Foreign Trade University

Learning Objectives
 Defining and understanding different types of
forecasting
 Clarifying features common of forecasting
 Understanding and able to apply quantitative and
qualitative approaches for demand forecasting
 Analyzing forecast accuracy
“GIANT” WAL-MART
 WAL-MART’S DATA WAREHOUSE
 500 terabytes: One of the world’s largest data
warehouses
 The systems track:
- Point-of-sale data at each store,
- Inventory levels by store,
- Products in transit,
- Market statistics,
- Customer demographics,
- Finance,
- Product returns,
- Supplier performance, etc...
 WAL-MART’S DATA WAREHOUSE
 The data are used  Data mining: demand-
for decision forecasting application
support:
- Analyzing trend,
- Managing Seasonal sales profile of
inventory, 100,000 products
- Understanding
customers. Market-basket analysis:
Relationship and
patterns in customer
purchases
Forecast
 Forecast – a statement about the future value of a
variable of interest
 We make forecasts about such things as weather,
demand, and resource availability
 Forecasts are important to making informed
decisions
 Forecast is familiar with the management science
(facts or conditions) and an art (past experiences...)
 Demand forecasting is one of the earliest and most
popular in OM
Two Important Aspects of Forecasts
 Expected level of demand
 The level of demand may
be a function of some
structural variation such ??
as trend or seasonal
variation
 Accuracy
 Related to the potential
size of forecast error
Forecast Uses
 Plan the system
 Generally involves long-range plans related to:
 Types of products and services to offer
 Facility and equipment levels
 Facility location
 Plan the use of the system
 Generally involves short- and medium-range plans related to:
 Inventory management
 Workforce levels
 Purchasing
 Production
 Budgeting
 Scheduling
Features Common to all Forecasts
 Forecasting techniques assume that the same
underlying causal system that existed in the past will
continue to exist in the future.

 Forecasts are rarely perfect.

 Forecasts for groups of items tend to be more accurate

than forecasts for individual items.

 Forecast accuracy decreases as the time horizon

increases.
Forecasting Time Horizons
• Short-range forecast

 Up to 1 year, generally less than 3 months

 Purchasing, job scheduling, workforce levels, job

assignments, production levels
Forecasting Time Horizons
 Medium-range forecast

 3 months to 3 years

 Sales and production planning, budgeting

Forecasting Time Horizons
 Long-range forecast

 3+ years

 New product planning, facility location, research

and development
Forecasts Are Not Perfect
 Forecasts are not perfect:
 Because random variation is always present,
there will always be some residual error, even if
all other factors have been accounted for.
Elements of a Good Forecast
The forecast
 Should be timely
 Should be accurate
 Should be reliable
 Should be expressed in meaningful units
 Should be in writing
 Technique should be simple to understand and use
 Should be cost-effective
Steps for Forecasting
 Determine the use of the forecast
 Select the items to be forecasted
 Determine the time horizon of the forecast
 Select the forecasting model/technique(s)
 Gather and analyze the data
 Make the forecast
 Validate and implement results
Approaches to Forecasting
• Qualitative methods

• Quantitative methods
Approaches to Forecasting
Qualitative methods

 Used when situation is vague and little data

exist
 New products
 New technology
 Includes soft information
 e.g., human factors, personal opinions,
hunches…
Approaches to Forecasting
Quantitative methods

 Used when situation is ‘stable’ and historical

data exist
 Existing products
 Current technology
 Involves mathematical techniques
 e.g., forecasting sales of LCD full HD
Overview of Qualitative Methods

 Executive opinions

 Sales force opinions

 Consumer survey

 Delphi method
 Executive Opinions
 Involves small group of high-
level experts and managers
 Group estimates demand by
working together
 Combines managerial
experience with statistical
models
 Relatively quick
 ‘Group-think’
disadvantage
 Sales Force Opinions
 Each salesperson projects his or her sales

 Combined at district and national levels

 Sales representatives know customers’ wants

 Tends to be overly optimistic, some of them can’t

identify want/need or demand
 Consumer Survey
 Ask customers about purchasing plans

