Empirical evaluation of sup-norm risk bounds for NW Kernel, Local Polynomial, Wavelet, and Spline estimators across regularity regimes.
| Estimator | h=0.02 Ratio | h=0.05 Ratio | h=0.1 Ratio | h=0.2 Ratio | h=0.3 Ratio | h=0.5 Ratio |
|---|---|---|---|---|---|---|
| NW Kernel | 0.049 | 0.094 | 0.118 | 0.138 | 0.154 | 0.189 |
| Local Poly | 0.049 | 0.094 | 0.117 | 0.138 | 0.153 | 0.189 |
| Wavelet | 0.275 | 0.302 | 0.303 | 0.278 | 0.257 | 0.204 |
| Spline | 0.150 | 0.176 | 0.200 | 0.237 | 0.274 | 0.384 |
| Finding | Details |
|---|---|
| Best Performance | NW Kernel and Local Poly achieve ratios below 0.05 at small bandwidths |
| Bounds Validity | All empirical risks remain well below theoretical bounds (ratio < 0.384) |
| Regime Behavior | Kernel methods excel at small h (r<h regime); splines degrade at large h |
| Wavelet Behavior | Wavelets show relatively stable ratios across bandwidths (~0.2-0.3) |