BM1710

FORECASTING
Forecasting Approaches
There are two (2) general approaches to forecasting, just as there are two (2) ways to tackle all decision
modeling; quantitative analysis or qualitative approach. Quantitative forecasts use a variety of
mathematical models that rely on historical data and/or associative variables to forecast demand.
Subjective or qualitative forecasts incorporate such factors as the decision maker’s intuition, emotions,
personal experiences, and value system in reaching a forecast (Heizer, Render, & Munson, 2017).

Overview of Qualitative Methods

• Jury of executive opinion

Under this method, the opinions of a group of high-level experts or managers, often in combination
with statistical models, are pooled to arrive at a group estimate of demand. Bristol-Myers Squibb
Company, for example, uses 220 well-known research scientists as its jury of executive opinion to
get a grasp on future trends in the world of medical research (Heizer, Render, & Munson, 2017).

• Delphi method
There are three (3) different types of participants in the Delphi method: decision makers, staff
personnel, and respondents. Decision makers usually consist of a group of five (5) to 10 experts who
will be making the actual forecast. Staff personnel assist decision makers by preparing, distributing,
collecting, and summarizing a series of questionnaires and survey results. The respondents are a
group of people, often located in different places, whose judgments are valued. This group provides
inputs to the decision makers before the forecast is made (Heizer, Render, & Munson, 2017).

EXAMPLE: The state of Alaska has used the Delphi method to develop its long-range economic
forecast. A large part of the state’s budget is derived from the million-plus barrels of oil pumped
daily through a pipeline at Prudhoe Bay. The large Delphi panel of experts had to represent all
groups and opinions in the state and all geographic areas (Heizer, Render, & Munson, 2017).

• Sales force composite

In this approach, each salesperson estimates what sales will be in his or her region. These forecasts
are then reviewed to ensure that they are realistic. Then they are combined at the district and
national levels to reach an overall forecast. A variation of this approach occurs at Lexus, where every
quarter, Lexus dealers have a “make meeting”. At this meeting, they talk about what is selling, in
what colors, and with what options, so the factory knows what to build (Heizer, Render, & Munson,
2017).

• Market survey
This method solicits input from customers or potential customers regarding future purchasing plans.
It can help not only in preparing a forecast but also in improving product design and planning for
new products. The consumer market survey and sales force composite methods can, however,
suffer from overly optimistic forecasts that arise from customer input (Heizer, Render, & Munson,
2017).

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 1 of 9
 BM1710

Time-Series Forecasting
Naïve Approach
The simplest way to forecast is to assume that demand in the next period will be equal to demand in the
most recent period. In other words, if sales of a product – say, Samsung cell phones – were 68 units in
January, we can forecast that February’s sales will also be 68 phones. For some product lines, this
approach is the most cost-effective and efficient objective forecasting model. At least it provides a starting
point against which more sophisticated models that follow can be compared (Heizer, Render, & Munson,
2017).

Moving Averages
A moving-average forecast uses a number of historical actual data values to generate a forecast. Moving
averages are useful if the assumption is that market demands will stay fairly steady over time. A 4-month
moving average is found by simply summing the demand during the past four (4) months and dividing by
four (4). With each passing month, the most recent month’s data are added to the sum of the previous
three (3) months’ data, and the earliest month is dropped. This practice tends to smooth out short-term
irregularities in the data series (Heizer, Render, & Munson, 2017).
∑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 𝑖𝑖𝑖𝑖 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 𝑛𝑛 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 =
𝑛𝑛
MONTH ACTUAL SHED SALES 3-MONTH MOVING AVERAGE
January 10

