MANAGEMENT SCIENCE 5.

Choose an alternative (decide)

 A body of knowledge that uses 6. Implement the decision.
quantitative approaches to
7. Evaluate the results.
decision making (Anderson et al.,
2019) An Alternate Classification of the
 Decision making based/using the Decision – Making
scientific method
Structuring the Problem
 A discipline that attempts to aid in
1. Define the Problem
managerial decision making
2. Identify the Alternatives
 Also known as: Operations
3. Determine the Criteria
Research, Decision Science,
Quantitative Methods, Analyzing the Problem
Quantitative Analysis, and
4. Evaluate the Alternatives
Business Analytics (Taylor 2006)
5. Choose an Alternative
Frederick W. Taylor (1856 -1915) 1. Single-criterion decision problems
– problems that have the objective
 Founder of Management Science
of finding the best solution with
 Initiated to use quantitative
respect to one criterion/standard
methods for management
2. Multicriteria decision problems –
Evolution of Management Science problems that involve more than
one criterion
 Started during the World War II
 Thrived in the business industry Analysis Phase of Decision Making
due to:
 Numerous methodological
developments (e.g., simplex Qualitative Analysis
method for solving linear
• based largely on the manager’s
programming problems)
judgment and experience
 A virtual explosion in
computing power • includes the manager’s intuitive
“feel” for the problem
Problem Solving (7 Steps)
• is more of an art than a science
1. Define the problem.
Quantitative Analysis
2.Determine the set of alternative
solutions. • analyst will concentrate on the
quantitative facts or data associated with
3. Determine the criteria for evaluating
the problem
alternatives.
• analyst will develop mathematical
4. Evaluate the alternatives.
expressions that describe the objectives,
constraints, and other relationships that expressions based on key
exist in the problem assumptions, estimates, or statistical
analyses
• analyst will use one or more
 Experimenting with
quantitative methods to make a
models
recommendation
 requires less time
The Role of qualitative and quantitative  is less expensive
analysis  involves less risk
 accurate model = accurate
conclusions and prediction
 Objective Function – a
mathematical expression that
describes the problem’s objective
 Constraints – a set of restrictions
or limitations
Why Quantitative Analysis Approach  Uncontrollable Inputs –
to Decision Making? environmental factors that are not
• The problem is complex. under the control of the decision
maker
• The problem is very important.  Decision Variables – controllable
• The problem is new. inputs; decision alternatives
specified by the decision maker,
• The problem is repetitive. such as the number of units of a
Quantitative Analysis Process product to produce
 Deterministic Model – if all
 procedures for finding the “best” uncontrollable inputs to the model
or optimal solution/s (Anderson et are known and cannot vary
al., 2019)  Stochastic (or Probabilistic)
1. Model Development - Model – if any uncontrollable are
representations of real objects or uncertain and subject to variation;
situations often more difficult to analyze.
 Iconic models - physical  Cost/benefit considerations must
replicas (scalar be made in selecting an
representations) of real objects appropriate mathematical model.
 Analog models - physical in  Less complicated and less precise
form, but do not physically model is more appropriate than a
resemble the object being more complex and accurate one
modeled (due to cost and ease of solution
 Mathematical models - represent real
world problems through a system of
mathematical formulas and
 Flowchart of the Process of  Accuracy cannot be assessed
Transforming Model Inputs to until solutions are generated.
Output  Small test problems having
known or at least expected,
solutions can be used for model
testing and validation.
 Model generates expected
solutions  model should be
used on the full-scale problem
 If inaccuracies or potential
2. Data Preparation shortcomings inherent in the
 not a trivial step; due to the time model are identified, take
required and the possibility of data corrective action such as:
collection errors  Collection of more-
 A model with 50 decision accurate input data
variables and 25 constraints could  Modification of the
have over 1300 data elements model
 fairly large data base is needed to 4. Report Generation
support a mathematical model  managerial report, based on the
 Information systems specialists results of the model, should be
might be needed. prepared
3. Model Solution  should be easily understood by the
 The analyst attempts to identify decision maker
the alternative (the set of decision  should include: the recommended
variable values) that provides the decision; other pertinent
“best” output for the model. information about the results (for
 “best” output is the optimal example, how sensitive the model
solution. solution is to the assumptions and
 If alternative does not satisfy all data used in the model)
