Abstract :
We consider in this paper the distribution of the cumulative size of recessions in 17 capitalist countries over the period 1871–1994, using data on annual percentage changes in real GDP. A recession is defined as a year in which GDP growth is negative, and the cumulative change is the change from peak to trough during a recession period. We examine both the whole sample and different partitions of the data.
The null hypothesis that the size distribution of recessions follows an exponential distribution is never rejected at the conventional level of statistical significance, p=0.05. However, there are always a small number of large recessions, no matter how the data is partitioned, which are not well fitted by a least-squares regression of the log of size on the rank of size. In other words, in a qualitative sense we see a bimodal distribution of recessions, with an exponential fit to the bulk of the data, and a second peak describing a small number of very large recessions.
A previously published agent-based economic theory model of the business cycle, calibrated purely on US data, is able with no change in its parameters to generate an exponential distribution of the size of recessions very close to that which is actually observed. In this model, information flows between agents on a completely connected network.
