Demand Forecast and Purchasing

CHAPTER 3
Fall 2023 1
 3 Demand Forecasts

Learning Objectives

At the end of this chapter, you will be able to:

• Explain the role of forecasting the demand in the organization;

• To distinguish independent demand from dependent demand;
• To identify and understand factors influencing the demand;
• To differentiate between the steps of forecasting process;
• To distinguish between qualitative and quantitative forecast methods and to
understand how they can complement each other;
• To understand the advantages and limitations of each of these methods.

2
 Demand Forecasts

• It is rarely possible to produce products/services instantaneously;

• It is therefore necessary to estimate in advance the quantities of these

products and services;

• Planning the resources to produce and obtain the stocks that will allow to
satisfy the demand;

• Demand forecasts are the starting point of all supply chain management
operations.
3
 Demand Forecasts

• Forecasting the market demand is the responsibility of the marketing

department (marketing consulting companies);

• Operations managers are responsible for :

1. Comprehension of market demand forecast and assessment of the
accuracy and the risks associated to it;

2. Forecasting the demand of various products used and consumed in their

department (supplies, maintenance products, etc.).

4
 Demand Forecasts Characteristics

• It can never be quite accurate;

• Long-term forecasts are less accurate than short-term forecasts;

• Forecasts are usually more accurate for a family of products than for
each separate product;

• The more volatile the demand is, the more it is difficult to forecast
the demand accurately.

5
Demand Forecasts-Demand Categories

Independent Dependent
Demand Demand
•Refers to finished •It’s about raw materials and
products sold to customers components used in the
or consumed internally. manufacture of finished
•It is necessary to forecast products.
•The quantity to be purchased
this demand to plan does not have to be forecasted,
resources and manage it is simply calculated from the
stocks. demand of finished products.
6
 Factors influencing the demand

1) Seasonality

2) Market conditions

3) Life cycle of the product/service

4) Internal factors

5) Residual (or random)

7
 Factors influencing the demand

1) Seasonality
• Vast majority of products and services demands is subject
to seasonal demand variations;

• We distinguish various seasonal variations :

 Per season (bicycles, suitcases, hotels)
 Per month (bank)
 Per week (restaurant, bar)
 Per day (metro)
8
 Factors influencing the demand

2) Market conditions
• All elements of the PESTEL environment are likely to have an influence on the market and
consequently on demand.
Example : Demand for electric cars
o Political environment : governments commitments to reduce greenhouse gas emissions,
o Economical environment : gasoline prices
o Social environment : increasing of schooling,
o Technological environment : life battery, recharge speed
o Ecological environment : pressures of environmentalists
o Legal environment: health and safety standards
• Demand fluctuation caused by the market may be expressed as a mid-and- long term , but can
also be expressed as short term, as in the case of actions of competing companies (promotions,
9
advertising).
 Factors influencing the demand

3) Life cycle of product/service

• Product/service life cycle: introduction, growth, maturity, and decline.
• Life cycle influences the demand :

10
 Factors influencing the demand

4) Internal factors
• Company’s decisions that can influence demand for the products
they produce or the services they provide:
- Discounts/promotions, advertising;
- Introduction of a substitute product;
- Expansion and choice of place to open a new branch;
- Capacity (frequent shortage of product/service).

11
 Factors influencing the demand

5) Residual (or random)

• Regroup all factors that have not been taken into account in the preceding
categories;
• The most problematic for operations managers because they represent:
 All factors causing demand variations due to chance and non-recurring
events (war, natural catastrophe, launch of new product, economic crisis,
competition, etc.) ;
 All factors whose causes analyses are difficult to discern;
• Face them by implementing an operational systems with flexible capacity,
accepting the facts : certain waste of resources, dissatisfied customers.
12
 How to forecast the demand?

1) Forecasting process

2) Forecasting methods

13
 Forecasting process

1) Define the problem

2) Collect data
3) Conduct an exploratory analysis of the data
4) Choose the most appropriate forecasting
method
5) Apply the selected method to forecast the
future demand and track results

14
 Forecasting process-Define the
problem

1) Define the problem

• Decide which Forecasting horizon to consider

(short, medium or long term) ;

• Determine the degree of accuracy required.

