Chapter2_LN精选.pdfVIP

This chapter introduces the theoretical properties of time series determined by mod- els, and they include the general linear process, the autoregressive (AR) model, the moving average (MA) model and the ARMA model. Without loss of generality, we set the mean of these time series to zero, i.e. µ = 0, in this chapter unless specified. (why?) • Suppose that Z = ∑ a for all t ∈ , where {a } ∼ ∑ W N (0, σ ) and σ ∞. (Is {Z } stationary?) 1. For the mean function, E(Z ) = ∑ E(a ) = 0. 2. For the variance function, var(Z ) = ∑ var(a ) = σ ∑ ∞. 3. For the auto-covariance function, cov(Z , Z ) = σ ∑ , k ≥ 0. . Hence, {Z } is stationary since the mean function and the auto-covariance function are both independent of t. Show that time series {Z } is also strictly stationary. • A time series {Z } is linear if the value of Z is a linear function of a white noise sequence. (Is the time series in Example 1 linear?) • A time series {Z } is causal if the value of Z is affected by the information up to now, i.e. Z is independent of the future i