Week 12_Forecasting

1. Demand Forecasting

Madiyar Sagnayev, MSc SCLM
Distribution of course materials is prohibited.

2. Levels of planning

3.

“The goal of supply chain management is very simple – to try to match
supply & demand. However, what makes this seemingly simple task so
difficult in reality is the presence of uncertainty.”
Martin Christopher, 2016, pg. 95

4. Demand Forecasting. Definitions

• “Forecasting is about predicting the future as accurately as possible, given all of the information available,
including historical data and knowledge of any future events that might impact the forecasts”
• We make forecasts every day (how long to travel to university, when to go the gym, etc.).
• Luckily estimating journey times has become a lot easier with tools such as Google Maps, (Google, 2020).
According to Cooper (2018) Google has been collecting data on traffic via its routing algorithms since 2011
giving past and current data. This is combined with data gathered from any via Apple and Android smart
phone users that have the Google Maps app open to give live traffic updates. The more users of its app
the greater the volume of data available and the more accurate the estimates of journey time become.
• Forecasting is the process that enables demand planners and supply chain professionals to estimate
customer demand for a product based on prior sales data and other contributing data factors.
• Creating a demand forecast helps companies better understand, predict, and plan their products and
make better business decisions. The demand plan is used to ensure that there will be enough product
supply to meet customer demand.
What other apps we have to forecast day to day
activities?

5. Forecast horizon

The forecast horizon is the length of time into the future for which forecasts are to be
prepared. These generally vary from short-term forecasting horizons (less than three
months) to long-term horizons (more than two years).
1.Short-range forecast: It is typically less than 3 months but has a time span
of up-to 1 year. Needed “for the scheduling of personnel, production and
transportation” (Hyndman & Athanasopoulos, 2018, p.13)
2.Medium-range forecast: It is typically 3 months to 1 year but has a time
span from one to three years. It is needed to ”purchase raw materials, hire
personnel, or buy machinery and equipment” (Hyndman &
Athanasopoulos, 2018, p.13)
3.Long-range forecast: This has a time span of three or more years. It is
needed to “take account of market opportunities, environmental factors
and internal resources” (Hyndman & Athanasopoulos, 2018, p.13)

6. Forecast horizon

Long-range forecast - aggregated data
In the automotive industry for a company like Ford this would mean forecasting the total number of vehicles to
be manufactured
Medium-range forecast - partially disaggregated
In a company like Ford a partially disaggregated forecast might mean breaking the forecast down by vehicle
type, vans, hatchbacks, sedans, multi-utility vehicle (MPV), sports-utility vehicles (SUV) etc.
Short-range forecast – totally disaggregated data
For Ford this would involve a forecast for every vehicle model and could also include forecasts for individual
options on each model, e.g. for each type of power unit offered, petrol engine, diesel engine, electric battery etc
The forecast horizon will vary from industry to industry, with nuclear power
generation needing one of the longest forecasting horizons to plan how to cope with
the nuclear facility at the end of its life.

7. Forecast error

• Forecasting is not an exact science and errors are inevitable
• Forecast error is the difference between forecast value and the actual value.
• “all forecasts are prone to error and the further ahead the forecast horizon is, the
greater the error” (Martin Christopher, 2016, p.95)

8. Types of demand

• Demand is independent if it is unrelated to demand for
any other product or service.
• Demand is dependent if it is derived from the demand
for another product or service.
• Independent demand needs to be forecast; however,
requirements for dependent demand are calculated
from the independent items.
Any examples of Independent and
Dependent Demand?

9. Why do we forecast?

• Lead-time gap
According to Christopher (2016, p.95) ”most organisations face a fundamental
problem: the time it takes to procure, make and deliver the finished product to a
customer is longer than the time the customer is prepared to wait for it”.
The customer’s order cycle refers to the length of time that the customer is prepared to wait, from
when the order is placed through to when the goods are received. This is the maximum period
available for order fulfilment. In some cases this may be measured in months but in others it is
measured in hours.

10. Why do we forecast?

In the conventional organization the only way to bridge the gap between the logistics lead-time (i.e. the
time taken to complete the process from procurement to delivered product) and the customer’s order
cycle (i.e. the period they are prepared to wait for delivery) is by carrying inventory. This normally
implies a forecast. Hence the way most companies address this problem is by seeking to forecast the
market’s requirements and then to build inventory ahead of demand. Unfortunately all our experience
suggests that no matter how sophisticated the forecast, its accuracy is always less than perfect. It has
been suggested that all mistakes in forecasting end up as an inventory problem – whether too much or
too little!
Reducing the gap can be achieved by shortening the logistics lead-time (end-to-end pipeline time)
whilst simultaneously trying to move the customer’s order cycle closer by gaining earlier warning of
requirements through improved visibility of demand.

