Demand Forecasting

Chapter 5
Umar Farooq
 Introduction – Demand Driven
Economy
• Organizations are moving from
supply driven to demand-driven
supply chain.
• Customers dictating to the
supplier
what products they desire and
when they need them delivered.
• Consumers have become more
demanding and discriminating.

• If you build it, they will come. Not anymore

 Introduction
• There are several ways to match supply and demand.
 Different Approaches to match supply-demand:
1) Supplier hold plenty of stock available for delivery at any
time.
 It is also expensive because of the cost of carrying inventory
 The possibility of write-downs at the end of the selling season
2) Use of flexible pricing is another approach.
3) Companies can also use overtime, subcontracting, or
temporary workers to increase capacity to meet demand.
4) Companies use the forecast approach to match its supply and
demand cycle.
 Introduction
Write-Down of prices

• “If customers don’t want the item, you may be

forced to write-down the item’s price below
what you paid. Thus, if the value of inventory
declines, your company incurs a financial loss.
Holding stock for too long increases your chance
the market price for the item will fall below
what you actually paid for it.”

So, hold plenty of stock is not a good idea

 Introduction
• Push and Pull Strategy:
• In the push system production orders begin upon
inventory reaching a certain level.
• While on the pull system production begins based
on demand (forecasted demand).
 Introduction
• It is important to manage demand, especially in pull
manufacturing environments.

• Improved forecasts benefit all trading partners to mitigates

supply-demand mismatch problems.
 Demand Forecasting
• Estimate about the future is very important in the changing
business scenario.
• Customer behavior, technologies available and need for
individual products is changing faster than ever.
• Accurate estimates provides strong base for planning and
sound business decisions.

Where to invest?
How much to invest?
When to invest?
 Demand Forecasting
• PREDICTIONS ABOUT THE FUTURE ARE DIFFICULT,.
• A forecast is an estimate of future demand & provides the basis
for planning decisions.
• The goal is to minimize forecast error (Supply-Demand errors).

What are the risks of wrong forecast in business?

• The benefits of better forecasts are

 lower inventories,
 reduced stock-outs,
 smoother production plans,
 reduced costs
 improved customer service.
Zara produces around 450 million
items a year.
How can it stay so efficient with
the sheer volume that passes
through its supply chain?
 Demand Forecasting - Zara
• Zara introduces 10,000 new designs every year.
• Distributes 5.2 million clothing articles per week to
a network of over 2200 stores in more than 96
countries.
• Their high product mix and vast global network
makes demand forecasting for Zara a challenging
endeavor.
• How can it stay so efficient with the sheer volume
that passes through its supply chain?
The secret behind Zara's retail
success
 Demand Forecasting
• Having accurate demand forecasts allows;
 The purchasing department to order the right
amount of products,
 The operations department to produce the right
amount of products at right time and
 The logistics department to deliver the right
amount of products at right time.
 Examples
Examples of Forecasting:

Traffic flow at a major junction

Pre-poll opinion survey amongst
voters
Weather forecast
 Demand Forecasting
Demand Forecasting
 Elements of a Good Forecast
A properly prepared forecast should meet the following
requirements:

1. The forecast should be accurate.

2. The forecast should be timely.
3. The forecast should be reliable.
4. The forecasting technique should be simple to understand
and use.
5. The forecast should be expressed in meaningful units.
6. The forecast should be in writing.

14
Steps involved in Forecast
 Demand Forecasting
Demand Forecasting
 Steps in the Forecasting Process
There are seven basic steps in the forecasting process:
1. Determine the purpose of the forecast.
2. Select the items to be forecast.
3. Establish a time horizon.
4. Select the forecasting technique.
5. Gather and analyze relevant data.

6. Prepare the forecast.

7. Monitor the results.

16
 Demand Forecasting
Demand Forecasting
Classification of Forecasting Methods
Forecasting

Quantitative Qualitative

Time Series Associative Executive

Models Models Opinions

Salesforce
Naive Techniques Techniques Techniques Simple Multiple
Forecast for Averaging for Trend for Seasonality Linear Regression Linear Regression Opinions

Consumer
Simple Trend Trend and Seasonal
Moving Average Exponential Smoothing Exponential Smoothing Surveys
Weighted
Moving Average
Trend
Projection
Trend
Projection
Market
Research
Simple
Exponential Smoothing
Market
Surveys

Market
Tests

Delphi
17
Method
 Forecasting Techniques

Qualitative

Forecasting
Quantitative

“If you don't have the numbers,

Time Series
it's just your opinion.”
 Forecasting Techniques

• Qualitative forecasting is based on opinion &

intuition.

