You know the feeling, it is only once you lost something that you realize how much you cared about it. Nowadays that we are affected by instability in the banking sector, we realize how good it was to have stable banks. How could we quantify this?
There is ample research on the impact of banking crises, but it treats data in a black or white fashion: either you are in a crisis or you are not. Pierre Monnin and Terhi Jokipii, however, use a continuous measure, the probability that banks would fail, in 18 OECD countries. Their panel VAR indicates clearly that bank instability leads to lower real GDP growth, more volatility of growth, and over-prediction of future growth. Looking at the numbers, the impact is not large, though: a one standard deviation shock to output increase the bank failure measure by 11% of its standard deviation, while it is 7% the other way around. While statistically significant, this does not strike me as economically significant. And one should not interpret too much these results, as VARs are only good to describe the data, but not good for understanding behavior and policy.
Of course, for such an exercise, details are also very important. For example, how do you measure the probability of default of a banking sector? Monnin and Jokinii model it as the probability that the whole banking sector would exercise an option to renege its debts. Why not use the Z-Score, which is available for individual banks and already widely used, and work from there? How sensitive are the results to the many choices the VAR econometrician has? There is always danger of data mining here, so having some theory to guide choices would be good.