 What consumers say, and what they actually do

are often different

 Sometimes difficult to answer or they can easily

be affected by others’ opinions, etc.
 Delphi Method
 Iterative group
process, continues
until consensus is Decision Makers
reached (Evaluate responses
and make decisions)
Staff
 3 types of participants (Administering
survey)
 Decision makers
 Staff
 Respondents
Respondents
(People who can make
valuable judgments)
 Delphi Method
Steps to do:
1. Choose respondents;
2. Create a questionnaire and send it to respondents;
3. Collect and analyze respondents’ answers, then re-do
the questions if possible;
4. Send a questionnaire to respondents again;
5. Collect and analyze answers secondly, if the answers
can satisfy decision makers, the survey can be
finished. If they can’t, the admin staffs have to redo
the questionnaire...
Overview of Quantitative Methods
1. Naive approach
2. Moving averages
3. Exponential smoothing Time-Series Forecasting
4. Trend projection (time
series trend)

5. Linear regression Associative Forecasting

Time Series Forecasting
 Forecasts that project patterns identified in recent
time-series observations
 Time-series – a time-ordered sequence of observations
taken at regular time intervals

 Assume that future values of the time-series can be

estimated from past values of the time-series
Time-Series Behaviors
 Trend
 Seasonality
 Cycles
 Irregular variations
 Random variation
Trends and Seasonality
 Trend
 A long-term upward or downward movement in
data
 Population shifts

 Changing income

 Seasonality
 Short-term, fairly regular variations related to the
calendar or time of day
 Restaurants, service call centers, and theaters all
experience seasonal demand
 Cycles and Variations
 Cycle
 Wavelike variations lasting more than one year
 These are often related to a variety of economic, political, or
even agricultural conditions
 Irregular variation
 Due to unusual circumstances that do not reflect typical
behavior
 Labor strike

 Weather event

 Random Variation
 Residual variation that remains after all other behaviors have
been accounted for
Time-Series Forecasting
Naive Method
 Assumes demand in next period is the same as demand
in most recent period
Ft = Dt-1
Ft : Demand forecasted at time t
Dt-1: Actual demand at time t-1
e.g., If January sales were 5000 products, then
February sales will be 5000 products
 Can be used with
 A stable time series
 Seasonal variations
 Trend
Naive Method
Advantages
 Sometimes cost effective and efficient
 Can be good starting point
 Suitable for demand forecasting with trend

Disadvantages:
 Naïve and not precise
 Not suitable for demand forecasting with
seasonal component => big error!
Time-Series Forecasting - Averaging
 These techniques work best when a series tends to vary
about an average
 Averaging techniques smooth variations in the data
 They can handle step changes or gradual changes in
the level of a series
 Techniques
1. Moving average
2. Weighted moving average
3. Exponential smoothing
Moving Average Method
 MA is a series of arithmetic means
 Used if little or no trend
 Used often for smoothing
 Provides overall impression of data over
time
n

D t −i
Ft = i =1
n
1. n → 
2. n is finite number
Moving Average Method
 As new data become available, the forecast is updated
by adding the newest value and dropping the oldest
and then re-computing the average
 The number of data points included in the average
determines the model’s sensitivity
 Fewer data points used—more responsive
 More data points used—less responsive
Moving Average Example
Making forecast for the following months:
Month Actual Demand
(100 units)
1 98
2 102
3 103
4 97
5 95
6
With n =3, making forecast for the next coming
months:

Month Actual Sales (units)

1 900
2 1100
3 1300
4 1050
5 950
6 1249
7
Weighted Moving Average
 Used when trend is present
 Older data usually less important
 Weights based on experience and intuition
n
Ft = D
i =1
t −i  t −i
Ft : Forecast at time t
Dt-i : Actual demand at time t - i
t-i : Weight at time t- i
n
Condition for t-i : 
i =1
t −i =1 0   t −i  1
Example: with n =3 and weights are t-1 = 0.5 ; t-2 =
0.3; t-3 = 0.2, making forecast for the following
months.

