Statistical measures of systemic risk: An application for the Turkish banking system

In this paper, we basically apply market data based statistical methods to measure systemic risk for the Turkish banking sector. In order to have a broad perspective on systemic risk with different dimensions, we employ four widely used systemic risk measures namely, MES, SRISK, CES and ΔCoVaR. First, aggregate versions of our systemic risk measures show the relative increase in systemic risk during 2000 - 2001 and 2008 crisis periods together with a pick-up in SRISK towards 2018-end. We test for predictive accuracy of SRISK as a conditional capital shortfall forecast using four cases of realized market downturns during crisis periods and results indicate that predicted SRISK levels of individual banks seem to be an acceptable estimate for realized capital shortfalls with some positive bias in particular. Tobit panel regressions of probability of defaults (PD) of individual banks on systemic risk measures indicate that, increased level of systemic risk is significantly associated with higher levels of PD up to 3-months horizon. Additionally, we compare model results in terms of their SIFI rankings which indicates that systemic risk measures have an importance in terms of ranking financial institutions based on risk characteristics beyond what can be observed by the ordinary market risk measures like VaR. As a way of comparing the relative reliability of systemic risk measures, we calculate guilt probabilities of banks associated with MES and ΔCoVaR and conclude that overall, MES and consequently MES based systemic risk metrics are relatively more reliable in terms of detecting the possible SIFIs in the Turkish banking system, albeit with a high degree of estimation risk.