Empirical Industrial Organization
Hidden city ticketing occurs when an indirect flight from city A to city C through connection node city B is cheaper than the direct flight from city A to city B. Then passengers traveling from A to B have an incentive to purchase the ticket from A to C but get off the plane at B. In this paper, I build a structural model to explain the cause and impact of hidden city ticketing. I collect empirical data from the Skiplagged webpage and apply global optimization algorithms to estimate the parameters of my model. I also conduct counterfactual analysis to shed some light on policy implications. I find that hidden city opportunity occurs only when airlines are applying a hub-and-spoke network structure, under which they intend to lower their flying costs compared to a fully connected network. I find that in the short run, hidden city ticketing does not necessarily decrease airlines' expected profits. Consumer welfare and total surplus always increase. In the long run, the welfare outcomes become more complicated. For some routes airlines have the incentive to switch from hub-and-spoke network to a fully connected one when there are more and more passengers informed of hidden city ticketing. During this process, firms always result in lower expected revenue, while consumers and the whole society are not necessarily better off.
“Maximum Likelihood Optimization via Parallel Estimating Gradient Ascent.” (joint with Yining Wang)
Abstract: Global optimization without access to gradient information is a central task to many econometric applications as the tool to obtain maximum likelihood estimators for very complicated likelihood functions. The estimating gradient descent framework is particularly popular, which uses local functional evaluation to build gradient estimates and perform gradient descent from multiple initial points. In this work, we study the problem of coordination between the multiple "threads" of estimating gradient descent in order to pause or terminate unpromising threads early. The high-level idea is to make predictions, either conservative or aggressive, on the potential progress of each estimating gradient descent threads and to compare them with the progress on other threads. We also test our proposed methodology on both synthetic data and real airline pricing data, and compare with competitive methods including the genetic algorithm and the pattern search algorithm. The numerical results show the effectiveness and efficiency of our proposed approach.
Abstract: This paper exploits large changes in the H-1B visa program and examines the effect of changes in H-1B admission levels on the likelihood that US natives major in STEM fields. Compare to impact on labor market outcomes, the possible impact of H-1B visa reforms on natives' college major choices indicate effect over longer horizons. I find some evidence that H-1B population adversely affect natives' choices in STEM fields when they enter the college and graduate from it. Both male and White subgroups have been negatively affected, and the native Asian subgroup suffer from the most dramatic crowding-out effect. Since foreign born Asian account for a large proportion of H-1B visa holders, there might be an interesting “Asian crowd out Asian” story.