February 12

March 13
(10 + 12 + 13)
April 16 = 12
3
(12 + 13 + 16)
May 19 = 14
3
(13 + 16 + 19)
June 23 = 16
3
(16 + 19 + 23)
July 26 = 19
3
(19 + 23 + 26)
August 30 = 23
3
(23 + 26 + 30)
September 28 = 26
3
(26 + 30 + 28)
October 18 = 28
3
(30 + 28 + 18)
November 16 = 25
3
(28 + 18 + 16)
December 14 = 21
3
(18+16+14)
 With that said, January of next year’s forecast will be: = 𝟏𝟏𝟏𝟏.
3

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 2 of 9
 BM1710

When a detectable trend or pattern is present, weights can be used to place more emphasis on recent
values. This practice makes forecasting techniques more responsive to changes because more recent
periods may be more heavily weighted. Choice of weights is somewhat arbitrary because there is no set
formula to determine them. Therefore, deciding which weights to use requires some experience. For
example, if the latest month or period is weighted too heavily, the forecast may reflect a large unusual
change in the demand or sales pattern too quickly (Heizer, Render, & Munson, 2017).
∑((𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑡𝑡 𝑓𝑓𝑓𝑓𝑓𝑓 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 𝑛𝑛)(𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 𝑖𝑖𝑖𝑖 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 𝑛𝑛))
𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑡𝑡𝑡𝑡𝑡𝑡 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 =
∑𝑊𝑊𝑊𝑊𝑊𝑊𝑊𝑊ℎ𝑡𝑡𝑡𝑡

WEIGHTS APPLIED PERIOD

3 Last month
2 Two (2) months ago
1 Three (3) months ago
6 Sum of weights

Forecast for this month:

(3 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚ℎ) + (2 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 2 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚ℎ𝑠𝑠 𝑔𝑔𝑔𝑔) + (1 × 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 3 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚ℎ𝑠𝑠 𝑎𝑎𝑎𝑎𝑎𝑎)
𝑆𝑆𝑆𝑆𝑆𝑆 𝑜𝑜𝑜𝑜 𝑡𝑡ℎ𝑒𝑒 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤ℎ𝑡𝑡𝑡𝑡

MONTH ACTUAL SHED SALES 3-MONTH MOVING AVERAGE

January 10

February 12

March 13
[(𝟑𝟑 × 𝟏𝟏𝟏𝟏) + (𝟐𝟐 × 𝟏𝟏𝟏𝟏) + (𝟏𝟏𝟏𝟏)]
April 16 = 𝟏𝟏𝟏𝟏
𝟔𝟔
[(𝟑𝟑 × 𝟏𝟏𝟏𝟏) + (𝟐𝟐 × 𝟏𝟏𝟏𝟏) + (𝟏𝟏𝟏𝟏)]
May 19 = 𝟏𝟏𝟏𝟏
𝟔𝟔
[(𝟑𝟑 × 𝟏𝟏𝟏𝟏) + (𝟐𝟐 × 𝟏𝟏𝟏𝟏) + (𝟏𝟏𝟏𝟏)]
June 23 = 𝟏𝟏𝟏𝟏
𝟔𝟔
[(𝟑𝟑 × 𝟐𝟐𝟐𝟐) + (𝟐𝟐 × 𝟏𝟏𝟏𝟏) + (𝟏𝟏𝟏𝟏)]
July 26 = 𝟐𝟐𝟐𝟐
𝟔𝟔
[(𝟑𝟑 × 𝟐𝟐𝟐𝟐) + (𝟐𝟐 × 𝟐𝟐𝟐𝟐) + (𝟏𝟏𝟏𝟏)]
August 30 = 𝟐𝟐𝟐𝟐
𝟔𝟔
[(𝟑𝟑 × 𝟑𝟑𝟑𝟑) + (𝟐𝟐 × 𝟐𝟐𝟐𝟐) + (𝟐𝟐𝟐𝟐)]
September 28 = 𝟐𝟐𝟐𝟐
𝟔𝟔
[(𝟑𝟑 × 𝟐𝟐𝟐𝟐) + (𝟐𝟐 × 𝟑𝟑𝟑𝟑) + (𝟐𝟐𝟐𝟐)]
October 18 = 𝟐𝟐𝟐𝟐
𝟔𝟔
[(𝟑𝟑 × 𝟏𝟏𝟏𝟏) + (𝟐𝟐 × 𝟐𝟐𝟐𝟐) + (𝟑𝟑𝟑𝟑)]
November 16 = 𝟐𝟐𝟐𝟐
𝟔𝟔
[(𝟑𝟑 × 𝟏𝟏𝟏𝟏) + (𝟐𝟐 × 𝟏𝟏𝟏𝟏) + (𝟐𝟐𝟐𝟐)]
December 14 = 𝟏𝟏𝟏𝟏
𝟔𝟔

