Part I:

Perform exploratory data analysis and briefly summarize your findings.

Part II:

To ensure that your modeling and decision making process and recommendations based on your models are robust and credible in a range of possible future outcomes, you will apply the validation approach two times (see Scenario 1 and Scenario 2 described below) and evaluate models’ predictive accuracy using validation MAPE for both scenarios. Then you will average MAPE’s from both scenarios for each model and obtain the average MAPE for each model.

If appropriate and applicable apply the Box-Cox transformation for models in R.

Scenario 1:

Training set: start = 01/1992, end=12/2007

Testing set:   start = 01/2008, end=12/2009

Scenario 2:

Training set: start = 01/1992, end=08/2018

Testing set:   start = 09/2018, end=08/2020

Report the results in Table 1 in the answer sheet.

Extra credit if you try different models that we have not discussed in class (both time series and machine learning ones)!

Part III:

The model that has smallest average MAPE is the champion model. You will need to use the champion model to forecast next 12 months of data. Whatever forecasting approach is used to estimate future patterns of data, the results will inevitably be subject to significant uncertainty. An understanding of how this uncertainty could affect any decisions is therefore about as important as the forecast itself. Thus report both future forecast and corresponding 95% prediction interval in Table 2 in the answer sheet. (Do not include decimal digits!)

Part IV:

Plot all historical data, future forecasts with prediction intervals, and fitted values.

Part V:

Do you consider that your modeling approach presents an accurate picture of current and future data patterns? Does your model need to be improved?

Part VI:

Briefly mention what situations could potentially distort predictions, cause challenges in finding reliable and accurate predictive models to forecast this data and discuss the ways to overcome these challenges.