Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: THEend8_
ECON 2022
Please work in groups of 2 or 3 and hand in your solution by April 10.
The object of forecasting is to produce a method to predict future values
of one or more series. Usually we wish to also have some measure of the
uncertainty of the forecast, and this is expressed in terms of a prediction
interval(s), e.g., you may give a 70% , 80% and 95% PI’s.
There is often an underlying model behind the forecast method, such that
the forecast method is optimal for that model. The prediction intervals are
often also based on this model.
In modelling, and in particular forecasting, you will want to choose be-
tween models to find the model that predicts best, or which model makes
better intuitive sense when two models are nearly equally good at prediction.
Note that you should avoid assessing how well a model predicts by seeing how
close the fitted values are to the observed yt. That is, choosing a model by
doing in-sample prediction. This also applies to choosing between predictive
models based on in-sample prediction. For that you need a validation sample;
or you do time series cross-validation; or cross-validation when appropriate.
Or you use AICC; note that when you transform the yt and fit a model to
the transformed variables, then the AICC must apply to the original yt, not
the transformed variables; i.e., it applies to the likelihood of yt.
Please pick a time series that exhibits both trend and seasonality, from
the PBS dataset (except for series A10), and develop a forecasting model for
TC, where the final model you present should satisfy the various criteria for
a good model specified in the book and the lecture notes. You should write
1
up your results as a consulting report, with the PBS as your client.
In your report, you should:
(a) Develop a forecasting model based on exponential smoothing. That
may mean considering several exponential smoothing models and then
choosing the best one using one or more criteria.
(b) Repeat for a forecasting model based on an ARIMA approach.
(c) Compare your preferred exponential smoothing model with your pre-
ferred ARIMA model using time series cross-validation.
Before you start, please send me the names of the people in your group,
as well as the identity of the time series chosen. If too many groups choose
a certain time series, we may direct you to choose a different one. In your
submission, please also report the role played by each person in the group.
The report should consist of a summary of what you achieved which is
half a page to 3/4 a page, the body of the report which contains the main
result and recommendations of the study, and is 3 to 4.5 pages in length.
You can also have a technical appendix with some extra maths and plots,
but make sure that the report (excluding the appendix) is coherent.