the model constraints =
REJECTED; INFEASIBLE MANAGEMENT SCIENCE
 If alternative satisfies all the TECHNIQUES
model constraints = FEASIBLE; 1. Linear programming –
candidate for the “best” solution maximizing the profit minimizing
 Most Common: Trial and Error the cost
Solution 2. Integer linear programming –
 Software packages are available same as linear programming^^
for solving mathematical models: with the additional requirement
Microsoft Excel & LINGO that some or all the decision
 Model Testing and Validation recommendations be integer
values.
3. Network models - specialized aspects of a business operation;
solution procedures for problems provides reliable guidelines for
in transportation system design, decision-making through
information system design, project prediction
scheduling 10. Goal programming – extension of
4. Project scheduling: PERT linear programming for solving
(Program Evaluation and Review multi-criteria decision problems,
Technique) and CPM (Critical usually within the framework of
Path Method) - help managers linear programming.
responsible for planning, 11. Analytic hierarchy process -
scheduling, and controlling multi-criteria decision-making
projects technique that permits the
5. Inventory models – provide means inclusion of subjective factors in
to determine how much to arriving at a recommended
order/order quantity; when to decision.
place an order; to minimize 12. Markov-process models - in
inventory related cost (*under the studying the evolution of certain
assumption that the demand is systems over repeated trials
known, certain, and constant)
Frequently Used Methods
6. Waiting line (or queuing) models -
help managers understand and  Linear programming
make better decisions by finding  Integer programming
the average time a person must  Network models (such as
wait in line to be served; operation transportation and transshipment
of systems involving waiting lines. models)
7. Simulation – systems are  Simulation
replicated with mathematical
TYPES OF FORECASTING
models; employs a computer
MODELS AND SIMPLE LINEAR
program to model the operation
REGRESSION ANALYSIS
and perform simulation
computations; common and Forecasts
popular because it can analyze
 Basic input in decision processes
very complex systems that cannot
because they provide information
be analyzed using other techniques
on future sales, demand, etc.
8. Decision analysis – several
(Stevenson, 2012)
decision-making techniques are
 a prediction of what will happen in
available and an uncertain pattern
the future (Anderson et al, 2016)
of future events (multiple
outcomes)
9. Forecasting methods – techniques
that can be used to predict future
 other variables included in the
model
Types of Forecasting Models
 include regression models and
(Render et al., 2016)
other more complex models
1. Qualitative Models  example: daily sales of bottled
 Based on judgmental or subjective  water might depend on the
factors average temperature, the average
 Lacks historical data humidity
 Relies on expert opinion and  Regression Analysis
personal experiences  two purposes: (1)
 Delphi method – allows to understand the
experts in different places to relationship between two
make forecasts; involves 3 variables, (2) to predict
participants (decision makers the value of one based on
 staff personnel  the other
respondents  applicability is
 Jury of executive opinion – virtually limitless (e.g.,
takes the opinions of a small level of education and
group of high-level managers, income, price of a house
often in combination with and the square footage,
statistical models, and results advertising expenditures
in a group estimate of demand and sales, etc.)
 Sales force composite – each  variable predicted
salesperson estimates what  dependent variable or
sales will be in his or her response variable
region; then reviewed and  the value of the
combined at the district and dependent variable is
national levels to reach an depending on the value of
overall forecast an independent variable
 Consumer market survey – (explanatory variable or
solicits input from customers predictor variable
or potential customers about 3. Time-Series Models
their future purchasing plans;  quantitative forecasting models/
it also improves product techniques that attempt to predict
design and new product the future values of a variable by
planning using only historical data on that
2. Causal Models one variable
 Quantitative forecasting models  these models are extrapolations of
wherein the variable to be forecast past values of that series
is influenced by or correlated with
Simple Linear Regression Model
  Only one independent variable, X response variable will change for
 The relationship between X and Y each one unit increase in the predictor
is described by a linear function variable+
 The changes in Y are related to o positive slope – tilting to the right /
changes in X o negative slope – tilting to the left \
o intercept - meaningful only if the
Population Regression Model predictor variable would reasonably
have a value equal to zero