• Quantitative forecasting uses mathematical

models & historical data to make forecasts.
 Time series models are the most frequently used
among all the forecasting models.
 Forecasting Techniques
Qualitative Forecasting Methods
•Based on intuition or judgmental evaluation.
•Uses expert judgment, rather than numerical analysis.
•Generally use
 when data are limited,
 Unavailable
 Not currently relevant.
Four qualitative models used are:
1. Jury of executive opinion
2. Delphi method
3. Sales force composite
4. Consumer survey
 Forecasting Techniques (Cont.)
Quantitative Methods
• Time series forecasting- based on the assumption
that the future is an extension of the past. Historical
data is used to predict future demand.
• Cause & Effect forecasting- assumes that one or
more factors (independent variables) predict future
demand.
It is generally recommended to use a combination
of quantitative & qualitative techniques.
 Forecasting Techniques (Cont.)
• Jury of executive opinion
Panel of experts in same field with
experience & working knowledge.
Exchange of ideas and claims.
Combines input from key
information sources.
Final decision is based on majority
or consensus, reached from
expert’s forecasts.
• Applicable for long-range planning
and new product introductions.
 Forecasting Techniques (Cont.)
• Advantages
Can be undertaken easily without the use of statistical tools.
Several individuals with considerable experience working
together

• Disadvantages
The statistics are not collected from the market.
If one member’s views dominate the discussion, then the
value and reliability of the outcome can be diminished
 Forecasting Techniques
Delphi method
 Very similar to jury of executives method
but this time members are both from inside
and outside of the company
 One coordinator is chosen by members of
the jury
 Group members do not physically meet.
 A questionnaire is given to each member of
the team which asks question.
 Translate opinion into some conclusion for
forecast.
 The summary of responses is then sent out
to all the experts
 2 to 3 cycles are undertaken
 Convergence and diversion is acceptable.
 Forecasts are revised until a consensus is
reached by all.
Forecasting Techniques (Cont.)
 Forecasting Techniques (Cont.)

• Advantages
Eliminates need for group meetings
Participants can change their opinions anonymously

• Disadvantages
Time consuming –reaching a consensus takes a lot of
time.
Participants may drop out.
 Forecasting Techniques (Cont.)
• Sales Force Composite
– Each sales-person makes a product-by-product forecast for their particular sales
territory.
– Is generated based on the sales personal’s knowledge of the market and
estimates of customer needs.

• Advantages
 Due to the proximity of the sales personnel to the consumers, the forecast tends
to be reliable
• Disadvantages
 Individual biases could negatively impact the effectiveness of forecast.
 some agents may give a lower forecast than the actual potential of sales
to easily achieve their target and get the money bonus from the company
on the extra sales generated.
 Forecasting Techniques (Cont.)
• Consumer Survey
• Involves asking consumer about there buying habits, new product
ideas and opinions about existing products.
• The survey is administered through telephone, mail, Internet, or
personal interviews
• Data collected from the survey are analyzed using statistical tools
and judgment

• Advantage: Simple and Easy

• Disadvantage: buyers might change their opinions
 Forecasting Techniques (Cont.)
Quantitative Methods
Time series forecasting
Cause & Effect forecasting (Causal)
 Forecasting Techniques (Cont.)
Time series forecasting
• The time series analysis method predicts the future sales by
analyzing the historical relationship between sales and time.
– use of a model to predict future values based on previously observed values.
Components of Time Series
Data should be plotted to detect for the following components:
Trend variations: increasing or decreasing trend
Cyclical variations: wavelike movements that are longer
than a year and are not of fixed period (e.g., business cycle)
Seasonal variations: show ups and downs that repeat over a
consistent/fix interval such as The effect of seasons – spring,
summer, Autumn and winter, or seasons
Random variations: due to unexpected or unpredictable
events; such as natural disasters (hurricanes, tornadoes, fire),
strikes and wars.
 Forecasting Techniques (Cont.)

Figure 1: Product Demand Charted Over 4 Years with a Growth Trend and Seasonality

Seasonal peaks Trend

component

Average demand
over four years

Random variation

Year Year Year Year

1 2 3 4
 Forecasting Techniques (Cont.)
• Trend variations: During the growth and
decline stages of the product life cycle, a
consistent trend pattern in terms of demand
growth or demand decline can be observed.
• Seasonal variations: A seasonal
pattern(e.g., quarter of the year, month of
the year, week of the month, day of the
week) exists when demand is influenced by
seasonal factors. i.e. short time variation
• Cyclical variations: Demand gradually
increases and decreases over an extended
period of time, such as years. Business
cycles (recession/expansion) product life
cycles influence this component of demand
 Forecasting Techniques (Cont.)
• Time Series Forecast: Time series models are the
most widely used (72 percent) and judgmental models are the
least used (11 percent).