Sales Actual Demand

1 900
2 1100
3 1300
4 1050
5 950
6 1250
7
 Exponential Smoothing
 Form of weighted moving average
 Weights decline exponentially
 Most recent data weighted most

Ft = Ft-1 + (Dt-1 - Ft-1) = Dt-1 + (1-)Ft-1

 Requires smoothing constant ()
 Ranges from 0 to 1
 Subjectively chosen
 Involves little record keeping of past data
 Choosing 
 The objective is to obtain the most accurate forecast no
matter the technique

 We generally do this by selecting alpha that gives us the

lowest forecast error
Forecast error (e)= Actual demand - Forecast value

= Dt - Ft
Example: Making forecast for following months with
 = 0.8 and it is assumed that D1 = F1

Month Actual Sales (units)

1 900
2 1100
3 1300
4 1050
5 950
6 1250
7
Making forecast with  = 0.1 or  = 0.4. Which  is
better? Assume that D1= F1.
Month, i Actual Forecast, Ft
Demand, α=0.1 α=0.4
Dt
Ft0.1 e Ft0.4 e
1 100
2 110
3 120
4 115
5 125
6
 Linear Trend
 Assumption
β1 ≠ 0
β1 is in the population regression model

y t = β 0 + β1x + ε
 Linear Trend
 Consider the time series data as bivariate
 The procedure for conducting the method
- Step 1: Take the physical time points as values of
the independent variable x (coded as 1, 2, 3, etc).
- Step 2: Take the data values themselves as values
of the dependent variable y
- Step 3: Calculate the regression line of y on x, y=
a+bx
- Step 4: Translate the regression line as t=a+bx,
where any given value of time point x will yield a
corresponding value of the trend, t.
Formulas
n n n
n xiyi −  xi  yi
n n n n

  yi −  xi  xiyi
xi
2

a= i =1
n
i =1 i =1
n
i =1
b= i =1
n
i =1
n
i =1

n xi − ( xi )
2 2
n  xi 2 − ( xi ) 2
i =1 i =1
i =1 i =1

or
xy − x y
b= a = y − bx
( x ) − ( x)
2 2
 Example
Estimate a trend line using regression analysis
Year Time Σx = 21
Period (x) Sales (y)
Σy = 275
Σxy = 1130
2015 1 20
2016 2 40  = 91
x 2

2017 3 30
2018 4 50
2019 5 70
2020 6 65
 Example
xy − x . y 188.333 − 3.5  45.833
b= = = 9.5714
x2
15.167 − (3.5) 2

a = y − bx = 45.833 − 9.5714  3.5 = 12.3358

 Translate the regression line as: t = a+bx

t (x = 1) = 12.3358 + 9.5714 * 1 = 21.9072
t (x = 2) = 12.3358 + 9.5714 * 2 = 31.4786
…
 Trend-Based Forecasting
 The linear trend model is:

t = 12.333 + 9.5714 x
Sales trend

80
70
60
50
sales

40
30
20
10
0
0 1 2 3 4 5 6 7
Year
 Trend-Based Forecasting
 Forecast for time period 7:
Time y= 12.333+9.5714(7)
Year Period (t) Sales
(y)
=79.33
2015 1 20
2016 2 40
2017 3 30
2018 4 50
2019 5 70
2020 6 65
2021 7 ??
Associative Forecasting
 Sometimes called as Causal relationship
 Used when changes in one or more independent
variables can be used to predict the changes in the
dependent variable.
 Most common technique is linear regression
analysis

 We apply this technique just as we did in the time

series example
Associative Forecasting
 Forecasting an outcome based on predictor
variables using the least squares technique

y = a + bx
where y= computed value of the variable to be
predicted (dependent variable)
a = y-axis intercept
b = slope of the regression line
x = the independent variable though to predict
the value of the dependent variable
Associative Forecasting
 Past data:
y1 y2 y3 ... yn
x1 x2 x3 ... xn
n: number of observations
Associative Forecasting
Y
y10

y8
y7
y9
Estimated
y6

y5
regression
y4 line
y3
y2

y1

0 1 1x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 X
Formulas
n n n
n xiyi −  xi  yi
n n n n

