Statistics 451 Spring 2015 Examination 2 1. The Box-Cox transformation is given by ( (Z ∗ +m)γ −1 t Zt = γ log(Zt∗ Name You must show all of your work γ 6= 0 + m) γ = 0 Usually we take m = 0 but there are some circumstances where we would use another value of m. Give an example if a situation where we would want to use m > 0. 2. If you had a time series process consisting of counting the number of events that are relatively rare (e.g., the number of serious automobile accidents in Ames in a month), what transformation might you expect to provide more constant variance than the original data. Explain why. 3. Consider the ARMA(1,1) model (1 − φ1 B)Zt = (1 − θ1 B)at As we know, this model can be expressed as Zt = ψ1 at−1 + ψ2 at−2 + ψ3 at−3 + · · · + at . (a) Derive expressions for ψ1 and ψ2 for this model. (b) Derive Cov(Zt , at−2 ) for this model. 1 4. A time series Zt can be described by a random walk Zt = Zt−1 + at . Assume that a realization Z1 , Z2 , . . . Z100 is available to compute forecasts and that a forecast is needed for Z103 , using Z100 as the forecast origin. (a) Give an expression for Z103 , based on this random walk model. (b) Give an expression that can be used to compute Zb100 (3), the forecast for Z103 . (c) Derive e100 (3), the forecast error in Zb100 (3). (d) Give an expression for the variance of e100 (3). (e) Give an expression for an approximate 95% prediction interval for Z103 . 2 5. Which diagnostic is most useful for detecting a violation of the assumption that the at values are independent? 6. The AR(1) model can be expressed by Zt = θ0 + φ1 Zt−1 + at , at ∼ nid(0, σa2 ). (a) Derive and expression for the variance σZ2 = Var(Z) for this model. (b) Derive expressions for the autocorrelation function values ρ1 and ρ2 for this model. (c) Explain how to compute (i.e., give a formula for) the first two true PACF values φ11 and φ22 for this model. 7. The AIC has been suggested as a useful tool for choosing among time series models. (a) Why is AIC a better criterion than choosing the model with the largest likelihood? (b) Explain why AIC should not be used exclusively in making such a choice. Be as specific as possible in your answer. 3 8. Suppose that the model for Zt is (1 − B)Zt = θ0 + at and that the working series is defined by Wt = (1 − B)2 Zt . (a) Derive an expression for the model for Wt (note that Zt (for any lags) should not appear in this expression). (b) Derive the mean of Wt . (c) Derive the variance of Wt . (d) Is Wt stationary? Why or why not? (e) Is Wt invertible? Why or why not? 4