1

Module

FORECASTING
I. Objectives
The objectives of the practicum are:
1. Understand the basic concepts and principles in forecasting.
2. Analyze and apply the right forecasting models based on demand pattern.
3. Compare between forecasting model’s results.
4. Calculate measures of forecasting accuracy and interpret the result.

II. Literature Review

A. Basic Concepts and Principles of Forecasting
A forecast is an estimate or prediction of the future. Variables to be
forecasted can be anything, most often is demand, where other variables could be
supply, price, capacities, et cetera. The importance of forecasting to operations
management cannot be overstated. Having a forecast of demand is essential for
determining how much capacity or supply will be needed to meet demand.
Forecast’s main purpose is to reduce the uncertainty of the future and make better
estimates of what will happen (Stevenson, 2012).
Consider the push/pull view of the supply chain, all push processes are
performed in anticipation of customer demand, and all pull process are performed
in response to customer demand. For push (make to stock) process, manager must
plan the level of production. For the pull (make to order) processes, a manager
must plan the level of capacity to make available. For both instances, the first step
that a supply chain manager must take is to forecast what customer demand will
be in the future (Chopra & Meindl, 2010).
A forecast is usually classified by the future time horizon that it covers.
Figure 1.1 illustrates the three-time horizon associated with forecasting and
typical forecasting problems encountered in operations planning associated with
each (Heizer, Render, & Munson, 2017).
• Short-range forecast: Short-range forecasting is crucial for day-to-day
planning. Short-range forecasts, typically measured in days or weeks, are
required for inventory management. Other planning with short-range forecast
includes purchasing, job scheduling, workforce levels, and production
planning. Shift scheduling or job assignments may require forecasts or
worker’s availabilities and preferences.
• Medium-range forecast: The medium-range forecast is measured in weeks or
months. Sales patterns for product families, requirements, production planning
and budgeting, and resource requirements are typical medium-range
forecasting problems encountered in operations management.
• Long-range forecast: Usually three years or more in timespan. Long-range
forecasts are used in planning for new products, capital expenditures, business
strategy, facility location or expansion, and research and development. One

Industrial Engineering Laboratory

BINUS University 1 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

example is long-term planning of capacity needs. When demands are expected

to increase, the firm must plan for the construction of new facilities and/or the
retrofitting of existing facilities with new technologies. Capacity planning
decisions may require downsizing in some circumstances. For example,
General Motors Corporation historically commanded about 45 percent of the
domestic car market. However, in the 1990s that percentage dropped to 35
percent. As a result, GM was force to significantly curtail its manufacturing
operations to remain profitable.

Source: (Nahmias & Olsen, 2015)

Figure 1.1 Forecast Horizons in Operation Planning

There are four main laws of forecasts (Bozarth & Handfield, 2016):
1. Forecasts are almost always wrong.
Even under the best of conditions, no forecasting approach can predict
the exact level of future demand, supply, or price. There are simply too many
factors that can ultimately affect these numbers. Rather, businesses should use
forecasting methods to get close estimates.
2. Forecasts for the near-term tend to be more accurate.
In near-term forecasts, the factors that affect the forecast variable are not
likely to change greatly. Main example for this law is gas prices. Given your
understanding of current economic and political conditions as well the current
price, you may feel reasonably comfortable predicting the price of gas for the
next period. However, what about 10 or 20 years from now? The effect of
variables could radically affect the price of gas in the long-term, including the
technological breakthroughs, demographic changes, and more.
3. Forecasts for groups of products or services tend to be more accurate.
Forecast for groups (aggregate) of products or services tend to be easier
and more accurate that it is to forecast for specific ones. This happens because
the demand, supply, or price of a specific item is usually affected by many
more factors. For example, the demand for dark green cars versus all cars.
Color fashion may affect the precise demand for green cars. However, when

Industrial Engineering Laboratory

BINUS University 2 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

looking at the overall demand, the impact of color fashion disappears. Higher
or lower demand for green cars is balanced out by demand for cars of other
colors. On a percentage basis, the error made in forecasting sales for an entire
product line is generally less than the error made in forecasting sales for an
individual item.
4. Forecasts are no substitute for calculated values.
Forecasts should be used only when better approaches to determining
the variable of interest are not available. To see what can go wrong when this
law is not followed, consider the experiences of a plant visited by one of the
authors. The plant made rubber products. Every Wednesday, the management
team would determine how many of each product would be made in the
coming week. From this production plan, the plant’s buyers could have easily
calculated exactly how much and what grades of war rubber would be needed.
Instead, the buyers chose to forecast rubber requirements. As a result,
sometimes the plant had too much rubber on hand, and at other times, it did
not have enough. In effect, the plant forecasted demand when it would have
been simpler and more accurate to calculate demand.
A company must understand such factors before it can select an
appropriate forecasting methodology. Figure 1.2 provides a road map to help
analyst figure the most appropriate forecasting methods. Forecasting methods
are classified according to the following types (Bozarth & Handfield, 2016):

Source: (Bozarth & Handfield, 2016)

Figure 1.2 Selecting a Forecasting Model

I. Qualitative Method
Qualitative forecasting method are forecasting methods based on
intuition or informed opinion. These methods are best used when historical
data are scarce, not available, of irrelevant.
• Market Surveys
Structured questionnaires submitted to potential customers. They
solicit opinions about product or potential products and often attempt to
estimate likely demand. A major drawback is that market surveys tend
to be expensive and time-consuming to perform.

Industrial Engineering Laboratory

BINUS University 3 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

• Panel Consensus Forecasting

A forecasting method that uses panels of experts to develop a
consensus forecast. This method brings the experts together to discuss
and develop forecast.
• Delphi Method
Similar to the panel consensus forecasting, but in contrast, the
Delphi Method has experts work individually to develop forecasts. The
individual forecasts are shared among the group, and then each
participant is allowed to modify his or her forecast based on information
from the other experts. This process is repeated until consensus is
reached.
• Life-Cycle Analogy Method
A method used when a product or service is new. The method
involves observation on the product while considering these stages:
introduction stage, growth stage, maturity stage, and decline stage. The
major questions that arise include the following: “how long will each
stage last?”, “how rapid will the growth be? how rapid will the decline
be?”, et cetera. One approach is to base the forecast for the new product
or service on the actual history of a similar product or service.
• Build-up Forecasts
A forecasting method that uses individuals across specific market
segments to estimate the demand within these segments. These forecasts
are then added up to get an overall forecast.
• Jury of Executive Opinions
Similar to build-up forecast method, however the individuals used
are groups of high-level experts or managers, often in combination with
statistical models, which will then be pooled to arrive at a group of
estimates of demand.
• Sales Force Composite
This method uses salesperson to estimate 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.

II. Quantitative Method

Chopra & Meindl (2010) defined three types of quantitative
forecasting method, such as:
• Time Series
Time series forecasting methods use historical demand to make a
forecast. There are based on the assumption that past demand history is
good indicator of future demand. These methods are most appropriate
when the environmental situation is stable and the basic demand pattern
does not vary significantly from one year to the next. These are the
simplest methods to implement and can serve as a good starting point
for a demand forecast.
• Causal
Causal forecasting methods involve assuming that the demand
forecast is highly correlated with certain factors in the environment (e.g.,
the state of the economy, interest rates). Causal forecasting methods find
this correlation between demand and environmental factors and use

Industrial Engineering Laboratory

BINUS University 4 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

estimates of what environmental factors will be to forecast future

demand.
• Simulation
Simulation forecasting methods imitate the customer choices that
give rise to demand to arrive at a forecast. Using simulations, a firm can
combine time series and causal methods to answer such questions as
these: What will the impact of a price promotion be? What will the
impact be of a competitor opening a store nearby?

