Boosting Multiple Testing for high-dimensional model selection
An R implementation of sequential variable selection for linear, logit, and probit specifications, with always-in controls and optional HAC inference.
The BoostingMultipleTesting package implements a sequential model-selection procedure for high-dimensional linear and binary-response regressions. Candidate variables are ranked through regression-based test statistics and admitted subject to a multiple-testing critical value.
| Primary function | boosting_glm() |
|---|---|
| Models | Linear regression, logit, and probit |
| Always-in variables | Supported through Xcond |
| Time-series inference | Optional Newey-West HAC standard errors for linear, logit, and probit models |
| Output | A logical vector identifying the selected columns of X |
Boosting with multiple testing
The implementation follows the model-selection frameworks developed for high-dimensional linear and binary-response regression.
Kapetanios, George, Vasilis Sarafidis, and Alexia Ventouri. 2026. Model Selection in High-Dimensional Linear Regression using Boosting with Multiple Testing. arXiv:2602.19705.
Grivas, Charisios, George Kapetanios, Zacharias Psaradakis, Vasilis Sarafidis, Marian Vavra, and Alexia Ventouri. 2026. Nonlinear Boosting with Multiple Testing in High-Dimensional Generalised Linear Models with Binary Responses. arXiv:2607.22440.
The package provides a common interface for linear, logit, and probit specifications. The local documentation separates the methodological description, reproducible simulations, and empirical use of transformed FRED-MD data.
One function, explicit inputs
Supply a numeric response and a candidate-regressor matrix. Optional conditioning variables remain in every auxiliary regression and are not part of the returned selection vector.
library(BoostingMultipleTesting)
selected <- boosting_glm(
y = y,
X = X,
Xcond = NULL,
pval = 0.05,
delta1 = 1,
link = "linear",
HAC = FALSE
)
colnames(X)[selected]