Modeling and Estimating Unpredictability with Applications in Political Economy

Open Access

Abstract

Social science theories often make predictions not only about the mean but also about the variance of the outcome of interest. For instance, comparative political scientists argue that democracies and non-democracies have, on average, the same rate of economic growth, but the former usually has less variance in the rate than the latter. Thus, both the mean and variance of economic performance can be functions of regime types. I review four important methods to model and estimate the error variance or its function: the naive two-stage estimation, variance function regression, joint maximum-likelihood estimation, and quantile regression (plus smoothing) estimation. Using simulated data, I compare the performance of these models when the sample size is small or large, when the variance function misspecification is mild or severe, and when the mean function is misspecified. I then apply these methods in two original studies: 1) How does county leaders' in-office time affect economic volatility in Chinese counties? 2) How does the pre-WTO economic unpredictability in Chinese municipalities affect foreign direct investment (FDI) inflows after China obtained its WTO membership in 2001?