11. Why do we forecast?

Summary
Production time > Demand time (This is the production time, or the ‘P time’ and the
time the customer is prepared to wait is the demand time, or ‘D time’. The gap between the P time and the D time is the
lead- time gap, P time > D time.)
The only way to bridge the gap is by carrying inventory.
All mistakes in forecast end up in inventory.
Whilst improving forecast accuracy is the desirable
goal, it may be that the answer is in reducing lead-time
gap.
Production time (P) > Demand time(D)

12.

ETO Products are designed and built to customer
specifications; examples according to Slack, et al. (2016, p.
325) include website development (service operation), large
construction projects
MTO Products are based on a standard design. The
component production and manufacture of the final product
is linked to the customised order placed by the final
customer. Examples according to Slack, et al. (2016, p. 325)
include a hair blow-dry bar (service) or house builder with a
standard design
ATO Products are built to customer specifications from a
stock of existing components and this usually entails a
modularised product design. Examples according to Slack, et
al. (2016, p. 325) include internet retail fulfilment (service)
and Dell computers. The construction industry has been
relatively slow to adopt this approach but the use of prefabricated ‘modules’ made in a factory and delivered to the
construction site complete for bathrooms and kitchens is
becoming more common, for example in student apartments.
As you move from Engineer-To-Order to MakeTo-Stock the lead-time gap increases, the
period for which you need to forecast, or the
forecast horizon, increases and forecast
accuracy declines
MTO Products are built against a sales forecast, and sold to
the customer from finished goods stock. Examples
according to Slack, et al. (2016, p. 325) include preserved
food production, such as canned vegetables, and domestic
appliances, such as a washing machine. MTS operations are
not possible in a pure service operation because there is
nothing physical to hold in stock.

13. How do we handle lead time gap? Managing D time

• To make forecasting, logistics and operations management easier we want
to do all we can to increase the D time and reduce the P time, thereby
reducing the lead-time gap and moving the decoupling point.
• Managing the demand time: differentiation strategy, marketing, dynamic
pricing, promotions.
• One of the most recognisable differentiated products where demand
exceeds supply is a Ferrari car. In July 2019 Ferrari commercial and
marketing boss Enrico Galliera stated that their unique models “have the
longest waiting list of
all products
our product
lines” atyou
four
fivetoyears,
What
as a customer
are to
ready
wait for(Burgess,
2019). Toyota have attempted to separate
their more differentiated cars
half a year?
by using the Lexus brand, as opposed to the Toyota brand.
• Other ways to influence demand include the use of dynamic pricing and
promotions. For example, in air travel with low cost airlines, such as
EasyJet, we are now accustomed to booking our flights early if we want
the lowest price and accept that the ticket price is ‘dynamic’.
Source: Christopher, M. (2016)

14. How do we handle lead time gap?

• Managing the production time: postponement strategy, late configuration,
improving visibility of demand
Source: Christopher, M. (2016)

15. What is Collaborative Planning, Forecasting & Replenishment (CPFR)?

What is Collaborative Planning, Forecasting & Replenishment
(CPFR)?
“CPFR is a supply chain strategy in which members of the supply chain work
towards best practices in planning the flow of production from the first link in the
chain to the last link in the chain. CPFR for distribution inventories may involve
planning for such activities as replenishment, vendor-managed inventory,
forecasting, assortment optimization, retail store clustering, and presentation
stock.” (American Production & Inventory Control Society (APICS), 2018a)
• Supply Chain partner collaboration
• Aligns goals and multiple S&OP across the supply chain
• Both customer and supplier participate in the forecast
• Can react to exceptions together
• Mitigates the Forrester (or Bullwhip) Effect

16. Factors that are related to the demand forecast

• Past demand
• Lead time of product replenishment
• Planned advertising or marketing efforts
• Planned price discounts
• State of the economy
• Actions that competitors have taken
Source: Chopra, 2016, pg. 192

17. What is involved in the forecasting process?

• Decide what the forecast is for, level of forecast accuracy needed and the
forecasting budget
• Select the items to be forecast
• Determine time horizon
• Gather and ‘pre-process’ the data (i.e. remove outliers)
• Select the forecasting technique(s)
• Make the forecast(s)
• Evaluate the forecast accuracy
• Review / validate /challenge assumptions and variables
• Feedback and revise

