Band Spectral Regression with Trending Data

Band spectral regression with deterministic and stochastic trends is considered. It is shown that conventional trend removal by regression in the time domain prior to band spectral regression leads to biased and inconsistent estimates of the parameters in a model with frequency dependent coefficients. Time domain and frequency domain procedures for dealing with this problem are examined. Trend removal in the frequency domain produces unbiased estimates and is recommended. An asymptotic theory is developed and the two cases of stationary data and cointegrated nonstationary data are compared. Efficient band spectral regression estimators and associated inferential methods are provided for models with deterministic and stochastic trends. Some supporting Monte Carlo evidence is presented. An empirical application to the present value model of stock prices is discussed. After removing trends in the frequency domain, we show that, while stock prices and dividends have significant coherence at low frequencies, transitory fluctuations in dividends (i.e., less than 3 years) do not have significant coherence with stock price movements.