Quantile function regression and variable selection for sparse models. Issue 4 (23rd April 2021)
- Record Type:
- Journal Article
- Title:
- Quantile function regression and variable selection for sparse models. Issue 4 (23rd April 2021)
- Main Title:
- Quantile function regression and variable selection for sparse models
- Authors:
- Yoshida, Takuma
- Abstract:
- Abstract : This article considers linear quantile regression and variable selection for high‐dimensional data. In general, an ordinary quantile regression estimator is obtained for a single, fixed quantile level. Therefore, the estimated coefficient does not have continuity with respect to the quantile level, and hence, the behaviour of the estimator and estimated active variable set could change rapidly for different but sufficiently close quantile levels. To obtain a stable estimator for a given quantile level, this study proposes a new quantile regression method to estimate the coefficient as a function of the quantile level of interest in a given region Δ ⊂ ( 0, 1 ), which is denoted quantile function regression. In quantile function regression, we approximate the coefficient function of the quantile level using a B ‐spline model, and hence, the estimated conditional quantile is continuous as it is a B ‐spline curve. To employ variable selection, a group lasso‐type sparse penalty is used to estimate a non‐zero coefficient function of the quantile level, which indicates the estimated active set that remains unchanged in Δ . Therefore, quantile function regression can achieve global variable selection. The proposed estimator exhibits an asymptotic rate of convergence and consistency in variable selection. Simulation studies and applications to real data further reveal that the proposed method yields good performance. Résumé : L'auteur considère la régression linéaireAbstract : This article considers linear quantile regression and variable selection for high‐dimensional data. In general, an ordinary quantile regression estimator is obtained for a single, fixed quantile level. Therefore, the estimated coefficient does not have continuity with respect to the quantile level, and hence, the behaviour of the estimator and estimated active variable set could change rapidly for different but sufficiently close quantile levels. To obtain a stable estimator for a given quantile level, this study proposes a new quantile regression method to estimate the coefficient as a function of the quantile level of interest in a given region Δ ⊂ ( 0, 1 ), which is denoted quantile function regression. In quantile function regression, we approximate the coefficient function of the quantile level using a B ‐spline model, and hence, the estimated conditional quantile is continuous as it is a B ‐spline curve. To employ variable selection, a group lasso‐type sparse penalty is used to estimate a non‐zero coefficient function of the quantile level, which indicates the estimated active set that remains unchanged in Δ . Therefore, quantile function regression can achieve global variable selection. The proposed estimator exhibits an asymptotic rate of convergence and consistency in variable selection. Simulation studies and applications to real data further reveal that the proposed method yields good performance. Résumé : L'auteur considère la régression linéaire quantile et la sélection de variables pour les données en haute dimension. Règle générale, un estimateur de régression quantile ordinaire est obtenu pour un seul quantile fixe. Ainsi, il n'y a pas de continuité des coefficients estimés par rapport au niveau du quantile, ce qui signifie que les estimés et les variables actives peuvent changer pour des niveaux de quantile différents, mais suffisament proches. Afin d'obtenir un estimateur stable pour un niveau de quantile donné, l'auteur propose une nouvelle méthode de régression dite de fonction quantile, qui estime le coefficient comme une fonction du niveau quantile dans une région donnée Δ ⊂ ( 0, 1 ) . Dans la régression de fonction quantile, la fonction coefficient du quantile est approximée par un modèle de B ‐spline, ce qui signifie que le quantile conditionnel est continu puisqu'il s'agit d'une courbe de B ‐spline. Pour la sélection de variable, une pénalité creuse de type lasso par groupe permet d'estimer une fonction coefficient non nulle pour les niveaux du quantile, indiquant l'ensemble de prédicteurs actifs qui demeurent inchangés dans Δ . Ainsi, la régression de fonction quantile permet de faire une sélection globale de variables. Les auteurs présentent des simulations et l'analyse de données réelles afin de révéler que la méthode proposée offre de bonnes performances. … (more)
- Is Part Of:
- Canadian journal of statistics. Volume 49:Issue 4(2021)
- Journal:
- Canadian journal of statistics
- Issue:
- Volume 49:Issue 4(2021)
- Issue Display:
- Volume 49, Issue 4 (2021)
- Year:
- 2021
- Volume:
- 49
- Issue:
- 4
- Issue Sort Value:
- 2021-0049-0004-0000
- Page Start:
- 1196
- Page End:
- 1221
- Publication Date:
- 2021-04-23
- Subjects:
- Adaptive group lasso -- ∖special t4ht@.B‐spline -- Quantile regression -- sparse model -- variable selection
Mathematical statistics -- Periodicals
519.5 - Journal URLs:
- http://archimede.mat.ulaval.ca/cjs/ ↗
http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1708-945X/issues ↗
http://www.jstor.org/journals/03195724.html ↗
http://onlinelibrary.wiley.com/ ↗
http://www.ingentaconnect.com/content/ssc/cjs ↗
http://www.mat.ulaval.ca/rcs/indexe.shtml ↗ - DOI:
- 10.1002/cjs.11616 ↗
- Languages:
- English
- ISSNs:
- 0319-5724
- Deposit Type:
- Legaldeposit
- View Content:
- Available online (eLD content is only available in our Reading Rooms) ↗
- Physical Locations:
- British Library DSC - 3035.760000
British Library DSC - BLDSS-3PM
British Library HMNTS - ELD Digital store - Ingest File:
- 24519.xml