Code for McClean et al. 2023 ``Double Cross-fit Doubly Robust Estimators: Beyond Series Regression''
The code conducts two analyses:
-
It generates data from Doppler nuisance functions and estimate the Expected Conditional Covariance with k-Nearest Neighbors. This analysis shows that undersmoothing (i.e., small k) will be optimal with double cross-fitting (i.e., independent training datasets).
-
It generates Holder smooth functions and esitmated the Expected Conditional Covariance with local polynomial regression and covariate-density-adapted local polynomial regression. The figures confirm the theoretical results in Theorems 1, 2, and 3 of the paper.