18. What is involved in the forecasting process?

• Marketing - Prediction of market characteristics, market share, market
price and trends in new product development (Research &
Development, R&D), promotions
• Sales - Ownership of the forecast numbers. Prediction of product
demand, and the addition of market intelligence.
• Production - Conversion of forecasted demand into inventory and
resource planning and use of forecast error in safety stock calculations.
• Finance - Budgeting. Predictions of cash flow, expenses and revenues.
• Personnel or human resources - Prediction of human resource
requirements and supply e.g. demographic trends. Predicting the level
of labour turnover, absenteeism and tardiness (or lateness), and
seasonal labour.
• General Management- Prediction of general economic trends that
affect the long-term future of the business. Business growth and
development.

19. What are the traditional forecast models

20. Forecasting methods

Forecasting
methods
Qualitative
Quantitative
Market surveys
Time Series
Delphi
Causal
Management
Judgment
Qualitative forecasting - subjective, based on the
opinion and the judgment of consumers and experts;
they may be appropriate (1) when past data is not
available or (2) when experts have market intelligence
that may affect the forecast or (3) there is a need to
forecast demand several years into the future in a
new industry.
Delphi: Several rounds of questionnaires are sent out
to the group of experts, and the anonymous
responses are aggregated and shared with the group
after each round. The experts are allowed to adjust
their answers in subsequent rounds, based on how
they interpret the “group response” that has been
provided to them. Since multiple rounds of questions
are asked and the panel is told what the group thinks
as a whole, the Delphi method seeks to reach the
correct response through consensus.
https://www.youtube.com/watch?v=Axq7hNLOal8
In which cases you will use Qualitative forecasting
method?

21. Forecasting methods

Forecasting
methods
Qualitative
Quantitative
Market surveys
Time Series
Delphi
Causal
Management
Judgment
• Time series: using historical
demand to make a forecast. It
is based on the assumption
that past demand history is a
good indicator of future
demand.
• Causal forecasting method
assume that the demand
forecast is highly correlated
with certain factors in the
environment.
Source: Chopra, 2016, pg. 192

22. What factors influence the selection of the forecast model or technique?

• Time period
• Availability of historical data
• Level of innovation in the
product or service
• Level of accuracy needed
• Cost / benefit ratio
• Amount of segmentation

23. What are the features and assumptions for quantitative forecasts?

Features:• data exists
• data is accurate
• events are not random
Assumptions:• the future is in some way related to the past
• the level of accuracy expected is considered reasonable

24.

25.

An upward trend line indicates growth. A downward trend points to
a decline in demand
Provide an example of upward / downward trend in KZ
market

26. An upward trend line indicates growth. A downward trend points to a decline in demand

Cyclical variation
Cyclical variation - this is variation of a cyclical nature that can be identified over a period of time. For
example, economic data may be affected by business cycles with a period varying between about 5 and 7
years, an example of world growth. Ideally, the historic cyclical element should be identified and removed
from past sales data. According to Hyndman & Athanasopoulos (2018, p. section 8.1) it is not considered
possible to forecast the cycle if the cycles are not of fixed length and in this situation the data should be
classified as ‘stationary’ (Data with only random or residual variation)

27. Cyclical variation

Seasonal variation
Provide an example of seasonal variation in KZ market
Seasonal variation – defined as variation within the year, often linked to weather, such as fluctuations in ice
cream sales. Other examples of ‘seasonality’ include demand from new University students for bed linen, kitchen
equipment before the start of the new academic year in October. The above figure shows the number of UK
residents visiting abroad and shows a clear seasonal variation, with a peak in the Summer months when UK
children are on holiday and most families take their main annual holiday. However, the data also appears to show
an upward trend.

28. Seasonal variation

Irregular variation
After the Icelandic
volcano, Eyjafjallajökull,
erupted in April 2010 in
Europe 100,000 flights
and 10 million passenger
journeys were cancelled.
Irregular variation - caused by irregular circumstances that do not reflect typical behavour, for example the
financial crash of 2007-2008. These data points might be considered as outliers and removed. An example of an
irregular event, a volcanic eruption in 2010, on the number of flights in Europe.

29. Irregular variation

Group work
• Provide examples of forecasting techniques / methods / approaches
used by Kazakhstani companies and compare them to global
companies (e.g. Magnum (KZ) VS ASDA (UK))
• What are the alternatives to forecast using excel spreadsheet exist
(e.g. AI tools, software)
20 min to prepare, 10 to present. (2 ppt slides)

30. Group work

Basic approach to demand forecasting
• Understand the objective of forecasting
• Integrate demand planning and forecasting throughout the supply
chain
• Identify the major factors that influence the demand forecast
• Forecast at the appropriate level of aggregation
• Establish performance and error measures for the forecast.