15
Forecasting process-Collecting data

2) Collect data

• Sales history (last weeks, months, years);

• Information gathered from customers (complaints, opinions, requests,

etc.);

• Information on the PESTEL environment (demographic statistics,

household income, regulations, market trend, market share, etc.);

• Competitors’ data (annual reports);

• Expert opinions (magazine, television, newspapers, articles, etc.). 16

 Forecasting process-Collecting data
2) Collecting data (example)

17
Source: http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/cst01/famil131a-eng.htm
 Forecasting process-Exploratory
analysis
3) Conduct an exploratory analysis of the data

• Extract and analyse the key data;

• Choose the analytical methods that allow more precision (e.g:

statistics (p, correlation, linear / exponential tendency, etc.);

• Identify and analyse factors that have more impact on demand

( season, product life cycle, market conditions, etc.). 18

 Forecasting process

4) Choose the most appropriate forecasting methods

(qualitative/quantitative)

• The choice of method depends on the available resources and information;

• The operations manager can do some tests to validate the choice of the

method (using sales data from previous years);

• The tests allow to choose the most optimal method, the one which gives

the most accurate results.

19
 Forecasting process

5) Apply selected methods to forecast future demand and track

results

• After using the method, check at the end of each forecast process
whether the method used is efficient or it is necessary to modify
it.

• Choose the best method and apply it for future forecast (weeks,
months, or years). 20
 Forecasting methods

Qualitative Quantitativ
methods e methods

Causal analysis Methods

Time series analysis

Methods
21
 Forecasting methods-Qualitative
methods

1) Qualitative methods

• Used when the company has little or no information (launch of a new product / service,
new company, Product diversification) ;

• Used also as a complementary method to nuance, enrich and validate the quantitative
methods ;

• Is based on the experience, opinions and expertise of actors who know well the field;

• May be expensive and long.

22
 Forecasting methods-Qualitative
methods

2) Qualitative methods’ categories

• Customers survey: determines the interest of consumers and their use of the products/services;

• Opinion of experts: through their experience, these internal actors have a very fair perception of market

trends as well as their impact on demand;

• Delphi method: it is a process that allows the collect opinions of anonymous (external) experts by sending

them a list of questions relating to the industry and its market. Their opinions are then compared with

those of the experts of the company.

• Historical analogies: is based on historical data of similar products (e.g CD/DVD)

23
 Forecasting methods-Quantitative
methods

• Quantitative methods

• Are divided into two groups: Causal Analysis and Time Series Analysis Methods;

• Causal analysis methods consist in establishing cause and effect relationships between certain

variables and the demand (simple and multiple regression analysis, and the correlation index

analysis);

• The time series analysis methods use past demand data to predict/forecast future demand. They rely

on the past to forecast the future, they neglect the changes of the environments;

• These methods are inexpensive, quick to use and appear to be more objective than qualitative

methods;

24
• The use of qualitative and quantitative methods makes the forecasts more reliable.
Quantitative methods-Time series Analysis

• Time series analysis methods:

1) Naïve forecasts

2) Moving Average

3) Weighted moving average

4) Simple exponential smoothing

5) Trend analysis

6) Analysis of seasonality 25
 Time series Analysis-Naïve forcasts

1) Naïve forecasts

• Estimate the demand for the next period from the demand of the current period;
o
Example : if the demand of January is 600 units, one will forecast a demand of 600 units for the next month of February,

• Because of the strong seasonality, we may take the value of the previous cycle
(January of last year);
o
Example : if the demand of last January (2017) is 600 units, one will forecast a demand of 600 units for the next month of
February (2018),

• This method is used only if the demand does not show pronounced trends and is
little subject to random variations.
*A Trend is general upward or downward movement of a variable over time.
*Random Variation is a fluctuation in data that is caused by uncertain or random occurrences. These random changes are generally very short-term, mere
bumps and dips on the road up or down a trend line. 26
Time series Analysis - Moving average

2) Moving Average
• Consists in averaging the demand of the "K“ last periods (week, month,
year).
• All periods have the same importance;
• The choice of the number of periods K to include in the average is
important.