B. Demand pattern
Analyzing time series means breaking down past data into components and
then projecting them forward. A time series has four components (Heizer, Render,
& Munson, 2017):
a. Level or Horizontal Pattern
The pattern has no trend and is stationary, the next value will be above
the mean or below it. This pattern can be represented by stable sales and
number of defects in production process.
b. Trend Pattern
Gradual upward or downward movement of the data over time.
c. Seasonal Pattern
Data pattern that repeats itself after a period of days, weeks, months, or
quarters. Seasonality that occurs for more than a year’s period can also mean
a cycle pattern.
d. Cyclical Pattern
Patterns in the data that occur every several years. They are usually tied
into the business cycle and are of major importance in short-term business
analysis and planning.

Source: (Reid & Sanders, 2010)

Figure 1.3 Demand Pattern

Industrial Engineering Laboratory

BINUS University 5 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

C. Time-series Forecasting Methods

Quantitative forecasting models use statistical methods and historical
data to predict future levels. Such forecasting models are considered objective
rather than subjective because they follow certain rules in calculation. Here,
time series consists of observations arranges in chronological order. Time
series forecasting models, then, are quantitative models that analyze time
series to develop forecasts (Bozarth & Handfield, 2016).

1. Naïve Approach (Last Period)

The simplest way to forecast is to assume that demand in the next
period will be equal to demand in the most recent period. Even though
logically it does not make any sense, the naïve approach is the most cost-
effective and efficient objective forecasting model for some product lines
(Heizer, Render, & Munson, 2017). Stated formally (Bozarth & Handfield,
2016):

Ft+1 = Dt
Where:
Ft+1 = Forecast for the next periode, t + 1
Dt = Demand for the current period, t

2. Moving Average
A Moving Average model derives a forecast by taking an average of
a set of recent demand values. By basing the forecast on more than one
observed demand value, the moving average model is less susceptible to
random swings in demand (Bozarth & Handfield, 2016). Another
perspective is a Moving Average model is used when demand has no
observable trend or seasonality. In the case, demand forecast is stated as
(Reid & Sanders, 2010):

a. Single Moving Average

∑ni=1 Dt+1-i
Ft+1 =
n
Where:
Ft+1 = Forecast for time period t+1
Dt+1-i = Actual demand for period t+1-i
n = Number of most recent demand observations used to develop
the forecast

b. Weighted Moving Average

n

Ft+1 = ∑ Wt+1-i Dt+1-i

i=1
Where:
Ft+1 : Forecast at period t+1
Wt+1-i : Weight assigned to the demand in period t+1-i
N : Number of the time series

Industrial Engineering Laboratory

BINUS University 6 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

Which is better? The shorter period or the longer period model used
in moving average model Generally speaking, it depends on the raw data.
The more randomness there is in the raw data, the more attractive the
smoothing and delayed reaction characteristics are. Longer period models
in moving average would be preferable in such a case, because long period
moving average models are more stable rather than short period moving
average models.
On the other hand, if rises or falls in demand are not random but
really do indicate changes in the underlying demand pattern, then a more
reactive and responsive model is more preferred. Short period moving
average models are better used in this case (Bozarth & Handfield, 2016).

3. Exponential Smoothing
The exponential smoothing model is a special form of the moving
average model in which the forecast for the next period is calculated as the
weighted average of the current period’s actual value and forecast (Bozarth
& Handfield, 2016)
a. Simple Exponential Smoothing
Simple exponential smoothing model is best used when there
seems to be no observable trend or seasonality.

Ft+1 = α.Dt + (1-α).Ft

Where:
Ft+1 = Forecast for the time period t+1 (the new forecast)
Ft = Forecast for time period t (the current forecast)
Dt : Actual demand for t period
α : smoothing constant, 0 < α < 1, defined by forecaster

The general rule for determining the α value is: the greater the
randomness in the time series data is, the lower the α value should be.
Conversely, the less randomness in the time series data, the higher the
α value should be. Higher α value means more responsiveness.

b. Exponential Smoothing with Trend Adjustment (Holt's Method)

Single exponential smoothing cannot respond when a trend is
present. A modified model of exponential smoothing to respond to
trend’s presence is called the Holt’s model. Holt’s model method is
appropriate when the demand is assumed to have a level and a trend in
the systematic component but no seasonality. As before, the smoothing
constants α (average smoothing constant) and β (trend smoothing
constant) are chosen to lie between 0 and 1. The new forecast formula
is (Heizer, Render, & Munson, 2017):

FITt = Ft + Tt

Computations for the Ft and Tt are as follows:

Ft = α(At-1) + (1-α).(Ft-1+Tt-1)

Tt = β (Ft – Ft-1) + (1-β).Tt-1

Industrial Engineering Laboratory

BINUS University 7 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

Where:
FITt = Forecast including trend in period t
Ft = Exponentially smoothed forecast average in period t
Tt = Exponentially smoothed trend in period t
At = Actual demand in period t
α = Smoothing constant for the average (0 ≤ α ≤ 1)
β = Smoothing constant for the trend (0 ≤ β ≤ 1)

The model requires initial specification of values for F0 and T0;

F0 might be set equal to the first month’s observation and T0 might be
set equal to the average monthly increase in observations from last year.

c. Linear Regression (Causal Models)

An approach to forecasting when there is a trend in the data is
linear regression. Linear regression is a statistical technique that
expresses the forecast variable as a linear function of some independent
variable. In the case of a time series model, the independent variable is
the time period itself. Linear regression works by using past data to
estimate the intercept term and slope coefficient for the following line
(Bozarth & Handfield, 2016):
ŷ = â + ̂bx
Where:
ŷ = Forecast for dependent variable y
𝑥 = Independent variable x, used to forecast y
â = Estimated intercept term for the line
̂b = Estimated slope coefficient for the line

â and b̂ are estimated using the raw time series data for variable y
(the dependent variable) and variable x (the independent variable):

[∑ni=1 xi ][∑ni=1 yi ]
∑ni=1 xi yi -
b̂ = n
n
[∑ x ]2
∑ni=1 x2i - i=1 i
n

â = 𝑦̅ - ̂b𝑥̅

Where:
(xi, yi) = matched pairs of observed (x, y) values
𝑦̅ = average y value
𝑥̅ = average x value
n = number of paired observations

Once the equation for ŷ = â + ̂bx has been estimated, the

forecaster can then plug-in values for x, the independent variable, to
generate forecast values, 𝑦̅

Industrial Engineering Laboratory

BINUS University 8 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

d. Exponential Smoothing Triple Parameter (Winters’ Method)

Winter’s model method is appropriate when the systematic
component of demand has a level, a trend, and a seasonal factor.
Assume periodicity of demand to be p. To begin, we need initial
estimates of level (So), trend (G0), and seasonal factors (C1, Cp). We
obtain these estimates using the procedure for forecast for future
periods were given by (Chopra & Meindl, 2010):

Ft,t+τ = (St + 𝜏Gt). Ct+τ-N

On observing demand for Period t + 1 we revise the estimates for

level, trend and seasonal factors as follows:

St = α.(Dt/Ct-N) + (1-α) . (St-1 + Gt-1)

Gt = β . (St - St-1) + (1 – β).Gt-1

Ct = γ.(Dt/St) + (1 – γ) . Ct-N

Where α is a smoothing constant for the level, 0< α < 1; β is a

smoothing constant for the trend, 0 < β < 1; and γ is a smoothing
constant for the seasonal factor, 0 < γ < 1.

