INDU 6211:

Production Systems and Inventory Control

Lecture 3-Chapter 4
Forecasting (continued)

Summer 2023

1
 AGENDA

• Introduction of forecasting
• Forecasting Models
• Constant Models
• Regression
1.Seasonality
2.Forecasting Models
3.Constant Models
4.Regression
2
Seasonality
 Seasonality may (or may not) be
relative to the general demand
trend
 Additive seasonal variation is
constant regardless of changes in
average demand
 Multiplicative seasonal variation
maintains a consistent
relationship to the average
demand (this is the more common
case)
3-3
Additive vs Multiplicative

3-4
Additive vs Multiplicative

3-5
 Additive vs Multiplicative
Average Multiplica
Seasonal Additive tive
Sales Forecast Forecast
Spring
Year 1 300 250 240
300 400 420
300 350 360
300 200 180
Spring
Year 2 350 300 280
350 450 490
350 400 420
350 250 210
Spring
Year 3 400 350 320
400 500 560
400 450 480
400 300 240
Spring
Year 4 450 400 360
450 550 630
450 500 540
450 350 270

3-6
 Seasonal Factor
 To account for seasonality within the
forecast, the seasonal factor is calculated
 The amount of correction needed in a time
series to adjust for the season of the year
Season Past Average Sales Seasonal Factor
Sales for Each Season
Spring 200 1000/4=250 Actual/Average=200/250=0.8
Summer 350 1000/4=250 350/250=1.4
Fall 300 1000/4=250 300/250=1.2
Winter 150 1000/4=250 150/250=0.6
Total 1000
3-7
 Seasonal Factor
 If we expect (forecast) next year’s sales to
be 1,100 units, the seasonal forecast is
calculated using the seasonal factors
Season Expected Average Sales for Seasonal Forecast
Sales Each Season Factor
Spring 1100/4=275 X 0.8 = 220
Summer 1100/4=275 X 1.4 = 385
Fall 1100/4=275 X 1.2 = 330
Winter 1100/4=275 X 0.6 = 165
Total 1,100

3-8
 Seasonality–Trend and
Season Estimate of trend, use
linear regression software
Quarter Amount
to obtain more precise
I – 2008 300 results
II – 2008 200
III – 2008 220
IV – 2008 530
I – 2009 520
Trend = 170 +55t
II – 2009 420
III – 2009 400
IV - 2009 700

3-9
 Seasonality–Trend and
Season
 Seasonal factors are calculated for each
season, then averaged for similar seasons
 Seasonal Factor = Actual/Trend

3-10
 Seasonality–Trend and
Season
 Forecasts are calculated by extending the
linear regression and then adjusting by the
appropriate seasonal factor
 FITS–Forecast Including Trend and Seasonal
Factors

3-11
 Decomposition Using Least
Squares Regression
1. Decompose the time series into its components
a. Find seasonal component
b. Deseasonalize the demand
c. Find trend component
2. Forecast future values for each component
a. Project trend component into future
b. Multiply trend component by seasonal
component

