1 / 51

Session 10a

Session 10a. Overview. Forecasting Methods Exponential Smoothing Simple Trend (Holt’s Method) Seasonality (Winters’ Method) Regression Trend Seasonality Lagged Variables. Forecasting. Analysis of Historical Data Time Series (Extrapolation) Regression (Causal)

hall
Télécharger la présentation

Session 10a

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Session 10a

  2. Overview Forecasting Methods • Exponential Smoothing • Simple • Trend (Holt’s Method) • Seasonality (Winters’ Method) • Regression • Trend • Seasonality • Lagged Variables Decision Models -- Prof. Juran

  3. Forecasting • Analysis of Historical Data • Time Series (Extrapolation) • Regression (Causal) • Projecting Historical Patterns into the Future • Measurement of Forecast Quality Decision Models -- Prof. Juran

  4. Measuring Forecasting Errors • Mean Absolute Error • Mean Absolute Percent Error • Root Mean Squared Error • R-square Decision Models -- Prof. Juran

  5. Mean Absolute Error Decision Models -- Prof. Juran

  6. e n å i Y = 1 i i n e n å i ˆ Y = 1 i i n Mean Absolute Percent Error = 100 % * MAPE = 100 % * Or, alternatively Decision Models -- Prof. Juran

  7. Root Mean Squared Error Decision Models -- Prof. Juran

  8. R-Square Decision Models -- Prof. Juran

  9. Trend Analysis • Part of the variation in Y is believed to be “explained” by the passage of time • Several convenient models available in an Excel chart Decision Models -- Prof. Juran

  10. Example: Revenues at GM Decision Models -- Prof. Juran

  11. You can right-click on the data series, and choose to superimpose a trend line on the graph: Decision Models -- Prof. Juran

  12. Decision Models -- Prof. Juran

  13. Decision Models -- Prof. Juran

  14. Decision Models -- Prof. Juran

  15. Decision Models -- Prof. Juran

  16. Decision Models -- Prof. Juran

  17. Decision Models -- Prof. Juran

  18. You can also show moving-average trend lines, although showing the equation and R-square are no longer options: Decision Models -- Prof. Juran

  19. Decision Models -- Prof. Juran

  20. Decision Models -- Prof. Juran

  21. Simple Exponential Smoothing Decision Models -- Prof. Juran

  22. Why is it called “exponential”? Decision Models -- Prof. Juran

  23. Example: GM Revenue Decision Models -- Prof. Juran

  24. In this spreadsheet model, the forecasts appear in column G. Note that our model assumes that there is no trend. We use a default alpha of 0.10. Decision Models -- Prof. Juran

  25. Decision Models -- Prof. Juran

  26. We use Solver to minimize RMSE by manipulating alpha. After optimizing, we see that alpha is 0.350 (instead of 0.10). This makes an improvement in RMSE, from 4691 to 3653. Decision Models -- Prof. Juran

  27. Decision Models -- Prof. Juran

  28. Exponential Smoothing with Trend:Holt’s Method Weighted Current Level Weighted Current Observation Weighted Current Trend Decision Models -- Prof. Juran

  29. Decision Models -- Prof. Juran

  30. Holt’s model with optimized smoothing constants. This model is slightly better than the simple model (RMSE drops from 3653 to 3568). Decision Models -- Prof. Juran

  31. Exponential Smoothing with Seasonality:Winters’ Method Decision Models -- Prof. Juran

  32. Weighted Current Seasonal Factor Weighted Seasonal Factor from Last Year Decision Models -- Prof. Juran

  33. Winters’ model with optimized smoothing constants. This model is better than the simple model and the Holt’s model (as measured by RMSE). Decision Models -- Prof. Juran

  34. Decision Models -- Prof. Juran

  35. Forecasting with Regression Decision Models -- Prof. Juran

  36. Decision Models -- Prof. Juran

  37. Decision Models -- Prof. Juran

  38. Which Method is Better? The most reasonable statistic for comparison is probably RMSE for smoothing models vs. standard error for regression models, as is reported here: The regression models are superior most of the time (6 out of 10 revenue models and 7 out of 10 EPS models). Decision Models -- Prof. Juran

  39. Decision Models -- Prof. Juran

  40. Decision Models -- Prof. Juran

  41. Time series characterized by relatively consistent trends and seasonality favor the regression model. If the trend and seasonality are not stable over time, then Winters’ method does a better job of responding to their changing patterns. Decision Models -- Prof. Juran

  42. Lagged Variables • Only applicable in a causal model • Effects of independent variables might not be felt immediately • Used for advertising’s effect on sales Decision Models -- Prof. Juran

  43. Example: Motel Chain Decision Models -- Prof. Juran

  44. Decision Models -- Prof. Juran

  45. Decision Models -- Prof. Juran

  46. Decision Models -- Prof. Juran

  47. Decision Models -- Prof. Juran

  48. Decision Models -- Prof. Juran

  49. Decision Models -- Prof. Juran

  50. Here are measures of model fit for the non-regression models: The regression model has a standard error of only 213, which is much better than any of the other models. Decision Models -- Prof. Juran

More Related