INTRODUCTION TO MODEL BUILDING
Model building
Models have been developed to advance market knowledge and to aid management decision
making. Marketing models will create generalizable insights in marketing phenomena. Often
mathematical models are used to numerically specify various components and their interrelations. This
form of model allows to quantify the effect of multiple forces, gives a reasonable representation of
a real-world system and gives the opportunity to explore a myriad of actions.
Benefits of model building
Can lead to better decisions
Indirect benefits:
1. Model forces to explain how the market works, leading to an improved understanding.
2. Models may work as problem-finding instruments.
3. Models can be instrumental in process improvement; decision makers deal with existing
information.
4. Models can help decide what information should be collected.
5. Models can guide research.
6. Models often allow management to pinpoint environmental changes faster than is possible
otherwise.
7. Models provide a framework for discussion.
8. Models may result in a beneficial reallocation of management time
The Model Building Process
Step 1: Opportunity Identification: evaluate the need of a model to improve decision making
Step 2: Model Purpose: define the use of the model
Step 3: Model Scope: model building for a specific or broader set of decisions?
Step 4: Data Availability
Step 5: Specification
Specify the variables to be included in the model and make
distinction between IV and DV.
Specify the functional relationship between the variables.
Step 6: Estimation
Determination of parameter estimates for a model.
Identify techniques to be applied for extracting estimates of the
model parameters from the data.
Choice for the technique depends on: kind of data available,
kind of variables, necessary assumptions and computational
effort.
Step 7: Validation
Assessing the quality or the success of the model.
Create criteria and measure the model upon them.
Step 8: Cost-Benefit Considerations: examine the cost-benefit trade-off
Step 9: Use
Step 10: Updating
,
,Model Specification
Specification is an important step in the model building process. The goal of this step is to express the
most important elements of a real-world system in one or more mathematical equations. Outcome is a
formula that summarizes the most important relationships of the phenomenon studied.
Model elements
Which elements to include in a model?
1. Dependent and independent variables single equation
Y = left-hand side of equation dependent / criterion variable
o brand level sales, market shares, profit, price
X = right-hand side of equation independent / predictor variables
o excluding important parameters causes omitted variable bias
o even though the user is interested in the effect of a few factors, all variables should be
included.
o relates to the discussion about model completeness
Each independent variable is associated with a parameter which indicates how strongly the
dependent variable responds to a one-unit change in that independent variable. The associated
parameter depends on the scaling of the associated independent variable (e.g. euros instead of
euro-cents). Right-hand side of the equation consists of a systematic part (e.g. intercept and
independent variables), which indicates how the dependent variable systematically varies with
the variables included in the model, and a stochastic part (e.g. error term).
2. Intercept constant term
The constant term reflects the average value of the dependent variable if all the independent
variables are zero, in practice, that will never happen. As price will not likely to become zero,
it is quite hard to interpret the intercept. If all the variables are mean-centered than the
intercept can be loosely interpreted as the “baseline level of sales”; at least it indicates the
average sales when the independent variables are at their mean level. The interpretation of the
constant term in a linear model depends on the centering and the scaling of the variables
included in the model.
3. Endogenous and exogenous variables system of equations
Effect of independent variables might also be driven by other variables.
System of equations = multiple equations in one formula
In case of a system of equations, you have multiple error terms because each equation has its
own error term. A system of equations allows for multiple dependent / endogenous variables
4. Disturbance term / error term stochastic part
Error-term to pick up whatever is not picked up by IVs, holds quality
Determines many of the statistical properties of the model
Used to accommodate random error, measurement error, missing variables error
Captures the errors that a model builder makes in specifying the model. Why?
1. Random error consumers do not behave rationally
2. Measurement error sample error, poor measurement instrument
3. Omitted variables error due to missing data; taken up by error term
4. Specification error wrong functional form: assume linearity while nonlinear
5. Reflects behaviour differences that are hard to model between people and time
, MODEL CRITERIA (LITTLE, 1970)
The model can be considered to be a “good ” model when the following criteria hold from a user’s
point of view simple, evolutionary, complete, adaptive, robust. The criteria considers the model
structure and ease of use. However, these criteria are focused on the model structure.