Ft=forecasts for t period

K= Number of periodsused in the calculation
Dt=Demande of the periodt

27
 Time series Analysis - Moving average
(continued)
2) Moving Average (continued)
• Example : a company specialized in the sale of sports shoes. The following table shows the
sales/demands of the last 12 weeks and the forecasts that could have been made from these data
using : k = 2, 3, and 4 weeks.
Weeks 1 2 3 4 5 6 7 8 9 10 11 12

Sales=demands 125 142 120 153 156 135 128 117 140 134 132 126

Forecasts at k = 2 134 131 137 155 146 132 123 129 137 133

Forecasts at k = 3 129 138 143 148 140 127 128 130 135

Forecasts at k = 4 135 143 141 143 134 130 130 131

Example of calculation : For k = 2, we calculate the forecast for period 3 by using the following formula (125 + 142) ÷
2; For period 4, the forecast is made by calculating (142 + 120) ÷ 2. For k = 3, one calculate the forecast of the period
4 by calculating (125 + 142 + 120) ÷ 3 and that of the period 5 by using the formula (142 + 120 + 153) ÷ 3. For k = 4,
the forecast for period 7 is calculated by using the formula (120 + 153 + 156 + 135) ÷ 4. All forecasts have been
rounded to the nearest integer.

28
 Time series Analysis - Moving average
(continued)
2) Moving Average (continued)

 These results don’t show a trend or a seasonality in sales (random trend);

 The larger K, the smaller the variation of the demand.
29
 Time series Analysis - Moving average
(continued)

2) Moving Average advantages/challenges

• This method makes it possible to reduce the background noise
created by the random variations of demand from one period to
another;
• It can lead to a certain slowness in responding to changes in demand;

• If K is small: forecasts will be influenced by random and non-random

variations of demand (used if the demand doesn’t vary too much).
• If K is large : forecasts will be less influenced by random variations of
demand, will be influenced by non-random variations of demand and
will also be slower to respond to a real trend (used if the demand
does vary too much).
30
 Time series Analysis - Weighted moving average

3) Weighted moving average

• We weight the data to give more importance to recent periods;

• The weights of demands of periods K numbers don’t have the same

importance;

• Note that all combinations of weights are possible, with one

condition that the total weights must be 100%.

31
𝐹 (𝑡+1)= 𝑤𝑒𝑖𝑔h𝑡1× 𝑑𝑒𝑚𝑎𝑛𝑑(𝑡)+ 𝑤𝑒𝑖𝑔h𝑡 2× 𝑑𝑒𝑚𝑎𝑛𝑑(𝑡 −1)+...+ 𝑤𝑒𝑖𝑔h𝑡(𝑡 − 𝑘)+1× 𝑑𝑒𝑚𝑎𝑛𝑑(𝑡 −𝑘)+1
 Time series Analysis - weighted moving average

3) Weighted moving average

• Example: if we calculate the weighted moving average of the last three weeks, we can decide that the
forecast of last week will have a weight of 50%, the previous week a weight of 30%, and the most distant
week a weight of 20 %. We will obtain the following table.

Weeks 1 2 3 4 5 6 7 8 9 10 11 12

Sales=demands 125 142 120 153 156 135 128 117 140 134 132 126

Forecasts at k = 3
129 138 143 148 140 127 128 130 135
Unweighted

Forecasts at k = 3
weighted at 128 141 148 145 136 124 131 132 134
50 % - 30 % - 20 %

Example of calculation : For unweighted forecasts, the forecasts for period 4 are calculated by
(125 + 142 + 120) ÷ 3; For the weighted forecasts, the forecasts for period 4 are calculated by
using the formula (50% × 120) + (30% × 142) + (20% × 125). The results were rounded to the
nearest integer.
32
Time series Analysis-Weighted moving average

3) Weighted moving average advantages/challenges

• It reflects more quickly the trends upwards or downwards (because the more

recent trends have more weight);

• The effect of random variations is lessened;

• Challenge:

- In addition to decide for the K value, it is also necessary to determine the

value of the weight to be allocated for each period;

- Several combinations are possible, it is necessary to choose the combination

which provides the best results, which may be long.

