Sixth session : Forecasting Methods 5
3. Tracking Signal (TSE) allows to check whether your model is biaised or not. If your
model is biased, then you under (or over-) estimate the actual demand repeatedly.
TSE =
e
i
i
n
=
1 -> cf. F21 cell
If there is no bias, the TSE should be small. We could decide the limits it should not exceed.
4. In the F19 cell, you can observe the sum of the errors (SE) = ei
i
n
=
1
Theoretically, if there is no bias, this sum should remain close to 0.
Check now the quality of your model.
PART III : FORECASTING
1. Forecastings can be done from the model you have built up in part 1 and validated in part
II. Observe these forecasts and retrieve these values from the mathematical formula
of your model.
2. For your forecast, you can build a confidence interval in which the actual demand
should fall :
Prob (lower limit LC < demand < upper limit UC) = α%
=> LC = forecast - z * σ = forecast - z * 1.25 * MAD
UC = forecast + z * σ = forecast + z * 1.25 * MAD
On your worksheet, you can determine this probability in the F23 cell.
N.B. : The standard deviation can be approximated by 1,25 * MAD.
For instance, set 95 in this cell and choose a wrong model (e.g. : for the exercise 1, you set
a0= 10, b0= 0, T = 1, c0 = 0 and your model is additive (a)). You can observe the
confidence interval in pink for the lower limit and in black for the upper limit on your
chart and on the grey part for the seventeenth period (LL17 and UL17). This interval is
very large.
Now enter the parameters of the best model for this exercise (cf. solutions). With a
probability of 95, you observe a smaller interval. This model is more accurate.
Now instead of 95, set 50 in the F23 cell. You can observe a smaller interval. It’s logical
because the chosen probability (50%) that the actual demand should belong to this
interval is smaller.
Thus, the breadth of the confidence interval can be influenced by two factors :
• the MAD, depending on the quality of your model;
• the confidence degree you have chosen for building your interval.