POM
Module 3: Demand Forecasting
Unit 3: Time Series Forecasting
Overview:
In this Unit, you will be working on a sample problem that involves calculating demand
forecasts using four time series forecasting models: naïve approach, simple moving
average, weighted moving average, and exponential smoothing. The computations will
be presented here step by step so that, hopefully, you will be able to do them using your
scientific calculator.
Module Objective:
After successful completion of this Unit, you should be able to:
• Calculate the demand forecast using naïve approach, simple moving
average, weighted moving average and exponential smoothing using your
scientific calculator;
Course Materials:
• Handout: Time Series Forecasting Exercise
Calculate:
TIME SERIES FORECASTING EXERCISE
Although our purpose in this exercise is only to determine the demand forecast for
Week 13, it would be good that we consider using not just one, but several, forecasting
models. Then, later, we will assess the accuracy of each of the forecasting models
considered.
The Week 13 forecast of the most accurate forecasting model will then be used as the
demand forecast.
For each of the forecasting models to be considered, we will first understand its
corresponding mathematical equation, so we will know how to go about doing the
computations. Let’s start with the simplest among these models: Naïve Approach.
Demand Forecasting Using the Naïve Approach (NA)
The mathematical equation for the Naïve Approach is:
where:
So, if we would like to forecast the demand for Week 13, then t = Week 13, and the
period immediately before that, t-1, is Week 12. If t = Wednesday, then t-1 is Tuesday.
If t = May, then t-1 is April. If t = 3rd Quarter, then t-1 is 2nd Quarter; and so on.
The NA model says that to determine the forecast for period t, just use the actual
demand for the period immediately before period t. Thus, from the given data, the NA
forecast for Week 13 would be 22,000 liters, since the actual demand for Week 12 is
22,000 liters.
22
Now, although we are ultimately concerned only with the NA forecast for Week 13, we
still need to determine the NA forecasts for the periods prior to Week 13, in this case
from Week 2, because we will be using these for calculating the accuracy of the NA
model later using MAD. Note that the earliest period for which we can forecast using NA
is Week 2, because we don’t know the actual demand prior to Week 1. Thus, our NA
forecasts would be:
17
21
19
23
18
16
20
18
22
20
15
22
Demand Forecasting Using the Simple Moving Average (SMA) Model
The mathematical equation for the Simple Moving Average model is:
where:
Before the SMA model can be used, the forecaster needs first to determine the averaging
period, n, which refers to the number of periods whose actual demand data will be used
in the calculation of the average demand.
This exercise requires that we: “Forecast demand using a 3-week simple moving average
model.” This means that we are to use n = 3. In actual situations, though, it is the
forecaster who decides what n to use.
The SMA model says that, if we would like to forecast the demand for a particular period
t, we just have to calculate the simple average of the actual demands for the n most
recent periods or for the n periods immediately before period t.
Because we are to use n = 3 in this exercise, we can only start SMA forecasting for Week
4 onwards; we cannot determine the SMA forecasts for Week 1, Week 2 and Week 3 for
lack of data.
The forecast for Week 4 is just the simple average of the actual demands for Week 1
through Week 3, thus:
𝐴𝐴1 + 𝐴𝐴2 + 𝐴𝐴3 17 + 21 + 19
𝐹𝐹4 = = = 19,000 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
3 3
The forecast for Week 5 is just the simple average of the actual demands for Week 2
through Week 4, thus:
𝐴𝐴2 + 𝐴𝐴3 + 𝐴𝐴4 21 + 19 + 23
𝐹𝐹5 = = = 21,000 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
3 3
The forecast for Week 6 is just the simple average of the actual demands for Week 3
through Week 5, thus:
𝐴𝐴3 + 𝐴𝐴4 + 𝐴𝐴5 19 + 23 + 18
𝐹𝐹6 = = = 20,000 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
3 3
… and so on.
Again, although we are ultimately concerned only with the SMA forecast for Week 13,
we still need to determine the SMA forecasts for the periods prior to Week 13, in this
case starting from Week 4, because we will be using these for calculating the accuracy
of the SMA model later using MAD. Thus, our SMA forecasts would be:
19
21
20
19
18
18
20
20
19
19
Now, try to arrive at these same values by doing the SMA computations on your own
using your scientific calculator.
Demand Forecasting Using the Weighted Moving Average (WMA) Model
The mathematical equation for the Weighted Moving Average model is:
To use the WMA model, the forecaster must first decide on two sets of data: (1) the
averaging period, n; and (2) the set of weights, Wi, to be multiplied to each Ai, the actual
demand data for each period i. This model requires that the value of each Wi should be
between 0 and 1; i.e., we could use either fractions or decimal numbers. We’d rather
use decimal numbers. Also, the sum of all Wi to be used must be equal to 1.
