The impacts of liquidity and estimation error on test statistics of CAPM

This dissertation conducts simulations to examine the pure effect of the beta estimation error (errors-in-variables, EIV) in Chapter 1, the pure liquidity effect in Chapter 2, and the joint effect of liquidity and estimation errors in Chapter 3 on univariate test statistics of CAPM in a Fama-MacBeth-type two-pass cross-sectional regression. The joint effect is the sum of the two pure effects and the interaction effect of liquidity and beta estimation error. To investigate the pure liquidity effect, Chapter 2 derives the liquidity equation based on the foundation of market microstructure. Chapter 2 also shows that some existing models can be presented as special cases of the liquidity model. The simulation results show that: (1) the pure EIV effect typically results in test power losses and test size increases; (2) the pure liquidity effect may result in test power losses and virtual test size increases; (3) the joint effect typically results in test power losses and virtual test size increases; and (4) the interaction effect generally brings about more test power losses and more test size increases, with few exceptions. This type of interaction suggests that the joint effect is generally augmented by the interaction and is greater than the sum of the two pure effects. Although CAPM is assumed to be true, the test results that are influenced by any of the pure EIV effect, the pure liquidity effect, and the joint effect may contradict CAPM with high probability under some sample sizes and/or market conditions.