Ir.

Haery Sihombing/IP
Sihombing/IP
Pensyarah Fakulti Kejuruteraan Pembuatan
Universiti Teknologi Malaysia Melaka
What is Forecasting?

Process of predicting a

6
;
future event
; Underlying basis of
all business decisions
??
; Production
; Inventory
; Personnel
; Facilities

FORECASTING

Forecasting and
Forecasting
Supply Chain Management
„ Accurate forecasting determines how much
„ Predicting the Future inventory a company must keep at various
„ Qualitative forecast points along its supply chain
methods „ Continuous replenishment
ƒ supplier and customer share continuously updated data
„ subjective
ƒ typically managed by the supplier
„ Quantitative forecast ƒ reduces inventory for the company
methods ƒ speeds customer delivery
„ based on mathematical „ Variations of continuous replenishment
formulas ƒ quick response
ƒ JIT (just-
(just-in-
in-time)
ƒ VMI (vendor-
(vendor-managed inventory)
ƒ stockless inventory
 Forecasting and TQM Types of Forecasting Methods
„ Accurate forecasting customer demand is a Depend on
key to providing good quality service
„ Continuous replenishment and JIT ¾ time frame
complement TQM ¾ demand behavior
ƒ eliminates the need for buffer inventory, which, in
turn, reduces both waste and inventory costs, a ¾ causes of behavior
primary goal of TQM
ƒ smoothes process flow with no defective items
ƒ meets expectations about on-
on-time delivery, which is
perceived as good-
good-quality service

Time Frame Forecasting Time Horizons

Indicates how far into the future is „ Short-

Short-range forecast
„ Up to 1 year, generally less than 3 months
forecast
„ Purchasing, job scheduling, workforce levels, job
¾ Short-
Short- to mid-
mid-range forecast assignments, production levels
„ typically encompasses the immediate future
„ Medium-
Medium-range forecast
„ daily up to two years
„ 3 months to 3 years
¾ Long-
Long-range forecast
„ Sales and production planning, budgeting
„ usuallyencompasses a period of time longer
than two years „ Long-
Long-range forecast
„ 3+ years
„ New product planning, facility location, research
and development
 Distinguishing Differences Influence of Product Life Cycle

¾ Medium/long range forecasts deal with Introduction – Growth – Maturity – Decline

more comprehensive issues and support
; Introduction and growth require longer
management decisions regarding planning
forecasts than maturity and decline
and products, plants and processes
; As product passes through life cycle,
¾ Short-
Short-term forecasting usually employs forecasts are useful in projecting
different methodologies than longer-
longer-term ; Staffing levels
forecasting ; Inventory levels
¾ Short-
Short-term forecasts tend to be more ; Factory capacity

accurate than longer-

longer-term forecasts

Product Life Cycle Product Life Cycle

Introduction Growth Maturity Decline Introduction Growth Maturity Decline

Best period to Practical to change Poor time to Cost control Product design Forecasting Standardization Little product
Company Strategy/Issues

increase market price or quality change image, critical and critical Less rapid differentiation
share image price, or quality development Product and product changes Cost
OM Strategy/Issues

critical process – more minor minimization

R&D engineering is Strengthen niche Competitive costs
Frequent reliability changes Overcapacity
critical become critical
product and Competitive Optimum in the
Defend market process design
position product capacity industry
Fax machines changes improvements
CD-
CD-ROM Increasing Prune line to
Short production and options stability of eliminate
Internet Drive-
Drive-through runs Increase capacity process items not
restaurants High production returning
Shift toward Long production
Color printers costs product focus runs good margin
Sales Limited models Reduce
3 1/2”
1/2” Enhance Product
Floppy Attention to distribution improvement capacity
Flat-
Flat-screen disks quality and cost cutting
monitors DVD

Figure 2.5 Figure 2.5

 Types of Forecasts Strategic Importance of Forecasting

; Human Resources – Hiring, training, laying

„ Economic forecasts off workers
„ Address business cycle – inflation rate,
; Capacity – Capacity shortages can result in
money supply, housing starts, etc. undependable delivery, loss of customers,
loss of market share
„ Technological forecasts
„ Predict rate of technological progress
; Supply-
Supply-Chain Management – Good supplier
relations and price advance
„ Impacts development of new products

„ Demand forecasts
„ Predict sales of existing product

Seven Steps in Forecasting The Realities!

1. Determine the use of the forecast ; Forecasts are seldom perfect

2. Select the items to be forecasted ; Most techniques assume an
3. Determine the time horizon of the underlying stability in the system
forecast ; Product family and aggregated
4. Select the forecasting model(s) forecasts are more accurate than
5. Gather the data individual product forecasts
6. Make the forecast
7. Validate and implement results
 Forecasting Approaches Forecasting Approaches

Qualitative Methods Quantitative Methods

; Used when situation is vague and ; Used when situation is ‘stable’
stable’ and
little data exist historical data exist
; New products ; Existing products
; Current technology
; New technology
; Involves intuition, experience ; Involves mathematical techniques
; e.g., forecasting sales of color
; e.g., forecasting sales on Internet
televisions

Qualitative Methods Overview of Qualitative Methods

o Management, marketing, purchasing,
and engineering are sources for
internal qualitative forecasts
o Delphi method
„ involves soliciting forecasts about technological
advances from experts
™ Jury of executive opinion
„ Pool opinions of high-
high-level executives, sometimes
augment by statistical models
™ Delphi method
„ Panel of experts, queried iteratively
™ Sales force composite
„ Estimates from individual salespersons are reviewed
for reasonableness, then aggregated
™ Consumer Market Survey
„ Ask the customer
 Jury of Executive Opinion Sales Force Composite

; Involves small group of high-

high-level managers
; Each salesperson projects his or her
; Group estimates demand by working
sales
together
; Combines managerial experience with ; Combined at district and national
statistical models levels
; Relatively quick ; Sales reps know customers’
customers’ wants
; ‘Group-
Group-think’
think’ ; Tends to be overly optimistic
disadvantage

Delphi Method Consumer Market Survey

; Iterative group Decision Makers

process, continues (Evaluate ; Ask customers about purchasing plans
until consensus is responses and
reached make decisions) ; What consumers say, and what they
actually do are often different
Staff
(Administering ; Sometimes difficult to answer
survey)