33
Time series Analysis - Simple exponential smoothing

4) Simple exponential smoothing

• Based on the principle of weighted moving average, but which is more refined and easier to use;
• Instead of considering the demands of the previous periods, it is based on the demand of the
most recent period but also considers that the forecast of the first period corresponds to its
demand;
• Doesn’t consider the K number of previous periods nor determining the weight relative for each
period;
• Choice of smoothing factor α (0.05-0.5);
• To calculate each Forecast, we use the following formula:

Ft + 1 = Ft + α(Dt − Ft)
Ft = Forecast of period t
Dt = demand of period t
α = choosed smoothing factor

34
 Time series Analysis - Simple exponential
smoothing(Continued)
4) Simple exponential smoothing
• Example : by using the formula [Ft + 1 = Ft + α(Dt − Ft)], we calculate the forecasts for differents weeks for
each α (0.1, 0.3, 0.5).

Weeks 1 2 3 4 5 6 7 8 9 10 11 12

Sales=demands 125 142 120 153 156 135 128 117 140 134 132 126

Forecast with
125 125 126,7 126,03 128,73 131,45 131,81 131,43 129,99 130,99 131,29 131,36
α = 0,1

Forecast with
125 125 130,1 127,07 134,85 141,19 139,34 135,94 130,25 133,18 133,42 133
α = 0,3

Forecast with
125 125 133,5 126,75 139,88 147,94 141,47 134,73 125,87 132,93 133,47 132,73
α = 0,5

 Week 1: It is assumed that the forecast of the week of departure (week 1) is equal to the demand.
 Week 2: by using the formula, we obtain (Dt-Ft) = 0, then the forecasts for week 2 become 125 whatever the value of
the smoothing factor α .
 Week 3: with the smoothing factor α = 0.1, the forecast is calculated by using the formula: forecast for week 2+
smoothing factor α x (week 2 demand-week 2 forecast) = 125 + 0.1x (142-125) = 126.7.
Here we keep the decimals to avoid errors (we rounded up at the end of the calculations).
35
 Time series Analysis - Simple exponential
smoothing (continued)
4) Simple exponential smoothing
Fig 1. Fig 2.

• Figure 1: for α=0.5 , the forecasts reflect the fluctuations in demand from previous periods. While at α=0.1 the fluctuations
disappear almost entirely. This represents an advantage in the case of random fluctuations.
• Figure 2: in the case where the demand shows a real trend , with a random variation during the first six weeks and a clear
downward trend, the forecasts made at α = 0.5 (green curve) quickly reflect this trend, while the forecasts at α = 0.1 react much
less strongly and slowly.
 The choice of the smoothing factor depends therefore on the context of the forecasts. 36
 Time series Analysis - Simple exponential
smoothing (continued)
4) Simple exponential smoothing advantages/challenges

• Frequently used method;

• Doesn’t require to determine neither a period K number nor a weight of each period;

• The unique choice is of smoothing factor α :

 If α is large: forecasts will react quickly to changes in demand (if these variations correspond to real
variations, this is an advantage, if they are caused by the random variations, it is a disadvantage). In
other words, if the demand doesn’t vary too much , we choose a high smoothing factor α.

 If α is small: the effects of random variation of the demand are reduced, but Forecasts will be slow
to respond to real changes in demand. In other words, if the demand varies too much, we choose a
low smoothing factor α.

37
Time series Analysis -Trend analysis

5) Trend analysis
• The previous methods of time series analysis respond to the
existence of trends, but they do not extrapolate the past
trend to establish future forecasts;
• Figure 2 shows that the data for the last six weeks showed a
downward trend that appeared to be linear. These data can
be used to calculate the rate of decline of demand.
38
 Time series Analysis - Trend analysis
5) Trend analysis

Weeks 7 8 9 10 11 12

Sales=demands 128 122 115 108 103 97

Decrease rate of
4,7 % 5,7 % 6,1 % 4,6 % 5,8 %
demand

• Example of calculation : for week 8, the decline rate is [(128 - 122) ÷ 128] = 4.7%; For period 9,
it corresponds to [ (122 - 115) ÷ 122] = 5.7%.
To calculate the average of decrease rate for all periods : [(4.7%+5.7%+6.1%+4.6%+5.8%) ÷ 5]=5.4%

• The rate of decrease is fairly constant, indicating a linear decrease which is on

average 5.4% per week;
• It could be expected that if this trend is maintained, demand in week 13 would be
92 corresponding to (97- (97x5.4%)=92);
• This analysis remains less accurate.
39
Time series Analysis - Analysis of seasonalities

6) Analysis of seasonalities

• Based on calculation of the demand average of each and of all previous

periods;

• It allows to calculate the seasonality index that helps to determine the

demand forecasts for the next periods.