D. Forecast Accuracy
As mentioned by Chopra & Meindl (2010), every instance of demand
has a random component. A good forecasting method should capture the
systematic component of demand but not the random component. The random
component manifests itself in the form of a forecast error. This forecast error
can show us how much we can rely on this forecast that we have made before.
The forecast accuracy can be measured by following method (Render, Stair,
& Hanna, 2018):

1. MAD (Mean Absolute Deviation)

It measures the total error in a forecast without regard a sign. The
formula is as following:

∑nt=1|Dt -Ft |
MAD =
n
Where:
Ft : Forecast in period t
Dt : Demand in period t
n : Number of errors

Industrial Engineering Laboratory

BINUS University 9 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

2. MSE (Mean Square Error)

It penalizes larger error. The MSE can be related to the variance of
the forecast error. The formula is as following:

∑nt=1 (Dt -Ft )2

MSE =
n

Where:
Ft : Forecast in period t
Dt : Demand in period t
n : Number of errors

3. MAPE (Mean Absolute Percentage Error)

It is the average of the absolute values of the errors expressed as
percentages of the actual values. MAPE can be useful in cases where the
item forecasted is measured in large quantity. The formula is as following:

|Ft -Dt |
∑nt=1
Dt
MAPE = x 100%
n

Where:
Ft : Forecast in period t
Dt : Demand in period t
n : Number of errors

4. Tracking Signal
It measures that if your model is working or not. The formula is as
following:

∑nt=1 (Dt -Ft ) ∑nt=1 (Dt -Ft )

TS = =
∑nt=1 |Dt -Ft | MAD
n

If the TS at any period is outside the range ±4, this is a signal that
the forecast is biased and is either under forecasting (TS ≤ -4) or over
forecasting (TS ≥ +4). In this case, a firm may decide to choose a new
forecasting method (Bozarth & Handfield, 2016).

Industrial Engineering Laboratory

BINUS University 10 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

III. Procedure
A. Case Example for Forecasting
Company XYZ is a new company that produce cotton shirts. The sales
of cotton shirts from 2 years lately become stable. Then, the marketing and
sales division had collected sales data for the production planning for one year
ahead and gave the data to the production division to determine the production
planning which start by using demand forecasting.

The table below is the sales data of train toys from January 2018 to
December 2019:
Table 1.1 Data Sales
No Period Sales No Period Sales
1 January 2018 1273 13 January 2019 1848
2 February 2018 1333 14 February 2019 1992
3 March 2018 1618 15 March 2019 1919
4 April 2018 1739 16 April 2019 1729
5 May 2018 1513 17 May 2019 1802
6 June 2018 1449 18 June 2019 1999
7 July 2018 1584 19 July 2019 2111
8 August 2018 1748 20 August 2019 1992
9 September 2018 1840 21 September 2019 1916
10 October 2018 1675 22 October 2019 1996
11 November 2018 1534 23 November 2019 2146
12 December 2018 1735 24 December 2019 2308

Remark:
The sales data on each month added with 2 last digits from group members’
student ID number.
Example: ABC xxxxxxxx22
DEF xxxxxxxx25
GHI xxxxxxxx63

Hence, the total sum is 22+25+63= 110

The sales for January 2018 became:

1273 + 110 = 1383*
*Applied it to all sales data.

Demand forecasting is using some methods such as: Moving average,

Holt’s Method, Time-Series Decomposition, and Winter Method. The
best forecasting calculation will be use for the next production planning.
0.00< α <1.00
0.00< β <1.00
0.00< γ <1.00

Industrial Engineering Laboratory

BINUS University 11 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

B. Procedure for Time Series Decomposition:

1. Using Microsoft Excel, copy all sales data of cotton shirt from January
2018 to December 2019 and sum with 110 to all sales data.
Table 1.2 Sales Data
Sales NIM Sales + NIM
1273 110 1383
1333 110 1443
1618 110 1728
1739 110 1849
1513 110 1623
1449 110 1559
1584 110 1694
1748 110 1858
1840 110 1950
1675 110 1785
1534 110 1644
1735 110 1845
1848 110 1958
1992 110 2102
1919 110 2029
1729 110 1839
1802 110 1912
1999 110 2109
2111 110 2221
1992 110 2102
1916 110 2026
1996 110 2106
2146 110 2256
2308 110 2418

2. Create a table with the name of column is number, period, sales, CMA,
dt bar, seasonal, forecast, error, MSE, ABS error, % error, MAPE, MAD,
bias, and TS. We will forecast for the next 12 Months from January 2020
until December 2020 (ignoring pandemic situation), so input the period
column from January 2018 until December 2019.

3. Copy the sales result from step number 1 to sales column in worksheet
from step number 2.

Industrial Engineering Laboratory

BINUS University 12 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

Table 1.3 TSD Worksheet

No Period Sales CMA dt bar Seasonal Forecast Error MSE ABS Error % Error MAPE MAD Bias TS
1 Jan-18 1383
2 Feb-18 1443
3 Mar-18 1728 1605.2
4 Apr-18 1849 1640.4
5 May-18 1623 1690.6
6 Jun-18 1559 1716.6
7 Jul-18 1694 1736.8
8 Aug-18 1858 1769.2
9 Sep-18 1950 1786.2
10 Oct-18 1785 1816.4
11 Nov-18 1644 1836.4
12 Dec-18 1845 1866.8
13 Jan-19 1958 1915.6
14 Feb-19 2102 1954.6
15 Mar-19 2029 1968
16 Apr-19 1839 1998.2
17 May-19 1912 2022
18 Jun-19 2109 2036.6
19 Jul-19 2221 2074
20 Aug-19 2102 2112.8
21 Sep-19 2026 2142.2
22 Oct-19 2106 2181.6
23 Nov-19 2256
24 Dec-19 2418
25 Jan-20
26 Feb-20
27 Mar-20
28 Apr-20
29 May-20
30 Jun-20
31 Jul-20
32 Aug-20
33 Sep-20
34 Oct-20
35 Nov-20
36 Dec-20

4. Make a graph for demand or sales data from January 2018 until
December 2019 by blocking the number and sales cells → insert → charts
→ scatter charts. The result is (chart can be chosen as preferred):

Sales Plot
3000

2500

2000

1500

1000

500

0
0 5 10 15 20 25 30

Figure 1.4 Sales Plot

5. Calculate the CMA (Center Moving Average) of five with formula

=(C2+C3+C4+C5+C6)/5. Fill this equation at the center of the five data,
for example if you have five data in column C2, C3, C4, C5, C6 the center
for these five data is C4. Drag the formula until the last cell to be included
in the formula. Notice this only works for odd-seasonal pattern. For even-
seasonal pattern, you have to calculate first the average of even-seasonal
sales and then calculate the center moving average from those averages.

6. Calculate level and trend, for the level find a value with formula
=INTERCEPT(Known y, Known x) and for the trend use

Industrial Engineering Laboratory

BINUS University 13 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

=SLOPE(Known y, Known x), where known y is CMA cells and known

x is the number of CMA cells.

