|
|
By including bank reactions, it offers a metric for judging the resilience of the banking sector that incorporates banks’ behaviour according to historically observed patterns. The paper also evaluates the effect of adding the feedback loops between the financial and real sectors, as well as the interactions between banks and other counterparties in financial and capital markets.
The stress test exercise illustrates the relevance of assumptions regarding banks’ behaviour, especially of a static vs. a dynamic balance sheet assumption. Compared with the results under the static balance sheet assumption, which is typically applied in the microprudential stress test exercise, under the dynamic balance sheet assumption this paper finds lower depletion in terms of the CET1 ratio, but higher depletion in terms of absolute system-wide CET1 capital. The lower depletion of capital ratios results from banks’ deleveraging that improves their capital ratios compared with the results under the static balance sheet assumption. In contrast, the higher depletion in terms of absolute system-wide capital is driven by the additional amplification mechanisms that increase losses and hence lower capital, though banks still deleverage and mute the effect on their capital ratios. The deleveraging, however, constitutes an externality and imposes costs in terms of an additional credit contraction that translates into lower economic activity. For the adverse scenario considered, a credit supply shock propagates back to the real economy and results in an additional 1.6 percentage point decrease in euro area cumulative GDP growth.
The proposed methodology for the macroprudential stress test will develop further. The results presented are based on many exogenous paths of bank-level variables that are intended to be endogenised in future versions of the model. This concerns for instance the market risk or selected components of funding costs. Further, semi-structural approach, while overcoming many modelling obstacles, is still not free from limitations imposed by generally low bank-level and loan-level data availability. For instance, separate estimation of many model equations may foster higher estimation errors. To address the latter concern at least partially, one of the appendices of the paper reports the results of an outside model validation of the reported results.