Abstract: This paper redefines the stability and time-varying characteristics of risk β-coefficient and conducts a statistic measurement on them. Through a two stage econometric test with both the classic CAPM and the conditional CAPM, a conclusion can be drawn that the risk β-coefficient is not reliable. The test result with more explaining variables shows that both the Model and its risk β coefficient are more significant; while the significance of other explaining variables is different in individual shares and in the industry, and no method has been found to identify this difference. The reason why the risk β coefficient is not reliable is that the decisive factor of the present yield rate should be able to reflect the future Knight Uncertainty, the method based on historic data cannot explain the present and future earnings.

Key words: risk β coefficient; reliability; conditional CAPM; Knight Uncertainty