Title: AMP meets expectiles: Testing heteroscedasticity and asymmetry in high dimensions
Authors: Jing Zhou - KU Leuven (Belgium) [presenting]
Hui Zou - University of Minnesota (United States)
Abstract: Heteroscedasticity is commonly observed in high-dimensional data and has been understudied. Specifically, testing heteroscedasticity and asymmetry of the error distribution in high dimensions remains underexplored. We introduce an expectile-based test that exploits the conditional distribution of the response variable at different expectile levels. We propose to use the approximate message-passing algorithm to perform an asymptotic analysis for the proposed test, assuming the sample size and the number of predictive variables follow a linear growth rate. The numerical performance of the proposed test will be validated by using simulated and real data.