A New Test for Speculative Bubbles Based on Return Variance Decompositions.

We suggest a new test for speculative stock market bubbles that has several advantages compared to earlier bubble tests. The test makes use of the fact that the variance of excess return innovations and the variance of (dividend news minus interest rate news minus excess returm news) will be equal if there is no bubble, and differ if there is a bubble. A VAR-model is used to estimate the variance decomposition, and the test is computed using bootstrap simulation. On US and UK data over the period 1919-1999, the test does not reject the no-buble hypothesis.