U.S. Income Volatility and Interstate Migration
This dissertation consists of three chapters. The first chapter explores how U.S. male income has evolved, ranging from 1979 to 2017. The research aims to decompose the income into the permanent component − the long-term average − and transitory component − the period-specific deviation from the average − since the two have different implications in practice. After constructing a pseudo panel using the Current Population Survey, we estimate the structure of income volatility using an extended semiparametric model proposed by Moffitt & Zhang (2018). The transitory variance fluctuated through the mid-1990s and declined until 2002. Since then, the transitory variance increased through 2013 and almost recovered to the level in the mid-1990s. Furthermore, we find a countercyclical pattern of gross volatility and transitory variance around the Great Recession. Second, the next chapter studies the income convergence of U.S. states. The decline of the U.S. income convergence rate across states has been widely accepted in the context of 𝛽𝛽-convergence or 𝜎𝜎-convergence. The chapter revisits the question of income convergence by applying a relative convergence test, which again demonstrates that U.S. regional incomes do not converge. Alternatively, income convergence has formed among four subgroups ‒ called clubs ‒ within which states have specific characteristics in common. The ordered logit regression analysis suggests that changes in labor demand and K-12 public school spending contribute to the formation and composition of convergence clubs. Finally, the third chapter uses Monte Carlo simulations to evaluate the efficacy of techniques used to estimate migration responses to redistributional taxes across states or countries. After constructing a synthetic dataset that incorporates real-world phenomena using Current Population Survey (CPS), the analysis imposes arbitrary migration elasticity. We then attempt to uncover the true elasticity using logit and difference-and-difference regressions, which are commonly applied in this literature. In the simple baseline scenario, estimators perform well, uncovering the true elasticity. When incorporating frictions into the data, however, both estimators are biased downward.