An Automatic Non‐parametric Spectrum Estimator

An estimator of the spectral density of a stationary process is obtained by approximating it with a step function. The positions of the level changes are determined using partitioning algorithms. The algorithms are stopped by using criteria such as AIC. In examples, the resulting estimates are shown to be good representations of the true spectra in most circumstances, even when the series are fairly short. In particular the estimates highlight spikes in the spectral density that are caused by periodicities.