This is a repository for theses that I wrote independently during my undergraduate and graduate studies.
This paper introduces new dynamic autoregressive probit models by including factors selected from a large number of predictors using principal component analysis, and applies them to predict recessions in the United States with different forecasting horizons. In addition, this paper adopts a ‘stepwise’ factor-selecting procedure, which mimics the idea of the Least Angle Regression algorithm with a customized version that can be applied to a binary response variable. Using McCracken's FRED-MD dataset from 1960 to 2016, the empirical results suggest that we should not impose a specific number on the number of predictors or factors selected. Further, extracting factors from the most informative predictors, rather than from all predictors, is more effective for recession predictions. We also find that different predictors or factors are most useful for different forecasting horizons, implying that it is preferable not to use the same variables for different forecasting horizons.
This paper explores the impact of having access to transportation infrastructure on regional economic growth, based on economic outcomes for non-metropolitan cities in China from 1997 to 2011. In particular, this paper uses both nonmetropolitan cities’ official economic data from Chinese Provincial Statistical Yearbooks and night lights data collected by U.S. Air Force Defense Meteorological Satellite Program (DMSP), of which the latter has been widely used as a proxy for economic activities. Adopting the “straight line” identification strategy to address the issue of endogeneity on the placement of transportation networks, this paper finds that cities more distant to the constructed straight lines have less access to transportation networks. Further, empirical results suggest that there exists a positive effect of transportation infrastructure on economic growth, and the magnitude of the effect becomes larger when using night lights remote sensing data rather than official GDP data.