||This thesis consists of three essays on empirical asset pricing around three themes: evaluating linear factor asset pricing models by comparing their misspecified measures, understanding the long-run risk on consumption-leisure to investigate their pricing performances on cross-sectional returns, and evaluating conditional asset pricing models by using the methodology of dynamic cross-sectional regressions. The first chapter is ``Comparing Asset Pricing Models: What does the Hansen-Jagannathan Distance Tell Us?''. It compares the relative performance of some important linear asset pricing models based on the Hansen-Jagannathan (HJ) distance using data over a long sample period from 1952-2011 based on U. S. market. The main results are as follows: first, among return-based linear models, the Fama-French (1993) five-factor model performs best in terms of the normalized pricing errors, compared with the other candidates. On the other hand, the macro-factor model of Chen, Roll, and Ross (1986) five-factor is not able to explain industry portfolios: its performance is even worse than that of the classical CAPM. Second, the Yogo (2006) non-durable and durable consumption model is the least misspecified, among consumption-based asset pricing models, in capturing the spread in industry and size portfolios. Third, the Lettau and Ludvigson (2002) scaled consumption-based CAPM (C-CAPM) model obtains the smallest normalized pricing errors pricing gross and excess returns on size portfolios, respectively, while Santos and Veronesi (2006) scaled C-CAPM model does better in explain the return spread on portfolios of U. S. government bonds. The second chapter (``Leisure, Consumption and Long Run Risk: An Empirical Evaluation'') uses a long-run risk model with non-separable leisure and consumption, and studies its ability to price equity returns on a variety of portfolios of U. S. stocks using data from 1948-2011. It builds on early work by Eichenbaum et al. (1988) that explores the empirical properties of intertemporal asset pricing models where the representative agent has utility over consumption and leisure. Here we use the framework in Uhlig (2007) that allows for a stochastic discount factor with news about long-run growth in consumption and leisure. To evaluate our long-run model, we assess its performance relative to standard asset pricing models in explaining the cross-section of returns across size, industry and value-growth portfolios. We find that the long-run consumption-leisure model cannot be rejected by the J-statistic and it does better than the standard C-CAPM, the Yogo durable consumption and Fama-French three-factor models. We also rank the normalized pricing errors using the HJ distance: our model has a smaller HJ distance than other candidate models. Our paper is the first, as far as we are aware, to use leisure data with adjusted working hours as a measure of leisure i. e. , defined as the difference between a fixed time endowment and the observable hours spent on working, home production, schooling, communication, and personal care (Yang (2010)). The third essay: ``Empirical Evaluation of Conditional Asset Pricing Models: An Economic Perspective'' uses dynamic Fama-MacBeth cross-sectional regressions and tests the performance of several important conditional asset pricing models when allowing for time-varying price of risk. It compares the performance of conditional asset pricing models, in terms of their ability to explain the cross-section of returns across momentum, industry, value-growth and government bond portfolios. We use the new methodology introduced by Adrian et al. (2012). Our main results are as follows: first we find that the Lettau and Ludvigson (2001) conditional model does better than other models in explaining the cross-section of momentum and value-growth portfolios. Second we find that the Piazessi et al. (2007) consumption model does better than others in pricing the cross-section of industry portfolios. Finally, we find that in the case of the cross-section of risk premia on U. S. government bond portfolios the conditional model in Santos and Veronesi (2006) outperforms other candidate models. Overall, however, the Lettau and Ludvigson (2001) model does better than other candidate models. Our main contributions here is using a recently developed method of dynamic Fama-MacBeth regressions to evaluate the performance of leading conditional CAPM (C-CAPM) models in a common set of test assets over the time period from 1951-2012.