Table 1.4 TSD Level and Trend

Level 1529.978045
Trend 29.08255639

7. Calculate dt bar (de-seasonalized demand), the dt bar equation based on

theory is:
dt bar = Level + (Trend x t).

t = time period, refers to the number in the excel (column A)

Note: use F4 to lock the cells holding the value of level and trend for
easier formula-dragging in excel. Lock in excel symbolized by “$”.

8. Calculate seasonal column by dividing between sales and dt bar, for

example use formula =sales/dtbar and drag down the result until
December 2019.

Table 1.5 TSD Worksheet 2

No Period Sales CMA dt bar Seasonal Forecast Error MSE ABS Error % Error MAPE MAD Bias TS
1 Jan-18 1383 1559.06 0.89
2 Feb-18 1443 1588.14 0.91
3 Mar-18 1728 1605.2 1617.23 1.07
4 Apr-18 1849 1640.4 1646.31 1.12
5 May-18 1623 1690.6 1675.39 0.97
6 Jun-18 1559 1716.6 1704.47 0.91
7 Jul-18 1694 1736.8 1733.56 0.98
8 Aug-18 1858 1769.2 1762.64 1.05
9 Sep-18 1950 1786.2 1791.72 1.09
10 Oct-18 1785 1816.4 1820.80 0.98
11 Nov-18 1644 1836.4 1849.89 0.89
12 Dec-18 1845 1866.8 1878.97 0.98
13 Jan-19 1958 1915.6 1908.05 1.03
14 Feb-19 2102 1954.6 1937.13 1.09
15 Mar-19 2029 1968 1966.22 1.03
16 Apr-19 1839 1998.2 1995.30 0.92
17 May-19 1912 2022 2024.38 0.94
18 Jun-19 2109 2036.6 2053.46 1.03
19 Jul-19 2221 2074 2082.55 1.07
20 Aug-19 2102 2112.8 2111.63 1.00
21 Sep-19 2026 2142.2 2140.71 0.95
22 Oct-19 2106 2181.6 2169.79 0.97
23 Nov-19 2256 2198.88 1.03
24 Dec-19 2418 2227.96 1.09
25 Jan-20 2257.04
26 Feb-20 2286.12
27 Mar-20 2315.21
28 Apr-20 2344.29
29 May-20 2373.37
30 Jun-20 2402.45
31 Jul-20 2431.54
32 Aug-20 2460.62
33 Sep-20 2489.70
34 Oct-20 2518.78
35 Nov-20 2547.87
36 Dec-20 2576.95

9. Calculate seasonal 1 (S1) until seasonal 5 (S5) value by averaging the

corresponding seasonal value from January 2018 until December 2019.
For example, to find the S1, average the first season of every cycle’s
value (see the cells highlighted in blue). The number of seasons is based
by CMA, known by analyzing the seasonal pattern.
Table 1.6 Seasonal
S1 0.911702
S2 0.956559
S3 1.040360
S4 1.089669
S5 0.994109

Industrial Engineering Laboratory

BINUS University 14 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

10. Find the forecast for January 2018 until December 2020 by multiplying
dt bar with the corresponding seasonal (S) value. Example, for dt bar
number 1, multiply with S1, number 2 with S2, number 3 with S3, etc.,
and repeat by multiplying number 6 with S1, and so on. Use formula
=ROUNDUP(dtbar*Seasonal,0).

11. Calculate the error value between sales data history and forecast value
from January 2018 until December 2019 with formula =sales column-
forecast column.

12. Calculate MSE (Mean Square Error) with using formula

=SUMSQ($first error$cells: error cells for t period)/number period.
Drag down the formula.

13. Find the absolute error from error column. Use formula =ABS(error
column).

14. Calculate %error by dividing between ABS error and sales data then
multiply with 100.
Table 1.7 TSD Worksheet 3
No Period Sales CMA dt bar Seasonal Forecast Error MSE ABS Error % Error MAPE MAD Bias TS
1 Jan-18 1383 1559.06 0.89 1422 -39 1521 39 2.82
2 Feb-18 1443 1588.14 0.91 1520 -77 3725 77 5.34
3 Mar-18 1728 1605.2 1617.23 1.07 1683 45 3158.33 45 2.60
4 Apr-18 1849 1640.4 1646.31 1.12 1794 55 3125 55 2.97
5 May-18 1623 1690.6 1675.39 0.97 1666 -43 2869.8 43 2.65
6 Jun-18 1559 1716.6 1704.47 0.91 1554 5 2395.67 5 0.32
7 Jul-18 1694 1736.8 1733.56 0.98 1659 35 2228.43 35 2.07
8 Aug-18 1858 1769.2 1762.64 1.05 1834 24 2021.88 24 1.29
9 Sep-18 1950 1786.2 1791.72 1.09 1953 -3 1798.22 3 0.15
10 Oct-18 1785 1816.4 1820.80 0.98 1811 -26 1686 26 1.46
11 Nov-18 1644 1836.4 1849.89 0.89 1687 -43 1700.82 43 2.62
12 Dec-18 1845 1866.8 1878.97 0.98 1798 47 1743.17 47 2.55
13 Jan-19 1958 1915.6 1908.05 1.03 1986 -28 1669.38 28 1.43
14 Feb-19 2102 1954.6 1937.13 1.09 2111 -9 1555.93 9 0.43
15 Mar-19 2029 1968 1966.22 1.03 1955 74 1817.27 74 3.65
16 Apr-19 1839 1998.2 1995.30 0.92 1820 19 1726.25 19 1.03
17 May-19 1912 2022 2024.38 0.94 1937 -25 1661.47 25 1.31
18 Jun-19 2109 2036.6 2053.46 1.03 2137 -28 1612.72 28 1.33
19 Jul-19 2221 2074 2082.55 1.07 2270 -49 1654.21 49 2.21
20 Aug-19 2102 2112.8 2111.63 1.00 2100 2 1571.7 2 0.10
21 Sep-19 2026 2142.2 2140.71 0.95 1952 74 1757.62 74 3.65
22 Oct-19 2106 2181.6 2169.79 0.97 2076 30 1718.64 30 1.42
23 Nov-19 2256 2198.88 1.03 2288 -32 1688.43 32 1.42
24 Dec-19 2418 2227.96 1.09 2428 -10 1622.25 10 0.41
25 Jan-20 2257.04 2244
26 Feb-20 2286.12 2085
27 Mar-20 2315.21 2215
28 Apr-20 2344.29 2439
29 May-20 2373.37 2587
30 Jun-20 2402.45 2389
31 Jul-20 2431.54 2217
32 Aug-20 2460.62 2354
33 Sep-20 2489.70 2591
34 Oct-20 2518.78 2745
35 Nov-20 2547.87 2533
36 Dec-20 2576.95 2350

15. Find MAPE (Mean Absolute Percentage Error) by calculating the

average of %error and using formula =AVERAGE($first
error$cells:error cells for t period). Drag down the formula.

16. Find MAD (Mean Absolute Deviation) by calculating the average of

ABS error and using formula =AVERAGE($ABS$firstcell:error
column for t period). Drag down the formula.

17. Calculate bias column with cumulative summation from error column
until December 2019, using formula =SUM($firsterror$cells:error
column for t period). Drag down the formula.

Industrial Engineering Laboratory

BINUS University 15 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

18. Calculate TS (Tracking Signal) by dividing bias with MAD, from

January 2018 until December 2019 by using = (bias column/ MAD
column).

