This dissertation comprises three papers examining questions in asset pricing, investigating the implications of new asset pricing theories on the cross-section and time series of asset prices. The papers are as follows: Chapter 1 studies how the fat-tailed distribution of US firm size generates extra risk premiums compared to the classical theory. The author refers to this fat tail as "granularity" and shows that it breaks the diversification of idiosyncratic risks assumed by arbitrage pricing theory (APT) to imply factor models. In the cross-section, large firms have higher idiosyncratic risk premiums than small firms despite having a lower level of risk. This finding explains the negative relation between idiosyncratic risk and risk premium, known as the "idiosyncratic risk premium puzzle." On aggregate, the level of granularity, measured by the Pareto distribution, explains market expected returns since it determines the under-diversification of idiosyncratic risk. Chapter 2 (joint work with Rossen Valkanov and Yan Xu) investigates the joint dynamics and predictability of asset returns for the equity, treasury, and foreign asset investment sectors, utilizing their respective valuation ratios constructed from their intertemporal budget constraints. We propose a new framework that enforces an aggregate accounting identity of the three sectors using a constrained estimation by the GMM method, which accounts for the cyclical movement of the whole economy. Our key finding shows that the government surplus-to-debt ratio negatively predicts the risk premium in the equity and foreign asset investment sectors. Our results suggest that incorporating data from all three sectors and imposing aggregate budget constraints can help to better identify how the fiscal policy adjustment channel propagates throughout the economy. Chapter 3 presents a model for modeling the correlation dynamics of stock returns using a conditional factor model. In this model, the employment of factors helps to reduce the estimation dimension by presenting the asset returns' covariance matrix as a quadratic function of the conditional covariance with factors. The factor structure allows for a closed-form solution for the inverse and determinant of the covariance matrix, which is convenient for computing the likelihood function and allocating a minimum variance portfolio. The model accurately fits the realized correlation among S&P 500 stocks computed from 5-minute data. It also generates out-of-sample minimum variance portfolios with a higher information ratio.