Prediction Using Regression

- main use of regressions is to make

predictions
- once we have a fitted regression
equation that shows the estimated
relationship between X and Y, any
Simple Linear Regression Equation
value can be plugged in (within the
range of our sample x values) to
obtain the prediction for Y

How to Form a Regression Equation

1. Relationship should be linear (see

scatterplot or scatter diagram or look
at the r or correlation coefficient; -1≤
R ≤ 1)
Simple Linear Regression Equation
(Render et al., 2016)

Formulas to compute slope (b 1) and the

intercept (b 0):

X=
∑ X = Average(mean) of X Values
n
Assumptions of Regression (L.I.N.E)
Y=
∑ Y = Average (mean)of Y Values  Linearity – the relationship between
n
X and Y is linear

b 1=
∑ ( X −X )(Y −Y ¿) ¿  Independence of errors – the error
values (difference between observed
∑ ( X −X )2 and estimated values) are statistically
b 0=Y −b1 X independent
 Normality of error – the error values
Interpreting an Estimated Regression are normally distributed for any given
Equation value of X
 Equal variance or
o slope - tells us how much, and in
homoskedasticity – the probability
what direction, the dependent or
 distribution of the errors has constant
variance

Assessing Fit: Coefficient of

Determination, R2

2 SSR regression ∑ of squares

r= =
SST total ∑ of squares
2
0≤r ≤1

Standard Error of Estimate (can be found

in Excel Data Analysis Table

Time Series and Forecasting: Overview

PAST
Historical Information
Comparing Standard Errors
PRESENT
- SYX is a measure of the variation of
observed Y values from the Analysis of the Past information to
regression line. uncover design or pattern or model.
- The magnitude of SYX should always FUTURE
be judged relative to the size of the Y
values in the sample data Forecasting or predicting possible
emerging event in the future based on the
Inferences about the slope using the t-test
model made at present based on the
- The t-test for a population slope is analyzed past information.
used to determine if there is a linear
relationship between X and Y.
- Null and alternative hypotheses Major Areas of Forecasting
H 0 : β 1 = 0 (no linear relationship)
H 1 : β 1 ≠ 0 (linear relationship does Demand Forecasting
exist)  Predicts timing and quantity of
demand of firm’s commodities
Economic Forecasting
 Predicts future business conditions
  Referencing to the underlying o Exchange rate, interest rate,
economic factors (i.e. inflation, inflation rate, national GDP
GDP, GNP, etc.) o Retail sales
Technology Forecasting o Electric power consumption
o Number of accidental fatalities
 Predicts possible technological
advancements in the future Time Series Patterns (Anderson et al.,
 Example: latest digital devices and 2018)
systems, future software, and  common types of data patterns
hardware, etc. that can be identified when
Uses of Time Series | Rahman (2015) examining a time series plot
include:
 To study the past behavior A. Horizontal Pattern - exists when
variable the data fluctuate randomly around
 To formulate policy decisions and a constant mean over time
planning of future operations  Stationary time series – refers
 To predict or estimate or forecast to a time series whose statistical
the behavior of the phenomenon in properties are independent of
future which is very essential for time
business planning 1. The process generating the data
 To compare the changes in the has a constant mean.
values of different phenomenon at 2. The variability of the time
different times series is constant over time
Time Series Patterns (Anderson et al.,
2016)
 Time series – sequence of
observations on a variable measured
at successive points in time
 The measurements may be taken
every hour, day, week, month, or
year, or at any other regular interval. ^^ time series plot for a stationary
 Pattern of data – aids in the time series will always exhibit a
understanding how the time series has
horizontal pattern with random
behaved in the past
fluctuations
 ^^ if behavior is expected to continue
in the future, it can be used in  Horizontal pattern with a shift –
selecting an appropriate forecasting shifts to a new level at some
method point in time due to changes in
Examples of Time Series business conditions; makes it
more difficult to choose an
o Stock price, Sensex appropriate forecasting method
 rush hour and peak levels at
restaurants; lunch and dinner time
 Time Series with a Seasonal
Pattern