19. The result for TSD Method:

Table 1.8 TSD Result
No Period Sales CMA dt bar Seasonal Forecast Error MSE ABS Error % Error MAPE MAD Bias TS
1 Jan-18 1383 1559.06 0.89 1422 -39 1521 39 2.82 2.82 39 -39 -1.00
2 Feb-18 1443 1588.14 0.91 1520 -77 3725 77 5.34 4.08 58.00 -116 -2.00
3 Mar-18 1728 1605.2 1617.23 1.07 1683 45 3158.33 45 2.60 3.59 53.67 -71 -1.32
4 Apr-18 1849 1640.4 1646.31 1.12 1794 55 3125 55 2.97 3.43 54.00 -16 -0.30
5 May-18 1623 1690.6 1675.39 0.97 1666 -43 2869.8 43 2.65 3.28 51.80 -59 -1.14
6 Jun-18 1559 1716.6 1704.47 0.91 1554 5 2395.67 5 0.32 2.78 44.00 -54 -1.23
7 Jul-18 1694 1736.8 1733.56 0.98 1659 35 2228.43 35 2.07 2.68 42.71 -19 -0.44
8 Aug-18 1858 1769.2 1762.64 1.05 1834 24 2021.88 24 1.29 2.51 40.38 5 0.12
9 Sep-18 1950 1786.2 1791.72 1.09 1953 -3 1798.22 3 0.15 2.25 36.22 2 0.06
10 Oct-18 1785 1816.4 1820.80 0.98 1811 -26 1686 26 1.46 2.17 35.20 -24 -0.68
11 Nov-18 1644 1836.4 1849.89 0.89 1687 -43 1700.82 43 2.62 2.21 35.91 -67 -1.87
12 Dec-18 1845 1866.8 1878.97 0.98 1798 47 1743.17 47 2.55 2.24 36.83 -20 -0.54
13 Jan-19 1958 1915.6 1908.05 1.03 1986 -28 1669.38 28 1.43 2.17 36.15 -48 -1.33
14 Feb-19 2102 1954.6 1937.13 1.09 2111 -9 1555.93 9 0.43 2.05 34.21 -57 -1.67
15 Mar-19 2029 1968 1966.22 1.03 1955 74 1817.27 74 3.65 2.16 36.87 17 0.46
16 Apr-19 1839 1998.2 1995.30 0.92 1820 19 1726.25 19 1.03 2.09 35.75 36 1.01
17 May-19 1912 2022 2024.38 0.94 1937 -25 1661.47 25 1.31 2.04 35.12 11 0.31
18 Jun-19 2109 2036.6 2053.46 1.03 2137 -28 1612.72 28 1.33 2.00 34.72 -17 -0.49
19 Jul-19 2221 2074 2082.55 1.07 2270 -49 1654.21 49 2.21 2.01 35.47 -66 -1.86
20 Aug-19 2102 2112.8 2111.63 1.00 2100 2 1571.7 2 0.10 1.92 33.80 -64 -1.89
21 Sep-19 2026 2142.2 2140.71 0.95 1952 74 1757.62 74 3.65 2.00 35.71 10 0.28
22 Oct-19 2106 2181.6 2169.79 0.97 2076 30 1718.64 30 1.42 1.97 35.45 40 1.13
23 Nov-19 2256 2198.88 1.03 2288 -32 1688.43 32 1.42 1.95 35.30 8 0.23
24 Dec-19 2418 2227.96 1.09 2428 -10 1622.25 10 0.41 1.88 34.25 -2 -0.06
25 Jan-20 2257.04 2244
26 Feb-20 2286.12 2085
27 Mar-20 2315.21 2215
28 Apr-20 2344.29 2439
29 May-20 2373.37 2587
30 Jun-20 2402.45 2389
31 Jul-20 2431.54 2217
32 Aug-20 2460.62 2354
33 Sep-20 2489.70 2591
34 Oct-20 2518.78 2745
35 Nov-20 2547.87 2533
36 Dec-20 2576.95 2350

Industrial Engineering Laboratory

BINUS University 16 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

C. Procedure for Moving Average:

1. Using Microsoft Excel, copy all sales data of cotton shirt from January
2018 to December 2019 and sum with 110 to all sales data.
Table 1.9 MA Sales Data
Sales NIM Sales + NIM
1273 110 1383
1333 110 1443
1618 110 1728
1739 110 1849
1513 110 1623
1449 110 1559
1584 110 1694
1748 110 1858
1840 110 1950
1675 110 1785
1534 110 1644
1735 110 1845
1848 110 1958
1992 110 2102
1919 110 2029
1729 110 1839
1802 110 1912
1999 110 2109
2111 110 2221
1992 110 2102
1916 110 2026
1996 110 2106
2146 110 2256
2308 110 2418

2. Create a table with the name of column is number, period, sales, forecast,
error, MSE, ABS error, % error, MAPE, MAD, bias, and TS. We will
forecast for the next 12 Months from January 2020 until December 2020,
so input the period column from January 2018 until December 2019.

3. Copy the sales result from step number 1 to sales column in worksheet
from step number 2.

4. Calculate the forecast value from the fifth period (May 2018) until the
last period (December 2019) by summing up previous period sales and
divide them with n, where n is the number of periods used. For example,
we can use the formula =ROUNDUP(((sales value at t period+sales
value at t+1 period+sales value at t+2 period+sales value at t+3
period)/4),0). Another simpler formula is
=ROUNDUP(AVERAGE(previous sales:previous sales at t-1);0)

Industrial Engineering Laboratory

BINUS University 17 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

5. Calculate the error value between sales data history and forecast value
from May 2018 until December 2019 with the formula: =sales cell-
forecast cell

6. Calculate MSE (Mean Square Error) by using the formula

=SUMSQ($firsterror$cell : error cell)/(period-n), where n is the
number of period used in moving average.

7. Find the absolute value of error by using formula =ABS(error cells).

8. Calculate %error by dividing ABS error with sales data, then multiply
with 100.

9. Find MAPE (Mean Absolute Percentage Error) by calculating the

average of %error. Use formula =AVERAGE($first %error
$cell:%error cell)

10. Find MAD (Mean Absolute Deviation) by calculating the average of

ABS error, use formula =AVERAGE($first ABS error $cell:ABS
error cell).

11. Calculate bias column by doing cumulative summation from the error
column. Use Formula =SUM($first error$cell:error cell).

