2. In Example 8.3 (the trapped lynx series), try using priors on the AR and MA coefficients based on the maximum likelihood solution but with the precision downweighted by 10. The maximum likelihood estimates from SPSS are

Also consider estimation with the priors as in the worked example but conditioning on the first three data points. Finally consider the model as in the worked example, including modelling of latent pre-series values, but introduce an error outlier mechanism such that with probability 0.05, some εt have variance 10 times σ2 ε . How do these options affect parameter estimates and one-step-ahead predictions?

Mean s.e. 2.07 0.126 P2 -1.77 0.200 P3 0.49 0.123 0.90 0.121 -0.09 0.141 -0.49 0.100 2.90 0.064