Moving Average, Smoothing vs. Sensitivity

In this assignment you will experiment with the number of periods to use in a moving average forecast. Using the Excel spreadsheet linked below, determine which moving average forecast (for example the MA(5), MA(10), MA(20), or any other number of periods you wish to try) minimizes the MAD of one-step forecasts for periods 31 through 100. You will work through this process for Time Series #1 in the attached workbook, then work through the process again for Time Series #2. The workbook linked below is currently using the MA(5) forecast for both datasets.

After you have finished, you should find that one of the datasets (Time Series #1 or Time Series #2) requires a relatively smooth forecast, or in other words a relatively large number of periods in the moving average in order to minimize the MAD. The other dataset requires a fairly sensitive forecast, or in other words a relatively small number of periods in the moving average forecast.

The video linked below explains the process you will use. Our spreadsheet will compute the MAD as the simple average of absolute forecasts errors, and does not update the MAD using exponential smoothing as discussed on page 303 of the text.

When you are finished, you should write a short paper (1-2 pages typed, double spaced) explaining your results. Your report should indicate:

• What MA forecast worked best for Time Series #1, and the MAD over periods 31-100 that resulted from that MA forecast
• What MA forecast worked best for Time Series #2, and the MAD over periods 31-100 that resulted from that MA forecast
• An explanation of why one of the datasets needed a larger number of periods in the moving average to minimize the MAD, while the other needed a smaller number of periods. In answering this question, you should be able to give a general characterization of the data characteristics that would prompt us to use a larger vs. smaller number of periods in a moving average forecast.
• Do you think a different forecasting method, such as a trend or seasonal model, should be used to forecast Time Series #2? Explain why or why not.