3e2a1b56-dafb-454d-87ad-86adea3e7b86

174
Part Two
Design
Table S6.3 Exponentially smoothed forecast calculated with smoothing constant 1 = 0.2
Week Actual demand Forecast
(thousands) (F t = 2A t−1 + (1 − 2)F t−1 )
(t) (A) (2 = 0.2)
20 63.3 60.00
21 62.5 60.66
22 67.8 60.03
23 66.0 61.58
24 67.2 62.83
25 69.9 63.70
26 65.6 64.94
27 71.1 65.07
28 68.8 66.28
29 68.4 66.78
30 70.3 67.12
31 72.5 67.75
32 66.7 68.70
33 68.3 68.30
34 67.0 68.30
35 68.04
The value of α governs the balance between the responsiveness of the forecasts to changes
in demand, and the stability of the forecasts. The closer α is to 0 the more forecasts will
be dampened by previous forecasts (not very sensitive but stable). Figure S6.4 shows the
Eurospeed volume data plotted for a four-week moving average, exponential smoothing
with α = 0.2 and exponential smoothing with α = 0.3.
Causal models
Causal models often employ complex techniques to understand the strength of relationships
between the network of variables and the impact they have on each other. Simple regression
models try to determine the ‘best fit’ expression between two variables. For example, suppose
an ice-cream company is trying to forecast its future sales. After examining previous demand,
Figure S6.4 A comparison of a moving-average forecast and exponential smoothing with the
smoothing constant a = 0.2 and 0.3