With high-quality household level asset holding data becoming available as well as the exponential increase in computing power, there is a growing literature that studies how households make investment decisions facing various types of uninsurable background risks. In this dissertation, I build theoretical models and conduct empirical studies to investigate different problems on household asset allocation. In chapter 1, I build a life-cycle model of portfolio choice with endogenous labor supply and a fixed cost of labor market participation to incorporate both the extensive and intensive mar- gins of labor supply decisions. I show that the risky asset holdings of young agents (agents younger than 45-year-old) are lower when compared to a model that only incorporates the intensive margin of labor supply. The risky asset holdings of young agents are further reduced and become hump-shaped when two additional features are included to the model: 1) endogenous Social Security accumulation and 2) a small possibility of a zero-income state. These two features increase the uncertainties faced by the agents while the fixed cost of labor market participation reduces the agents's ability to use labor supply to buffer against future income uncertainties. My model therefore reduces the gap between the empirical observations on household risky asset holdings and the predictions made by life-cycle models with endogenous labor supply. In chapter 2, we build a three-period model to study asset allocation ("how much to invest") and location ("which account to use") consequences when an economic agent has internal habit formation utility and has access to both an illiquid but tax-favored retirement account and a taxable personal account. We show that the incentive to maintain high consumption relative to the habit level and the restriction of only having access to the personal account before retirement induces the agent to hold a safer portfolio in her personal account and a riskier portfolio in her retirement account, in accordance with empirical findings on retirement asset allocation. We also show that retirement asset allocation and location decisions are affected by bequest motives and employer match, providing policy implications for retirement plan designers. In chapter 3, I provide updated estimations of the age profiles of stock market participation and risky share in the United States using data from the Panel Study of Income Dynamics (PSID). This chapter is motivated by the recent findings of Fagereng, Gottlieb, and Guiso (2017) on Norwegian data that the age profiles of stock market participation rate and risky share become closer to theoretical predictions when they employ more precise empirical strategies to identify the age, cohort and year effects, control for demographic variables and use a Heckman selection model to control for the endogeneity of stock market participation decision. I apply the same empirical strategies in Fagereng et al. (2017) on the U.S. data. I find that the age profile of stock market participation rate is increasing over the life cycle instead of hump-shaped. The estimated conditional risky share, after controlling for selection, is higher than the risky shares reported in previous papers and it is slightly increasing over the life cycle.