• Common Time Series approaches are

naïve forecast,
simple moving average,
weighted moving average and
exponential smoothing
 Time Series Forecast
Naive forecast:
• The estimate for the next period is equal to the actual demand for the
immediate past period.

Ft+1 = At

Where Ft+1 = forecast for period t+1

At = actual demand for period t

• This method is inexpensive to understand and develop.

• The method may not generate accurate forecasts.
Example- Naïve Forecast
 Simple Moving Average Forecast
• Calculates the overall trend in a data set using
historical data.
• Works well if demand is stable over time and is
extremely useful for forecasting long-term trends .
• For Example: if you have sales data for a twenty-year
period, you can calculate a five-year moving average,
a four-year moving average, a three-year moving
average and so on.
• Stock market analysts often use a 50 or 200 day
moving average to help them see trends in the stock
market and forecast where the stocks are headed.
 Simple Moving Average Forecast
• Puts equal weight on each of the historical results being used

Forecast = Sum of last n demands

n
Example-Moving Average Forecast
Example-Moving Average Forecast

Each of the observations used to compute the forecasted

value is weighted equally.
 Weighted Moving Average Forecast
• The weighted average of the n-period observations, using
unequal weights.
• The weights should be non-negative and sum to one.
• Weighted Moving Average puts more weight on recent data and
less weight on old data.
Weighted Moving Average Forecast

Fourth most recent period = 0.1

Third most recent period = 0.2
Second most recent period =0.3
Most recent period = 0.4
Total 1.0
 Exponential Smoothing Forecast
• A type of weighted moving average. Only two data points are needed.
• The forecast for next period’s demand is the current period’s forecast
adjusted by a fraction of the difference between the current period’s
actual demand and forecast.
• This approach requires less data than the weighted moving average
method because only two data points are needed.
• This method is suitable for forecasting data with no trend or seasonal
pattern.
Ft+1 = Ft+(At - Ft) or
Ft+1 = At + (1 – ) Ft
Where
Ft+1 = forecast for Period t + 1
Ft = forecast for Period t
At = actual demand for Period t
 = a smoothing constant (0 ≤  ≤1)
Exponential Smoothing Forecast

Ft+1 = Ft+(At - Ft)

Smoothing Constant
0≤α≤1
 Exponential Smoothing Forecast
The forecast for next period’s demand is the current period’s forecast
adjusted by a fraction of the difference between the current period’s
actual demand and forecast.

Next period Current period Difference b/w Current period’s

forecast forecast actual demand and forecast

Ft+1 = Ft+(At - Ft)

Ft+1 = At + (1 – ) Ft
Exponential Smoothing Forecast

At Ft
 Exponential Smoothing Forecast
Calculate the forecast for period 3 using the exponential smoothing
method. Assume the forecast for period 1 is 1600. Use a smoothing
constant (α) of 0.3.

F11600
= A1 1-α = 1-0.3 = 0.7
1600
1780 Ft+1 = At + (1 – ) Ft
Ft+1 = 0.3(At) + 0.7(Ft)
F2 = 0.3(1600) + 0.7(1600)
F2 = 1600
F3 = 0.3(2200) + 0.7(1600)
F3 = 1780
 Linear Trend Forecast
• Linear Trend Forecast

• Linear trend forecasting is used to impose a line of

best fit to time series historical data.
• Purpose is to find the best possible trend for future
demand forecasts.

Lets Visualize it first

Linear Trend Forecast
 Linear Trend Forecast
Y = mx + b
Ŷ = b1x + b 0
where
Y = forecast or dependent variable
x = time variable
b = intercept of the line
m = slope of the line
Demand
m = Rise/Run
m = ∑xy – nx y
∑x2 – n (x)2
Time
Example: Linear Trend Forecast

Y = mx + b
Example: Linear Trend Forecast

Y = mx + b

x = ∑x / n =3.5
y = ∑y / n =2233.4
Example: Linear Trend Forecast

x = 3.5
y = 2233.4

m = 285

Y = mx + b
Example: Linear Trend Forecast

y = mx + b x = 3.5
Y = 2233.4
b = y - mx = 1235.9
M = 285
Y = mx + b
Y = 285x + 1235.9

Forecast for 7th month or July = Y = 285(7) + 1235.9 = 3230.9

 Class Activity
• The demand for toys produced by the Miki Manufacturing
Company is shown in the table below.