12. Calculate TS (Tracking Signal) by dividing bias cells with MAD cells.

Industrial Engineering Laboratory

BINUS University 18 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

13. The result for Moving Average Method:

Table 1.10 MA Result
No Period Sales Forecast MA2 Error MSE ABS Error %Error MAPE MAD Bias TS
1 Jan-18 1383
2 Feb-18 1443
3 Mar-18 1728 1413 315 99225 315 18.23 18.23 315 315 1.00
4 Apr-18 1849 1586 263 84197 263 14.22 16.23 289 578 2.00
5 May-18 1623 1789 -166 65316.667 166 10.23 14.23 248 412 1.66
6 Jun-18 1559 1736 -177 56819.75 177 11.35 13.51 230.25 235 1.02
7 Jul-18 1694 1591 103 47577.6 103 6.08 12.02 204.8 338 1.65
8 Aug-18 1858 1627 231 48541.5 231 12.43 12.09 209.167 569 2.72
9 Sep-18 1950 1776 174 45932.143 174 8.92 11.64 204.143 743 3.64
10 Oct-18 1785 1904 -119 41960.75 119 6.67 11.02 193.5 624 3.22
11 Nov-18 1644 1868 -224 42873.556 224 13.63 11.31 196.889 400 2.03
12 Dec-18 1845 1715 130 40276.2 130 7.05 10.88 190.2 530 2.79
13 Jan-19 1958 1745 213 40739.182 213 10.88 10.88 192.273 743 3.86
14 Feb-19 2102 1902 200 40677.583 200 9.51 10.77 192.917 943 4.89
15 Mar-19 2029 2030 -1 37548.615 1 0.05 9.94 178.154 942 5.29
16 Apr-19 1839 2066 -227 38547.214 227 12.34 10.11 181.643 715 3.94
17 May-19 1912 1934 -22 36009.667 22 1.15 9.52 171 693 4.05
18 Jun-19 2109 1876 233 37152.125 233 11.05 9.61 174.875 926 5.30
19 Jul-19 2221 2011 210 37560.824 210 9.46 9.60 176.941 1136 6.42
20 Aug-19 2102 2165 -63 35694.611 63 3.00 9.24 170.611 1073 6.29
21 Sep-19 2026 2162 -136 34789.421 136 6.71 9.10 168.789 937 5.55
22 Oct-19 2106 2064 42 33138.15 42 1.99 8.75 162.45 979 6.03
23 Nov-19 2256 2066 190 33279.19 190 8.42 8.73 163.762 1169 7.14
24 Dec-19 2418 2181 237 34319.636 237 9.80 8.78 167.091 1406 8.41
25 Jan-20 2337
26 Feb-20
27 Mar-20
28 Apr-20
29 May-20
30 Jun-20
31 Jul-20
32 Aug-20
33 Sep-20
34 Oct-20
35 Nov-20
36 Dec-20

Note: We can see the value of the last tracking signal is above ±4,
meaning the forecast result may be biased or not accurate. Based
from known theory, since the history sales plotting has a fixed
seasonality and trend pattern, a smaller moving average period
maybe more accurate because of its responsiveness. If we use a
moving average with 2 period, the result of the last tracking
signal value is -8.41, still not accurate but it is actually better
than the four-period moving average. Know the pattern before
using a forecast method. You can even try to calculate two-
period moving average to hone your skills at excel.

Industrial Engineering Laboratory

BINUS University 19 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

D. Procedure for Holt’s:

1. Using Microsoft Excel, copy all sales data of cotton shirt from January
2018 to December 2019 and sum with 110 to all sales data.
Table 1.11 Holt Sales Data
Sales NIM Sales + NIM
1273 110 1383
1333 110 1443
1618 110 1728
1739 110 1849
1513 110 1623
1449 110 1559
1584 110 1694
1748 110 1858
1840 110 1950
1675 110 1785
1534 110 1644
1735 110 1845
1848 110 1958
1992 110 2102
1919 110 2029
1729 110 1839
1802 110 1912
1999 110 2109
2111 110 2221
1992 110 2102
1916 110 2026
1996 110 2106
2146 110 2256
2308 110 2418

2. Create a table with the name of columns are number, period, sales, level,
trend, forecast, error, MSE, ABS error, % error, MAPE, MAD, bias, and
TS. We’ll try to forecast for the next 12 Months from January 2020 until
December 2020. Input the number and period columns from January
2018 until December 2019.

3. Copy the sales result from table 1.11 to sales column in the worksheet.

4. Calculate level and trend at t=0 (0 period). Use formula

=INTERCEPT(Known y, Known x) for calculating level’s value and
use =SLOPE(Known y, Known x), for calculating trend’s value.
Known y is sales value and known x is the number of periods.
Table 1.12 Holt Level and Trend
Level Trend
1,491.98 32.10

Industrial Engineering Laboratory

BINUS University 20 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

5. After finding level’s and trend’s value for t=0, find their value for the
next period until the last sales data history. Calculate the level value by
using the formula =(α x period t sales) + ((1-α) x (previous period of
level + previous period of trend)). Next, calculate the trend value by
using formula = β x (period t level - the previous period level) + ((1-β)
x previous period of trend). Note that α and β values must be locked
when doing calculation.

6. Calculate the forecast value for period 1-24 by using formula

=ROUNDUP(Levelt-1 + Trendt-1,0) then for period 25-36, use formula
=ROUNDUP(levelt=24 + (trendt=24 x (t=25 – t=24)),0). Note that
levelt=24, trendt=24, and t=24 must be locked when doing calculation. Drag
down the formula.

7. Calculate the error value between sales data history and forecast value
from January 2016 until December 2017 with formula = sales-forecast.

8. Calculate MSE (Mean Square Error) with using formula

=SUMSQ($firsterror$cell: error cell)/number period.

9. Find absolute error from error column. Use formula =ABS(error cell).

10. Calculate %error by dividing ABS error with sales data, then multiply
with 100.

11. Find MAPE (Mean Absolute Percentage Error) by averaging %error

column, use formula =AVERAGE($first %error $cell:%error cell).

12. Find MAD (Mean Absolute Deviation) by averaging error column, using
formula =AVERAGE($first ABS error$cell:ABS error cell).

13. Calculate bias column by doing cumulative summation from the error
column. Use Formula =SUM($first error$cell:error cell).

14. Calculate TS (Tracking Signal) by dividing bias with MAD, from

January 2018 until December 2019 by using formula =(bias column/
MAD column).

Industrial Engineering Laboratory

BINUS University 21 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

15. The result of Holts 1 method:

Table 1.13 Holts Result

No Period Sales Level Trend Forecast Error MSE ABS Error % Error MAPE MAD Bias TS
0 1,491.98 32.10
1 Jan-18 1383 1,509.98 30.69 1,525 -142 20,164 142 10.27 10.27 142 -142 -1.00
2 Feb-18 1443 1,530.90 29.72 1,541 -98 14,884 98 6.79 8.53 120 -240 -2.00
3 Mar-18 1728 1,577.36 31.39 1,561 167 19,219.00 167 9.66 8.91 135.67 -73 -0.54
4 Apr-18 1849 1,632.78 33.79 1,609 240 28,814.25 240 12.98 9.93 161.75 167 1.03
5 May-18 1623 1,662.21 33.36 1,667 -44 23,438.60 44 2.71 8.48 138.20 123 0.89
6 Jun-18 1559 1,681.91 31.99 1,696 -137 22,660.33 137 8.79 8.53 138.00 -14 -0.10
7 Jul-18 1694 1,711.91 31.79 1,714 -20 19,480.29 20 1.18 7.48 121.14 -34 -0.28
8 Aug-18 1858 1,755.14 32.94 1,744 114 18,669.75 114 6.14 7.31 120.25 80 0.67
9 Sep-18 1950 1,804.27 34.56 1,789 161 19,475.44 161 8.26 7.42 124.78 241 1.93
10 Oct-18 1785 1,833.44 34.02 1,839 -54 17,819.50 54 3.03 6.98 117.70 187 1.59
11 Nov-18 1644 1,845.11 31.78 1,868 -224 20,761.00 224 13.63 7.58 127.36 -37 -0.29
12 Dec-18 1845 1,873.70 31.46 1,877 -32 19,116.25 32 1.73 7.10 119.42 -69 -0.58
13 Jan-19 1958 1,910.45 31.99 1,906 52 17,853.77 52 2.66 6.76 114.23 -17 -0.15
14 Feb-19 2102 1,958.40 33.59 1,943 159 18,384.29 159 7.56 6.81 117.43 142 1.21
15 Mar-19 2029 1,995.69 33.96 1,992 37 17,249.93 37 1.82 6.48 112.07 179 1.60
16 Apr-19 1839 2,010.58 32.05 2,030 -191 18,451.88 191 10.39 6.72 117.00 -12 -0.10
17 May-19 1912 2,029.57 30.74 2,043 -131 18,375.94 131 6.85 6.73 117.82 -143 -1.21
18 Jun-19 2109 2,065.18 31.23 2,061 48 17,483.06 48 2.28 6.48 113.94 -95 -0.83
19 Jul-19 2221 2,108.87 32.48 2,097 124 17,372.16 124 5.58 6.44 114.47 29 0.25
20 Aug-19 2102 2,137.41 32.08 2,142 -40 16,583.55 40 1.90 6.21 110.75 -11 -0.10
21 Sep-19 2026 2,155.15 30.65 2,170 -144 16,781.29 144 7.11 6.25 112.33 -155 -1.38
22 Oct-19 2106 2,177.82 29.85 2,186 -80 16,309.41 80 3.80 6.14 110.86 -235 -2.12
23 Nov-19 2256 2,212.50 30.33 2,208 48 15,700.48 48 2.13 5.97 108.13 -187 -1.73
24 Dec-19 2418 2,260.35 32.09 2,243 175 16,322.33 175 7.24 6.02 110.92 -12 -0.11
25 Jan-20 2,293
26 Feb-20 2,325
27 Mar-20 2,357
28 Apr-20 2,389
29 May-20 2,421
30 Jun-20 2,453
31 Jul-20 2,485
32 Aug-20 2,518
33 Sep-20 2,550
34 Oct-20 2,582
35 Nov-20 2,614
36 Dec-20 2,646