  yi −  xi  xiyi
xi
2

a= i =1
n
i =1 i =1
n
i =1
b= i =1
n
i =1
n
i =1

n xi − ( xi )
2 2
n  xi 2 − ( xi ) 2
i =1 i =1
i =1 i =1

or
xy − x y
b= a = y − bx
( x ) − ( x)
2 2
 Example
 The data below Cost of No of products
relates cost of advertisement sold
advertising (VND (VND m) (1000 units)
m)to number of 1 2
products sold (1000 3 8
units). 4 9
 Prepare the 5 15
regression model and 6 15
make a conclusion 7 20
about this relation. 9 23
12 25
 Forecast the number 14 22
of products sold if 15 36
cost of ads is 20 m.
VND
Correlation Analysis
 Correlation is a technique used to measure the
strength of the relationship between two variables.

 The stronger the correlation, the better the

relationship or the better fit the regression line and
vice versa.
Correlation Coefficient (r)
 The correlation coefficient is used to measure the
strength of the linear relationship between two
variables
 The product moment correlation coefficient is
calculated using the formula:

n xy −  x y
r=
n x − ( x) n y − ( y ) 
2 2 2 2
Correlation Coefficient (r)
-1 r  1

r < 0 : negative relationship between x and y

r > 0 : positive relationship between x and y

r→ 1/-1: strong linear relationship

Forecast Accuracy
 Mean absolute deviation (MAD)

MAD =
 Actual t − Forecast t
n

MAD weights all errors evenly

 Mean squared error (MSE)

 (Actual − Forecast t )
2

MSE = t

n −1
MSE weights errors according to their squared values
 The smaller MAD and MSE are, the better they are.
Forecast Accuracy
Actualt − Forecast t
 Actual t
100
MAPE =
n
MAPE weights errors according to relative error
Forecast Error Calculation
Actual Forecast (A-F)
Period
(A) (F) Error |Error| Error2 [|Error|/Actual]x100
1 107 110 -3 3 9 2.80%

2 125 121 4 4 16 3.20%

3 115 112 3 3 9 2.61%

4 118 120 -2 2 4 1.69%

5 108 109 1 1 1 0.93%

Sum 13 39 11.23%

n=5 n-1 = 4 n=5

MAD MSE MAPE

= 2.6 = 9.75 = 2.25%

Calculate MAD, MSE
Period Actual Forecast Error |e| e²
Demand (D-F)

1 217 215
2 213 216
3 216 215
4 210 214
5 213 211
6 219 214
7 216 217
8 212 216
Exercise 1
UK outward passenger movements by sea
(millions):

Qtr 1 Qtr 2 Qtr 3 Qtr 4

Year 1 2.2 5.0 7.9 3.2
Year 2 2.9 5.2 8.2 3.8
Year 3 3.2 5.8 9.1 4.1

Using regression model to extract the trend line

and making forecast for 4 quarters of year 4?
 Exercise 2
 The data below Cost of No of products
relates cost of advertisement sold
advertising (VND (VND m) (1000 units)
m)to number of 1 2
products sold (1000 3 8
units). 4 9
 Prepare the 5 15
regression model and 6 15
make a conclusion 7 20
about this relation. 9 23
12 25
14 22
15 36
Exercise 3
A director of industrial pump company has to choose one
of following methods in order to make demand forecasting.
If he takes MAD as standard measure, which method is
better, method 1 or method 2?

Month Actual demand Forecast Forecast

(Method 1) (Method 2)

1 492 488 495

2 470 484 482
3 485 480 478
4 493 490 488
5 498 497 492
6 492 493 493
Exercise 4
A CEO considers whether his company can take the
unemployment rate to measure the number of products sold
or not. Write down the regression model displaying the
relationship between those variables and make conclusion
based on your results.

Period 1 2 3 4 5 6 7 8 9 10
Unemployment 1.3 2.0 1.7 1.5 1.6 1.2 1.6 1.4 1.0 1.1
rate, %
Number of10 6 5 12 10 15 5 12 17 20
products sold
(1000 units)