Research
Fields of Interest: Environmental Economics, Public Economics, and Urban/Spatial Economics
Beyond the Neighborhood: Regional Amenities in Sorting Models
Equilibrium sorting models treat amenities as localized; they are enjoyed equally within a geographic unit and not at all outside of it. However, many environmental amenities generate diffuse benefits across broad regions. Incorporating such regional amenities into sorting models requires solving a fundamental problem: at what rate do benefits attenuate with distance and when are they negligible? I demonstrate empirically that arbitrary decisions made by the researcher about an amenity's market extent can cause estimates of its value to vary by orders of magnitude. I address this problem by endogenizing value attenuation with the results of a recreation demand study, microfounding a spatial index that estimates the exposure of any given location to a complete network of amenities. To demonstrate, I estimate a model of residential sorting across Iowa and find not only that people have significant willingness to pay for improvements in lake access ($5,945 for a one standard deviation increase), but that failing to account for them threatens the exclusion restriction of the standard sorting instrument; failing to account for them attenuates the coefficient on price by 9%. In counterfactuals I simulate the creation of a lake, and my estimated model shows that interregional variation in supply elasticity allows the resulting demand shift to be absorbed by higher-elasticity locations, softening the price response everywhere and raising concerns for the ability of hedonic models to fully capitalize the value of regional goods even in low-elasticity places. Finally, I show the importance of spatially modeling environmental goods by simulating cleanup projects at two lakes that differ sharply in substitute availability: the isolated lake's cleanup is worth roughly 20 times more than the identical water-quality improvement at the lake with substitutes nearby.
Darkening the Mood: Are People Willing to Pay for Dark Skies?
The invention of the lightbulb is widely credited as the primary driver of widespread residential electrification; to this day, when new communities are connected to power grids, the first purchases are often lighting devices. Indeed, light is so correlated with wealth and productivity that economists regularly use remote-sensed night sky brightness as a proxy for GDP in hard-to-reach locations. Yet the popular use of the term “light pollution” implies some unease with the trend of lighting up the night. This paper uses nationwide cell phone mobility data to quantify willingness to pay for sky darkness through a recreation demand study of campground visitation. Exploiting spatial variation in sky brightness from remote-sensing data and temporal variation from lunar phase and position, modeled with a full astronomical brightness model, I estimate a willingness to pay of roughly $10–15 for a one magnitude per square arcsecond increase in darkness on an ordinary night, rising to approximately $25 during a week containing a meteor shower. I find that the increases in light pollution since 2019 incur $833 million in recreation damages annually.
When Can We Trust Prediction Markets?
Prediction markets are increasingly cited as real-time probability estimates, but existing evaluations have been limited to small numbers of markets or comparisons against a single alternative forecast. This paper proposes a marketwide bias index, varying between 0 (perfect probability prediction) and 1 (as good as random guessing), that measures systemic miscalibration, and applies it to a sample of Polymarket contracts: 43,028 markets closing between April 15 and September 15, 2025, with 82.5 million minute-level price observations. I find substantial miscalibration overall (0.172) alongside striking heterogeneity across categories — crypto markets are nearly perfect (0.096) while golf markets are severely biased (0.601). Generally, sports markets underperform non-sports markets throughout. Bias also falls as markets approach resolution and as trading volume rises. I undertake an exercise to determine optimal observations by volume, time to close, and a sports indicator to minimize systemic miscalibration, and find optimal weighting cuts miscalibration by roughly in half (0.074), suggesting that prediction market prices are best treated not as probabilities but as signals whose reliability is predictable from observable market characteristics.