In this exercise, we are asked to use n = 4, and the weights 0.1, 0.2, 0.3 and 0.4. Notice
that the weights to be used are decimal numbers and their sum is exactly equal to 1.0.
Also, notice that the n to be used determines how many weights should be used. So, if
n = 3, there should be 3 weights, which could be any of these sets of weights: {0.1, 0.3,
0.6}; {0.2, 0.3, 0.5}; and so on. If n = 2, there should be 2 weights, which could be any
of these sets of weights: {0.1, 0.9}; {0.2, 0.8}; {0.3, 0.7}; and so on.
Also, take note that in every situation where we are to use the WMA model; i.e., not just
in this exercise, we must multiply the heaviest weight to the most recent data in the set
of Ai, multiply the second heaviest weight to the second most recent Ai data, …, and so
on, and multiply the lightest weight to the least recent Ai in the set.
Since we are to use n = 4, we can only start forecasting for Week 5 onwards. To compute
the forecast for Week 5, the set of actual demand data Ai and the set of weights Wi to be
used are: Ai = {17, 21, 19, 23}, which are the actual demand data for Weeks 1, 2, 3 and
4, respectively; and Wi = {0.1, 0.2, 0.3, 0.4}. Thus,
𝐹𝐹5 = (0.1)(17) + (0.2)(21) + (0.3)(19) + (0.4)(23) = 20,800 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
To compute the forecast for Week 6, the set of actual demand data Ai and the set of
weights Wi to be used are: Ai = {21, 19, 23, 18}, which are the actual demand data for
Weeks 2, 3, 4 and 5, respectively; and Wi = {0.1, 0.2, 0.3, 0.4}. Thus,
𝐹𝐹6 = (0.1)(21) + (0.2)(19) + (0.3)(23) + (0.4)(18) = 20,000 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
To compute the forecast for Week 7, the set of actual demand data Ai and the set of
weights Wi to be used are: Ai = {19, 23, 18, 16}, which are the actual demand data for
Weeks 3, 4, 5 and 6, respectively; and Wi = {0.1, 0.2, 0.3, 0.4}. Thus,
𝐹𝐹7 = (0.1)(19) + (0.2)(23) + (0.3)(18) + (0.4)(16) = 18,300 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙
… and so on.
Again, although we are ultimately concerned only with the WMA forecast for Week 13,
we still need to determine the WMA forecasts for the periods prior to Week 13, in this
case from Week 5, because we will be using these for calculating the accuracy of the
WMA model later using MAD. Thus, our WMA forecasts would be:
20.8
20.0
18.3
18.7
18.2
19.8
20.2
18.2
19.5
Now, try to arrive at these same values by doing the WMA computations on your own
using your scientific calculator.
Demand Forecasting Using the Exponential Smoothing (ES) Model
The mathematical equation for the Exponential Smoothing model is:
To use the ES model, the forecaster must first decide on the exponential smoothing
constant, α. This model requires that the value of α should be between 0 and 1; i.e., we
could use either fractions or decimal numbers. We’d rather use decimal numbers. In
this exercise, we are required to use α = 0.3.
When using the ES model, we can only start forecasting if there’s a value for the demand
forecast for the first period, F1. In some forecasting problems, F1 might be given. For
example, the problem may state that “the exponentially smoothed forecast for the first
period is 20,000 liters,” in which case F1 = 20 (thousand liters). However, in this exercise,
F1 is not given, so what to do? We simply use the actual demand for the first period, A1
= 17 (thousand liters) as our ES forecast for the first period; thus, F1 =17 (thousand
liters).
17
So, henceforth, we can now find the ES forecasts for Week 2 through Week 13. Let’s
state our results in one decimal place.
F2 = F1 + α (A1 - F1) = 17 + 0.3 (17 – 17) = 17.0
F3 = F2 + α (A2 – F2) = 17 + 0.3 (21 – 17) = 18.2
F4 = F3 + α (A3 – F3) = 18.2 + 0.3 (19 – 18.2) = 18.4
F5 = F4 + α (A4 – F4) = 18.4 + 0.3 (23 – 18.4) = 19.8
… and so on.
Notice that we cannot compute the ES forecast for Week 13 unless we calculate the ES
forecasts for the previous periods, which is just fine since we will be using these data
anyway for calculating the accuracy of the ES model later using MAD. Thus, our ES
forecasts for Weeks 2 through 13 would be:
17.0
17.0
18.2
18.4
19.8
19.3
18.3
18.8
18.6
19.6
19.7
18.3
19.4
Now, try to arrive at these same values by doing the ES computations on your own
using your scientific calculator.
Keep on practicing.
In Module 3: Unit 4, we will learn how to evaluate the accuracy of these four forecasting
models using MAD and answer the remaining Items 5 and 6 in this exercise.
Review:
Activities/Assessments:
Activity __.
