The objective of this dissertation is to study the dynamics of size and value risk premia in an equilibrium model with belief dependent preferences and to analyze the impact of investor attention on asset pricing. There is ample evidence that size and value risk premia are non-constant and vary over the business cycle. Empirical patterns, however, are unknown and traditional equilibrium models cannot fit the observed dynamic patterns. The representative agent model with belief dependent preferences is known to fit both unconditional moments such as the equity premium as well as times-series features of volatilities and market prices of risk. The basic model is extended to capture the dynamics of size and value risk premia. The representative agent in this model is a rational Bayesian decision maker who updates her beliefs continuously when new information arrives. However, information processing costs are non-zero and opportunity costs of non-continuous updating of beliefs are higher during times of crisis. In the second part of this dissertation, the representative agent model with beliefs dependent preferences is extended to incorporate the notion of investor attention. The attention version of the model is shown to increase the dynamic fit of equilibrium asset pricing quantities by dampening the volatility of bond yields, market prices of risk, and stock volatility. As such the inattention version of the model with belief dependent preferences is shown to improve the intertemporal fit. Chapter 1 provides a overview of existing studies about the dynamics of size and value risk premia and investor attention. Chapter 2 investigates the dynamic features for size and value risk premia. An asset pricing model with regime dependent risk aversion and incomplete information about economic regimes is introduced to derive closed-form formulas for market prices of risk, asset prices, their volatilities, and risk premia of value and size style indices. Both size and value risk premia vary across normal, recession and boom periods. The premia amplify in recession times but tend to reverse or disappear during boom times. Such findings match the historical performances of small-minus-big (SMB) and high-minus-low (HML) portfolios. Chapter 3 integrates investor attention into regime-switching learning model with regime-dependent risk aversion. The model provides a good fit to the time series of stock volatility, bond volatility and bond yields. Investor attention at the aggregate level is captured by a new representative agent measure which combines the continuously updated beliefs about regimes of a rational Bayesian decision maker with those of a decision maker using steady state regime probabilities. The new representative agent measure can capture the scenario where investor updates her beliefs about economic regimes according to time-varying attention to the available market information. Equilibrium asset pricing quantities are obtained in closed form in the extended model with investor attention. Unconditional asset pricing model moments match their empirical counterparts including the equity premium, the stock volatility and the correlations between stock returns and consumption and dividends. Dynamics features of the data can be well captured. Stock and bond volatilities, bond yield and interest rate time series all have smaller mean square errors compared to the model which does not consider investor attentions. The scale and volatilities for these financial time series are also close to real financial data.