(Received August 16, 1993; revised October 9, 1995)
Abstract. A non-Gaussian state-space model is proposed to estimate a switching trend from serial data taken at equally spaced intervals. A procedure to detect structural changes in a linear trend is also proposed. The results of a simulation study conducted to check the performance of the detection procedure are shown. A numerical illustration is provided using economic time series data.
Key words and phrases: Akaike information criterion, edge detection, non-Gaussian state-space model, smoothing, structural change, switching trend.