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 3 of 9
 BM1710

Exponential Smoothing
Exponential smoothing is another weighted-moving-average forecasting method. It involves very little
record keeping of past data and is fairly easy to use. The basic exponential smoothing formula can be
shown as:

𝑁𝑁𝑁𝑁𝑁𝑁 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 = 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑜𝑜𝑑𝑑 ′ 𝑠𝑠 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 + 𝛼𝛼 (𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑑𝑑′ 𝑠𝑠 𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑

− 𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑑𝑑 ′ 𝑠𝑠 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓)

Where:

α = weight, or smoothing constant, chosen by the forecaster, that has a value greater than or equal to
0 (≥0) and less than or equal to 1 (≤1)

The equation above can also be written mathematically as:

𝐹𝐹𝑡𝑡 = 𝐹𝐹𝑡𝑡−1 + 𝛼𝛼(𝐴𝐴𝑡𝑡−1 − 𝐹𝐹𝑡𝑡−1 )

Where:

𝑭𝑭𝒕𝒕 = 𝑛𝑛𝑛𝑛𝑛𝑛 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓

𝑭𝑭𝒕𝒕−𝟏𝟏 = 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑑𝑑′ 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠
𝜶𝜶 = 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠ℎ𝑖𝑖𝑖𝑖𝑖𝑖 (𝑜𝑜𝑜𝑜 𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤ℎ𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡) 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 (0 ≤ 𝛼𝛼 ≤ 1)

𝑨𝑨𝒕𝒕−𝟏𝟏 = 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑑𝑑′ 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑

EXAMPLE:

In January, a car dealer predicted February demand for 142 Ford Mustangs. Actual February
demand was 153 autos. Using a smoothing constant chosen by management of α = .20, the
dealer wants to forecast March demand using the exponential smoothing model.

𝐹𝐹𝑡𝑡 = 𝐹𝐹𝑡𝑡−1 + 𝛼𝛼(𝐴𝐴𝑡𝑡−1 − 𝐹𝐹𝑡𝑡−1 ) = 142 + .20(153 − 142) = 144.2

 This means that the forecast for March is 144.

Measuring Forecast Error

The overall accuracy of any forecasting model – moving average, exponential smoothing, or other – can
be determined by comparing the forecasted values with the actual or observed values.

𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 = 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷 − 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 = 𝑨𝑨𝒕𝒕 − 𝑭𝑭𝒕𝒕

Several measures are used in practice to calculate the overall forecast error. These measures can be used
to compare different forecasting models, as well as to monitor forecasts to ensure they are performing
well. Three of the most popular measures are mean absolute deviation (MAD), mean squared error
(MSE), and mean absolute percent error (MAPE).

