0000000000243152
AUTHOR
Paolo Frumento
A penalized approach to covariate selection through quantile regression coefficient models
The coefficients of a quantile regression model are one-to-one functions of the order of the quantile. In standard quantile regression (QR), different quantiles are estimated one at a time. Another possibility is to model the coefficient functions parametrically, an approach that is referred to as quantile regression coefficients modeling (QRCM). Compared with standard QR, the QRCM approach facilitates estimation, inference and interpretation of the results, and generates more efficient estimators. We designed a penalized method that can address the selection of covariates in this particular modelling framework. Unlike standard penalized quantile regression estimators, in which model selec…
Parametric estimation of non-crossing quantile functions
Quantile regression (QR) has gained popularity during the last decades, and is now considered a standard method by applied statisticians and practitioners in various fields. In this work, we applied QR to investigate climate change by analysing historical temperatures in the Arctic Circle. This approach proved very flexible and allowed to investigate the tails of the distribution, that correspond to extreme events. The presence of quantile crossing, however, prevented using the fitted model for prediction and extrapolation. In search of a possible solution, we first considered a different version of QR, in which the QR coefficients were described by parametric functions. This alleviated th…
Non-crossing parametric quantile functions: an application to extreme temperatures
Quantile regression can be used to obtain a non-parametric estimate of a conditional quantile function. The presence of quantile crossing, however, leads to an invalid distribution of the response and makes it difficult to use the fitted model for prediction. In this work, we show that crossing can be alleviated by modelling the quantile function parametrically. We then describe an algorithm for constrained optimisation that can be used to estimate parametric quantile functions with the noncrossing property. We investigate climate change by modelling the long-term trends of extreme temperatures in the Arctic Circle.
Robust estimation and regression with parametric quantile functions
A new, broad family of quantile-based estimators is described, and theoretical and empirical evidence is provided for their robustness to outliers in the response. The proposed method can be used to estimate all types of parameters, including location, scale, rate and shape parameters, extremes, regression coefficients and hazard ratios, and can be extended to censored and truncated data. The described estimator can be utilized to construct robust versions of common parametric and semiparametric methods, such as linear (Normal) regression, generalized linear models, and proportional hazards models. A variety of significant results and applications is presented to show the flexibility of the…