B. Trend Pattern – gradual shifts or

movement to relatively higher or
lower values
 Trends result in these
factors:
1. population increases or D. Trend and Seasonal Pattern –
decreases includes both a trend and seasonal
2. shifting demographic pattern; used be used with a
characteristics of the forecasting method that can deal
population with trend and seasonality
3. Improving technology  Time Series Decomposition –
4. changes in consumer used to separate a time series
preferences into trend and seasonal
 Trends may be steady and components
predictable, increasing, E. Cyclical Pattern - exists if the
decreasing, or staying the time series plot shows an
same (Doane and Seward, alternating sequence of points
2016) below and above the trend line
 Time Series with Linear that lasts for more than a year or
Trend Pattern when the data exhibit rises and
falls that are not of a fixed period
 Cyclical effects are often
combined with long-term trend
effects aka trend cycle effects
SELECTING A FORECASTING
METHOD (Anderson et a, 2016)
 The underlying pattern - an
C. Seasonal Pattern - recognized by
important factor in selecting a
observing recurring patterns;
forecasting method
exists when a series is influenced
 Time series plot – one of the first
by seasonal factors or in regular
analytic tools employed
interval (i.e., daily, monthly,
 OBJ of Forecasting: for the error
quarterly); can also be less than
to be as slight as possible
one year like daily traffic volume:
 Forecast Accuracy – a way to WHERE:
determine if forecasting technique is
n = is the number of periods of the
appropriate
time series
Forecast Error=¿
k = is the number of periods at the
Forecast accuracy=Maximum(0,100 %−Forecast error)
beginning of the time series for which
a forecast cannot be computed
o + forecast error - the forecasting
method underestimated the actual C. Mean Absolute Percentage
value Error (MAPE) – obtained by
o - forecast error - the forecasting getting the average of the absolute
method overestimated the actual value of percentage forecast
value
o forecast bias – the persistent
errors, ( )
et
Yt
100

tendency for forecasts to be

greater or less than the actual
values of a time series WHERE:
Measures of Forecast Accuracy n = is the number of periods of the
A. Mean Absolute Error (MAE) – time series
obtained by getting the average of k = is the number of periods at the
the absolute value of the forecast beginning of the time series for
errors, e i which a forecast cannot be computed
Forecast Error
MAPE= (100 %)
Actual Value

WHERE:
n = number of periods of the time Comparing the Three Measures
series
k = number of periods at the MAE and MSE
beginning of the time series for  measures that avoid the
which a forecast cannot be problem of positive and
computed negative forecasting errors
B. Mean Squared Error (MSE) – offsetting one another
obtained by getting the average of  the size depends upon the
the squared forecast errors, e i scale of the data
 difficult to make comparisons
of different time intervals or
to make comparisons across
different time series
 MAPE has the advantage
 ALL THREE MEASURES
 All three simply measure how THINGS TO REMEMBER:
well the forecasting method can
 forecast accuracy may not assure
forecast historical values of the
of good forecasts.
time series
 forecast accuracy gives an idea of
 the smaller the value of the
how close the forecasted values,
forecast accuracy, the more
predicted by the estimated model,
accurate the forecast
or be alike with the actual values
of the corresponding historical
past data.
Simple Forecasting Methods
 MAD computes absolute
1. Naïve Forecasting Method – difference.
most recent past observation as the  MAPE computes relative to the
forecasted value for the next totality of the data and commonly
period used when comparing cases of
2. Average of All Historical Data – different size or degree/amount.
average of all previous data as the (e.g case 1 in millions; case 2 in
forecasted value for the next tens or units)
period  ** the lower the value of any of
the 3 measures, the better quality
Example: of forecast accuracy
Forecast the gasoline time series using the Moving Averages and Exponential
naïve method. Then, Compute MAE, MSE,
Smoothing
and MAPE. Compare the results.
 Three forecasting methods that are
appropriate for a time series with a
horizontal pattern:
 Moving Averages
 Weighted Moving Averages
 Exponential Smoothing
 Called smoothing methods
because their objective is to
Example 2: Rosco Drugs smooth out random fluctuations in
the time series
Sales of Comfort brand headache medicine  Most appropriate for short-range
for the past 10 weeks at Rosco Drugs are forecasts
shown below.
Moving Averages
 If Rosco uses the naïve forecast
method to forecast sales for weeks 2-
10, what are the resulting MAE,
MSE, and MAPE values?
- Uses the average of the recent k Example: Moving Average
data values in the time series as If Rosco Drugs uses a 3-period moving
the forecast for the next period average to forecast sales, what are the
forecasts for weeks 4-11?
Σ(most recent k data values)
Y^ t +1=
k
- Each observation in the moving
average calculation receives the
same weight
- Moving is used because every time
a new observation becomes
available for the time series, it
replaces the oldest observation in
Weighted Moving Averages
the equation
- This will result to a change in 1. Select the number of data values
average as new observations to be included in the average
become available 2. Choose the weight for each of the
- Kinda like compound average data values
(from compound interest) 2.1 The more recent observations
are typically given more
Application of Moving Averages
weight than older observations
1. Select the order k (number of time 2.2 For convenience, the weights
series values should equal to 1
1.1 Smaller value of k will track 2.3 Example: 3WMA = .2(110)
shifts in a time series more + .3(115) + .5(125) = 119
quickly than a larger value of 2.4 .2 + .3+ .5 = 1; 125 is the most
k recent observation
1.2 If more past observations are
Example: Weighted Moving Average
considered relevant, then a
larger value of k is better
2. Uses several values during recent
past to develop forecasts
3. Useful for forecasting items that
are relatively stable and do not
display any pronounced behavior
(e.g., a trend or a seasonal
 Simple and weighted moving
pattern).
averages are effective in
4. The longer the moving average
smoothing out fluctuations to
period, the smoother the data will
provide stable estimates
be
 Problems of moving averages
 1. Increasing n does smooth  Y^ t = forecast of the time series for
fluctuations better but it makes period t
the method less sensitive to  ∝ = smoothing constant (0 ≤ ∝≤ 1)
real changes  And let Y^ 2=Y 1 (to initiate
2. It cannot pick up trends very computations
well. Since they are averages,
they will always stay within Rewritten Formula for Exponential
past level and will not predict Smoothing Forecast:
a change.
Y^ t +1=∝ Y t + ( 1−∝) Y^ t

