This paper contributes to the literature by assessing the predictive power of systemic risk measures (SRM) in the context of bank defaults. Quarterly forward-looking probabilities of default (PD) are estimated for a heterogeneous population of 22,751 banks using a model consisting of only idiosyncratic variables. Under the developed hypothesis, the accuracy of the model should improve, when combining measures of idiosyncratic and systemic risk. Surprisingly, β and SRISK do not improve the performance to exceed a random draw.  ΔCoVaR and MES enhance the accuracy of the model to 99.97%, respectively 99.90%. However, only the latter is statistically significant. In investigating explanations for the findings, bank size becomes a likely determinant. It is suggested that depending on size, banks allot different shares of their risk budget to credit and market risk. The applied metrics capture market risk more accurately than credit risk, which might explain their poor performance, as the majority of banks is rather small. The findings challenge the adequacy of current measures of systemic risk. Furthermore, it allows regulators to prioritize their use of SRM, and identify ailing banks ahead of a possible default. The results are time invariant, and robust against structural distortions such as size and winsorization.