A Bayesian approach for determining the optimal semi-metric and bandwidth in scalar-on-function quantile regression with unknown error density and dependent functional data

Han Lin Shang*

*Corresponding author for this work

Research output: Contribution to journalArticle

6 Citations (Scopus)

Abstract

Extreme values are often associated with tails of a cumulative distribution function, and the study of extreme values and their predictions is an important research topic in climate problems. Through a regression approach, we consider a scalar-on-function nonparametric regression to estimate and predict conditional quantiles, where the regression function can be estimated by the functional Nadaraya–Watson estimator. The accuracy of such an estimator crucially depends on the optimal selections of semi-metric and bandwidth parameters. A Bayesian approach is proposed to simultaneously estimate the bandwidths in the regression function and kernel-form error density. As a by-product of the Bayesian approach, marginal likelihood is used to select the optimal semi-metric. In both independent and dependent functional data, a series of simulation studies shows that the proposed Bayesian approach outperforms the functional cross validation for estimating the regression function, and it performs better than the likelihood cross validation for estimating the error density. The proposed Bayesian approach is utilised in the extreme value analysis for predicting the recurrence interval of maximum temperature at Melbourne Airport, in Australia.
Original languageEnglish
Pages (from-to)95-104
Number of pages10
JournalJournal of Multivariate Analysis
Volume146
DOIs
Publication statusPublished - Apr 2016
Externally publishedYes

Keywords

  • Extreme value prediction
  • Functional kernel regression
  • Kernel-form error density
  • Markov chain Monte Carlo

Fingerprint Dive into the research topics of 'A Bayesian approach for determining the optimal semi-metric and bandwidth in scalar-on-function quantile regression with unknown error density and dependent functional data'. Together they form a unique fingerprint.

Cite this