; 3 types of participants
; Decision makers Respondents
; Staff (People who can
make valuable
; Respondents judgments)
 Overview of Quantitative Time Series Forecasting
Approaches
1. Naive approach ; Set of evenly spaced numerical data
2. Moving averages ; Obtained by observing response
Time-
Time-Series
3. Exponential Models variable at regular time periods
smoothing ; Forecast based only on past values
4. Trend projection ; Assumes that factors influencing past
Associative and present will continue influence in
5. Linear regression Model future

Time Series Components Components of Demand

Trend
component
Demand for product or service

Seasonal peaks
Trend Cyclical
Actual
demand

Average
demand over
Random four years
Seasonal Random variation
| | | |
1 2 3 4
Year
 Demand Behavior Forms of Forecast Movement
„ Trend
„ a gradual, long-
long-term up or down movement of

Demand
Demand
demand
„ Random variations Random
„ movements in demand that do not follow a pattern movement

„ Cycle Time Time

(a) Trend (b) Cycle
„ an up-
up-and-
and-down repetitive movement in demand
„ Seasonal pattern
„ an up-
up-and-
and-down repetitive movement in demand

Demand
Demand
occurring periodically

Time Time
(c) Seasonal pattern (d) Trend with seasonal pattern

Trend Component Seasonal Component

; Regular pattern of up and down

; Persistent, overall upward or
fluctuations
downward pattern
; Due to weather, customs, etc.
; Changes due to population,
technology, age, culture, etc. ; Occurs within a single year
; Typically several years duration Number of
Period Length Seasons
Week Day 7
Month Week 4-4.5
Month Day 28-31
Year Quarter 4
Year Month 12
Year Week 52
 Cyclical Component Random Component

; Repeating up and down movements ; Erratic, unsystematic, ‘residual’

residual’
; Affected by business cycle, political, fluctuations
and economic factors ; Due to random variation or
unforeseen events
; Multiple years duration
; Short duration and
; Often causal or nonrepeating
associative
relationships

0 5 10 15 20 M T W T F

Forecasting Methods Forecasting Process

‰ Qualitative 1. Identify the
purpose of forecast
2. Collect historical
data
3. Plot data and identify
patterns
„ use management judgment, expertise, and
opinion to predict future demand 6. Check forecast 5. Develop/compute 4. Select a forecast
accuracy with one or forecast for period of model that seems
‰ Time series more measures historical data appropriate for data

„ statistical
techniques that use historical 7.
demand data to predict future demand Is accuracy of
forecast
No 8b. Select new
forecast model or
acceptable?
‰ Regression methods adjust parameters of
existing model

„ attempt to develop a mathematical relationship Yes

between demand and factors that cause its 8a. Forecast over 9. Adjust forecast based 10. Monitor results
and measure forecast
on additional qualitative
behavior planning horizon
information and insight accuracy
 Time Series Moving Average
„ Assume that what has occurred in the past will ™ Naive forecast
continue to occur in the future „ demand the current period is used as next
„ Relate the forecast to only one factor - time period’
period’s forecast
„ Include ™ Simple moving average
ƒ moving average „ stable demand with no pronounced behavioral
ƒ exponential smoothing patterns
ƒ linear trend line
™ Weighted moving average
„ weights are assigned to most recent data

Naive Approach Naïve Approach

ORDERS
FORECAST
MONTH PER MONTH
; Assumes demand in next period is the
Jan 120 -
same as demand in most recent period Feb 90 120
Mar 100 90
; e.g., If October sales were 90, then Apr 75 100
November sales will be 90 May 110 75
June 50 110
; Sometimes cost effective and efficient July 75 50
Aug 130 75
Sept 110 130
Oct 90 110
Nov - 90
 Simple Moving Average Simple Moving Average

; MA is a series of arithmetic means

n
; Used if little or no trend
6
i=1
Di
; Used often for smoothing
MAn =
n ; Provides overall impression of data over
where time
n =number of periods in the
moving average
Di = demand in period i ™ demand in previous n periods
Moving average = n

3-month Simple Moving Average 5-month Simple Moving Average

EXAMPLE EXAMPLE

3
ORDERS MOVING ORDERS MOVING
MONTH PER MONTH AVERAGE
6 Di MONTH PER MONTH AVERAGE 5
i=1
MA3 = 6 Di
Jan 120 – 3 Jan 120 – i=1
Feb 90 – Feb 90 – MA5 =
5
Mar 100 – 90 + 110 + 130 Mar 100 –
Apr 75 103.3 = 3 Apr 75 –
90 + 110 + 130+75+50
May 110 88.3 May 110 – =
June 50 95.0 June 50 99.0
5
July 75 78.3 = 110 orders July 75 85.0
Aug 130 78.3 for Nov Aug 130 82.0 = 91 orders
Sept 110 85.0 Sept 110 88.0 for Nov
Oct 90 105.0 Oct 90 95.0
Nov - 110.0 Nov - 91.0
 Smoothing Effects Weighted Moving Average

150 –

125 –
5-month
ƒ Adjusts WMAn = 6 Wi Di
i=1 i=1
100 –
moving
average where
Orders

75 –
method to Wi = the weight for period i,
50 – 3-month
more closely between 0 and 100
percent
25 –
Actual reflect data
| | | | | | | | | | |
fluctuations 6 Wi = 1.00
0–
Jan Feb Mar Apr May June July Aug Sept Oct Nov
Month

Weighted Moving Average Weighted Moving Average

Example 1

; Used when trend is present MONTH WEIGHT DATA

; Older data usually less important August 17% 130
; Weights based on experience and September 33% 110
October 50% 90
intuition 3
November Forecast WMA3 = 6 Wi Di
™ (weight for period n)
n) i=1

Weighted x (demand in period n)

n)
= (0.50)(90) + (0.33)(110) + (0.17)(130)
moving average = ™ weights
= 103.4 orders
 Potential Problems With
Example 2 Weights Applied Period
Moving Average
3 Last month
2 Two months ago
1 Three months ago
6 Sum of weights ; Increasing n smooths the forecast but
makes it less sensitive to changes
Actual 3-Month Weighted
Month Shed Sales Moving Average ; Do not forecast trends well
January 10 ; Require extensive historical data
February 12
March 13
April 16 [(3 x 13)
13) + (2 x 12)
12) + (10)]/6 = 121/6
(10)]/6
May 19 [(3 x 16) + (2 x 13) + (12)]/6 = 141/3
June 23 [(3 x 19) + (2 x 16) + (13)]/6 = 17
July 26 [(3 x 23) + (2 x 19) + (16)]/6 = 201/2

Moving Average And Exponential Smoothing

Weighted Moving Average

Weighted ƒ Averaging method

30 – moving
average ƒ Weights most recent data more strongly
25 –
ƒ Reacts more to recent changes
Sales demand

20 – Actual
sales ƒ Widely used, accurate method
15 –
Moving
10 – average

5 –
| | | | | | | | | | | |
J F M A M J J A S O N D
 Exponential Smoothing Exponential Smoothing (cont.)