Desirable Value for the Smoothing

Comparing Forecasting Techniques Constant
 Based on MAE, 3-period moving  Substantial random variability –
average technique is the most smaller value (nearer to 0) is
accurate preferred
 Based on MSE, 3-period moving  Little random variability – a
average technique is the most larger value for a is preferred
accurate because forecast errors are more
 Based on MAPE, weighted likely to represent a change in the
moving average is the most level of the time series
accurate  Choose a value of ∝ that
minimizes the MSE
Exponential Smoothing
Example: Exponential Smoothing
- Special case of a weighted moving
averages method If exponential smoothing is used on
- Selects only the weight for the gasoline sales time series, which
most recent observation smoothing constant (0.1 or 0.8) gives a
- Weights for other data values are better forecast?
computed automatically and
become smaller as the
observations grow older
- Weighted average of all the
observations in the time series

Y^ t +1=∝ Y t + ( 1−∝) Y^ t

Where:

 Y^ t +1= forecast of the time series

for period t + 1
 Y t = actual value of the time series
in period t
 error between the trend line
forecasts and the actual observed
values of the time series
 independent variable: time period;
dependent variable:
actual/observed values in the time
series
Formula for Trend Projection

Exponential smoothing with ∝=0.8 Y^t =b 0+ b1 t

provided less accurate forecasts than ES
WHERE:
with ∝=0.1
Y^t = linear trend forecast in period t
(forecasted value)
Linear Trend Projection
b 0 = intercept of the linear trend line at t +
Regression 0
 Forecasting technique for time b 1 = slope of the linear trend line
series exhibiting TREND patterns
 Simplest form is linear regression t = period
 Causal method of forecasting Formula for Trend Projection Equation
 A mathematical relationship is
Y^t =b 0+ b1 t
developed between a dependent
variable and some other factor that
affects the variable
 dependent variable  displays an
obvious trend, a least squares
regression line or linear trend line
 can be used to forecast the
dependent variable
Example: Bicycle Sales
Linear Trend Line
Consider the data for bicycle sales
 a linear regression model that
over the past 10 years. The linear trend
relates a dependent variable to
line provides a reasonable approximation
time
of the long-run movement of the series.
Least Squares Use linear trend projection to forecast the
sales for years 11 and 12.
 also used in regression analysis
 determines the unique trend line
forecast or linear trend projection
which minimizes the mean square
Solution: