First-Degree Price Discrimination Using Big Data

Person-specific pricing had until recently rarely been observed. The reason, that reservation values were unobtainable, may no longer be true now that massive datasets tracking detailed individual behavior exist. Hence, a fundamental change in the way goods are priced may be underway. I investigate this claim in one context. I show demographics, which in the past could be used to personalize prices, poorly predict which consumers subscribe to Netflix. By contrast, modern web-browsing data, with variables such as visits to Amazon.com and internet use on Tuesdays variables which reflect behavior do substantially better. I then present a model to estimate demand and simulate outcomes had personalized pricing been implemented. The model is structural, derived from canonical theory models, but resembles an ordered Probit, allowing a new method for handling massive datasets and addressing overfitting and high dimensionality. Simulations show using demographics alone to tailor prices raises profits by 0.8%. Including nearly 5000 potential website browsing explanatory variables increases profits by much more, 12.2%, increasingly the appeal of tailored pricing, and resulting in some consumers paying double the price others do for the exact same product. Implications for the overall economy and its structure are discussed. (JEL: D42, L130) (