some PAPERS I LIKE
On the spatial impact of global warming (Rossi-Hansberg and Desmet) | 2017, Journal of Urban Economics, LINK
A macroeconomic model with a financial sector (Brunnermeier and Sannikov) | 2014, AER, LINK
Speculative fever: investor contagion in the housing bubble (Bayer, Mangum and Roberts) | 2021, AER, LINK
Dynamic interpretation of emerging risks in the financial sector (Hanley and Hoberg) | 2019, Review of Financial Studies, LINK
Asset markets with heterogenous information (Kurlat) | 2016, Econometrica, LINK
CHARTS
State-space vizualization
PANDEMICS: I worked on a computational macro model with [Susceptible, Infected, Z=Recovered] module. This model uses value function iteration to solve the dynamic problem of a Susceptible agent – when Susceptible your consumption today affects your consumption tomorrow, since in consuming you increase your chance of becoming infected.
Based on the idea of a coauthor, I developed this chart to show how the value function changes across the state space (can be represented on the unit simplex as S+I+Z=1). It shows that states with high Infected lead to lower lifetime utility realizations – since in such states the chance of also becoming infected is high. This creates a computational incentive to avoid consumption when these states are at high likelihood.
This chart is useful to corroborate the intuition baked into the model. It is also useful as a diagnostic tool to make sure your optimization is creating solutions you would expect.
Balance testing
ASYLUM: In my work on migration and the US asylum system, our work entails using judges as instruments since institutionally judges are randomly assigned. However, it is important to validate that the judges are indeed randomly assigned.
Usually simple approaches are used but our context required us to develop a better method. We ran thousands of Judge A – Judge B mean difference tests and examined how often there was a significant difference between mean characteristic of cases seen by Judge A and those seen by Judge B across all judge pairs within a court and year.
The charts above were our way to summarize this mass of information to communicate in how many judge pairs the test appeared to fail. We would expect that few of the tests failed i.e. few showed a significant difference between the mean case characteristic for Judge A and that for Judge B. The charts plot the CDFs of these test p-values for three different case characterstics – applicant gender, or nationality being Chinese or Northern Triangle (two large nationality groups). The charts show that in most cases only a small percentage of cases (y-axis) have p-values less than 0.1 (x-axis) – e.g. of mean difference tests of judge pairs where judge A has between 40% and 60% male cases, 18% of cases fail the test at the 10% (or 0.1) level of significance. This helped us to understand if certain groups failed balance e.g. weekend cases.