Problem Set 2: Forecasting
1) Chopra and Meindl (CM) Discussion Questions 7.1, 7.7, and 7.10
2) Using the data on the right, fit a four-period moving average model (use periods 3
1-4 for the initial estimate of L4), an exponential smoothing model with a=0.1 4 (assume that L =40), and exponential smoothing model with a=0.6 (assume that 5
0 6
L0=40). Compare forecasts with actual demand for periods 5-12 using MAD, MSE,
bias, and TS and comment with some detail on the forecast fit. Using each
7 8 9
forecasting methods, forecast demand for periods 13-15. What do you think of 10
using L0=40? What is the impact? Why do you think I chose this?
3) If you regress a line through this same data, the intercept is ~ 42 and the slope is
11 12
~ 4.3. Using these values in a ‘static’ forecast model of the form F=L+nT, compare your forecasts to actual demand for periods 5-12 using MAD, MSE, bias, and TS and comment on the fit of the model. Forecast periods 13-15.
4) Now using L0 =42 and T0 =4.3, forecast demand for periods 1-12 using Holts model (a=0.2, ß=.4), updating your factors each period when new data becomes available. Does this model fit better than the other models (MAD, MSE, bias, and TS)? Comment. Forecast periods 13-15.
5) Why do you suppose most ‘real-world’ systems use these time series forecasting methods (like exponential smoothing) and not a regression like the one above? Think carefully about the principles of the exponential forecasting model.