Here is a condensed version of my UChicago BA thesis. 

I study Gibrat's Law, an empirical regularity which asserts cities' population growth rates are independent of their population sizes. While the existing literature only replicates this regularity, I further test for conditions that may disrupt Gibrat's Law. To do so, I modified a structural theory of citizens' location choices and implemented a simulation micro-founded on that theory.  In the theory and simulation, utility-maximizing citizens optimally choose to move across cities. Citizens' migration updates population parameters for each city, which in turn affects cities' productivity and induce further migration. 

I ran the simulation in Python to find that citizens' heterogeneous preferences, costly migration, and serially-correlated productivity shocks to cities can disrupt Gibrat's Law. As an empirical case, I noted that Chinese cities were an outlier to Gibrat's Law: larger Chinese cities in fact grew slower. I reasoned that restrictive hukou residential permits caused this pattern. To verify this narrative, I calibrated my simulation to Chinese data, modeling the hukou as a one-time migration cost to replicate China's empirical outlier down to 1% error. 

Developing countries often try to ease growth stresses on their largest cities. My research identifies conditions that may slow these cities' growth. My simulation can further examine welfare implications and counterfactuals of policies that limit urban growth. For example, I can compare simulation worlds with vs. without hukou. 

This is a paper I wrote in a UChicago Economics PhD course, which I took as an undergraduate. 

I studied why an inverted "U" shape commonly characterizes the empirical relationship between the number of firms in a market and the per-firm intensity of innovation.  To explain this inverted "U" shape, I constructed a theory in which firms choose to differentially invest in producing a regular and innovative good. Initially, firms invest more in making the innovative good to escape Cournot competition in the regular good. However, a larger number of firms in the market can render the race to innovate first too costly and less worthwhile, thereby reducing per-firm innovation. I then tested my model against innovation (patent) data.

The literature has not yet reached a complete consensus on how to define and estimate submarkets. In this paper, we demonstrated how machine learning (ML) methods, combined with spatial data science, can identify NYC Airbnb submarkets. We then used logit models to estimate demand in our defined submarkets. 

My main contribution was defining the Airbnb submarkets. I built a pipeline of feature selection (LASSO), dimension reduction (PCA), and clustering  (KMeans, GIS spatial clustering with SKATER) to process Airbnb location and property attributes into submarket definitions. I then helped prepare data to run demand estimation in each identified submarket. 

We showed that demand estimation results differ substantially between submarkets entirely defined on NYC boroughs (the "default") vs. defined on our ML pipeline that combines both location and property attributes. 

I assisted Dr. Wang on her working paper, now public on the Boston Fed website. We studied which firms borrowed from the MSLP, a pandemic-era Fed program to avail credit to businesses, and how the program impacted recipient firms' financial health. 

I processed firms' confidential credit data and ran logistic regressions to identify firms most likely to borrow from the MSLP. Moreover, because relatively few firms borrowed out of the population, I computed and implemented an econometric rare-event bias correction to our estimates. Finally, I helped produce publication-ready LaTeX tables with summary statistics and regression results. 

This paper could be thought of as a thesis for my GIS minor. 

I ran hedonic regressions for NYC Airbnb prices, comparing results from conventional regressions vs. regressions that involve spatial econometrics. For example, I used a spatial lag model and weights matrix to show that each Airbnb's price depended on the attributes of Airbnbs nearby - an insight not immediately apparent in the conventional regressions. To collect data and create results, I used spatial database systems (e.g. PostGIS), spatial econometrics packages in Python, and advanced spatial data visualization tools. 

I led the Panama section of a US National Science Foundation-funded project led overall by Prof. Patricia Solis at Arizona State University (ASU). In this project, we studied the state and potential of crowdsourced (volunteered) spatial data initiatives in Central America. 

Panama's geographc institute, Tommy Guardia, had an outdated database for street names in Panama City, which hindered infrastructure improvement plans. I spearheaded a volunteered spatial data initiative to update this street name database. Technically, I coded the data collection app, set up data servers, and created name-matching algorithms. More broadly, I partnered with Tommy Guardia and reached out to University of Panama students to kickstart data collection. I used my Spanish skills to complete most of my work, from coding user interfaces to interviewing stakeholders and presenting my findings. 

I synthesized my findings into the Panama section of this publication by ASU's Knowledge Exchange for Resilience, and presented my case study to the Pan American Institute of Geography and History.