Abstract
We develop a method for uniform valid confidence bands of a nonparametriccomponent $f_1$ in the general additive model $Y=f_1(X_1)+\ldots + f_p(X_p) +\varepsilon$ in a high-dimensional setting. We employ sieve estimation andembed it in a high-dimensional Z-estimation framework allowing us to constructuniformly valid confidence bands for the first component $f_1$. As usual inhigh-dimensional settings where the number of regressors $p$ may increase withsample, a sparsity assumption is critical for the analysis. We also runsimulations studies which show that our proposed method gives reliable resultsconcerning the estimation properties and coverage properties even in smallsamples. Finally, we illustrate our procedure with an empirical applicationdemonstrating the implementation and the use of the proposed method inpractice.