Re-Examining Variance-Bounds Tests for Asset Prices

The hypothesis of market efficiency is typically rejected by standard variance-bounds tests which assume stationary asset prices. A number of researchers, however, argue that tests used in previous studies are inappropriate since asset prices appear to be generated by nonstationary processes. In this paper, we propose a regression-based variance-bounds test that is valid when the asset price is an integrated process. We apply this test to annual U.S. data over the 1889 to 1985 sample period using measures of the perfect-foresight price constructed from a nonlinear asset-pricing equation that allows for a stochastic discount parameter. The results suggest that the data appear consistent with a version of the efficient-market hypothesis detailed in this paper