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 4 of 9
 BM1710

Mean Absolute Deviation

∑ | 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 − 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 |
𝑀𝑀𝑀𝑀𝑀𝑀 =
𝑛𝑛
EXAMPLE:

During the past eight (8) quarters, the Port of Baltimore has unloaded large quantities of grain from
ships. The port’s operations manager wants to test the use of exponential smoothing to see how well
the technique works in predicting tonnage unloaded. He guesses that the forecast of grain unloaded
in the first quarter was 175 tons. Two (2) values of α are to be examined: α = .10 and α = .50

 To solve for the MAD, solve first for the forecast per period using exponential smoothing:

ACTUAL TONNAGE FORECAST WITH α

QUARTER FORECAST WITH α = .10
UNLOADED = .50
1 180 175 175
2 168 175.00 + .10(180 − 175) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟓𝟓𝟓𝟓
3 159 175.50 + .10(168 − 175.50) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟕𝟕𝟕𝟕
4 175 174.75 + .10(159 − 174.75) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟏𝟏𝟏𝟏
5 190 173.18 + .10(175 − 173.18) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟑𝟑𝟑𝟑
6 205 173.36 + .10(190 − 173.36) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟎𝟎𝟎𝟎
7 180 175.02 + .10(205 − 175.02) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟎𝟎𝟎𝟎
8 182 178.02 + .10(180 − 178.02) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟐𝟐𝟐𝟐
9 ? 178.22 + .10(182 − 178.22) = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟓𝟓𝟓𝟓

To evaluate the accuracy of each smoothing constant, we can compute forecast errors in terms of
absolute deviations and MADs:

ACTUAL FORECAST ABSOLUTE ABSOLUTE

FORECAST
QUARTER TONNAGE WITH α = DEVIATION FOR α DEVIATION FOR
WITH α = .50
UNLOADED .10 = .10 α = .50
1 180 175 5.00 175 175
2 168 175.50 7.50
3 159 174.75 15.75
4 175 173.18 1.82
5 190 173.36 16.64
6 205 175.02 29.98
7 180 178.02 1.98
8 182 178.22 3.78
Sum of absolute deviations: 82.45

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 5 of 9
 BM1710

82.45
∑ |𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷𝐷| 𝑴𝑴𝑴𝑴𝑴𝑴 =
𝑴𝑴𝑴𝑴𝑴𝑴 = 8
𝑛𝑛 = 𝟏𝟏𝟏𝟏. 𝟑𝟑𝟑𝟑

Mean Squared Error

∑(𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒)2
𝑀𝑀𝑀𝑀𝑀𝑀 =
𝑛𝑛

ACTUAL FORECAST ABSOLUTE

QUARTER TONNAGE WITH α = DEVIATION FOR α (ERROR)2
UNLOADED .10 = .10
1 180 175 5.00 52 = 𝟐𝟐𝟐𝟐
2 168 175.50 7.50 (−7.5)2 = 𝟓𝟓𝟓𝟓. 𝟐𝟐𝟐𝟐
3 159 174.75 15.75 (−15.75)2 = 𝟐𝟐𝟐𝟐𝟐𝟐. 𝟎𝟎𝟎𝟎
4 175 173.18 1.82 (1.82)2 = 𝟑𝟑. 𝟑𝟑𝟑𝟑
5 190 173.36 16.64 (16.64)2 = 𝟐𝟐𝟐𝟐𝟐𝟐. 𝟖𝟖𝟖𝟖
6 205 175.02 29.98 (29.98)2 = 𝟖𝟖𝟖𝟖𝟖𝟖. 𝟖𝟖𝟖𝟖
7 180 178.02 1.98 (1.98)2 = 𝟑𝟑. 𝟗𝟗𝟗𝟗
8 182 178.22 3.78 (3.78)2 = 𝟏𝟏𝟏𝟏. 𝟐𝟐𝟐𝟐
Sum of errors squared 1526.52
∑(𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹 𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒)2 1526.52
𝑀𝑀𝑀𝑀𝑀𝑀 = 𝑀𝑀𝑀𝑀𝑀𝑀 = = 𝟏𝟏𝟏𝟏𝟏𝟏. 𝟖𝟖
𝑛𝑛 8

 A low MSE is better because MSE exaggerates errors through squaring.

