【时间】2019年6月12日（周三）下午 2：30
Yan Huang is an Assistant Professor of Business Technologies at Tepper School of Business, Carnegie Mellon University. In Yan's research, she uses quantitative methods to examine the economic and social impacts of technologies (especially AI and crowd-based technologies) and the mechanisms behind them, and, based on these understandings, recommends more productive use of technologies and more effective design of technology platforms and applications. Yan holds a bachelor's degree from Tsinghua University and a Ph.D. from Carnegie Mellon University. Prior to joining the Tepper School, she was an Assistant Professor of Technology and Operations at Ross School of Business, University of Michigan.
【Speaker】 Yan Huang, Assistant Professor of Business Technologies at Tepper School of Business, Carnegie Mellon University.
【Topic】Crowd, Lending, Machine, and Bias
【Time】Wednesday, June 12, 2019, 14:30
【Venue】Room 453, Weilun Building, Tsinghua SEM
【Organizer】Department of Management Science and Engineering
【Abstract】 Prediction markets have shown great predictive accuracy in many cases, and the wisdom of crowd is well recognized. Can machine learning algorithms outperform crowd decisions? In this paper, we overcome the "selective-label" problem, and show that a reasonably sophisticated machine learning model predicts loan default more accurately than the crowd investors on a P2P lending platform. Our results suggest potentially large welfare gains from replacing crowds with machines: policy simulations show an 8.13% increase in return on investment for investors with a decreased average interest rate for borrowers when machine predictions are used to set the interest rate, or a 6.39% increase in return on investment for investors with an increased funding opportunity for borrowers with few alternative funding options when machine predictions are used to select loans. Our results also reveal that both the machine learning algorithm and the crowd are biased in decision making against minorities and female. We propose a general and effective “de-bias”method, and show that the de-biased algorithm, which suffers a loss in prediction accuracy, still leads to a higher total return as compared with the crowd decisions.
|