• What is the trend line?

• What is the forecast for period 13?
Class Activity

Y = mx + b

Y=? X=? m=?

 Class Activity

Trend Line for first 12 months = Y = 1236.4 + 286.7X

Y = 1236.4 + 286.7(13)

Forecast for period 13 = Y = 4964 Toys

 Cause and Effect Forecasting
• Cause and Effect forecasting assumes that
one or more factors are related to demand
and that the relationship between cause and
effect can be used to estimate future
demand.
• The cause-and-effect models have a cause
(independent variable or variables) and an
effect (dependent variable).
• Cause and Effect Forecasting Models
Simple Linear Regression Forecast
Multiple Linear Regression Forecast
 Cause and Effect Forecasting
• Regression analysis is a statistical technique that attempts to
explore and model the relationship between two or more variables.
• For example: An analyst may want to know if there is a relationship
between road accidents and the age of the driver.
• Simple Linear Regression Forecast: Demand is dependent on
only one variable.
Ŷ = b0 + b1x
Y = mx + b
Y = forecast or dependent variable
x = time variable Explanatory or independent Variable
b = intercept of the line
m = slope of the line

x variable is no longer time but instead an explanatory variable of demand.

For example, demand could be dependent on the size of the advertising budget.
 Cause and Effect Forecasting
• Multiple Regression Forecast: When several
explanatory variables are used to predict the dependent variable
• There is one variable to be forecast and several predictor variables.
Intercept Independent variables

Ŷ = b0 + b1x1 + b2x2 + . . . Bkxk + Error

Y = c + m1x1 + m2x2 + m3x3 + …+ mkxk + error

where
Ŷ = forecast or dependent variable
xk = kth explanatory or independent variable
b0 or m = intercept of the line
bk or mk = regression coefficient of the independent
variable xk
 Cause and Effect Forecasting
• Example: Banks score loan customers based on a lot of
personal information. A sample of 500 customers from an
Australian bank provided the following information.
Cause and Effect Forecasting
 Forecast Error

• Forecast error is the difference between the actual

quantity and the forecast.
• The ultimate goal of any forecasting endeavour is to have
an accurate and unbiased forecast.
• Forecast error can be expressed as:

Forecast error, et = At - Ft
where
et = forecast error for Period t
At = actual demand for Period
Ft = forecast for Period t
 Forecast Error
• There are several measures of forecasting accuracy
follow:

Mean absolute deviation (MAD)

Mean squared error (MSE)
Mean absolute percentage error (MAPE)
 Forecast Error
• Mean absolute deviation (MAD):
• It is the average of the absolute value, or the difference between
actual values and their average value.
• The mean absolute deviation of a dataset is the average distance
between each data point and the mean. It gives us an idea about the
variability in a dataset.

where et = At -Ft
et = forecast error for period t
At = actual demand for period t;
n = number of periods of evaluation
 Forecast Error
The difference between actual values and their average value.
 Forecast Error
• Mean absolute percentage error (MAPE)-Provides a perspective
of the true magnitude of the forecast error.

where
et = forecast error for period t
At = actual demand for period t;
n = number of periods of evaluation
 Forecast Error
• Mean squared error (MSE)-
• The mean squared error tells you how close a regression line is to a
set of points.
• It does this by taking the distances from the points to the
regression line (these distances are the “errors”) and squaring them.
• The squaring is necessary to remove any negative signs.

Where
et = forecast error for period t
n = number of periods of evaluation
 Forecast Error
• Running Sum of Forecast Errors (RSFE)-
• Indicates bias in the forecasts or the tendency of a forecast to be
consistently higher or lower than actual demand.
• The RSFE tells us whether our forecast is biased to always be too
high, or always be to low.
n
Running Sum of Forecast Errors, RSFE = e
t 1
t

Where
et = forecast error for period t

How big is a big enough error that we should do something about

it?
 Forecast Error
• Tracking signal determines if forecast is within acceptable
control limits.
• If the tracking signal falls outside the pre-set control limits,
there is a bias problem with the forecasting method and an
evaluation of the way forecasts are generated is warranted.
• If Tracking Signal > 3.75 then there is persistent under
forecasting.
• If this is less than -3.75 then, there is persistent over-
forecasting
• Positive tracking signals indicate that demand is greater than
forecast.
• Negative signals mean that demand is less than forecast.
RSFE
Tracking Signal =
MAD
Given the following data, compute the tracking
signal and decide whether or not the forecast
should be reviewed.