„ Form of weighted moving average Ft +1 = DD

DDt + (1 - D)Ft
„ Weights decline exponentially where:
„ Most recent data weighted most Ft +1 = forecast for next period
„ Requires smoothing constant (D) Dt = actual demand for present period
„ Ranges from 0 to 1 Ft = previously determined forecast for
present period
„ Subjectively chosen
D=
D= weighting factor, smoothing constant
„ Involves little record keeping of past
data
OR

Exponential Smoothing (cont.) Effect of Smoothing Constant

New forecast = last period’

period’s forecast 0.0 dDd
Dd 1.0
+ D (last period’
period’s actual demand
– last period’
period’s forecast)
forecast) If D= 0.20Dt + 0.80 Ft
D= 0.20, then Ft +1 = 0.20

Ft = Ft – 1 + D(At – 1 - Ft – 1)
If D= 0Dt + 1 Ft 0 = Ft
D= 0, then Ft +1 = 0
Forecast does not reflect recent data
where Ft = new forecast
Ft – 1 = previous forecast If D= 1Dt + 0 Ft =Dt
D= 1, then Ft +1 = 1
D = smoothing (or weighting) Forecast based only on most recent data
constant (0 d D t 1)
 Exponential Smoothing (Į=0.30) Exponential Smoothing (cont.)

FORECAST, Ft + 1
PERIOD MONTH DEMAND (D = 0.3) (D = 0.5)
PERIOD MONTH DEMAND F2 = DD1 + (1 - D)F1
1 Jan 37 – –
1 Jan 37 = (0.30)(37) + (0.70)(37)
2 Feb 40 37.00 37.00
2 Feb 40 = 37
3 Mar 41 37.90 38.50
3 Mar 41
4 Apr 37 38.83 39.75
4 Apr 37 F3 = DD2 + (1 - D)F2
5 May 45 38.28 38.37
5 May 45 = (0.30)(40) + (0.70)(37) 6 Jun 50 40.29 41.68
6 Jun 50
= 37.9 7 Jul 43 43.20 45.84
7 Jul 43
8 Aug 47 43.14 44.42
8 Aug 47
F13 = DD12 + (1 - D)F12 9 Sep 56 44.30 45.71
9 Sep 56
= (0.30)(54) + (0.70)(50.84) 10 Oct 52 47.81 50.85
10 Oct 52
= 51.79 11 Nov 55 49.06 51.42
11 Nov 55
12 Dec 54 50.84 53.21
12 Dec 54
13 Jan – 51.79 53.61

Exponential Smoothing (cont.) Adjusted Exponential Smoothing

70 –

60 – Actual D = 0.50
AFt +1 = Ft +1 + Tt +1
50 – where
T = an exponentially smoothed trend factor
Orders

40 –
D = 0.30
30 – Tt +1 = E(Ft +1 - Ft) + (1 - E) Tt
where
20 – Tt = the last period trend factor
10 –
E=
E= a smoothing constant for trend
| | | | | | | | | | | | |
0–
1 2 3 4 5 6 7 8 9 10 11 12 13
Month
 Adjusted Exponential Smoothing Adjusted Exponential Smoothing:
(ȕ=0.30) Example

FORECAST TREND ADJUSTED

PERIOD MONTH DEMAND T3 = E(F3 - F2) + (1 - E) T2 PERIOD MONTH DEMAND Ft +1 Tt +1 FORECAST AFt +1

1 Jan 37 = (0.30)(38.5 - 37.0) + (0.70)(0) 1 Jan 37 37.00 – –

2 Feb 40 = 0.45 2 Feb 40 37.00 0.00 37.00
3 Mar 41 3 Mar 41 38.50 0.45 38.95
4 Apr 37 AF3 = F3 + T3 = 38.5 + 0.45 4 Apr 37 39.75 0.69 40.44
5 May 45 = 38.95 5 May 45 38.37 0.07 38.44
6 Jun 50 6 Jun 50 38.37 0.07 38.44
7 Jul 43 T13 = E(F13 - F12) + (1 - E) T12 7 Jul 43 45.84 1.97 47.82
8 Aug 47 = (0.30)(53.61 - 53.21) + (0.70)(1.77) 8 Aug 47 44.42 0.95 45.37
9 Sep 56 9 Sep 56 45.71 1.05 46.76
= 1.36
10 Oct 52 10 Oct 52 50.85 2.28 58.13
11 Nov 55 11 Nov 55 51.42 1.76 53.19
12 Dec 54 AF13 = F13 + T13 = 53.61 + 1.36 = 54.96 12 Dec 54 53.21 1.77 54.98
13 Jan – 53.61 1.36 54.96

Adjusted Exponential Smoothing

Forecasts Linear Trend Line

70 –
Adjusted forecast ( E =
60 – 6xy - nxy
Actual y = a + bx b =
6x2 - nx2
50 –
a = y-bx
where
Demand

40 –
a = intercept where
30 – (D = 0.50)
Forecast (D b = slope of the line n = number of periods
x = time period
20 – 6x
y = forecast for x = = mean of the x values
demand for period x n
10 –
6y
| | | | | | | | | | | | |
y = n = mean of the y values
0–
1 2 3 4 5 6 7 8 9 10 11 12 13
Period
 Least Squares Example Least Squares Example (cont.)
x(PERIOD) y(DEMAND) xy x2
1 73 37 1
2 40 80 4 78
3 41 123 9 x = = 6.5
12
4 37 148 16
557
5 45 225 25 y = = 46.42
6 50 300 36 12
7 43 301 49 ¦xy - nxy 3867 - (12)(6.5)(46.42)
b = = =1.72
8 47 376 64 ¦x2 - nx2 650 - 12(6.5)2
9 56 504 81
10 52 520 100
11 55 605 121 a = y - bx
12 54 648 144 = 46.42 - (1.72)(6.5) = 35.2
78 557 3867 650

