevant in the context of the EU-wide stress tests, where institutions often use different credit risk variables (and levels of data aggregation) to estimate the impact of the common methodology and macroeconomic scenario on their capital level and supervisors rely on different models to quality assure and validate banks’ results. More generally, there is also a need to improve the availability and quality of the data to be used for stress testing purposes.

The success of new technologies depends on both the firms' investment and consumers' usage decisions. We study this problem in a shared ATM network. Inefficiencies may arise because banks coordinate investment, and consumers may not make proper use of the network. Based on an empirical model of ATM investment and demand, we find that banks substantially underinvested in ATMs, in contrast with earlier findings of strategic overinvestment in the United States. Furthermore, ATM usage was too low, because regulation prohibited fees for cash withdrawals. A direct promotion of investment improves welfare, but fees for branch cash withdrawals would be more effective.