Actual Forecast
Month Sales Sales
1 8 10
2 11 10
3 12 10

70
Tracking signal is computed as the running sum of forecast error
(RSFE) divided by MAD. We compute RSFE by summing up
the forecast errors over time. Forecast errors for January is the
difference between its actual and forecast sales. RSFE for
January is equal to the cumulative forecast errors.

e
Actual Forecast Forecast
Month Sales Sales Error RSFE t
1 8 10 -2 -2 t 1

2 11 10
3 12 10
4 14 10

Forecast Error = RSFE = -2

Actual – Forecast =
8 -10 = -2
71
Forecast errors for February is the difference between its actual
and forecast sales. RSFE for February is equal to the cumulative
forecast errors of January and February.

Actual Forecast Forecast

Month Sales Sales Error RSFE
1 8 10 -2 -2
2 11 10 1 -1
3 12 10
4 14 10

Forecast Error = RSFE = -2 + 1

Actual – Forecast = = -1

11 -10 = 1
72
Forecast errors for March is the difference between its actual and
forecast sales. RSFE for March is equal to the cumulative
forecast errors of January, February and March.

Actual Forecast Forecast

Month Sales Sales Error RSFE
1 8 10 -2 -2
2 11 10 1 -1
3 12 10 2 1
4 14 10

Forecast Error = RSFE = -1 + 2

Actual – Forecast = = 1

12 -10 = 2
73
Forecast errors for April is the difference between its actual
and forecast sales. RSFE for April is equal to the
cumulative forecast errors of January, February, March
and April.

Actual Forecast Forecast

Month Sales Sales Error RSFE
1 8 10 -2 -2
2 11 10 1 -1
3 12 10 2 1
4 14 10 4 5

Forecast Error =
RSFE = 1 + 4
Actual – Forecast =
= 5
14 -10 = 4
74
 MAD for January is computed by averaging the absolute
errors over time. Tracking signal for January is computed
by dividing its RSFE by MAD.

Actual Forecast Absolute Tracking

Month Sales Sales RSFE Error MAD Signal
1 8 10 -2 2 2 -1
2 11 10 -1
3 12 10 1
4 14 10 5

Absolute Error = MAD = 2 TS = RSFE/MAD

Absolute (Actual – Forecast) = = -2/2 = -1

Absolute(8 -10) = 2
75
MAD for February is computed by averaging the absolute
errors of January and February. Tracking signal for
February is computed by dividing its RSFE by MAD.

Actual Forecast Absolute Tracking

Month Sales Sales RSFE Error MAD Signal
1 8 10 -2 2 2 -1
2 11 10 -1 1 1.5 -0.67
3 12 10 1
4 14 10 5

Absolute Error = MAD = (2+1)/2 TS = RSFE/MAD

Absolute (Actual – Forecast) = = 1.5 = -1/1.5 = -0.67

Absolute(11 -10) = 1
76
MAD for March is computed by averaging the absolute
errors of January, February and March. Tracking signal for
March is computed by dividing its RSFE by MAD.

Actual Forecast Absolute Tracking

Month Sales Sales RSFE Error MAD Signal
1 8 10 -2 2 2 -1
2 11 10 -1 1 1.5 -0.67
3 12 10 1 2 1.67 0.6
4 14 10 5

Absolute Error = MAD = (2+1+2)/3 TS = RSFE/MAD

Absolute (Actual – Forecast) = = 1.67 = 1/1.67 = 0.6

Absolute(12 -10) = 2
77
MAD for April is computed by averaging the absolute
errors of January, February, March and April. Tracking
signal for April is computed by dividing its RSFE by MAD.

Actual Forecast Absolute Tracking

Month Sales Sales RSFE Error MAD Signal
1 8 10 -2 2 2 -1
2 11 10 -1 1 1.5 -0.67
3 12 10 1 2 1.67 0.6
4 14 10 5 4 2.25 2.22

Absolute Error =
TS = RSFE/MAD
Absolute (Actual – Forecast) =
MAD = (2+1+2+4)/4 = 5/2.25 = 2.22
Absolute(14 -10) = 4
= 2.25

78
Since the tracking signals for months January to April are
within +/- 4, the forecast needs not be reviewed.

Tracking
Month Signal
1 -1
2 -0.67
3 0.6
4 2.22

79
 Tracking signal

Signal exceeded limit

Tracking signal
Upper control limit = +4MAD
+

0 0 MAD

-
Lower control limit = -4MAD

Time