Recent estimates of the intergenerational correlation of income in the United States are centered around 0.6. Existing empirical work is only able to explain about half of this correlation. The first chapter of this dissertation provides a behavioral explanation that accounts for almost half of the unexplained correlation. Heterogeneous agents in the model are loss averse and must choose their education level after learning their “earning ability” and inheriting a reference level of consumption and bequest from the previous generation. These agents make education choices in part to avoid losses relative to reference consumption in the first and second periods of their lives. Agents with high inherited reference consumption choose high levels of education in order to avoid losses in the second period and are therefore likely to have high income and consumption themselves. Those with very low reference consumption are likely to get more education than those in the middle of the reference consumption distribution, as they are less likely to experience a loss in the first period. I find support for this U-shaped education decision rule using the NLSY97 data set. The dissertation also tries to answer the question of why black and white workers display significant differences in their labor market outcomes. Black workers tend to have less education and earn lower income than their white counterparts at each level of education. The second chapter explores three possibilities (wage discrimination, lower earning ability, and low aspirations) for these gaps within the framework of a model with loss aversion and inherited reference consumption. When people have loss-averse preferences, low aspirations lead to lower levels of chosen education. Loss aversion and low aspirations can lead to education outcomes similar to those caused by outright discrimination or lower earnings ability. When combined with wage discrimination the model can also help explain the larger poverty trap and lower affluence net in black families as opposed to white families. Simulation results compare favorably to intergenerational quintile transition rates in the literature. The model takes many generations to reach educational equality after a period of wage discrimination is ended.