Bubbles and Crashes: Empirical Evidence

Abstract

Stock market crashes are rare events. This complicates a thorough quantitative empirical analysis of crashes and their probable causes. We introduce the concept of an industry crash and study the presence of these crashes in 48 US industry indexes over the period 1926 to 2004. The concept of an industry crash enlarges the sample of crashes available for study substantially. This large sample of crashes allows us to test several theoretical hypotheses recently put forward in the literature on the relation between bubbles and crashes. Our empirical evidence shows that bubbles - periods of strong outperformance - double the likelihood of a crash. This relation is stronger for more severe crashes: for the twenty percent largest industry crashes the presence of a bubble triples the crash probability. Our results also confirm theoretical results that crashes are more likely when bubble growth is stronger, but that the time a bubble takes to develop seems unrelated to crash likelihood. Last but not least, we verify whether these results for industry crashes carry over to general stock market crashes using shorter time series of stock market indexes for 49 countries starting in 1969. We find that they do.

Keywords: bubbles, crashes, skewness

JEL Classification: G10, G14, C14