UCL School of Management is delighted to welcome Xinlong Li, Toronto to host a seminar discussing “How Does A Firm Learn in A Changing World? The Case of Prosper Marketplace”.
Most marketing and economic research makes the assumption that models built on historical data can predict consumer choice in the future. However, consumer tastes might change over time, especially due to the digital innovations which have revolutionized the world. Therefore, treating all historical data equally may lead to a seriously misspecified model for the current market. This problem is referred to as Concept Drift in machine learning. In this paper, using a rich micro-level dataset, machine learning techniques (Naïve Bayes Classifier, Ensemble Modeling, and Hidden Markov Model) and structural modeling, we find evidence that Prosper Marketplace, which is one of the largest online peer-to-peer (P2P) lending platforms in the U.S., recognizes the concept drift problem when developing their model to assess borrowers’ risk and set interest rates. More specifically, we find that Prosper selectively uses the data based on a moving window. In the counterfactual experiments, we demonstrate that we can improve Prosper’s revenue by changing the way Prosper uses the data.