1981 年 34 巻 4 号 p. 455-464
A new method for the analysis of non-stationary time series with trends is proposed. Integrated autoregressive models are used to eliminate the trend component. An adaptive procedure of fitting the non-stationary autoregressive models which is based on the concept of the local stationarity is applied for the identification of the stationarity of the data. This procedure enables us to divide the non-stationary data into several locally stationary series. This method is useful for the prediction of the future values and the detection of the change of the statistical properties in the data. Feasibility of this method is discussed and it is successfully applied to the series of the strain change, the ground tilt and the number of the earthquakes.