31. Basic approach to demand forecasting

Types of time-series forecasting
• Simple moving average
• Exponential smoothing
• Weighted moving average

32. Types of time-series forecasting

Simple moving average
• This method is used when demand has no observable trend or seasonality.
• Systematic component (expected value of demand) = level
• To compute the new moving average, we simply add the latest observation.
• The moving average corresponds to giving the last N periods of data equal weight when
forecasting and ignoring all data older than this new moving average. As we increase N,
the moving average becomes less responsonsive to the most recently observed demand.
Source: Chopra, 2016, pg. 192

33. Simple moving average

Simple moving average. Example
• The Agricultural Market Report published by DEFRA indicates weekly
sales of wheat cereals in Great Britain over the four weeks of April
2019 to be 38, 35, 77 and 90 thousand tons. Calculate the sales
forecast for the first week of May using a 4-period moving average?
What is the forecast error if the sale in the first week of May turns out
to be 80 thousand tons?

34. Simple moving average. Example

• We need to forecast for Period 5 at the end of Period 4.
•T=4
• The forecast for demand in Period 5 is (38+35+77+90)/4 = 60
• Forecast error is 60 – 80 = -20

35. Simple moving average. Example

How does simple moving average treat data from past time periods?
Assigns equal weighting to all past periods
Weight
1/n
n
...
3
2
1

36. How does simple moving average treat data from past time periods?

How do you choose the value of n in simple moving average?
The larger n is (the more observations included) the smoother the result
- it increases the likelihood to eliminate randomness
But,
The longer the length of the moving average the more terms (and therefore
information) is lost in the process of averaging
- it might smooth out cycles or non random variations of interest

37. How do you choose the value of n in simple moving average?

Simple exponential smoothing
• This method is appropriate when demand has no observable trend or seasonality.
• Simple moving average assumes that all past observations have equal weight and it does
not use data beyond the n periods. Exponential smoothing overcomes these problems
and includes the forecasts made for the previous period and the actual demand for this
period in the forecast for the next period.
• So, the benefit is that it takes the most recent observations into account and weights them
accordingly. For example, if we are looking at four years of demand forecasts,
April/May/June of 2014 will likely be weighted differently in April/May/June of 2019. The
exponential smoothing method takes this into account and enables us to plan inventory
more efficiently by placing greater weight on more relevant, recent data.
• Another benefit is that spikes in the data aren’t quite as detrimental to the forecast. In
exponential smoothing, the most recent forecast has the greatest weight and therefore
should be the most accurate in predicting demand, as opposed to the moving averages
method where the weight for each period is fixed.

38. Simple exponential smoothing

Simple moving average
Simple exponential smoothing

39.

Simple exponential smoothing. Example
• Weekly sales for wheat cereals in
Great Britain has been 38, 35, 77
and 90 thousand tons over the 4
weeks of April 2009. Calculate
the sales forecast for period 5
(first week of May) using simple
exponential smoothing with
a=0,1.

40. Simple exponential smoothing. Example

Actual Demand
Forecast
period 1
38.00
60.00
period 2
35.00
57.80
period 3
77.00
55.52
period 4
90.00
57.67
period 5
60.90
a
0.10

41. Simple exponential smoothing. Example

How to choose best smoothing constant?
• Most textbooks provide general recommendations on the magnitude of
the smoothing constants. For example, both Schroeder, Rungtusanatham,
& Goldstein (2013) and Jacobs & Chase (2013) suggest values of α between
0.1 and 0.3.
• Heizer & Render (2011) and Stevenson (2012) advocate a wider range: 0.05
to 0.50. Chopra & Meindl (2013) prescribe α values no larger than 0.20.
• Most textbooks also recommend that smoothing constants be chosen so
that forecasts are more accurate, with accuracy measured by Mean
Absolute Deviation (MAD), Mean Squared Error (MSE), Mean Absolute
Percent Error (MAPE), or some other summary metric. For example,
Chopra & Meindl, while prescribing values of α no larger than 0.2, go on to
say “in general, it is best to pick smoothing constants that minimize the
error term that a manager is most comfortable with from among MAD,
MSE, and MAPE.”

