What are the best indicators of systemic risk?

After the financial crisis of 2007-2008, the need of reliable indicators of financial stability became increasingly important. In a first stage, we review the current literature about ten systemic risk indicators (measures/early warnings). The second part is dedicated to an empirical study where we evaluate the performance of some indicators in order to know which is the best measure/early warning of systemic risk. We test only five indicators from the US and Europe that we consider as simple and easily available to everyone: market volatility (VIXX & VSTOXX), interbank rates (Ted spread), Yield curve slope, Return of the bank equity indices and CDS indices (CDX IG, Itraxx & Itraxx Financials). These monthly data are the independent variables for the US and Europe by means of which we try to see whether we can predict systemic risk. We test the performance of each of these indicators from 0 to 6 lags (in months). When the indicators are tested with 0 lags, we are testing the performance of the indicator as measure of systemic risk. When the indicators are tested from 1 to 6 lags, we are testing the performance of the indicator as early warning of systemic risk. We also choose two systemic stress indicators: the Composite Indicator of Systemic Stress (CISS) for Europe and the Cleveland Financial Stress Index (CFSI) for the US. They act as dependent variables which will tell us whether there is or there is no systemic risk. We use three methods to test the performance of the indicators. The first one is the Logit regression and we take the highest McFadden R² and Count R² as criteria of selection of the best performing indicator of systemic risk. The second one is the KLR signal approach and we take the smallest noise-to-signal ratio as criterion of selection of the best performing indicator of systemic risk. The third and last one is the Granger causality test and we take the Akaike Information Criterion as criterion of selection. After having computed the results for all these three methods, the purpose is to see if their results are consistent.