Preliminary Draft
Abstract
As technology enables more frequent price changes, dynamic demand consequences from pricing decisions become increasingly important to firms. Consumers’ reference dependence on past prices is believed to have such a dynamic effect. I provide the first field evidence of the causal reference price effect through a large-scale multi-stage experiment in partnership with a food vending machine company. In a first stage, I create long-term exogenous cross-sectional variation in prices to initialize reference prices. In a second stage, I then periodically implement a common sale price to isolate the impact of the reference price on demand and price-sensitivity. I use the data to test for and measure the reference price effect non-parametrically: a 10% increase in the first-stage reference price improves the second-stage net revenue and quantity sales by about 1 - 3%. To flesh out the managerial implications, I then use the experimental data to estimate a micro-founded structural model of demand with reference prices based on Kőszegi and Rabin (2006). Using these demand estimates, I conduct a case study simulating counterfactual promotional policies under two scenarios: firm accounting for reference dependence versus ignoring it with a myopic model. I find that being reference-aware yields 6.56% higher profits through less frequent but more strategic promotions. The key mechanism is that myopic firms misattribute loss aversion to price sensitivity, leading to excessive promotions that erode reference prices and long-term profitability.

Presented at 2025 ISMS Marketing Science Conference. Draft is coming soon!
Abstract
We investigate how data ownership structures affect the implementation of personalized pricing in consumer packaged goods (CPG) markets, where retailers typically control customer data through loyalty programs. Using a game-theoretic model calibrated with IRI data from five CPG categories, we analyze the strategic interactions between oligopolistic manufacturers and a monopolistic retailer. Our model incorporates a multi-stage game where the retailer can offer personalized pricing services to manufacturers for a fee. We employ machine learning techniques, specifically Bayesian Bootstrapped Lasso, to estimate heterogeneous consumer preferences and simulate equilibrium outcomes under different pricing regimes. Our findings reveal that full personalized pricing where every brand gets personalized, does not always emerge as an equilibrium outcome, particularly when retailers charge differential access fees. However, the equilibrium outcome almost always increases consumer welfare compared to uniform pricing, with progressive distributional effects benefiting lower-income consumers more. The results suggest that market structure and the misalignment between retailer and manufacturer incentives play crucial roles in determining personalized pricing adoption. These insights have important implications for public policy debates surrounding data privacy and price discrimination.

Draft, Under Review.
Abstract
This paper uses a sequence of three randomized controlled trials (RCTs) to support a small online business’ decision of whether and how to implement AI in the creation of email marketing content. To accommodate the small samples available for testing in the small-business setting, the authors leverage recent developments in frequentist statistical decision theory. The RCTs comprise three test policy cells with email content created by (i) salaried writers (“human”), (ii) a large-language model (“LLM”), and (iii) a “hybrid” combination of a human editing the content created by the LLM, respectively. When a “no email” control policy is included, all three test cells approximately double the gross profits from orders relative to the control cell. Across the RCTs, the hybrid cell varies by whether it is edited by a salaried writer or the marketing team. The LLM cell is varied by whether the AI is pre-trained using historic emails or uses a prompt-based generative pre-trained transformer (“GPT”). Across all three RCTs, an AI policy with automated content generation is always selected over the human cell, based on total annual profit net of related labor and software overhead. Most notably, the prompt-based GPT outperforms the human writers by 8 to 9%.

Abstract
We use variation in the growth of broadband to study how the rise of information technology has shaped advertising. From 2010 to 2019, broadband rates increased from 70% to 80%. At the same time, TV advertising remained steady while spending as a proportion of advertising has decreased in other channels. Our results show that a 10 percentage point increase in broadband increases spending in offline advertising in that DMA by approximately 1.2%. Similarly, a 10 percentage point increase in broadband increases spending in TV advertising in that DMA by approximately 1.1%. We show that broadband decreases print advertising in most DMAs.

Abstract
The COVID-19 pandemic caused supply chain disruption and an unprecedented surge in demand due to panic-buying and stockpiling. Both factors have contributed to the skyrocketed grocery prices during and post-COVID in the US, which increased the cost of living across the nation. We utilize nearly 2 decades of scanner data to assess the change in the consumer price index in the grocery industry and examine the differential effects of inflation on households with different income levels. Low-income households could be more sensitive to price and switch to cheaper products more quickly, but there might be limited options they could turn to. Therefore, the differential impact of inflation is ambiguous. To better understand consumers’ price sensitivity, we also examine CPI at the package level, which helps understand consumers’ various response to “shrinkflation.”