Non-parametric estimation
How much structure can be recovered without specifying a functional form?
Question
How much structure can be recovered without specifying a functional form?
Positioning
This is a laboratory experiment in non-parametric estimation, not a new applied model.
Kernel density estimation and Nadaraya–Watson regression are treated as two faces of the same smoothing problem. Bandwidth selection is the parameter that governs smoothness, bias and variance.
Method
The experiment asks how much structure can be recovered from data without imposing a parametric family or a specified functional form.