Linear trend line y = 35.2 + 1.72x

Seasonal Adjustments
Forecast for period 13 y = 35.2 + 1.72(13) = 57.56 units

70 – The multiplicative seasonal model can

modify trend data to accommodate seasonal
60 –
Actual variations in demand
50 –
Demand

1. Find average historical demand for each season

40 –
Linear trend line 2. Compute the average demand over all seasons
30 –
3. Compute a seasonal index for each season
20 – 4. Estimate next year’
year’s total demand
10 – | | | | | | | | | | | | |
5. Divide this estimate of total demand by the
1 2 3 4 5 6 7 8 9 10 11 12 13 number of seasons, then multiply it by the
Period
0– seasonal index for that season
 Seasonal Adjustments Seasonal Adjustment (cont.)
DEMAND (1000’
(1000’S PER QUARTER)
YEAR 1 2 3 4 Total

ƒ Repetitive increase/ decrease in demand 2002 12.6 8.6 6.3 17.5 45.0
2003 14.1 10.3 7.5 18.2 50.1
ƒ Use seasonal factor to adjust forecast 2004 15.3 10.6 8.1 19.6 53.6
Total 42.0 29.5 21.9 55.3 148.7

Di
Seasonal factor = Si = D1 42.0 D3 21.9
¦D S1 = =
¦D 148.7
= 0.28 S3 = =
¦D 148.7
= 0.15

D2 29.5 D4 55.3
S2 = = = 0.20 S4 = = = 0.37
¦D 148.7 ¦D 148.7

Seasonal Adjustment (cont.) Forecast Accuracy

For 2005
„ Forecast error
„ difference between forecast and actual
y = 40.97 + 4.30x
4.30x = 40.97 + 4.30(4) = 58.17 demand
„ MAD
„ mean absolute deviation
SF1 = (S
(S1) (F
(F5) = (0.28)(58.17) = 16.28 „ MAPD
SF2 = (S
(S2) (F
(F5) = (0.20)(58.17) = 11.63
„ mean absolute percent deviation
SF3 = (S
(S3) (F
(F5) = (0.15)(58.17) = 8.73
SF4 = (S „ Cumulative error
(S4) (F
(F5) = (0.37)(58.17) = 21.53
„ Average error or bias
 Mean Absolute Deviation (MAD) MAD Example

PERIOD DEMAND, Dt Ft (D =0.3) (Dt - Ft) |Dt - Ft|

1 37 37.00 – –
2 40 37.00 3.00 3.00
6_ Dt - Ft _ 3 41 37.90
6_ D38.83
3.10 3.10
MAD = 4 37 t - Ft _ -1.83 1.83
n 5 MAD
45 = n38.28 6.72 6.72
6 50 40.29 9.69 9.69
where 7 43 53.3943.20 -0.20 0.20
=
t = period number 8 47 1143.14 3.86 3.86
9 56 44.30 11.70 11.70
Dt = demand in period t
10 52 = 4.85 47.81 4.19 4.19
Ft = forecast for period t 11 55 49.06 5.94 5.94
n = total number of periods 12 54 50.84 3.15 3.15
»» = absolute value 557 49.31 53.39

Other Accuracy Measures Comparison of Forecasts

Mean absolute percent deviation (MAPD)

¦|Dt - Ft| FORECAST MAD MAPD E (E)
MAPD =
¦Dt Exponential smoothing (D(D=
= 0.30) 4.85 9.6% 49.31 4.48
Exponential smoothing (D(D=
= 0.50) 4.04 8.5% 33.21 3.02
Cumulative error
Adjusted exponential smoothing 3.81 7.5% 21.14 1.92
E = ¦et (D=
D= 0.50, E=
E= 0.30)
Linear trend line 2.29 4.9% – –
Average error
¦et
E= n
 Forecast Control Tracking Signal Values
Tracking signal DEMAND FORECAST, ERROR ¦E = TRACKING
PERIOD Dt Ft Dt - Ft ¦(Dt - Ft) MAD SIGNAL
„ monitors the forecast to see if it is biased
1 37 37.00 – – – –
high or low 2 40 37.00 3.00 3.00 3.00 1.00
3 41 37.90 3.10 6.10 3.05 2.00
¦(Dt - Ft) E 4 37 38.83 -1.83 4.27 2.64 1.62
Tracking signal = = 5 45 38.28
Tracking 6.72 for period
signal 10.99 3 3.66 3.00
MAD MAD 6 50 40.29 9.69 20.68 4.87 4.25
7 43 43.20 -0.20
„1 MAD § 0.8 ɛ 6.10 20.48 4.09 5.01
8 47 TS3 = 3.86 =24.34
43.14 2.00 4.06 6.00
9 56 44.30 3.05 36.04
11.70 5.01 7.19
„ Control limits of 2 to 5 MADs are used most
10 52 47.81 4.19 40.23 4.92 8.18
frequently 11 55 49.06 5.94 46.17 5.02 9.20
12 54 50.84 3.15 49.32 4.85 10.17

Tracking Signal Plot Statistical Control Charts

3V –
¦(Dt - Ft)2
Tracking signal (MAD)

2V – Exponential smoothing (D = 0.30) V= n-1

1V –

0V –

-1 V – ƒ Using V we can calculate statistical

-2 V – Linear trend line control limits for the forecast error
-3 V – ƒ Control limits are typically set at r 3V
| | | | | | | | | | | | |
0 1 2 3 4 5 6 7 8 9 10 11 12
Period
 Statistical Control Charts Regression Methods

„ Linear regression
18.39 –
UCL = +3V „a mathematical technique that relates a
12.24 – dependent variable to an independent
6.12 – variable in the form of a linear equation
Errors