40
Time series Analysis - Analysis of seasonalities

6) Analysis of seasonalities
• Example: The following table shows sales of new vehicles in Canada.

Months Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec.

Year 1 83 512 101 788 148 052 152 187 157 082 156 891 150 800 138 210 137 349 125 731 118 521 114 376

Year 2 86 320 98 147 156 452 162 613 152 064 168 119 144 171 143 653 137 660 129 049 124 466 117 863

• We calculate the demand average of each month then we calculate the demand
average of all months.
• This will help us to calculate the seasonality index.

Demand average of period t *100%

Seasonality Index =
Demand average of all periods

41
Time series Analysis - Analysis of seasonalities
6) Analysis of seasonalities
The following table shows the results of calculations for each month : demand average, seasonality index,
forecasts for the third year.
Average of
monthly
Months Jan. Feb. Mar. Apr. May. Jun. Jul. Aug. Sep. Oct. Nov. Dec.
demand
average
Year 1 83 512 101 788 148 052 152 187 157 082 156 891 150 800 138 210 137 349 125 731 118 521 114 376
133 545
Year 2 86 320 98 147 156 452 162 613 152 064 168 119 144 171 143 653 137 660 129 049 124 466 117 863

Average 84 916 99 968 152 252 157 400 154 573 162 505 147 486 140 932 137 505 127 390 121 494 116 120

Seasonality
63,59 % 74,86 % 114,01 % 117,86 % 115,75 % 121,69 % 110,44 % 105,53 % 102,97 % 95,39 % 90,98 % 86,95 %
Index *(SI)

Forecasts
87 118 102 558 156 194 161 468 158 578 166 715 151 303 144 576 141 069 130 684 124 643 119 122
Year 3

 Average: the average of January correspond to: 84916=[(83512+86 320)÷2].

 Seasonality Index (SI): for January is calculated by using the formula: 63.59%=(84916÷133545)x100%. The results
showed that the sales are lower in the fall and winter (October to February) (where the SI is inferior than 100%)
and they are higher in spring and summer (SI is superior than 100%).
 Forecasts for year 3: by suggesting that the trend analysis of last years allowed to predict for the third year
1 644 000 sales of vehicles, which is translated by a monthly average of 137 000 vehicles (1644000/12). To calculate the
forecasts for January of the third year, we use the following calculation: 63.59%x137 000=87 118 vehicles. 42
 Time series Analysis - Analysis of seasonalities
6) Analysis of seasonalities advantages/challenges

• More used by industries and organizations whose demand is influenced by

significant seasonalities (weekly, monthly, or daily).

• It helps to forecast the demand for each hour, day, week, month, etc.

• These forecasts will be useful in planning resources and determining

schedules for staff and the need for part-time employees.

43
 Error Measurement

• Error: is the difference between the actual demand

and the forecasted demand for each period.

• Error measurement: helps to choose the optimal

method.

Error (ou deviation) = Demand − Forecast

44
 Error Measurement-Average of Absolute
Deviation

1. Average of absolute deviation

Weeks 1 2 3 4 5 6 7 8 9 10 11 12

Sales=demands 125 142 120 153 156 135 128 117 140 134 132 126

Forecasts at k = 3
129 138 143 148 140 127 128 130 135
Unweighted

Absolute Deviations 24 18 8 20 23 13 6 2 9

Forecats at k = 3
weighted 128 141 148 145 136 124 131 132 134
at 50 % - 30 % - 20 %

Absolute Deviations 25 15 13 17 19 16 3 0 8

• The average absolute deviation for the unweighted moving average method is 13.6 (by calculating [24 + 18
+ 8 + 20 + 23 + 13 + 6 + 2 + 9) ÷ 9], while the weighted average is 12.9 [ (25+15+13+17+19+16+3+0+8)/9].
These results indicate that weighted moving average method is more appropriate.

