Relaxing the Rational Expectations Assumption: Data-based and Model-based Approaches
Abstract
The fundamental importance of beliefs about future outcomes in decision-making suggests that an accurate characterization of these beliefs is important for understanding individuals' behavior and for evaluating the counterfactuals typically needed for policy analysis. Traditionally, many researchers have been using some form of Rational Expectations (RE) assumptions to characterize these beliefs. However, empirical evidence suggests that the RE assumption might not hold in many contexts, and that incorrectly imposing the RE assumption can lead to biased policy predictions. Motivated by these findings, I explore alternative approaches to conducting economic analysis without imposing the RE assumption.
Chapters 2 and 3 of my thesis, which are co-authored with Todd and Ralph Stinebrickner, utilize unique survey expectations data from the Berea Panel Study (BPS) to characterize college students' beliefs about various future outcomes. Specifically, in Chapter 2, we characterize how much uncertainty about post-college income is present for students at college entrance and how quickly this uncertainty is resolved. Measuring an individual's income uncertainty by the variance of the distribution describing her beliefs about earnings at age 28, we find that, on average, students resolve roughly one-third of the income uncertainty present at the time of entrance during college. Consistent with the finding that the majority of initial income uncertainty remains at the end of college, We find that uncertainty about college GPA and field of study, which are the two primary income-influencing factors that are realized in college, can only account for about 19% to 27% of students' initial income uncertainty.
Chapter 3 provides a concrete example that illustrates the importance of quantifying the resolution of students' (income) uncertainty during college. By entering college, students have the option to decide whether to remain in college after receiving relevant new information. We show that the value of this option of receiving new information is determined by a student's dropout probability and how much uncertainty is resolved before the decision is made. Taking advantage of longitudinal expectations data from the BPS, we find that students have accurate perceptions about the amount of income uncertainty that is resolved during college but vastly underestimate the probability of dropping out of school. Consequently, on average, they underestimate this option value by 65%.
Chapter 4 proposes an alternative, model-based approach to jointly nonparametrically identify individuals' beliefs and the decision rule, which is a function that maps beliefs to decisions. My method can be applied to signal-based learning models, where individuals use signals to update their beliefs about an unknown permanent factor and repeatedly make decisions based on these beliefs. The econometrician observes individuals' decisions and the signals they receive at each period. Using data from the BPS, I apply my method to estimate the relationship between college students' study time and their beliefs about academic productivity as measured by the ratio of semester GPA to study time. I find that expectations about own academic productivity have a negative effect on study time. The RE assumption is rejected at a 10% level for a subgroup of students. Incorrectly imposing the RE assumption would lead to a substantially larger estimate of the effect of expectations about academic productivity on college study time.