0–

-6.12 – „ Correlation
-12.24 – „a measure of the strength of the
LCL = -3V relationship between independent and
-18.39 –
dependent variables
| | | | | | | | | | | | |
0 1 2 3 4 5 6 7 8 9 10 11 12
Period

Linear Regression Linear Regression Example

x y
(WINS) (ATTENDANCE) xy x2
y = a + bx a = y-bx
6xy - nxy
6xy 4 36.3 145.2 16
b = 6 40.1 240.6 36
6x2 - nx2
6x 6 41.2 247.2 36
where 8 53.0 424.0 64
a = intercept 6 44.0 264.0 36
b = slope of the line 7 45.6 319.2 49
5 39.0 195.0 25
6x
6x 7 47.5 332.5 49
x = = mean of the x data
n
49 346.7 2167.7 311
6y
6y
y = n = mean of the y data
 Linear Regression Example (cont.) Linear Regression Example (cont.)
Regression equation Attendance forecast for 7 wins
49 y = 18.46 + 4.06x y = 18.46 + 4.06(7)
x= = 6.125
8 60,000 – = 46.88, or 46,880
346.9
y= = 43.36
8 50,000 –

¦xy - nxy2
b= 40,000 –

Attendance, y
¦x2 - nx2
(2,167.7) - (8)(6.125)(43.36) 30,000 –
= (311) - (8)(6.125)2 Linear regression line,
20,000 – y = 18.46 + 4.06x
4.06x
= 4.06
10,000 –
a = y - bx
= 43.36 - (4.06)(6.125) | | | | | | | | | | |
= 18.46 0 1 2 3 4 5 6 7 8 9 10
Wins, x

Correlation and Coefficient of Computing Correlation

Determination
ƒ Correlation,
Correlation, r n¦ xy - ¦ x¦ y
r=
ƒ Measure of strength of relationship [n¦ x2 - (¦ x)2] [n
[n¦ y2 - (¦ y)2]
ƒ Varies between -1.00 and +1.00
(8)(2,167.7) - (49)(346.9)
r=
ƒ Coefficient of determination,
determination, r2 [(8)(311) - (49)2] [(8)(15,224.7) - (346.9)2]
ƒ Percentage of variation in dependent
variable resulting from changes in the r = 0.947
independent variable
Coefficient of determination
r2 = (0.947)2 = 0.897
 Multiple Regression

Study the relationship of demand to two or

more independent variables

y = E0 + E1x1 + E2x2 … + Ekxk THE END

where
E0 = the intercept
E1, … , Ek = parameters for the
independent variables
x1, … , xk = independent variables
 Exponential Smoothing
; Form of weighted moving average
• EXERCISE : TUTORIAL
; Weights decline exponentially
; Most recent data weighted most
; Requires smoothing constant (D)
; Ranges from 0 to 1
; Subjectively chosen
; Involves little record keeping of past
data

Exponential Smoothing Exponential Smoothing

Example

New forecast = last period’

period’s forecast Predicted demand = 142 Ford Mustangs
+ D (last period’
period’s actual demand Actual demand = 153
– last period’
period’s forecast)
forecast) Smoothing constant D = .20

Ft = Ft – 1 + D(At – 1 - Ft – 1)

where Ft = new forecast

Ft – 1 = previous forecast
D = smoothing (or weighting)
constant (0 d D t 1)
 Exponential Smoothing Exponential Smoothing
Example Example

Predicted demand = 142 Ford Mustangs Predicted demand = 142 Ford Mustangs
Actual demand = 153 Actual demand = 153
Smoothing constant D = .20 Smoothing constant D = .20

New forecast = 142 + .2(153 – 142) New forecast = 142 + .2(153 – 142)
= 142 + 2.2
= 144.2 § 144 cars

Effect of Smoothing Constants Impact of Different D

225 –

Weight Assigned to Actual D = .5

200 – demand
Most 2nd Most 3rd Most 4th Most 5th Most
Demand

Recent Recent Recent Recent Recent

Smoothing Period Period Period Period Period
Constant (D) D(1 - D) D(1 - D) 2 D(1 - D) 3 D(1 - D)4
175 –
D = .1 .1 .09 .081 .073 .066

D = .5 .5 .25 .125 .063 .031 D = .1

150 – | | | | | | | | |
1 2 3 4 5 6 7 8 9
Quarter
 Choosing D Common Measures of Error
The objective is to obtain the most
accurate forecast no matter the Mean Absolute Deviation (MAD)
MAD)
technique ™ |actual - forecast|
MAD =
We generally do this by selecting the n
model that gives us the lowest forecast
error Mean Squared Error (MSE)
MSE)
Forecast error = Actual demand - Forecast value errors)2
™ (forecast errors)
= At - Ft MSE =
n

Common Measures of Error Comparison of Forecast Error

Rounded Absolute Rounded Absolute
Actual Forecast Deviation Forecast Deviation
Tonnage with for with for
Quarter Unloaded D = .10 D = .10 D = .50 D = .50
Mean Absolute Percent Error (MAPE)
MAPE)
1 180 175 5 175 5
2 168 176 8 178 10
n 3 159 175 16 173 14
100 ™ |actuali - forecasti|/actuali 4
5
175
190
173
173
2
17
166
170
9
20
MAPE = i=1
6 205 175 30 180 25
n 7 180 178 2 193 13
8 182 178 4 186 4
84 100
 Comparison of Forecast Error Comparison of Forecast Error
™ (forecast errors) 2
™ |deviations|
Rounded Absolute Rounded Absolute Rounded Absolute Rounded Absolute
MADActual
= Forecast Deviation Forecast Deviation
MSE = Actual Forecast Deviation Forecast Deviation
Tonage n
with for with for Tonage n
with for with for
Quarter Unloaded D = .10 D = .10 D = .50 D = .50 Quarter Unloaded D = .10 D = .10 D = .50 D = .50
1
For D =
180
.10 175 5 175 5 1
For D180
= .10 175 5 175 5
2 168 = 84/8 176= 10.50 8 178 10 2 = 1,558/8
168 = 194.758
176 178 10
3 159 175 16 173 14 3 159 175 16 173 14
4 For D175
= .50 173 2 166 9 4 For D175
= .50 173 2 166 9
5 190 173 17 170 20 5 190 173 17 170 20
6 205 = 100/8
175 = 12.50
30 180 25 6 = 1,612/8
205 = 201.50
175 30 180 25
7 180 178 2 193 13 7 180 178 2 193 13
8 182 178 4 186 4 8 182 178 4 186 4
84 100 84 100
MAD 10.50 12.50

