I am a PhD candidate in Quantitative Marketing. I will be graduating in June 2020.
Job Market Paper
In this paper I show that firms strategically take actions in order to improve their ratings due to the rating's platform rounding system. In my context, Yelp displays ratings to the nearest half star, as is common practice on many ratings platforms. Since the true average rating is not shown, firms have an incentive to remain just above the rounding threshold to have a higher displayed rating. However once they pass the threshold the incentive to improve their ratings drops. I study this in the context of auto repair, a market with a lot of uncertainty on the consumer side, and thus a good market to study reviews, which improve consumer knowledge. I first show that consumers value ratings in this market and that firms' revenue and number of consumers increases with increases in their displayed rating. To overcome identification issues of ratings and quality and to show the causal estimate, I utilize a regression discontinuity and instrumental variable strategy. Because ratings have a significant effect on demand, firms should pay attention to their ratings. I next show evidence that firms take on actions when they are close to roundings thresholds, as can be seen with bunching behavior at these thresholds. Next I build a structural model to quantify these actions and I consider how the firms' incentives would change under different ratings displays.
Information tracking services, such as Fitbit, MyFitnessPal, and Credit Karma, are gaining in popularity as data becomes more easily available and accessible. These services often attract users through the notion that information can help them reach their goals. However, using data from a consumer finance company where individuals sign up to receive information about their credit report, we find that retention is the lowest for individuals who presumably would benefit from information the most - those who have low credit scores. This paper explores when individuals change their demand for information and the impact of information on financial health. Specifically, we first document a causal link between credit score trajectories and the demand for information. A decline in credit score decreases the likelihood the individual checks her credit score in the future. Surprisingly, this "information avoidance" might be rational. Second, through variation induced by the firm’s email campaigns and A/B email copy tests, we show that individuals with a declining credit score experience a further causal decline if they are exogenously nudged to check their credit report.We find the effect of information on future credit score is negative for individuals who had a declining credit score prior to checking information, compared to those who did not have a declining score. In addition, this effect is the most negative for individuals with low credit scores. This finding suggests that encouraging people to access information when their credit scores are declining can actually worsen their financial health.
The increasing amount of data available to consumers has most likely aided in decision-making. However, it has also created an opportunity for sellers to design the information landscape that consumers navigate. This paper develops a novel fully dynamic search model for alternatives with multiple characteristics, and reports estimation results for an online used car seller. The model allows characterizing search over alternatives with multiple characteristics that may be distributed arbitrarily. It also allows for a rich set of consumer search behaviors, including piecemeal search within and arbitrary paths across alternatives. We estimate the model using clickstream data on the website of a used car seller. The dataset tracks incremental search actions as well as test-drive reservations. The estimated fundamentals are then used to consider the effects of different information design policies. We find that the choice of the characteristics to be made available to consumers upfront may have conversion implications ranging from -0.39% to +1.65%. The perfect information scenario increases conversion rates by 10.4$\%$. Finally, we compare our model with the knowledge gradient model of learning, and show that taking forward-looking behavior into account explains the moments of the data better, and that the models’ likelihoods are significantly different.
Work in Progress
with Anastasica Buyalskaya, Colin Camerer, Wes Hartmann, Peter Landry, Ryan Webb