Little (1975) adds easy to control and easy to communicate by the other criteria
1. Simple
All models are simplified/stylized/abstract representations of real-world phenomena. One way in
which model simplicity can be achieved by the model builder is by keeping the number of variables
small, and only include important phenomena. A simple model results in manageable and estimable
models. Calls for parsimony of models. There should be a modest number of parameters with simple
structure; which might mean that linear or linearizable models are preferred to non-linearizable ones.
Why?
o Important to explain the model
o Reduce the risk of overfitting (too complex)
o More parameters = more uncertainty
Achieve model simplicity by:
o Not too many explanatory variables keep number of variables small
o Not too many interactions only include important phenomena
o Simple mathematical forms
How?
1. Eliminating less relevant variables
Including all possible variables interferes with the simplicity criterion and possibly
also introduces estimation problems due to multicollinearity or lack of degrees of
freedom. First reduction can be realized by excluding least relevant variables.
Potentially this leads to omitted variable bias, but the bias will be small if the
variables are truly not relevant.
2. Clustering of variables
For example, if a large number of brands assumed to influence the performance of the
brand of interest, they may be aggregated into one variable called “competition”.
Advertising expenditures across various types of media might be aggregated into
“total advertising expenditures”.
3. Introducing relative variables
4. Phasing variables over different levels
5. Constraining parameter values
2. Complete (good source of criticism: useful for EXAM!)
For a model to be a useful decision-support tool, it has to represent all relevant elements of the
problem being studied. May be in conflict with simplicity, there is a trade-off (always strive for
parsimonious)
Why?
o A model should capture most relevant phenomena
o Should represent all relevant element of the problem to be a useful decision-support tool
How?
o Building on existing knowledge
o Include all relevant independent variable (classes with simple)
o Balance completeness and simpleness
o Did you include all instruments / competitors / environment / dynamic / other marketing
variables effects?
Completeness is relative to the problem, to the organization, and to the user.
Model building
Models have been developed to advance market knowledge and to aid management decision
making. Marketing models will create generalizable insights in marketing phenomena. Often
mathematical models are used to numerically specify various components and their interrelations. This
form of model allows to quantify the effect of multiple forces, gives a reasonable representation of
a real-world system and gives the opportunity to explore a myriad of actions.
Benefits of model building
Can lead to better decisions
Indirect benefits:
1. Model forces to explain how the market works, leading to an improved understanding.
2. Models may work as problem-finding instruments.
3. Models can be instrumental in process improvement; decision makers deal with existing
information.
4. Models can help decide what information should be collected.
5. Models can guide research.
6. Models often allow management to pinpoint environmental changes faster than is possible
otherwise.
7. Models provide a framework for discussion.
8. Models may result in a beneficial reallocation of management time
The Model Building Process
Step 1: Opportunity Identification: evaluate the need of a model to improve decision making
Step 2: Model Purpose: define the use of the model
Step 3: Model Scope: model building for a specific or broader set of decisions?
Step 4: Data Availability
Step 5: Specification
Specify the variables to be included in the model and make
distinction between IV and DV.
Specify the functional relationship between the variables.
Step 6: Estimation
Determination of parameter estimates for a model.
Identify techniques to be applied for extracting estimates of the
model parameters from the data.
Choice for the technique depends on: kind of data available,
kind of variables, necessary assumptions and computational
effort.
Step 7: Validation
Assessing the quality or the success of the model.
Create criteria and measure the model upon them.
Step 8: Cost-Benefit Considerations: examine the cost-benefit trade-off
Step 9: Use
Step 10: Updating
,
,Model Specification
Specification is an important step in the model building process. The goal of this step is to express the
most important elements of a real-world system in one or more mathematical equations. Outcome is a
formula that summarizes the most important relationships of the phenomenon studied.
Model elements
Which elements to include in a model?
