Quadratic-inverse estimates of transfer functions

Estimating transfer functions between time series is essentially regression in the frequency domain but significant differences occur when the available data sequences are short. The author suggests an estimate of transfer functions on narrow, overlapping frequency bands using a quadratic-inverse formalism. (The quadratic-inverse coefficients apply to both an orthogonal expansion of the spectrum and to a components-of-variance-like decomposition of the covariances between the multiple window eigencoefficients.) A method is suggested to include the effects of line components and similar `high leverage´ effects explicitly