Working Paper: NBER ID: w23083
Authors: Thomas Philippon; Pierre Pessarossi; Boubacar Camara
Abstract: We provide a first evaluation of the quality of banking stress tests in the European Union. We use stress tests scenarios and banks’ estimated losses to recover bank level exposures to macroeconomic factors. Once macro outcomes are realized, we predict banks’ losses and compare them to actual losses. We find that stress tests are informative and unbiased on average. Model-based losses are good predictors of realized losses and of banks’ equity returns around announcements of macroeconomic news. When we perform our tests for the Union as a whole, we do not detect biases in the construction of the scenarios, or in the estimated losses across banks of different sizes and ownership structures. There is, however, some evidence that exposures are underestimated in countries with ex-ante weaker banking systems. Our results have implications for the modeling of credit losses, quality controls of supervision, and the political economy of financial regulation.
Keywords: stress tests; banking supervision; macroeconomic shocks; credit losses
JEL Codes: G01; G18; G2; G21; G28; G32
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
Cause | Effect |
---|---|
stress tests are informative and unbiased on average (C52) | model-based losses serve as good predictors of realized losses (C52) |
logistic-linear model fits the stress data well (C29) | indicates a direct relationship between estimated losses and actual losses (C51) |
2014 stress tests improved in quality compared to the 2011 tests (G28) | centralized supervision leads to more accurate disclosures (G38) |
estimated exposures predict variations in bank stock returns following macroeconomic announcements (E44) | indicates market participants' agreement with the model's predictions (E17) |
predicted outcomes compare against realized macroeconomic data (E17) | provides unbiased predictions of realized losses (G17) |