Abstract

A procedure based on density estimation is suggested in the paper to discriminate trend stationary processes about local linear time trends from difference stationary processes. A 'rule of thumb' is constructed to detect the suitability of a segmented trend representation, and a regression analysis is used to identify the number and the dates of structural breaks. The U.S. series of nominal wages over the period 1900-1970 is analysed according to the assumption the dynamics are driven by exogenous shocks which occur infrequently. In a multivariate domain, implications of segmented trend modeling for cointegration theory are also briefly considered.