45
 Exercise 1
True or False
1. The independent demand is not forecasted, while the dependent demand is forecasted;
2. It is possible to obtain accurate demand forecasts if sufficient time and effort are devoted to
it;
3. The longer you forecast the demand in advance, the more accurate are the forecasts;
4. The forecasts are more accurate for a product than for a family of objects;
5. The demand decreases in function of the product cycle;
6. The forecast horizon is related to the required accuracy level;
7. The qualitative and quantitative forecasts methods are complementary;
8. The exploratory analysis includes factors analysis;
9. Delphi method allows to collect information from internal experts;
10. The qualitative method is useful only when we don’t have access to data.
46
 Exercise 2

Source: https://www.youtube.com/watch?v=N0HBUjzLCAw 47
 Exercise 2 (continued)
1. It’s about demand forecast for which service? And what is the forecasting
horizon in this case?
2. Who is responsible for undertaking the air-service demand forecasts?
3. What are the main forecasted demands?
4. What information is used to carry out the forecasts of passenger and
cargo demand?
5. What are the results of the preliminary analysis?
6. What are the factors that are taken into account and analysed to forecast
the demand for air traffic?
7. After undertaking the whole analysis, what are the results of the Air traffic
forecasts for 2030? 48
 Exercise 3
Bakery Matins Dores

The bakery Matins Dores produces honey donuts that it distributes through a grocery network in the Montreal area. Donuts are produced by
using an ancestral secret recipe, which requires that the dough should be prepared several days in advance (the aging of the dough is the
secret of the lightness and special taste of these donuts).
Located in the east of the city, the company cooks and packs the donuts during the night according to the order book of the day. The delivery
trucks depart at four o'clock every morning (except on Sundays) in order to supply customers with fresh donuts of the day.
The company being closed on Sundays. Saturday delivery includes orders for the next two days.
Mr. Horton, who has been the owner of the company for a few months, is facing several problems of overstocking or under-production of the
dough due to difficulty in forecasting the demand.
One of two things, one must throw out dough produced in too large quantity, or one lacks dough to produce in sufficient quantity. In order to
remedy the situation, he considers the sales of the last four weeks (see the table below) and tries to analyse them.
Present graphically the daily sales of donuts over the last four weeks. Comment the appearance of the donuts request at Matins Dores.
Daily sales (in dozens of donuts)

Days Four weeks ago Three weeks ago Two weeks ago Last week
Monday 2200 2400 2300 2400
Tuesday 2000 2100 2200 2200
Wednesday 2300 2400 2300 2500
Thursday 1800 1900 1800 2000
Friday 1900 1800 2100 2000
Saturday 2800 2700 3000 2900
49
 Exercise 4
Following the results he obtained by testing different quantitative methods to predict the demand of skateboards, Elias Chamma decides to use the
exponential smoothing method with a smoothing factor α=0.2. In the following year (from Month 16 to 27), he obtained the results reproduced in the table
below. Months Realized sales Forecasted sales
Exponential smoothing α=0.2
12 16 980 16 980
13 17 280 16 980
14 17 030 17 040
15 17 230 17 088
16 16 950 17 076
17 16 540 17 051
18 16 750 16 949
19 17 020 16 909
20 16 740 16 931
21 16 420 16 893
22 16 650 16 798
23 16 450 16 768
24 16 230 16 704
25 16 360 16 609
26 16 400 16 559
27 16 320 16 527

To help Elias to make it clear, prepare an Excel spreadsheet that can repeat the calculations that Elias made the year before:
Establish forecasts from week 16 to week 27 by using:
the moving average at k = 3 and at k = 4,
the weighted moving average at k = 4, with a weighting of 50% -25% -15% -10%;
the exponential smoothing with α = 0.2 and with α = 0.5, (note: since we already have past forecasts, the new forecasts must continue using
sales from the end of the previous year and the forecast of the week 15=17088).
According to the obtained results, calculate the average of the absolute deviation and determine if Elias should continue using the exponential
50
smoothing with α = 0.2? If yes, why? If not, which method should he used to obtain the optimal forecasts results? Justify your answer.
Synthesis

51