From RIETI by Esteban Rossi-Hansberg:
Economies experience a variety of local shocks: Natural disasters and weather related disruptions, discoveries of natural resources, disruptions of particular trade routes, corporate bankruptcies and other shocks to local firms, and labor strikes are a few examples. Moreover, industry shocks are local since industries are unevenly distributed in space. Many of these local shocks are too small to affect the aggregate economy and to have an impact that goes beyond the very town, firm, or industry that it affects directly. However, other shocks can have widespread effects on states, other sectors, and, importantly, the aggregate economy.
Japan experienced this first-hand with the economic effect of the earthquake and tsunami, as did cities like New Orleans in the case of Hurricane Katrina, or states like Michigan with the troubles of U.S. automakers, or California with inventions in the high-tech industry. Examples of local shocks abound, but do these shocks have sizable aggregate implications? Can we gauge and measure the effect of these shocks on aggregate outcomes? What does the elasticity of aggregate productivity, output, and employment depend on? Understanding the answer to these questions is essential for the design of policies to alleviate and manage the effect of shocks to the economy.
The potential importance of idiosyncratic shocks in generating aggregate fluctuations has long been recognized. Nevertheless, these types of shock were dismissed as an important source of aggregate fluctuations due to arguments based on the "law of large numbers". The idea is simply that the sum of these shocks averages out in the aggregate as we add a large number of 'small' shocks. This view has been recently challenged in a number of ways, both theoretically and empirically. For example, Gabaix (2011) argues that, if the units that receive the shocks are firms, these shocks will not average out unless the number of firms is unrealistically large. The reason is that the size distribution of firms is well approximated by a Pareto distribution and such a distribution, with its fat upper tail, implies that the law of large numbers converges very slowly; too slow for the number of firms in the U.S. to be large enough for idiosyncratic firm-level shocks not to matter. So the shocks to some of the large U.S. firms, like General Motors Company or Walmart, have a large impact on aggregate output.