01 Econ 326 Fundamentals
01 Econ 326 Fundamentals
01 Econ 326 Fundamentals
Jonathan Graves
12/15/22
Models in Economics
What do we mean when we talk about “models” in economics? To get us started with this
question, let’s think about what we already know.
A model is only as good as we build it - so how do we build models? And understand them?
1
Figure 1: Title
2
Figure 3: Outline
Figure 4: Data
3
Figure 5: Marbles
4
Figure 6: Research Questions
Figure 7: Models
5
Figure 8: Examples
Figure 9: Correlation
6
Figure 10: Classification
7
Predictive versus Inferential Models
A key concept in econometrics concerns the purpose to which a model is applied: predictive
or inferential. Imagine we have a very general model:
Figure 11: Prediction
8
𝑦 = 𝑓(𝑥|𝛽)
When we estimate a model, we try to figure out what 𝑓 is relying on data about 𝑥 and 𝑦 -
usually by evaluated different choices for 𝛽.
• When we estimate the model, we will come up with some optimal value of 𝛽 - let’s call
it 𝛽.̂
• This also allow us to calculate fitted values of 𝑦 using this value: 𝑦 ̂ = 𝑓(𝑥|𝛽)̂
– We can also figure out the error of the model: 𝑦 − 𝑦 ̂
The key issue we have here is which of these outcomes to do we care about?
Predictive Models
When we use a model to try and figure out 𝑦 ̂ we refer to this as a predictive model.
• In general, most data science applications (e.g. machine learning, AI) are predictive in
nature.
• We are interested in creating a really flexible, accurate model - and don’t care about how
it does that prediction.
Inferential Models
When we use a model to try and figure out 𝛽 ̂ we refer to this as an inferential model.
9
Example
In econometrics, we actually take this idea one step further. Generally, economists are inter-
ested in testing theoretical models: i.e. formal descriptions, from economic theory, of how
an economic situation operates.
• The extent to which an empirical model connects to the underlying economic theory is
called structure
• Models which have very close connection to a model from economic theory are called
structural models
• Models which test an economic theory indirectly are called reduced form models
This is where the terminology “theoretical” versus “empirical” model comes from. Theoretical
models are economic models while an empirical model is an econometric model designed to
test the theory.
Example
Generally, structural models are much more complex and difficult to estimate. So why do we
use them?
This is critically important in economics, because “the past can be a misleading guide to
current affairs”
10
Lucas Critique
In economic, this forms the basis of a famous argument called the Lucas critique - articulated
by Robert Lucas (1976).
Basically, many (all?) economic models involve agents who make decisions, and those decisions
are based on both past outcomes, and their expectations of future outcomes.
Structural models avoid this problem by identifying invariant properties and estimating them,
allowing for this kind of counterfactual estimation.
Example
Suppose we are trying to forecast how much revenue a tax on gasoline will earn the govern-
ment.
Better Example
One final element of models concerns what kinds of questions we want to answer with them.
• One important category of questions are causal questions: that is, questions about cause
and effect
11
Figure 13: Machine Uprising - (c) Zach Weinersmith SMBC 2022
12
Ď Tip
These can normally be phased as “does A cause B?” or “to what extent does A cause
B?”
• Many important economic questions are causal in nature, but not all. Other types of
questions are associative (or correlational) in nature.
It’s important to remember that both kinds of questions are valuable, but we need to use
models that are appropriate for them.
Causal Models
We refer to econometric models that are design to study cause-and-effect questions as causal
models.
• These are nested within a framework which allows us to relate specific parts of the model
(e.g. parameters) with cause-and-effect relationships
• In general, causal models require data with specific properties in order to be used
• One classic example is an experiment: this is a causal model, but it relies on experi-
mental data in order to work.
– This experimental data also needs special properties (e.g. randomization)
Example
Learning Objectives
The key things you should be able to do after this unit are:
• Explain why economists use models, and the difference between a theoretical and empir-
ical model
• Describe some stylized types of models (e.g. correlation, prediction) and their uses
• Explain the difference between inferential and predictive models, and determine which
model is appropriate in different situations
• Explain structure in a model, and what reduced form versus structural models mean
• Understand the Lucas crituqe, and give examples of how it applies, and why structural
models are important in economics
• Explain what a causal model is, and how they are different from other models
• Identify whether a question is causal or associative in nature
13
Figure 14: Body Image Article
14