TIME SERIES - TAKE HOME EXAM July 2004
1.
Discuss the identified ARIMA models in pages 41-46. Do you think they are appropriate based on graphs, SACF and SPACF? Would you use the mentioned tools in a different way in order to clarify the results?2.
Write the ARIMA-no outliers models for Ugashik, Cooper and Situk. Obtain the covariance function, the SACF and the -weights for Ugashik and Situk.3.
Write the Eventual Forecast Equation for Skeena. How are the transitory and permanent components?4.
The authors only provide RSE as diagnostic tool. Discuss if that measure provides a good measure of fit and describe other diagnostics that would be of help.5.
Write and describe the intervention model for Ugashik. If we wished to forecast the next three years for that river, what outliers should still be considered?6.
Compare the ARIMA-no outliers models for Chignik, Alitak and Karluk. The three have the same identified model but could you comment on the differences (or similarities) based on the estimated parameters?7.
In the outlier discussion pages 46-48 the authors use the original series (not log- transformed) to match the identified outliers. Is that a correct procedure? Why?8.
Authors claim that the series needed a log-transformation. Still, from graphs in page 39, some series seem to have a quadratic shape. What do you think? Is there any way to know if the log-transformation is really needed? Could you describe a reliable method to know whether the quadratic transformation is more appropriate?9.
In page 37 authors describe the way missing observations are treated. Do you think this is an optimal interpolator? Suppose we have that, for Ugashik, z(1958)=0,6 millions of catches and z(1960)=1,3 millions of catches, what would be the optimal interpolated value for z(1959) taking the ARIMA model without outliers in page 45?10.
The 9 series can be aggregated into 3 regions (as described in page 37). Are there similarities in the graphs/models/outliers for the rivers of the same region?Could the authors get any new information by analysing the aggregated series?