Mean Absolute Percent Error

A problem with both the MAD and MSE is that their values depend on the magnitude of the item being
forecast. If the forecast item is measured in thousands, the MAD and MSE values can be very large. To
avoid this problem, we can use the mean absolute percent error (MAPE).

|𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 − 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖 |
∑𝑛𝑛𝑖𝑖=1 100
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖
𝑀𝑀𝑀𝑀𝑃𝑃𝑃𝑃 =
𝑛𝑛

ACTUAL FORECAST ABSOLUTE

QUARTER TONNAGE WITH α = DEVIATION FOR α PERCENTAGE
UNLOADED .10 = .10
5
1 180 175 5.00 100 � � = 𝟐𝟐. 𝟕𝟕𝟕𝟕%
180
7.5
2 168 175.50 7.50 100 � � = 𝟒𝟒. 𝟒𝟒𝟒𝟒%
168
15.75
3 159 174.75 15.75 100 � � = 𝟗𝟗. 𝟗𝟗𝟗𝟗%
159

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 6 of 9
 BM1710

1.82
4 175 173.18 1.82 100 � � = 𝟏𝟏. 𝟎𝟎𝟎𝟎%
175
16.64
5 190 173.36 16.64 100 � � = 𝟖𝟖. 𝟕𝟕𝟕𝟕%
190
29.98
6 205 175.02 29.98 100 � � = 𝟏𝟏𝟏𝟏. 𝟔𝟔𝟔𝟔%
205
1.98
7 180 178.02 1.98 100 � � = 𝟏𝟏. 𝟏𝟏𝟏𝟏%
180
3.78
8 182 178.22 3.78 100 � � = 𝟐𝟐. 𝟎𝟎𝟎𝟎%
182
Sum of errors squared 44.75%
|𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 − 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖 |
∑𝑛𝑛𝑖𝑖=1 100 44.75%
𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝑖𝑖 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = = 𝟓𝟓. 𝟓𝟓𝟓𝟓%
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = 8
𝑛𝑛

 MAPE expresses the error as a percent of the actual values, undistorted by a single large value.

Associative Forecasting: Regression and Correlation Analysis

Simple Linear Regression

ŷ= 𝑎𝑎 + 𝑏𝑏𝑏𝑏
Where: ŷ = dependent variable
𝑎𝑎 = y-axis intercept

𝑏𝑏 = slope of the regression line

𝑥𝑥 = independent variable

EXAMPLE:
Nodel Construction Company renovates old homes in West Bloomfield, Michigan. Over time, the
company has found that its dollar volume of renovation work is dependent on the West Bloomfield
area payroll. Management wants to establish a mathematical relationship to help predict sales.

NODEL’S SALES (IN $ AREA PAYROLL (IN $

MILLIONS), y BILLIONS), x
2.0 1
3.0 3
2.5 4
2.0 2
2.0 1
3.5 7

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 7 of 9
 BM1710

Solution:
SALES, y PAYROLL, x x^2 y^2 xy
2.0 1 1 4 2.0
3.0 3 9 9 9.0
2.5 4 16 6.25 10.0
2.0 2 4 4 4.0
2.0 1 1 4 2.0
3.5 7 49 12.25 24.5
∑𝑦𝑦 = 15 ∑𝑥𝑥 = 18 ∑𝑥𝑥 2 = 80 ∑𝑦𝑦 2 = 39.5 ∑𝑥𝑥𝑥𝑥 = 51.5

∑𝑥𝑥 18
𝑥𝑥� = = = 𝟑𝟑
𝑛𝑛 6
∑𝑦𝑦 15
𝑦𝑦� = = = 𝟐𝟐. 𝟓𝟓
𝑛𝑛 6
∑𝑥𝑥𝑥𝑥 − 𝑛𝑛𝑥𝑥̅ 𝑦𝑦� 51.5 − 6(3)(2.5)
𝑏𝑏 = = = . 𝟐𝟐𝟐𝟐
∑𝑥𝑥 2 − 𝑛𝑛𝑥𝑥̅ 2 80 − 6(32 )
𝑎𝑎 = 𝑦𝑦� − 𝑏𝑏𝑥𝑥̅ = 2.5 − (. 25)(3) = 𝟏𝟏. 𝟕𝟕𝟕𝟕

