I've been very frustrated for a long time now with Krugman's abuse of the term "natural experiment" and his really, really bad habit of throwing around scatterplots irresponsibly, but this post is good in that he actually highlights what could be described as a natural experiment. Shock deleveraging could probably be considered one: specifically, it's a "regression discontinuity design" of sorts (my favorite quasi-experimental method, if I had to choose one). I'd have some lingering concerns about whether this would actually work as one associated with bubble build-ups before the shock. But it's definitely in the ballpark in a way that scatterplots of government spending and GDP growth are not.
Out of curiosity does anyone know of any work on macro shocks as regression discontinuity designs?