Testing for multiple bubbles with daily data

Abstract

A new methodology for testing and dating economic bubbles based on a sign test with recursive median adjustment is presented. The methodology, originally proposed by Soo and Shin (2001) to detect random walks, is well-suited, theoretically, to deal with the many features of high-frequency financial time series such as leptokurtosis, conditional heteroskedasticity and heavy tails. The approach is very pragmatic and relies upon an analysis of the integration order of the analyzed series. This paper presents an applicationof the method to the North American stock market and the findings concerning the origination and collapsing of dates for the bubbles are consistent with those identified through the application of the previous theoretical literature.