Research
What is the value of attention? Supply and demand estimation of attention in a mobile phone setting.:
I study the digital market for attention in a freemium mobile game where users choose between paying with money or by watching 30-second video ads. Using unique event-level data, I estimate consumer’s supply elasticity of attention. In the aggregate, a one percent higher price increases the share of payments by users watching videos by 2.2 percent. A substantial part is due to individual heterogeneity in tastes. Accounting individual heterogeneity reduces the elasticity to 0.5. The individual elasticities vary throughout the day, peaking in the evening. By analyzing the prices advertisers pay for each individual ad shown to consumers, I find advertisers pay more for users less likely to watch ads, with a stronger effect among Android users than iOS users. Finally, I estimate the willingness to accept a 30-second ad at 0.15 euros. This is of similar magnitude as previous estimates of consumer’s valuation of time.
Market Definitions in the Real-estate Agents Market, A Data-driven Approach Using Statistical Learning. (with Adam Lindhe):
This paper introduces a novel method to define geographic markets using machine learning.
Using an unsupervised learning approach, we cluster sales based on customers’ locations, with each cluster representing a distinct market. The novelty of our method is that we leverage the identity of the seller for each observation to capture market structures that cannot be captured by today’s methods. We integrate the assumption that sellers focus on a few geographic markets into our Bayesian framework and empirically implement the method using a Gibbs sampler. Estimating the geographic markets for real estate agents in Stockholm, our algorithm does significantly better in correctly classifying sales than the baseline K-means algorithm, achieving a Dice score of $0.78$ compared to $0.67$. We find that the number of markets each agent works in is distributed more similarly to the industry knowledge in our classification than in the baseline comparison. Our method classifies the markets such that market concentration, as measured by the Herfindahl-Hirschman Index (HHI), is closer to the market concentration calculated using the correct classification than the baseline K-means, thereby improving understanding of market power and competition dynamics. Finally, we investigate the correct number of clusters and find that, in our example, it corresponds to the established knowledge of the market’s geographic structure.
Work in Progress
School openings and teacher mental health. (with Helena Svaleryd and Jonas Vlachos)
Loneliness, Alzheimer and Financial Distress. (with Marieke Bos and Andrew Herzberg )
The impact of remedial education on educational achievements. (with Shubhaa Bhattacharyya)