Conventional tests of capital asset pricing model usually assume that fl, a
measure of the systematic risk, is stable over time. Nonetheless, empirica
l investigations generally find that fl tends to be volatile over time. Thi
s paper formulates novel empirical models for testing the capital asset pri
cing model (CAPM) by allowing 8 to be drawn from two distinct regimes, or t
o be time varying. The econometric method via the Gibbs sampler with data a
ugmentation algorithm is applied to the data from Taiwan Stock Market to es
timate and test the model. Copyright (C) 2001 John Wiley & Sons, Ltd.