I think there's a way to do it via bootstrapping; and Simen Gaure might have implemented it in his R package for FEs; but it might involve some coding

2/ The paper calls time on the idea a systemically important firm needing an emergency loan is likely "solvent but illiquid"—in none of the cases was liquidity alone a "cure.”

It also highlights (literally!) other takeaways and notable design features for future crisis-fighters.

Can you split it by theory vs empirics? (asking chatgpt based on the abstract for instance)

I prefer not to delve into opinions (particularly given my job), but what I've seen from looking at historical bank runs is that joint runs+failures never took place in a vacuum (due to sunspot, etc.). In most cases it appears obvious that the bank would have failed even with no run. But, ...

BTW, we are not saying runs are unimportant. They might have large consequences, from causing banks to hoard cash (ex-ante) to reducing the supply of credit (ex-post)

Our statement is very specific about bank runs not being the ultimate cause of bank failures, even though they often occur together.

Why would you use a 800 parameter model when you can fine-tune a 8000000000 parameter language model to output the sentences "this is risky" and "this is not risky" instead?