第11讲时间序列分析二.pdfVIP

Introduction  In general, the most important component of a time series and therefore the most often studied is the trend.  We most often are interested in the trend because it can help us to make short- and long-range forecasts.  As we saw in Lecture 10, one way to identify the trend is to smooth the time series using either the moving average or exponential smoothing technique.  In this lecture, we will consider another method of estimating the trend, which is to use least-squares regression techniques to fit trend lines to the data.  We will consider three trend models: the linear trend model, the quadratic model, and the exponential trend model.  Finally, we will look at autoregressive models. The linear trend model  The linear trend model may be stated as:  Y = β + β X + ε t 0 1 t t  where:  Y = data value in time period t t  X = time period t t  β , β = intercept and slope parameters, respectively 0 1   = error term in period t t  Approach:  1. Generate the Xt values by re cing the years with discrete time periods starting from Xt = 1, 2, 3, ..., n.  2. Regress the time-series values (Y ) the time t periods (X ). t Example 1  The data for thi ample is the deseasonalised quarterly Australian beer production data for the period Sep-2005 to Dec-2009 generated from Example 6, Lecture 10.  1. Estimate a linear trend model using the data.  2. Use the model to generate a forecast for the March 2010 quarter. Data for Example 1 Quarter Y X Sep-05 429.84 1 Dec-05 422.56 2 Mar-06 439.02 3 Jun-06 42