3-12
 Decomposition Using Least
Squares Regression
Period Quarter Actual Average of Same Quarter of Seasonal
Demand Each Year Factor
1 I 600 (600+2400+3800)/3=2266.7
2 II 1,550
3 III 1,500
Calculate average of
same period values
4 IV 1,500
5 I 2,400
6 II 3,100
7 III 2,600
8 IV 2,900
9 I 3,800
10 II 4,500
11 III 4,000
12 IV 4,900
Total 33,350
3-13
 Decomposition Using Least
Squares Regression
Period Quarter Actual Average of Same Quarter of Seasonal
Demand Each Year Factor
1 I 600 (600+2400+3800)/3=2266.7
2 II 1,550 (1550+3100+4500)/3=3050
3 III 1,500 (1500+2600+4000)/3=2700
4 IV 1,500 (1500+2900+4900)/3=3100
5 I 2,400
6 II 3,100
7 III 2,600
8 IV 2,900
9 I 3,800
10 II 4,500
11 III 4,000
12 IV 4,900
Total 33,350
3-14
 Decomposition Using Least
Squares Regression
Period Quarter Actual Average of Same Quarter Seasonal Factor
Deman of Each Year
d
1 I 600 (600+2400+3800)/3=2266.7 2266.7/(33350/12)=0.82
2 II 1,550 (1550+3100+4500)/3=3050
3 III 1,500 (1500+2600+4000)/3=2700
4 IV 1,500 (1500+2900+4900)/3=3100
5 I 2,400
Calculate seasonal factor
6 II 3,100
for each period
7 III 2,600
8 IV 2,900
9 I 3,800
10 II 4,500
11 III 4,000
12 IV 4,900
Total 33,350 3-15
 Decomposition Using Least
Squares Regression
Period Quarter Actual Average of Same Quarter Seasonal Factor
Demand of Each Year
1 I 600 (600+2400+3800)/3=2266.7 2266.7/(33350/12)=0.82
2 II 1,550 (1550+3100+4500)/3=3050 3050/(33350/12)=1.10
3 III 1,500 (1500+2600+4000)/3=2700 2700/(33350/12)=0.97
4 IV 1,500 (1500+2900+4900)/3=3100 3100/(33350/12)=1.12
5 I 2,400 0.82
6 II 3,100 1.10
7 III 2,600 0.97 Seasonal
8 IV 2,900 1.12 factors
9 I 3,800 0.82 repeat each
10 II 4,500 1.10
year
11 III 4,000 0.97
12 IV 4,900 1.12
Total 33,350
3-16
 Decomposition Using Least
Squares Regression
Period Quarter Actual Seasonal Deseasonalized Demand
Demand Factor (Actual/Seasonal Factor)
1 I 600 0.82 600/0.82=735.7
2 II 1,550 1.10 1550/1.10=1412.4
3 III 1,500 0.97 1500/0.97=1544.0
4 IV 1,500 1.12 1500/1.12=1344.8
5 I 2,400 0.82 2942.6
Calculate deseasonalized
6 II 3,100 1.10 2824.7
demand for each period
7 III 2,600 0.97 2676.2
8 IV 2,900 1.12 2599.9
9 I 3,800 0.82 4659.2
10 II 4,500 1.10 4100.4
11 III 4,000 0.97 4117.3
12 IV 4,900 1.12 4392.9

3-17
 Least Squares Regression
for Deseasonalized Data
Period Deseasonalized SUMMARY OUTPUT

Demand Regression Statistics

Use linear regression to fit
1 735.7 Multiple R 0.929653282 trend line to
R Square 0.864255225
Adjusted R Square 0.850680748
deseasonalized data
2 1412.4
Standard Error 512.8180268
3 1544.0 Observations 12

4 1344.8 ANOVA
df SS MS F Significance F
5 2942.6 Regression 1 16743469.64 16743469.64 63.66766059 1.20464E-05
Residual 10 2629823.286 262982.3286
6 2824.7 Total 11 19373292.92

7 2676.2 Coefficients Standard Error t Stat P-value

Intercept 555.0045455 315.6176776 1.758471039 0.109173704
8 2599.9 Period 342.1800699 42.88399775 7.979201751 1.20464E-05

9 4659.2
10 4100.4
11
12
4117.3
4392.9
Y= 555.0 + 342.2x
3-18
 Create Forecast by Projecting
Trend and Reseasonalizing
Period Quarter Y from Regression Seasonal Forecast
Factor
13 I 555+342.2*13=5003.5 X 0.82 = 4102.87
14 II 555+342.2*14=5345.7 X 1.10 = 5880.27
15 III 555+342.2*15=5687.9 X 0.97 = 5517.26
16 IV 555+342.2*16=6030.1 X 1.12 = 6753.71

Project Linear Trend Project Seasonality

3-19
 Forecast Evaluation

 Is the forecast too high or too low?

 Mean Error (bias)
 What is the magnitude of the forecast error?
 Mean Absolute Deviation (MAD)
 Standard Deviation of forecast error = 1.25*MAD

 Measuring both bias and MAD is critical to

understanding the quality of the forecast

3-20
 Forecast Evaluation
n

 ( ActualDemand i  ForecastDemand i )
Mean Error (bias )  i 1

n
n

 ActualDemand i  ForecastDemand i
Mean Absolute Deviation ( MAD)  i 1

n
i  period number
n  number of periods of data

3-21
 Forecast Evaluation
Mean Square Error (MSE)

Mean Absolute Percent Error (MAPE)

3-22
 Example
The demand manager of Maverick Jeans is responsible for ensuring
sufficient warehouse space for the finished jeans that come from the
production plants. It has occasionally been necessary to rent public
warehouse space, something that Maverick would like to avoid. In order to
estimate the space requirements the demand manager is evaluating
moving-average forecasts. The demand (in 1,000 case units) for the last
fiscal year is shown below

3-23
 Example

a. Use a three-month moving average to estimate the month-in-advance forecast of

demand for months 4–12 and generate a forecast for the first month of next year.
Calculate the average forecast error and mean absolute error.

b. Use a three-month weighted moving average with weights of 0.6, 0.3,

0.1 (most recent to last recent, respectively) to calculate month-in-
advance forecasts for months 4–12 and forecast for the first month of next
year. Calculate the average forecast error and mean ab- solute error.

c. Compare the average forecast error and MAD for the forecasting
methods in parts a and b. Based on these error calculations, which of
the two forecast methods would you recommend? Why?

