Figure 19.20 shows time plots of monthly sales of six types of Australian wines (red, rose, sweet

Question:

Figure 19.20 shows time plots of monthly sales of six types of Australian wines (red, rose, sweet white, dry white, sparkling, and fortified) for 1980– 1994. [Data are available in AustralianWines.csv, Source: Hyndman and Yang (2018).] The units are thousands of liters. You are hired to obtain short-term forecasts (two to three months ahead) for each of the six series, and this task will be repeated every month.

a. Which forecasting method would you choose if you had to choose the same method for all series? Why?

b. Fortified wine has the largest market share of the above six types of wine. You are asked to focus on fortified wine sales alone and produce as accurate as possible forecasts for the next two months.

• Start by partitioning the data using the period until December 1993 as the training set.

• Apply Holt-Winters exponential smoothing to sales with an appropriate season length (use smoothing parameters α = 0.2, β = 0.001, γ = 0.001 with multiplicative seasonality

image text in transcribed

d. Use the LSTM deep learning technique to forecast the fortified wine series. As before, use the period until December 1993 as the training period and the rest as the holdout period. Set the forget gate bias initialization parameter in the Add LSTM Layer operator to 2, and use 3 neurons for that layer.
i. Report the RMSE and MAPE for the holdout period.
ii. Change the deep learning network architecture by replacing the LSTM layer with a simple RNN using the Add Recurrent Layer operator. Report the RMSE and MAPE for the holdout period.
iii. Which of the two models would you select for forecasting future values? Why?

Fantastic news! We've Found the answer you've been seeking!