Conditional Quantile Estimation with Neural Network Structure

Open Access
Feng, Yijia
Graduate Program:
Master of Science
Document Type:
Master Thesis
Date of Defense:
Committee Members:
  • Dr Runze Li, Thesis Advisor
  • Runze Li, Thesis Advisor
  • Nonparametric Regression
  • Quantile Regression
  • Neural Network
  • Robust Estimation
In this thesis, we apply neural network method to estimate nonparametric conditional quantile under the quantile regression loss, i.e., the check loss function. The proposed robust neural network (RNN) method integrates quantile regression and neural network together, and is a useful modelling tool. We further apply an majorization-minization (MM) algorithm (Hunter & Lange, 2000) to deal with the minimization of RNN. Monte Carlo simulation study is conducted to examine the performance of the proposed robust neural network. From our simulation results, we found that the RNN method is promising. The proposed procedures are illustrated and compared with two popular nonparametric methods, local linear and regression splines, by a real data example in credit card portfolio analysis.