3-24
 Example
Absolute
Month Demand Forecast Error Error Month Demand Forecast Error Absolute
1 20 Error
1 20
2 18
2 18
3 21
3 21
4 25 20 5 5
4 25 20 5 5
5 24 21 3 3 5 24 20 4 4
6 27 23 4 4 6 27 23 4 4
7 22 25 -3 3 7 22 25 -3 3
8 30 24 6 6 8 30 25 5 5
9 23 26 -3 3 9 23 26 -3 3
10 20 25 -5 5 10 20 25 -5 5
11 29 24 5 5 11 29 27 2 2
12 22 24 -2 2 12 22 23 -1 1
1 23
1 24
SUM 10 36
SUM 8 32
MEAN 1.11 4
MEAN 0.9 3.6

3-25
 Example
The three month weighted moving average performed better
than the three month moving average on both measures of
forecast error. Sometimes there can be contradictions in
these two measures , but in this case the weighted moving
average is the best on both measures.

3-26
 Example
 The most widely used measure of deviation or dispersion in statistics is the
standard deviation. MAD also measures deviation (error) from an expected result
(the forecast). When the forecast errors are distributed normally, there is a direct
relationship between the two measures that can be used to develop statistical
insights and conclusions. The standard deviation of the errors is arithmetically
related to MAD by Equation:

This quantity can be used to evaluate the safety stocks to cover the
uncertainty of the forecast.

3-27
 Example
 Example: Suppose the standard deviation of errors of a forecast
model for a product is 10.32. If you consider the safety stock at
3 times this standard deviation (3 * 10.32) = 30.96 or
approximately 31 units of product, the risk of stockout will be
reduced by 0.3%. This means: to have a 99.7% chance of never
being out of stock.

3-28
 Aggregating Forecasts

 The SOP process reconciles differences in

forecasts from various sources
 Customer/product knowledge
 Sum of individual product detailed forecasts
(by product family, for example)
 SOP result is an aggregate demand forecast
 Long-term and/or aggregate forecasts are
more accurate than short-term, detailed
forecasts
3-29
 Pyramid Forecasting

 One means of aggregating and

disaggregating forecasts is pyramid
forecasting
 Ensures consistency as the forecast sources
are integrated
 Provides a logical framework for summing
lower level forecasts and distributing higher
level forecast changes to individual products

3-30
Pyramid Forecasting

3-31
 Pyramid Forecasting

The procedure used in implementing the approach often begins with individual
product item forecasts at level 3, which are rolled up into forecasts for product lines
shown as level 2. We then aggregate forecasts for product lines into a total
business forecast (in dollars) at level 1 in Figure 3.18. Once the individual item and
product line forecasts have been rolled up and considered in finalizing the top
management forecast (plan), the next step is to force down (constrain) the product
line and individual item forecasts, so they’re consistent with the plan.

3-32
Pyramid Forecasting

11 individual product items are divided into

two product lines.
Two of these items, X1 and X2, form
product line X, while the remaining
products, Z1 through Z9, are included in
product line Z.
These two product lines, X and Z, represent
the firm’s entire range of products.
unit prices and initial forecasts for each
level is shown in the figure.

3-33
 Pyramid Forecasting

The ratio between the roll-up forecast at level 1 ($778,460) and the
management total ($900,000) is used to make the adjustment.

3-34
 External Information
 Activities or conditions that may invalidate the
assumption that history is a good predictor must
be accounted for in the forecasting process
 Special promotions, product changes,
advertising, competitors’ actions
 Changes to forecasting process may be needed
 Change exponential smoothing parameter to
place more (or less) emphasis on recent history
 Forecast more frequently to identify conditions
that result in higher forecast errors
3-35
 Principles
 Forecast models should be as simple as possible.
Simple models often outperform more complicated
approaches.
 Inputs (data) and outputs (forecasts) must be
monitored for quality and appropriateness.
 Information on the sources of variation (seasonality,
market trends, company policies) should be
incorporated into the forecasting system.
 Forecasts from different sources must be reconciled
and made consistent with company plans and
constraints. 3-36
 Quiz – Chapter 3
 Regression analysis where the relationship between variables
is a straight line is called _______ _______.

 In a time series analysis, time is the _________ variable.

 In an exponential smoothing forecast, a higher level of alpha

(α) will place more emphasis on recent history (T/F).

 Mean error of a forecast provides information concerning the

forecast’s ________.

3-37
Questions

38