This procedure estimates the "d" parameter for fractional integration using Robinson's Gaussian Semiparametric Estimator from Robinson(1992), "Semiparametric analysis of long-memory time series", Annals of Statistics
, vol 22, 515-539. Typically, you will difference the series first apply this to that. If fractional differencing is appropriate, a full difference overshoots, so the estimate will generally be negative. That is, if you apply this to (1-L)y and get an estimate of d=-.2, the conclusion is that (1-L)^.8 y is stationary where the .8=1-.2.
It estimates "d" by examining the spectral density at low frequencies, where "low" is determined by the power of the number of observations set using the POWER
( options ) series start endOptionsPOWER
=power of nobs to use as frequencies
=title of report ["Robinson GSE, Series xxx"]Variables Defined
|%%D ||estimated value of D |
|%%DSE ||estimated standard error of D |