【Metrics Seminar】Utility-Maximizing Binary Prediction via Nearest Neighbor Method

  • 研討會日期 : 2022-11-22
  • 時間 : 14:30
  • 主講人 : Professor Jiun-Hua Su (蘇俊華)
  • 地點 : Conference room B110
  • 主持人 : Professor Le-Yu Chen
  • 演講者簡介 : Professor Jiun-Hua Su received his Ph.D. in Economics from University of California, Berkeley in 2018. He is currently an Assistant Research Fellow at the Institute of Economics, Academia Sinica. His research interests are Econometrics and Machine Learning.
  • 演講摘要 : We propose a nonparametric k-nearest neighbor prediction rule under the framework of utility-maximizing binary prediction with possibly many predictors. The proposed prediction rule is constructed based on the predictors selected by various variable selection methods. We establish that the proposed prediction rule, allowing for the data-dependent selection of parameter k, is utility consistent under regularity assumptions. Simulation results suggest that the proposed prediction rule, along with the adaptive component selection and smoothing operator (COSSO) in Lin and Zhang (2006), is utility consistent and outperforms the complexity-penalized utility-maximizing prediction rule in Su (2021).