Thus, the estimated regression equation is:

𝑦𝑦� = 1.75 + .25𝑥𝑥
Or:
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = 1.75 + .25(𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝)
If: the local chamber of commerce predicts that the West Bloomfield area payroll will be $6 billion next
year, we can estimate sales for Nodel with the regression equation:

𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 (𝑖𝑖𝑖𝑖 $ 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚) = 1.75 + .25(6) = 1.75 + 1.50 = 3.25

Therefore,

𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 = $3,250,000
Standard Error of the Estimate

To measure the accuracy of the regression estimates, we must compute the standard error of the
estimate.

It measures the error from the dependent variable, y, to the regression line, rather than to the mean.

∑𝑦𝑦 2 − 𝑎𝑎∑𝑦𝑦 − 𝑏𝑏∑𝑥𝑥𝑥𝑥

𝑆𝑆𝑦𝑦,𝑥𝑥 = �
𝑛𝑛 − 2

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 8 of 9
 BM1710

The interpretation of the standard error of the estimate is similar to the standard deviation; namely, ±1
standard deviation = .6827. So there is a 68.27% chance of sales being ±$306,000 from the point estimate
of $3,250,000.

Correlation coefficients for Regression Lines

𝑛𝑛∑𝑥𝑥𝑥𝑥 − ∑𝑥𝑥∑𝑦𝑦
𝑟𝑟 =
�[𝑛𝑛∑𝑥𝑥 2 − (∑𝑥𝑥)2 ][𝑛𝑛∑𝑦𝑦 2 − (∑𝑦𝑦)2 ]

Forecasting in the Service Sector

Forecasting in the service sector presents some unusual challenges. A major technique in the retail sector
is tracking demand by maintaining good short-term records.

EXAMPLE: A barbershop catering to men expects peak flows on Fridays and Saturdays. Indeed, most
barbershops are closed on Sunday and Monday, and many call in extra help on Friday and Saturday. A
downtown restaurant, on the other hand, may need to track conventions and holidays for effective short-
term forecasting (Heizer, Render, & Munson, 2017).

In the cases of specialty retail shops, such as flower shops, many have other unusual demand patterns
which will differ depending on the holiday. When Valentine’s Day falls on a weekend, flowers can’t be
delivered to offices, and those romantically inclined are likely to celebrate with outings rather than
flowers. If a holiday falls on a Monday, some of the celebration may also take place on the weekend,
reducing flower sales. However, when Valentine’s Day falls in a midweek, busy midweek schedules often
make flowers the optimal way to celebrate. Because flowers for mother’s Day are to be delivered on
Saturday or Sunday, this holiday forecast varies less. Due to special demand patterns, many service firms
maintain records of sales, noting not only the day of the week but also unusual events, including the
weather, so that patterns and correlations that influence demand can be developed (Heizer, Render, &
Munson, 2017).

REFERENCES
Heizer, J., Render, B., & Munson, C. (2017). Operations management: Sustainability and supply chain
management (12th ed.). Boston: Pearson Education Inc.

Krajewski, L. J., Malhotra, M. K., & Ritzman, L. P. (2016). Operations management: Processes and supply
chains (11th ed.). Essex: Pearson Education Limited.

Schroeder, R. & Goldstein, S. M. (2016). Operations management in the supply chain: Decisions and cases
(7th ed.). Dubuque: McGraw-Hill Education.

Stevenson, W. J. (2015). Operations management (12th ed.). New York: McGraw Hill Education.

02 Handout 1 *Property of STI

 student.feedback@sti.edu Page 9 of 9