Case Analysis

Instructions:
Please compose a brief report on the Forecasting Analysis you completed for this case. The report should be no more than two to three pages, should be typed in Microsoft Word, and should be submitted via PLATO. Each student is required to submit their own unique write up. The report itself should be targeted to an executive who is reasonably well versed in the subject area, but may not understand the finer points behind the calculations.
In your executive report, please provide a summary of your findings from Parts I, II, and III. Your summary should contain answers to the questions that follow from the respective parts of the case. However, you should not simply list out the answers to the questions. This is a summary of what you found, and should be written as if it were a brief narrative that was telling the story of this part to the executive team. The questions that follow are merely to help guide you in deciding what to include in your report.

Part I Guide Questions:
What are the four components to a time series of data?
The data clearly move upward through time, so a trend component is present. Of the other three components, which do you observe in the data?
Does the trend appear more linear in the full dataset or the limited dataset? Do any of our potential forecasting models rely on linear trend assumptions? How might this affect our predictions?
Given the time series components you identified above, which forecasting model should provide the best estimates for upcoming demand? Use what you know about the models from the notes/text, and which time series components these models are designed to address.
For instance, purely random data with no trend is best modeled with Moving Average, as this is designed to respond gently to transitory shocks. Since the data HAS a trend, this will NOT be the answer to number 4.

Part II Guide Questions:
What is bias? What might it indicate if the bias is consistently positive on average?
On your template for the Moving Average (MA) model, the Bias you observe should be positive and relatively large (>1B). Why does this make sense given the trend we observe in the data? What does this suggest about the MA model?
On your template for the Exponential Smoothing (ES) model, which alpha value provided the best model for the data? Was this value higher or lower than the other values? What does this high/low value of alpha (the smoothing constant) mean?
Was it possible to get a more accurate forecast using the TAES compared to MA and ES? How can you tell (i.e. what elements from the case file help us compare models)? Think about how we compare models and their effectiveness. Also, please explain why you might expect the TAES to yield better results – think about which components it addresses compared to the other two models.
Examine the MAD for the Decomposition. How does it compare to the other models we have used? Why might this be the case – that is, what essential element of the Decomposition is violated with this input data?

Part III Guide Questions:
How do your MAD and MAPE compare for TAES modeling when using the full dataset vs. the limited data set? When is MAD larger, and when is MAPE larger?
Given that MAD is an absolute measure, while MAPE is a percentage error, why might your answer to question 1 make sense? You should use what you know about what MAD (a measure of distance between forecast and actual value) and MAPE (a measure of the error as a percentage) are. It might help to calculate the average quarterly sales figure for the full dataset and compare it to the average for the limited dataset.
Given that we are now comparing models with two different input datasets, our standard method of comparing models (MAD) is no longer the correct approach. The MAPE will assess the accuracy in percentage terms, which is not affected when different input data is used, while MAD is. Thus, MAPE is a better measure of model accuracy. Using the MAPE, which model provides the best estimate for a forecast of next period’s sales demand? Why is it the best (i.e. how does its MAPE compare to the other models’ MAPEs)?
Using the model you identified as best, what is your forecast for Amazon’s sales in Q1 of 2019? What sales figure should the executive team plan for?