Abstract
In a rapidly changing world, older data is not as informative as the most recent data. This is known as the concept drift problem in statistics and machine learning. How does a firm adapt in such an environment? To address this research question, we propose a generalized revealed preference approach. We argue that by observing a firm’s choices, we can uncover the way the firm uses the past data to make business decisions. We apply this approach to study how Prosper Marketplace, an online P2P lending platform, adapts in order to address the concept drift problem. More specifically, we develop a two-sided market model, where Prosper uses the past data and machine learning techniques to assess borrowers' and lenders' preferences, and then classify their loans by risk ratings and set their interest rates accordingly to maximize her expected profit. By observing Prosper's choices over time, we find evidence that Prosper likely assigns different weights to past data points depending on how close the economic environments that generate the data are to the current economic environment. In the counterfactual, we demonstrate that Prosper may not be using the past data optimally, and it could improve its revenue by 9.02% if it uses the Ensemble Hidden Markov model proposed in our study.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call