Comparison
n
of Forecast Error Comparison of Forecast Error
100 ™ |deviationi|/actuali
MAPE = i =Rounded
1 Absolute Rounded Absolute Rounded Absolute Rounded Absolute
Actual Forecast Deviation Forecast Deviation Actual Forecast Deviation Forecast Deviation
Tonage with n for with for Tonnage with for with for
Quarter Unloaded D = .10 D = .10 D = .50 D = .50 Quarter Unloaded D = .10 D = .10 D = .50 D = .50
1
For D
180
= .10 175 5 175 5 1 180 175 5 175 5
2 168 = 45.62/8
176 = 5.70%
8 178 10 2 168 176 8 178 10
3 159 175 16 173 14 3 159 175 16 173 14
4 D = .50 173
For 175 2 166 9 4 175 173 2 166 9
5 190 173 17 170 20 5 190 173 17 170 20
6 205 = 54.8/8
175 = 6.85%
30 180 25 6 205 175 30 180 25
7 180 178 2 193 13 7 180 178 2 193 13
8 182 178 4 186 4 8 182 178 4 186 4
84 100 84 100
MAD 10.50 12.50 MAD 10.50 12.50
MSE 194.75 201.50 MSE 194.75 201.50
MAPE 5.70% 6.85%
 Exponential Smoothing with Exponential Smoothing with
Trend Adjustment Trend Adjustment
When a trend is present, exponential
smoothing must be modified Ft = D(At - 1) + (1 - D)(F
)(Ft - 1 + Tt - 1)

Forecast exponentially exponentially Tt = E(Ft - Ft - 1) + (1 - E)Tt - 1

including (FITt) = smoothed (Ft) + (Tt) smoothed
trend forecast trend
Step 1: Compute Ft
Step 2: Compute Tt
Step 3: Calculate the forecast FITt = Ft + Tt

Exponential Smoothing with Exponential Smoothing with

Trend Adjustment Example Trend Adjustment Example
Forecast Forecast
Actual Smoothed Smoothed Including Actual Smoothed Smoothed Including
Month(
Month(t) Demand (At) Forecast, Ft Trend, Tt Trend, FITt Month(
Month(t) Demand (At) Forecast, Ft Trend, Tt Trend, FITt
1 12 11 2 13.00 1 12 11 2 13.00
2 17 2 17
3 20 3 20
4 19 4 19
5 24 5 24 Step 1: Forecast for Month 2
6 21 6 21
7 31 7 31 F2 = DA1 + (1 - D)(F1 + T1)
8 28 8 28 F2 = (.2)(12) + (1 - .2)(11 + 2)
9 36 9 36
10 10 = 2.4 + 10.4 = 12.8 units
Table 4.1 Table 4.1
 Exponential Smoothing with Exponential Smoothing with
Trend Adjustment Example Trend Adjustment Example
Forecast Forecast
Actual Smoothed Smoothed Including Actual Smoothed Smoothed Including
Month(
Month(t) Demand (At) Forecast, Ft Trend, Tt Trend, FITt Month(
Month(t) Demand (At) Forecast, Ft Trend, Tt Trend, FITt
1 12 11 2 13.00 1 12 11 2 13.00
2 17 12.80 2 17 12.80 1.92
3 20 3 20
4 19 4 19
5 24 Step 2: Trend for Month 2 5 24 Step 3: Calculate FIT for Month 2
6 21 6 21
7 31 T2 = E(F2 - F1) + (1 - E)T1 7 31 FIT2 = F2 + T1
8 28 T2 = (.4)(12.8 - 11) + (1 - .4)(2) 8 28 FIT2 = 12.8 + 1.92
9 36 9 36
10 = .72 + 1.2 = 1.92 units 10 = 14.72 units
Table 4.1 Table 4.1

Exponential Smoothing with Exponential Smoothing with

Trend Adjustment Example Trend Adjustment Example
Forecast 35 –
Actual Smoothed Smoothed Including
Month( Actual demand (At)
Month(t) Demand (At) Forecast, Ft Trend, Tt Trend, FITt 30 –
Product demand

1 12 11 2 13.00
25 –
2 17 12.80 1.92 14.72
3 20 15.18 2.10 17.28 20 –
4 19 17.82 2.32 20.14
5 24 19.91 2.23 22.14 15 –
6 21 22.51 2.38 24.89 10 – Forecast including trend (FITt)
7 31 24.11 2.07 26.18
8 28 27.14 2.45 29.59 5 –
9 36 29.28 2.32 31.60
0 – | | | | | | | | |
10 32.48 2.68 35.16
1 2 3 4 5 6 7 8 9
Table 4.1 Figure 4.3
Time (month)
 Trend Projections Least Squares Method

Values of Dependent Variable

Fitting a trend line to historical data points Actual observation Deviation7
to project into the medium-
medium-to-to-long-
long-range (y value)

Linear trends can be found using the least Deviation5 Deviation6

squares technique Deviation3

y^ = a + bx
Deviation4

Deviation1
where y^ = computed value of the variable to Deviation2
be predicted (dependent variable) Trend line, y^ = a + bx
a = y-
y-axis intercept
b = slope of the regression line
x = the independent variable Time period Figure 4.4

Least Squares Method Least Squares Method

Equations to calculate the regression variables
Values of Dependent Variable

Actual observation Deviation7

(y value)

Deviation5 Deviation6 y^ = a + bx
Deviation3 Least squares method
minimizes the sum of the
Deviation
squared errors (deviations)
6xy - nxy
4
b=
Deviation1
6x2 - nx2
Deviation2
Trend line, y^ = a + bx
a = y - bx
Time period Figure 4.4
 Least Squares Example Least Squares Example
Time Electrical Power Time Electrical Power
Year Period (x) Demand x2 xy Year Period (x) Demand x2 xy
1999 1 74 1 74 1999 1 74 1 74
2000 2 79 4 158 2000 2 79 4 158
2001 3 80 9 240 The
2001 trend
3 line is 80 9 240
2002 4 90 16 360 2002 4 90 16 360
2003 5 105 25 525 2003 y^ 5= 56.70 + 10.54x
105 25 525
2004 6 142 36 852 2004 6 142 36 852
2005 7 122 49 854 2005 7 122 49 854
™x = 28 ™y = 692 ™x2 = 140 ™xy = 3,063 6x = 28 6y = 692 6x2 = 140 6xy = 3,063
x=4 y = 98.86 x=4 y = 98.86