Note: In this example, the α and β value used are 0.1 and 0.1,
respectively.

Industrial Engineering Laboratory

BINUS University 22 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

E. Procedure for Winters:

1. Using Microsoft Excel, copy all sales data of cotton shirt from January
2018 to December 2019 and sum with 110 to all sales data.

2. Create table with the name of columns are number, period, sales, level,
trend, seasonal, forecast, error, MSE, ABS error, % error, MAPE, MAD,
bias, and TS. We will try to forecast for the next 12 Months from January
2020 until December 2020. Input the period column from January 2018
until December 2019.

3. Determine the alpha (α), beta (β), and gamma (γ) value to be used, then
copy S1-S5 value from the TSD method. In this example, we’ll use a 0.5
value for all α, β, and γ.
Table 1.14 Alpha, Beta, Gamma, and S Value
S1 0.911701718
S2 0.956559401
S3 1.040359853
S4 1.089669416
S5 0.994109168

4. Calculate level at t=0 using formula =INTERCEPT(Known y, Known

x) and calculate trend at t=0 using formula =SLOPE(Known y, Known
x), where known y is sales value from t=1 until t=24 and known x is the
time period from 1-24.

5. Copy the S1-S5 value to seasonal column for the first period until the
fifth period. Use 2 decimals after comma.
Table 1.15 S Value
0.91
0.96
1.04
1.09
0.99

6. After calculating the level and trend value for t=0, now calculate their
value for the next period until the last sales data history. Calculate the
level value first by using the formula =(α x (sales/seasonal)) + ((1-α) x
(previous period of level + previous period of trend)). Next, calculate
the trend value using formula = (β x (level at period t - the previous
period level)) + ((1-β) x previous period of trend). Note that alpha and
beta values must be locked when doing calculation.

7. Calculate seasonal value for the sixth period until 25th period by using
formula =(γ x (sales 1st period/level at 1st period)) + ((1-γ) x S1) then
drag the result until 25th period.

8. Calculate forecast value by using formula = ROUNDUP((level at t-1

period + trend at t-1 period) x Seasonal,0) for 1st -24th period.

Industrial Engineering Laboratory

BINUS University 23 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

9. Find forecast value at 25th – 36th period using formula

=ROUNDUP((level at last sales data + trend at last sales data*(t=25
– t=24))*seasonal value,0). For the 25th period, use the last seasonal
value (at period 25) and for 26th period forward, use S1-S5 for every 5
period. Note that level value at last sales data, trend at the last sales data,
and t=24 must be locked when doing calculation.

10. Calculate the error value between sales data history and forecast value
from January 2018 until December 2018 with formula = sales-forecast.

11. Calculate MSE (Mean Square Error) by using formula =SUMSQ($first

error$cell: error cell)/number period.

12. Find absolute error from error column. Use formula =ABS(error cell).

13. Calculate %error by dividing ABS error with sales data, then multiply
with 100.

14. Find MAPE (Mean Absolute Percentage Error) by averaging the %error,
using formula =AVERAGE($first %error$cell :error cell at t period).

15. Find MAD (Mean Absolute Deviation) by averaging the ABS error,
using formula =AVERAGE($first ABS error $cell:error cell at t
period).

16. Calculate bias column with cumulative summation from error column
until December 2019, by using formula =SUM($first error $cell:error
cell at t period).

17. Calculate TS (Tracking Signal) by dividing bias with MAD, from

January 2018 until December 2019. Use formula = (bias column/ MAD
column).

Industrial Engineering Laboratory

BINUS University 24 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

18. The result of Winters Method:

Table 1.16 Winters Result
No Period Sales Level Trend Seasonal Forecast Error MSE ABS Error % Error MAPE MAD Bias TS
1491.98 32.10
1 Jan-18 1383 1520.52 30.32 0.91 1390 -7 49 7 0.51 0.51 7 -7 -1
2 Feb-18 1443 1529.68 19.74 0.96 1484 -41 865.00 41 2.84 1.67 24.00 -48 -2.00
3 Mar-18 1728 1605.19 47.63 1.04 1612 116 5062.00 116 6.71 3.35 54.67 68 1.24
4 Apr-18 1849 1674.83 58.63 1.09 1802 47 4348.75 47 2.54 3.15 52.75 115 2.18
5 May-18 1623 1683.04 33.42 0.99 1724 -101 5519.20 101 6.22 3.77 62.4 14 0.22
6 Jun-18 1559 1714.23 32.31 0.91 1564 -5 4603.5 5 0.32 3.19 52.83 9 0.17
7 Jul-18 1694 1764.90 41.49 0.95 1660 34 4111.00 34 2.01 3.02 50.14 43 0.86
8 Aug-18 1858 1780.90 28.75 1.06 1912 -54 3961.63 54 2.91 3.01 50.63 -11 -0.22
9 Sep-18 1950 1793.75 20.80 1.10 1985 -35 3657.56 35 1.79 2.87 48.89 -46 -0.94
10 Oct-18 1785 1818.72 22.88 0.98 1777 8 3298.20 8 0.45 2.63 44.80 -38 -0.85
11 Nov-18 1644 1824.06 14.11 0.91 1676 -32 3091.45 32 1.95 2.57 43.64 -70 -1.60
12 Dec-18 1845 1885.17 37.61 0.95 1756 89 3493.92 89 4.82 2.76 47.42 19 0.40
13 Jan-19 1958 1893.01 22.72 1.05 2021 -63 3530.46 63 3.22 2.79 48.62 -44 -0.91
14 Feb-19 2102 1920.35 25.03 1.09 2092 10 3285.43 10 0.48 2.63 45.86 -34 -0.74
15 Mar-19 2029 2007.54 56.11 0.98 1908 121 4042.47 121 5.96 2.85 50.87 87 1.71
16 Apr-19 1839 2047.10 47.84 0.91 1869 -30 3846.06 30 1.63 2.77 49.56 57 1.15
17 May-19 1912 2036.31 18.52 0.97 2026 -114 4384.29 114 5.96 2.96 53.35 -57 -1.07
18 Jun-19 2109 2038.83 10.52 1.04 2143 -34 4204.94 34 1.61 2.89 52.28 -91 -1.74
19 Jul-19 2221 2040.43 6.06 1.09 2241 -20 4004.68 20 0.90 2.78 50.58 -111 -2.19
20 Aug-19 2102 2078.98 22.30 1.00 2038 64 4009.25 64 3.04 2.79 51.25 -47 -0.92
21 Sep-19 2026 2173.70 58.51 0.90 1896 130 4623.10 130 6.42 2.97 55.00 83 1.51
22 Oct-19 2106 2221.18 53.00 0.95 2128 -22 4434.95 22 1.04 2.88 53.50 61 1.14
23 Nov-19 2256 2223.27 27.54 1.04 2362 -106 4730.65 106 4.70 2.96 55.78 -45 -0.81
24 Dec-19 2418 2233.67 18.97 1.09 2456 -38 4593.71 38 1.57 2.90 55.04 -83 -1.51
25 Jan-20 1.00 2261
26 Feb-20 2050
27 Mar-20 2183
28 Apr-20 2399
29 May-20 2541
30 Jun-20 2356
31 Jul-20 2135
32 Aug-20 2274
33 Sep-20 2498
34 Oct-20 2644
35 Nov-20 2451
36 Dec-20 2221