42. How to choose best smoothing constant?

16000
14000
12000
DEMAND
10000
8000
6000
4000
2000
0
1
4
7
10
13
16
19
22
25
28
31
34
37
40
WEEK
Actual Demand
Forecast Alpha 0.3
Forecast Alpha 0.5
Forecast Alpha 0.7
43
46
49
52

43.

Weighted moving average
• Weighted moving average assigns different, not equal weights to the historic data.
• It is a simple & quick technique to set weights to cope with a trend or seasonality.
• In certain cases, it might be beneficial to put more weight on the observations that are closer to the
time period being forecast.

44. Weighted moving average

Weighted moving average. Example
The demand for defense
machinery for a certain
project is given each
month as follows.
The defense officer is
asked to forecast the
demand for the 11th
month using three period
moving average technique.
Use a weighting scheme
of 0.5, 0.3, 0.2
1
120
2
110
3
90
4
115
5
125
6
117
7
121
8
126
9
132
10
128
11
???

45. Weighted moving average. Example

46. Weighted moving average. Example

Patterns present
Forecasting technique
Stationary
Simple moving average
Exponential smoothing
With a trend
Weighted moving average with sum of weights >1 for
upward trend or <1 for downward trend
With seasonality
With a trend & seasonality
Weighted moving average
Holts-Winters
Weighted moving average
Holts-Winters

47.

What is forecast accuracy?
• Forecast are considered accurate when errors are low
• It is important to distinguish between errors and noise
• Noise is uncontrollable and will cause error
• Optimal forecasts using time series will understand the
pattern perfectly and the error will be caused only by
noise….but this is difficult to tell!
• Optimal forecasts will identify if there is a ‘turning point’ or
major change to the pattern of demand

48. What is forecast accuracy?

What is forecast bias?
• Bias is the persistent tendency to err in the
same direction
• It can be a result of lack of knowledge about
the data pattern
• It can be unintentional
• It can also be intentional (for example,
where a high level of service is important,
forecasts may be biased high)

49. What is forecast bias?

How can bias and accuracy combine?
Less accurate but unbiased
Forecast
Demand
More accurate but biased
Forecast
Demand

50. How can bias and accuracy combine?

How do you measure accuracy?
• Accuracy is measured by the error between the forecast and demand
• Errors can be positive or negative
Demand
+1
+4
-3
-2
Time

51. How do you measure accuracy?

What different measures of accuracy are
there?
• Mean Absolute Deviation (MAD)
• Mean of the absolute variation between the actual demand
and the forecast
• Mean Absolute Percentage Error (MAPE)
• Mean of the absolute percentage variation between actual
demand and the forecast divided by the actual demand
• Mean Squared Error (MSE)
• Mean of the squared variation between forecast and actual
demand

52. What different measures of accuracy are there?

How to calculate forecast error?
• Mean error: the average error is calculated. A large positive ME
means that forecast is underestimating demand; large negative ME
means that forecast is overestimating demand.
mean = sum of all values/number in the set
• For which industries positive ME is better than negative ME?

53. How to calculate forecast error?

• Mean square error: the errors are squared
before they are averaged. The underlying
assumption is that doubling an error will
quadruple the consequence.

54. How to calculate forecast error?

• Mean absolute deviation: the absolute value of
the errors is averaged and the positive and
negative errors do not cancel each other out. The
value of MAD should be as small as possible. This
measure may be appropriate when costs of
having too much demand and the costs of having
too much supply are similar.

55. How to calculate forecast error?

• Mean absolute percentage error: it is a
good technique when the underlying
forecast has significant seasonality and
demand varies considerably from one
period to the next. The absolute error
is expresses as a percentage of the
actual demand.

56. How to calculate forecast error?

Characteristics of forecast
• Forecasts are always innacurate and should thus incude both the expected value of the
forecast and a measure of forecast error. Consider 2 car dealers with following forecasts:
one of them expects sales to range between 900 and 1 100; the second one expects
sales between 100 and 1 900 units.
• Long-term forecasts are usually less accurate than short-term forecasts; that is, longterm forecasts have a larger standard deviation of error relative to the mean than shortterm fore- casts
• Aggregate forecasts are usually more accurate than disaggregate forecasts. What is
easier to forecast: GDP for any given year or a yearly revenue for a company with less
than a 2% error?
• The farther the supply chain is from the consumer, the greater the distortion of
information it receives.
Source: Chopra, 2016, pg. 190

57. Characteristics of forecast

Homework
• Think around how to choose the best smoothing constant for
exponential smoothing forecast technique.
• Explore Holts-Winter & other time-series forecasting techniques.
• Try to use weighted moving average for the data in the Excel.