™xy - nxy 3,063 - (7)(4)(98.86) 6xy - nxy 3,063 - (7)(4)(98.86)

b= = = 10.54
10.54 b= = = 10.54
10.54
™x2 - nx2 140 - (7)(42) 6x2 - nx2 140 - (7)(42)

a = y - bx = 98.86 - 10.54(4) = 56.70 a = y - bx = 98.86 - 10.54(4) = 56.70

Least Squares Example Least Squares Requirements

Trend line,
160 –
150 – y^ = 56.70 + 10.54x 1. We always plot the data to insure a
140 – linear relationship
Power demand

130 –
120 – 2. We do not predict time periods far
110 –
100 – beyond the database
90 –
80 – 3. Deviations around the least
70 – squares line are assumed to be
60 –
50 –
random
| | | | | | | | |
1999 2000 2001 2002 2003 2004 2005 2006 2007
Year
 Seasonal Variations In Data Seasonal Index Example
Demand Average Average Seasonal
The multiplicative seasonal model can Month 2003 2004 2005 2003-
2003-2005 Monthly Index
modify trend data to accommodate Jan 80 85 105 90 94
seasonal variations in demand Feb 70 85 85 80 94
Mar 80 93 82 85 94
1. Find average historical demand for each season Apr 90 95 115 100 94
May 113 125 131 123 94
2. Compute the average demand over all seasons Jun 110 115 120 115 94
3. Compute a seasonal index for each season Jul 100 102 113 105 94
Aug 88 102 110 100 94
4. Estimate next year’
year’s total demand
Sept 85 90 95 90 94
5. Divide this estimate of total demand by the Oct 77 78 85 80 94
number of seasons, then multiply it by the Nov 75 72 83 80 94
seasonal index for that season Dec 82 78 80 80 94

Seasonal Index Example Seasonal Index Example

Demand Average Average Seasonal Demand Average Average Seasonal
Month 2003 2004 2005 2003-
2003-2005 Monthly Index Month 2003 2004 2005 2003-
2003-2005 Monthly Index
Jan 80 85 105 90 94 0.957 Jan 80 85 105 90 94 0.957
Feb 70 85 85 80 94 Feb 70 85 85 80 94 0.851
Mar 80 93 average
82 85 monthly demand
2003-2005 94 Mar 80 93 82 85 94 0.904
Seasonal90index95= 115
Apr 100 94 Apr 90 95 115 100 94 1.064
average monthly demand
May 113 125 131 123 94 May 113 125 131 123 94 1.309
= 90/94 = .957
Jun 110 115 120 115 94 Jun 110 115 120 115 94 1.223
Jul 100 102 113 105 94 Jul 100 102 113 105 94 1.117
Aug 88 102 110 100 94 Aug 88 102 110 100 94 1.064
Sept 85 90 95 90 94 Sept 85 90 95 90 94 0.957
Oct 77 78 85 80 94 Oct 77 78 85 80 94 0.851
Nov 75 72 83 80 94 Nov 75 72 83 80 94 0.851
Dec 82 78 80 80 94 Dec 82 78 80 80 94 0.851
 Seasonal Index Example Seasonal Index Example
Demand Average Average Seasonal
Month 2003 2004 2005 2003-
2003-2005 Monthly Index 2006 Forecast
Jan 80 85 105 90 94 0.957 140 – 2005 Demand
Feb 70 85 Forecast
85 for802006 94 0.851 130 – 2004 Demand
Mar 80 93 82 85 94 0.904 2003 Demand
120 –
Apr 90Expected
95 115annual demand
100 = 1,200
94 1.064

Demand
May 113 125 131 123 94 1.309 110 –
Jun 110 115 120 1,200 115 94 1.223 100 –
Jul Jan 113
100 102 x .957 = 96 94
12 105 1.117 90 –
Aug 88 102 110 100 94 1.064
1,200 80 –
Sept 85 90
Feb 95 x90.851 = 85 94 0.957
Oct 77 78 85 12 80 94 0.851 70 –
| | | | | | | | | | | |
Nov 75 72 83 80 94 0.851
J F M A M J J A S O N D
Dec 82 78 80 80 94 0.851
Time

San Diego Hospital San Diego Hospital

Trend Data Seasonal Indices

10,200 – 1.06 –
1.04
Index for Inpatient Days

1.04
10,000 – 1.04 – 1.03
1.02
Inpatient Days

9745 1.02 – 1.01

9,800 – 9702 1.00
9616 9659 0.99
9573 9766 1.00 –
9,600 – 9530 9680 9723 0.98
9594 9637 0.98 – 0.99
9551
9,400 –
0.96 – 0.97 0.97
0.96
9,200 – 0.94 –
9,000 – | | | | | | | | | | | | 0.92 – | | | | | | | | | | | |
Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec
67 68 69 70 71 72 73 74 75 76 77 78 67 68 69 70 71 72 73 74 75 76 77 78
Month Month
Figure 4.6 Figure 4.7
 San Diego Hospital
Associative Forecasting
Combined Trend and Seasonal Forecast
Used when changes in one or more
10,200 –
10068 independent variables can be used to predict
9949
10,000 – 9911 the changes in the dependent variable
Inpatient Days

9,800 – 9764 9724

9691
9,600 –
9572 Most common technique is linear
9520 9542 regression analysis
9,400 –
9411
9265 9355
9,200 –
| | | | | | | | | | | |
We apply this technique just as we did
9,000 –
Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec in the time series example
67 68 69 70 71 72 73 74 75 76 77 78
Month
Figure 4.8

