University of Alberta



Top of Form

[pic][pic][pic]

Bottom of Form

|[pic]WINTER 2007 MGTSC 352 LEC B1 > DOCUMENTS > LABS > LAB 3 - FORECASTING (TES, SLR) |

|[pic] |

[pic]Lab 3 - Forecasting (TES, SLR)

|[pic] |Lab 3 - Forecasting (TES, SLR) | |

| |Lab3_TES and SLRwSI (439.5 Kb) | |

| |Lab 3 - Forecasting January 26, 2007 | |

| | | |

| | | |

| | | |

| |Agenda: | |

| |1. Triple Exponential Smoothing | |

| |1.1 Installing Solver | |

| |2. Simple Linear Regression | |

| |  | |

| |Triple Exponential Smoothing | |

| |In order to complete this lab, you will need to download the file attached above. This data represents the monthly sales of garden equipment | |

| |taken from the US Census Bureau. Given such data, how would you compute the forecasts for periods 157 to 159?  | |

| |Propagate or copy this formula down in order to initialize the next 11 periods (cells G3:G13). Notice that cell E13 is an absolute reference. | |

| |1) Initialization | |

| |We will use the first p=12 periods of past data (D1, D2, ... D12) for initialization. Recall that p is the number of seasons. | |

| | | |

| |Level: Let the initial level (L) be the average of the first 12 periods | |

| |[pic] | |

| | | |

| |Trend: Tp=(Dp+1-D1)/p | |

| |[pic] | |

| |Seasonality: Si=Di/L. | |

| |[pic] | |

| |2) Learning | |

| |Recall that LS, TS, and SS are the smoothing constants for level, trend, and seasonality, respectively. We will refer to these cells often in our| |

| |formulas so it can save us time to name them. | |

| |Level: Lt=LS*(Dt/St-p) + (1-LS)*(Lt-1 + Tt-1). | |

| |[pic] | |

| |Trend: Tt=TS*(Lt - Lt-1) + (1 - TS)*Tt-1. | |

| |[pic] | |

| |Seasonality: St=SS*(Dt/Lt) + (1 - SS)*St-p. | |

| |[pic] | |

| | | |

| |Copy the formulas in cells E14, F14, and G14 down to row 157. This will take care of the learning from periods 13 through 156. | |

| |3) Prediction | |

| |At time t, the one-step forecast is Ft+1=(Lt + Tt)*St+1-p. | |

| |[pic] | |

| |Propagate this formula down to row 157. | |

| |Now we will make a "real" forecast": | |

| |A k-step forecast at time t is Ft+k=(Lt + k*Tt) St+k-p for k ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download