1. Dependent and independent variables single equation
Y = left-hand side of equation dependent / criterion variable
o brand level sales, market shares, profit, price
X = right-hand side of equation independent / predictor variables
o excluding important parameters causes omitted variable bias
o even though the user is interested in the effect of a few factors, all variables should be
included.
o relates to the discussion about model completeness
Each independent variable is associated with a parameter which indicates how strongly the
dependent variable responds to a one-unit change in that independent variable. The associated
parameter depends on the scaling of the associated independent variable (e.g. euros instead of
euro-cents). Right-hand side of the equation consists of a systematic part (e.g. intercept and
independent variables), which indicates how the dependent variable systematically varies with
the variables included in the model, and a stochastic part (e.g. error term).
2. Intercept constant term
The constant term reflects the average value of the dependent variable if all the independent
variables are zero, in practice, that will never happen. As price will not likely to become zero,
it is quite hard to interpret the intercept. If all the variables are mean-centered than the
intercept can be loosely interpreted as the “baseline level of sales”; at least it indicates the
average sales when the independent variables are at their mean level. The interpretation of the
constant term in a linear model depends on the centering and the scaling of the variables
included in the model.
3. Endogenous and exogenous variables system of equations
Effect of independent variables might also be driven by other variables.
System of equations = multiple equations in one formula
In case of a system of equations, you have multiple error terms because each equation has its
own error term. A system of equations allows for multiple dependent / endogenous variables
4. Disturbance term / error term stochastic part
Error-term to pick up whatever is not picked up by IVs, holds quality
Determines many of the statistical properties of the model
Used to accommodate random error, measurement error, missing variables error
Captures the errors that a model builder makes in specifying the model. Why?
1. Random error consumers do not behave rationally
2. Measurement error sample error, poor measurement instrument
3. Omitted variables error due to missing data; taken up by error term
4. Specification error wrong functional form: assume linearity while nonlinear
5. Reflects behaviour differences that are hard to model between people and time
, MODEL CRITERIA (LITTLE, 1970)
The model can be considered to be a “good ” model when the following criteria hold from a user’s
point of view simple, evolutionary, complete, adaptive, robust. The criteria considers the model
structure and ease of use. However, these criteria are focused on the model structure.
Little (1975) adds easy to control and easy to communicate by the other criteria
1. Simple
All models are simplified/stylized/abstract representations of real-world phenomena. One way in
which model simplicity can be achieved by the model builder is by keeping the number of variables
small, and only include important phenomena. A simple model results in manageable and estimable
models. Calls for parsimony of models. There should be a modest number of parameters with simple
structure; which might mean that linear or linearizable models are preferred to non-linearizable ones.
Why?
o Important to explain the model
o Reduce the risk of overfitting (too complex)
o More parameters = more uncertainty
Achieve model simplicity by:
o Not too many explanatory variables keep number of variables small
o Not too many interactions only include important phenomena
o Simple mathematical forms
How?
1. Eliminating less relevant variables
Including all possible variables interferes with the simplicity criterion and possibly
also introduces estimation problems due to multicollinearity or lack of degrees of
freedom. First reduction can be realized by excluding least relevant variables.
Potentially this leads to omitted variable bias, but the bias will be small if the
variables are truly not relevant.
2. Clustering of variables
For example, if a large number of brands assumed to influence the performance of the
brand of interest, they may be aggregated into one variable called “competition”.
Advertising expenditures across various types of media might be aggregated into
“total advertising expenditures”.
3. Introducing relative variables
4. Phasing variables over different levels
5. Constraining parameter values
2. Complete (good source of criticism: useful for EXAM!)
For a model to be a useful decision-support tool, it has to represent all relevant elements of the
problem being studied. May be in conflict with simplicity, there is a trade-off (always strive for
parsimonious)
Why?
o A model should capture most relevant phenomena
o Should represent all relevant element of the problem to be a useful decision-support tool
How?
o Building on existing knowledge
o Include all relevant independent variable (classes with simple)
o Balance completeness and simpleness
o Did you include all instruments / competitors / environment / dynamic / other marketing
variables effects?
Completeness is relative to the problem, to the organization, and to the user.
