Abstract
This paper proposes plug-in bandwidth selection for kernel density estimation with discrete data via minimization of mean summed square error. Simulation results show that the plug-in bandwidths perform well, relative to cross-validated bandwidths, in non-uniform designs. We further find that plug-in bandwidths are relatively small. Several empirical examples show that the plug-in bandwidths are typically similar in magnitude to their cross-validated counterparts.
Original language | English (US) |
---|---|
Pages (from-to) | 199-214 |
Number of pages | 16 |
Journal | Econometrics |
Volume | 3 |
Issue number | 2 |
DOIs | |
State | Published - Jun 1 2015 |
Keywords
- Bandwidth selection
- Discrete variable
- Kernel
- Nonparametric
- Plug-in
ASJC Scopus subject areas
- Economics and Econometrics