The local linear ridged estimator for E(Y|X) and the local quadratic ridged estimator for E'(Y|X) (the first derivative) with one-d covariate.
- Fully data-driven bandwidth selection using the plug-in idea with cross validation pilot bandwidth.
- There is a ridge term in the denomiator of the final estimate to prevent a too small denominator.
- The kernel is Gaussian.
X: n*1 covariate vector
Y: n*1 outcome vector
[a,b]: the interval for which we estimate E(Y|X), [q_0.025 and q_0.975] as default
t: the vector where we evaluate the estimator
f1: the vector of estimated E(Y|X=x)
f2: the vector of estimated E'(Y|X=x)
- Fan, J and Gijbels, I. (1996). Local polynomial modelling and its applications, vol 66. CRC Press.
- Lin, Z and Yao, F. (2020). Functional regression on the manifold with contamination. Biometrika.