"Open(?) AI" Latest available draft: November 2024. [Dropbox]
Open AI models allow users to download the parameters and fine-tune the model for a specific purpose. When would the developer of such a ``foundation'' model prefer to do this?
Embed software customization in a framework of general and niche products. Fine-tuning makes a model more niche.
Examine the distortions in the direction and quantity of innovation that arise when the developer community fine-tunes vs. when a founder designs the model privately.
"Economic Inequality and Market Power" Latest available draft: November 2024. [Dropbox][SSRN] [Slides]
Microfound the idea that poor consumers are price sensitive because dollars mean more. Nest this in a standard partial equilibrium IO framework that includes mixed logit.
Now, there is causal mapping from wealth to demand via price sensitivity, and possibly correlated wealth and tastes.
Develop a set of axioms that are equivalent to an inequality-adjusted consumer surplus that is utilitarian in nature. In other words: marginal social welfare weights are the price coefficients.
Second: develop sufficient conditions for increases in inequality to increase markups. In future: empricial exercise.
"Algorithmic Competition, with Humans" Latest available draft: October 2024.
[Dropbox] [Old Slides]
Pricing algorithms = delegation of pricing to automatic rules. This implies commitment and can soften competition.
If managers can override algorithms, commitment is eroded. I show firms can generally design override-proof pricing algorithms that raise prices.
The form these algorithms take depends on primitives (e.g., demand) and equilibrium selection rules, but they usually entail collusion by algorithm or one-sided override Stackelberg-type pricing.
Extensions to positive override costs, dynamic pricing, and demand or marginal cost uncertainty. Uncertainty can break algorithms and cause Bertrand reversion if override is costless.
Overall, commitment and prediction are substitutes: an algorithm featuring one or the other is enough to sustain supracompetitive prices.
How should an environmental regulator target costly inspections? Examine a regulation that targets past-violator plants (escalation) and plants co-owned with past-violator plants (linking).
Estimate structural model with dynamic moral hazard (hidden pollution actions) and interdependent payoffs via linked regulation.
Methodological innovation: use continuation value sufficiency to collapse the state space
Targeting historic bad actors is better than random inspections, and linking is better than just targeting. Massive efficiency improvements!
Why are some firms better at predicting demand than others when setting prices?
New methodology leveraging common-knowledge demand shifts. to identify information structure and mapping from information to prices (price policy functions).
Apply the method to hotel industry. Turns out, hotels vary widely in demand prediction abilities. This matters for productivity, and it doesn't tend to advantage big chains with big data capabilities.
"Volatility, Uncertainty, and Hotel Capacity" Latest available draft: June 2020.
[Dropbox] [Slides]
Distinguish between demand volatility (fixed capacity, adjustable prices) and uncertainty (fixed capacity, nonresponsive prices) and the differential effects on the incentives to build more capacity.
Theory gives two countervailing effects: insurance effect means uncertainty makes you build more to avoid stockouts, while productivity effect means uncertainty makes capacity less valuable and makes you build less.
The insurance effect dominates if and only if capacity is costly to build.
Test the theory using data from the hotel industry. Here, capacity is cheap, so the productivity effect dominates. More uncertain markets build less.
In Progress
"Inequality, Market Power, and the Pain of Supply Shocks: Evidence from the Egg Crisis" Working. Email me if you're interested!
The 2022 Bird Flu Pandemic dramatically impacted egg supply. What role did inequality between poor amd wealthy consumers play in amplifying these effects?
Build on my previous work and a novel dataset to (1) separately estimate the joint distribution of wealth and tastes and the causal effect of wealth on egg demand; (2) quantify the economic incidence of the pandemic on rich vs. poor consumers and adjust estimates of harm accordingly; (3) examine the role inequality played in pass-through.
"Automatic Price Cuts" Early stage
Gasoline pricing has a peculiar pattern: prices decline by uniform amounts in regular intervals before jumping irregularly.
Develop a theory where firms prefer to set automatic price cuts punctuated by managerial override to explain these patterns.
Resting Papers
"Informational Complementarities and IT Arms Races" Draft available upon request.
Examine firms' information-gathering in equlibrium. Theory suggests that if one firm gets better at predicting demand its rivals may try to do the same.
In the hotel industry, equilibrium effects account for over half the variation in demand prediction quality across hotels.
This suggests that if one or two large hotel chains achieve an improvement in demand forecasting ability, they may incite an IT arms race where everyone scrambles to improve as well.