Industrial Engineering Laboratory

BINUS University 25 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

F. Disaggregate Forecasting:
Disaggregate forecast is used to find forecasting for component that are
divided to find the proportion of each component.

Figure 1.5 Product Structure

1. Example below is the demand for each period of each coloured pen which
will be used for disaggregate forecast.

Table 1.17 Demand Each Components

Demand
Pen
Jan-18 Feb-18 Mar-18 Apr-18 May-18 Jun-18 Jul-18 Aug-18 Sep-18 Oct-18 Nov-18 Dec-18
Blue Pen 452,130 402,840 497,631 511,963 601,230 552,756 613,215 655,813 634,515 619,005 671,556 716,638
Black Pen 227,240 204,350 248,982 271,325 276,940 247,600 294,232 319,275 291,415 273,025 306,957 339,550
Red Pen 237,174 214,410 259,735 271,605 286,874 257,660 304,985 319,555 301,349 283,085 317,710 329,610

2. Sum up the demand of each coloured pen per period by using formula =
SUM(Demand of Blue Pen at period t=1:Demand of Red Pen at
period t=1). Alternatively, the formula =(Demand of blue pen at period
t=1 + Demand of Black Pen at period t=1 + Demand of Red Pen at
period t=1) can be used.

Table 1.18 Total Demand

Demand
Jan-18 Feb-18 Mar-18 Apr-18 May-18 Jun-18 Jul-18 Aug-18 Sep-18 Oct-18 Nov-18 Dec-18
Total Sales 916,544 821,600 1,006,348 1,054,893 1,165,044 1,058,016 1,212,432 1,294,643 1,227,279 1,175,115 1,296,223 1,385,798

3. Before calculating the disaggregate forecasting, calculate first the

forecast of summed up demand from table 1.18. Choose the forecasting
method that has the smallest margin of error. Tabel 1.19 shows the
forecast result by using Holt’s method with α and β values are 0.1.
Table 1.19 Forecast Result
Demand Forecast
Pen
Jan-19 Feb-19 Mar-19 Apr-19 May-19 Jun-19 Jul-19 Aug-19 Sep-19 Oct-19 Nov-19 Dec-19
Total Sales 1,408,119 1,450,079 1,492,038 1,533,998 1,575,957 1,617,917 1,659,876 1,701,836 1,743,795 1,785,755 1,827,714 1,869,674

Industrial Engineering Laboratory

BINUS University 26 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

4. Calculate proportion (in percentage) for each coloured pen of every

period by using the formula =(demand of a disaggregate item/total
demand of the corresponding item)*100. For example:

The proportion of black pen at period Jan-18:

Proportion = 452,130 / (452,130 + 227,240 + 237,74)
Proportion = 452,130 / 916,544
Proportion = 0.49 = 49%

Table 1.20 Proportion Each Components

Proportion (%)
Pen
Jan-18 Feb-18 Mar-18 Apr-18 May-18 Jun-18 Jul-18 Aug-18 Sep-18 Oct-18 Nov-18 Dec-18
Blue Pen 49% 49% 49% 49% 52% 52% 51% 51% 52% 53% 52% 52%
Black Pen 25% 25% 25% 26% 24% 23% 24% 25% 24% 23% 24% 25%
Red Pen 26% 26% 26% 26% 25% 24% 25% 25% 25% 24% 25% 24%

5. Calculate the disaggregate demand forecast for each coloured pen by

using formula = ROUNDUP(proportion of a disaggregate item at
respective period *forecast for the next respective period,0).

Table 1.21 Demand Forecast

Demand Forecast
Pen
Jan-19 Feb-19 Mar-19 Apr-19 May-19 Jun-19 Jul-19 Aug-19 Sep-19 Oct-19 Nov-19 Dec-19
Blue Pen 694,624 710,991 737,801 744,484 813,285 845,274 839,520 862,081 901,559 940,667 946,915 966,865
Black Pen 349,117 360,667 369,148 394,554 374,618 378,630 402,818 419,694 414,061 414,901 432,819 458,110
Red Pen 364,379 378,422 385,090 394,961 388,055 394,014 417,539 420,062 428,176 430,189 447,981 444,700

Industrial Engineering Laboratory

BINUS University 27 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

G. Forecasting Using Minitab:

Minitab is a statistical software, where one of its function is to help
forecast an amount of data. Time series, moving average, double exponential
smoothing, and triple exponential smoothing are supported in Minitab. In this
example, we will try to forecast with Double Exponential Smoothing method
and check the similarity between our calculation from excel and from
Minitab.

Steps:
1. Using Minitab, copy all sales data of cotton shirt from January 2018 to
December 2019 and sum with 110 to all sales data
2. At the ribbon bar, select Stat > Time Series > Double Exponential
Smoothing.
3. In Variable, double click the column which hold the data that will be
forecasted.
4. In Weights to Use in Smoothing, select specified weights. Enter 0.1 for
level (α) and 0.1 for trend (β). If Optimal ARIMA is selected, Minitab
will use the weight that minimizes the sum of squared residuals in an
ARIMA (0,2,2) model. ARIMA stands for AutoRegressive Integrated
Moving Average. Never get too dependent at Optimal ARIMA function,
because sometimes our own specified weights can return smaller error
margins than the Optimal ARIMA itself.
5. Check Generate forecasts. In Number of forecasts, enter 12. Where 12
represent the number of forecast for the next 12 periods.
6. Click OK.
7. The result of Forecast by using Minitab:

Figure 1.6 Forecast Result

Note that the calculations from Excel does not differ to far from the
Minitab results. If there are any difference however, it is usually caused by
the value in decimals and their multiplicative characteristics.

Industrial Engineering Laboratory

BINUS University 28 Revision:
PRACTICUM MODULE
ISYE6101 – PRODUCTION & OPERATION ANALYSIS

Figure 1.7 Forecast Pattern

Industrial Engineering Laboratory

BINUS University 29 Revision: