For several years, I have been assigning my students a forecasting project using data from their company. Since I want students to learn about seasonality, I require the data must be four complete cycles of the measurement. That is, I ask for 48 months, 16 quarters, 20 or 28 days, or 8 or 24 hours over 4 days. Most students are able to obtain data from their company and for those who cannot I ask them to find acceptable data on the web or ask a classmate for data.
I have the students first present three graphs: a graph of the data over 48 months, a graph of the annual data over the 4 years, and a stacked graph of the data over 12 months for each of the 4 years. For the first two graphs, I have them apply Excel’s easy-to-use option to identify the trend line. I also ask the students to identify the ratio of the trends. They expect annual to monthly trend to be near 12, and are surprised when it is closer to 144. I have the students use the stacked graph to identify seasonality.
I then require the students to run their data through most of the models in the Heizer-Render textbook. Using POM for Windows (which comes free with the book), it is very easy to change from model to model, to determine the best n for moving averages and the best alpha for exponential smoothing. It also is straightforward to run the decomposition methods that are in POM, which will determine seasonal factors as in the textbook. For each model, students are to identify the bias, MAD, MSE, standard error, and MAPE –and select the best forecasting method based on the error measures. Finally, using the method they have selected, I ask them to identify forecasts for the next 12 months and the seasonal factors for each month.
The students very much like applying forecasting to data from their own company. They also appreciate the value of the graphs, the ease in changing from one method to another in POM, and that the different methods will yield different results.