|
Essays on Asset Pricing Models
This paper studies the role of time-varying liquidity in explaining the future asset returns. Using the innovations in liquidity and its volatility as liquidity risk measures, the methodology of Shanken (1990) is adopted in a multifactor asset pricing framework. The resulting conditional model allows the factor loadings to vary over time with the liquidity variables, hence, distinguish the risk and non-risk components of their explanatory power. The merit of the conditional model is its ability to directly test whether the innovations in liquidity and volatility capture time-variation in the risk of an asset and whether they contain further information after controlling for changes in risk.
The analysis covers the period of January 1964 - December 2008, and considers daily, weekly, and monthly data for ten liquidity-sorted portfolios. I estimate the innovations in liquidity using a recursive AR(1) specification. To model the time-varying volatility of liquidity, I employ a recursive GARCH (1,1) process for daily, and ARCH(2) process for weekly, monthly frequencies. First, I estimate the seemingly unrelated predictive regressions for ten portfolios using least-squares method, since it provides efficient estimates in our framework. The results show
that, the lagged innovations in market-wide liquidity have a significant impact on the expected portfolio returns. Second, I investigate the source of the time-variation in expected returns by using the conditional multifactor regressions. I find that both the innovations in liquidity and its volatility are strongly associated with changes in assets' risk. After controlling for the time-varying risk, the explanatory power of the liquidity variables disappear at weekly and monthly horizons. However, the daily innovations in liquidity convey information about the future prices beyond the risk explanation.
This paper develops exact distribution-free tests of unconditional mean-variance efficiency.
These new tests allow for unknown forms of non-normalities, conditional heteroskedasticity, and other non-linear temporal dependencies among the absolute values of the error terms in the asset pricing model. Exactness here rests on the assumption that the joint temporal error density is symmetric around zero. This still leaves open the possibility of return distribution asymmetry via coskewness with the benchmark portfolio. A simulation study shows that the new tests have very good power relative to that of many commonly used tests. The inference procedures developed are further illustrated by tests of the mean-variance efficiency of a market index using a forty-two-year sample of monthly returns on ten U.S. equity portfolios.
"Liquidity Risk and Intertemporal Portfolio Choice: A Nonparametric Approach"
"Yield Curve and Macroeconomic Variables: An Intertemporal Asset Pricing Model"
|