Associative Forecasting Associative Forecasting

Example
Forecasting an outcome based on Sales Local Payroll
($000,000), y ($000,000,000), x
predictor variables using the least squares
2.0 1
technique 3.0 3
y^ = a + bx 2.5 4 4.0 –
2.0 2
2.0 1 3.0 –
Sales

where y^ = computed value of the variable to 3.5 7

be predicted (dependent variable) 2.0 –
a = y-
y-axis intercept
b = slope of the regression line 1.0 –
x = the independent variable though to | | | | | | |
predict the value of the dependent 0 1 2 3 4 5 6 7
variable Area payroll
 Associative Forecasting Associative Forecasting
Example Example
Sales, y Payroll, x x2 xy
2.0 1 1 2.0 y^ = 1.75 + .25x
.25x Sales = 1.75 + .25(payroll
.25(payroll))
3.0 3 9 9.0
2.5 4 16 10.0
2.0 2 4 4.0
If payroll next year
4.0 –
2.0 1 1 2.0
is estimated to be
$600 million, then: 3.25
3.5 7 49 24.5 3.0 –

Sales
™y = 15.0 ™x = 18 ™x2 = 80 ™xy = 51.5
Sales = 1.75 + .25(6) 2.0 –
™xy - nxy 51.5 - (6)(3)(2.5) Sales = $325,000
x = ™x/6 = 18/6 = 3 b= = = .25 1.0 –
™x2 - nx2 80 - (6)(32)
| | | | | | |
y = ™y/6 = 15/6 = 2.5 a = y - bx = 2.5 - (.25)(3) = 1.75 0 1 2 3 4 5 6 7
Area payroll

Standard Error of the Estimate Standard Error of the Estimate

; A forecast is just a point estimate of a

future value ™(y - yc)2
Sy,x =
; This point is 4.0 – n-2
actually the 3.25
mean of a
Sales

3.0 –
where y = y-value of each data point
probability 2.0 –
distribution yc = computed value of the dependent
1.0 – variable, from the regression
equation
| | | | | | |
0 1 2 3 4 5 6 7 n = number of data points
Area payroll
Figure 4.9
Standard Error of the Estimate Standard Error of the Estimate

Computationally, this equation is ™y2 - a™y - b™xy = 39.5 - 1.75(15) - .25(51.5)

Sy,x =
considerably easier to use n-2 6-2

Sy,x = .306 4.0 –

™y2 - a™y - b™xy
Sy,x = 3.25

Sales
3.0 –
n-2
The standard error
of the estimate is 2.0 –
We use the standard error to set up $30,600 in sales 1.0 –
prediction intervals around the
| | | | | | |
point estimate 0 1 2 3 4 5 6 7
Area payroll

Correlation Correlation Coefficient

; How strong is the linear n6xy - 6x6y
relationship between the r=
variables? [n6x2 - (6x)2][n
][n6y2 - (6y)2]
; Correlation does not necessarily
imply causality!
; Coefficient of correlation, r,
measures degree of association
; Values range from -1 to +1
y y
Correlation Coefficient Correlation

n™xy - ™x™y
; Coefficient of Determination, r2,
r= measures the percent of change in
[n™x2
(a) Perfect positive x
- (™x)2][n
][n™y 2 - (™y)2]
(b) Positive x y predicted by the change in x
correlation: correlation:
r = +1 0<r<1 ; Values range from 0 to 1
y y ; Easy to interpret

For the Nodel Construction example:

r = .901
(c) No correlation: x (d) Perfect negative x
correlation:
r2 = .81
r=0
r = -1

Multiple Regression Analysis Multiple Regression Analysis

If more than one independent variable is to be In the Nodel example, including interest rates in
the model gives the new equation:
used in the model, linear regression can be
extended to multiple regression to
accommodate several independent variables y^ = 1.80 + .30x
.30x1 - 5.0x
5.0x2

An improved correlation coefficient of r = .96

y^ = a + b1x1 + b2x2 … means this model does a better job of predicting
the change in construction sales

Computationally, this is quite Sales = 1.80 + .30(6) - 5.0(.12) = 3.00

complex and generally done on the Sales = $300,000
computer
 Monitoring and Controlling Monitoring and Controlling
Forecasts Forecasts
Tracking Signal
Tracking RSFE
=
; Measures how well the forecast is signal MAD
predicting actual values
; Ratio of running sum of forecast errors ™(actual demand in
(RSFE) to mean absolute deviation (MAD) period i -
forecast demand
; Good tracking signal has low values
Tracking in period i)
; If forecasts are continually high or low, the =
forecast has a bias error
signal ™|actual - forecast|/n)
forecast|/n)

Tracking Signal Tracking Signal Example

Cumulative
Signal exceeding limit Absolute Absolute
Actual Forecast Forecast Forecast
Qtr Demand Demand Error RSFE Error Error MAD
Tracking signal
Upper control limit 1 90 100 -10 -10 10 10 10.0
+ 2 95 100 -5 -15 5 15 7.5
3 115 100 +15 0 15 30 10.0
Acceptable
0 MADs range 4 100 110 -10 -10 10 40 10.0
5 125 110 +15 +5 15 55 11.0
– 6 140 110 +30 +35 30 85 14.2
Lower control limit

Time
 Tracking Signal Example Adaptive Forecasting
Cumulative
Tracking Absolute Absolute
Actual Signal
Forecast Forecast Forecast It’
It’s possible to use the computer to
(RSFE/MAD)
Qtr Demand Demand Error RSFE Error Error MAD

1 90-10/10
100= -1 -10 -10 10 10 10.0 continually monitor forecast error and
2 95 100= -2 -5
-15/7.5 -15 5 15 7.5 adjust the values of the D and E
3 115 0/10
100
= 0 +15 0 15 30 10.0 coefficients used in exponential
4 100-10/10
110= -1 -10 -10 10 40 10.0 smoothing to continually minimize
5 125
+5/11110
= +0.5+15 +5 15 55 11.0
forecast error
6 140 110= +2.5
+35/14.2 +30 +35 30 85 14.2
This technique is called adaptive
The variation of the tracking signal smoothing
between -2.0 and +2.5 is within acceptable
limits

Focus Forecasting
Developed at American Hardware Supply,
focus forecasting is based on two principles:

1. Sophisticated forecasting models are not

always better than simple models
2. There is no single techniques that should
be used for all products or services

This approach uses historical data to test

multiple forecasting models for individual items
The forecasting model with the lowest error is
then used to forecast the next demand