Research

Work in Progress

What is the value of attention? Supply and demand estimation of attention in a mobile phone setting:

I investigate the digital market for attention, documenting large heterogeneity in the use of attention as payment in mobile games. By using a detailed event-level dataset, I estimate user elasticities for a 30-second Rewarded Video, which is my unit of attention. In the aggregate, the elasticity of a Rewarded Video is 2.2 with regard to the in-game currency of the mobile game. When accounting for individual heterogeneity the elasticity reduces to 0.8. I further find that the elasticity of a Rewarded Video is the largest during evenings. By relating the use of Rewarded Videos to the price paid by the advertiser to show the ad to the user, I find that the attention of individuals who watch more Rewarded Videos is valued less by the advertisers. Finally, I model the willingness to pay to avoid a Rewarded Video to 9.71 in-game coins, corresponding to about 0.1 Euro. This is of the same magnitude as the value of time, as previously documented for the value of travel time.

Market Definitions in the Real-estate Agents Market, A Data-driven Approach Using Statistical Learning (with Adam Lindhe):

We define geographic markets using a machine learning clustering algorithm. The algorithm is an unsupervised learning algorithm that uses data on customers’ location and seller identity to define markets. The novelty of our method is that we leverage the seller identity of each observation to improve on pure distance-based measures of markets. We impose the intuition that a seller will focus on a few geographic markets and incorporate it into a Bayesian method. We empirically implement the method using a Gibbs sampler and estimate the geographic markets for real estate agents in Sweden. We find that our algorithm does significantly better than the standard geographic clustering algorithm, K-means. We get a Dice score of 0.78 compared to 0.67. We also find a distribution of the number of markets each agent works in closer to that of the validation set. The baseline K-means market definition indicates a lower market concentration than our market definition, where ours is closer to that of the validation set. Finally, we investigate the correct number of clusters and find that it in our example corresponds to the general knowledge of how the market is geographically structured.