Introduction to Econometrics
Organisation and Introduction
David Preinerstorfer
Week 1
David Preinerstorfer Introduction to Econometrics Week 1 1 / 38
Organisational details
Organisation of STAT-S-301
David Preinerstorfer Introduction to Econometrics Week 1 2 / 38
Organisational details Contacts and communication
Instructor: David Preinerstorfer
E-mail address: [Link]@[Link]
Office hour: upon appointment.
Teaching assistant: Elise Petit ([Link]@[Link]).
Student teaching assistants: Isaure D’Hennezel, Alexandre
Laasman.
Schedule: [Link]
Course platform: Course material and announcements will be
posted on the Université virtuelle (UV) [Link]
David Preinerstorfer Introduction to Econometrics Week 1 3 / 38
Organisational details Structure of the course
Two parts:
1 Lectures: David Preinerstorfer.
2 Exercise classes: Elise Petit, and student teaching assistants.
Weekly lectures Tuesday 16:00 - 18:00.
See [Link] for the dates and location.
Weekly exercise classes: Information sessions (in groups) next
week; exercise classes start in Week 3.
How to sign up for groups, date + location, etc. will be announced
on UV and per e-mail today (17.09.2019) evening.
Contact for exercise class related questions is Elise Petit.
David Preinerstorfer Introduction to Econometrics Week 1 4 / 38
Organisational details Problem sets
Problem sets
Problem sets are posted on UV (starting in Week 2).
You can mark the problems you solved (and you can present in class)
in UV before the deadline.
The problem set is then discussed in the exercise classes in the
following week.
David Preinerstorfer Introduction to Econometrics Week 1 5 / 38
Organisational details Exercise classes
Exercise classes
You can sign up for one of the exercise classes via UV (info session).
Before the exercise class you mark problems you have solved in UV.
In the exercise classes, for each problem, the instructors will randomly
select a student that has marked that problem to present it on the
blackboard (or computer).
The instructor might ask clarifying questions concerning the solution.
If the student is unable to present the solution, the instructor needs
to delete all problems the student has marked for this week.
David Preinerstorfer Introduction to Econometrics Week 1 6 / 38
Organisational details Grading
Grading is based on (i) a written final exam, and (ii) the performance
in the exercise classes.
Performance in exercise classes = percentage of problems successfully
solved out of all problems discussed in the weekly exercise classes
(“successfully solved” = the student was present in the exercise class,
has marked the problem as solved, and the marking was not erased
by the instructor due to an improper presentation of a solution by the
student).
If this percentage is ≥ 50%, the final grade is determined via the
written final exam (which then carries 70 %) and the evaluation of
the exercise classes (which then carries 30 %).
If the percentage of successfully solved exercises is < 50%, the final
grade is determined exclusively via the written final exam (which then
carries 100 %).
The written exam will be in January 2020, and there will be a written
resit written final exam in August/September 2020.
David Preinerstorfer Introduction to Econometrics Week 1 7 / 38
Organisational details Exam
The exam will consist of about 30 questions (with short answers).
Some of the questions will be single-choice questions.
Some of the questions will require you to do computations/proofs.
Some of the questions will require you to know
definitions/assumptions/estimators etc. I talked about in class.
You will be able to do similar problems in some of the lectures.
David Preinerstorfer Introduction to Econometrics Week 1 8 / 38
Organisational details Topics
The main reference for this course is:
Stock, J. H. and Watson, M. (2003). Introduction to econo-
metrics. Boston: Addison Wesley.
1 Linear regression models (estimation, testing, confidence
intervals, with a particular emphasis on how to interpret results).
2 Model specification and model selection (omitted variables,
selection of variables, specific regressors).
3 Heteroskedasticity and autocorrelation robust inference.
4 Panel data.
5 Regression models for binary outcomes.
6 Endogeneity and instrumental variables.
7 An introduction to time series models and methods.
David Preinerstorfer Introduction to Econometrics Week 1 9 / 38
Outline
Plan for today is to consider the question:
What is Econometrics, and why is it relevant for economists?
In particular:
1 Some questions that require econometric methods.
2 What is causality.
3 Different types of data.
Reference: Stock and Watson, Chapter 1.
David Preinerstorfer Introduction to Econometrics Week 1 10 / 38
Econometrics
WHAT IS ECONOMETRICS?
SOME QUESTIONS REQUIRING ECONOMETRIC METHODS
David Preinerstorfer Introduction to Econometrics Week 1 11 / 38
Questions we try to answer with econometric methods Class size
Q1: Does reducing class size improve school education?
Common sense (?): Reducing class size ⇒ each student gets more
attention ⇒ basic learning is enhanced.
Basic learning = reading, writing, basic mathematics.
Reducing class size costs money: more teachers, more classrooms.
Decision maker needs to balance costs against benefits.
What exactly is the effect of reducing class size on basic learning?
We need data, and methods to extract knowledge from data.
David Preinerstorfer Introduction to Econometrics Week 1 12 / 38
Questions we try to answer with econometric methods Racial discrimination
Q2: Is there racial discrimination in the home loan market?
Many factors might influence the decision if one gets a loan:
amount of the loan,
loan/income ratio,
credit history, ...
But: US lending institutions must not take race into account when
deciding to grant or deny a request for a mortgage:
⇒ Applicants who are identical in all ways but their race should be
equally likely to have their mortgage applications approved.
Otherwise: racial discrimination.
David Preinerstorfer Introduction to Econometrics Week 1 13 / 38
Questions we try to answer with econometric methods Racial discrimination
Q2 ctd
Dataset (1990, Federal Reserve Bank of Boston):
28 % of black and hispanic applicants and 9 % of white applicants
were denied.
Do these data indicate a racial bias in mortgage lending?
We have seen: many variables (other than race) are relevant.
⇒ A direct comparison of percentages above is debatable.
Need to quantify the change in probability of getting a loan
depending on race while “keeping other characteristics constant”.
David Preinerstorfer Introduction to Econometrics Week 1 14 / 38
Questions we try to answer with econometric methods Cigrarette taxes
Q3: Increase cigarette tax by which amount?
Costs of smoking: medical expenses for smokers and nonsmokers via
secondhand cigarette smoke.
Costs are born by all members of a society.
Possible interventions for cutting consumption ...
1 smoke-free public places and workplace.
2 ban advertising.
3 public education and information campaigns.
Another possibility: Increase taxes on cigarettes.
See also the article:
Sandford, A. (2003). Government action to reduce smoking. Respirology,
8, 7-16.
David Preinerstorfer Introduction to Econometrics Week 1 15 / 38
Questions we try to answer with econometric methods Cigrarette taxes
Q3 ctd
Basic Economics: Cigarette prices increase ⇒ consumption
decreases.
But by what amount much should one increase the tax rate?
Need to know price elasticity of demand for cigarettes.
Price elasticity of demand: change in quantity demanded resulting
from 1 % increase in price.
Need to analyze data on cigarette consumption and prices.
David Preinerstorfer Introduction to Econometrics Week 1 16 / 38
Questions we try to answer with econometric methods Prediction
Q4: Prediction of future events:
– How will housing prices develop in Brussels?
– What is the weather going to be like on the weekend?
– Will the football national team win its next game?
Macroeconomics and financial economics:
What will the price inflation rate be next year?
Economists at the Federal Reserve Banks are responsible for keeping
the rate of price inflation under control.
David Preinerstorfer Introduction to Econometrics Week 1 17 / 38
Questions we try to answer with econometric methods Prediction
Q4 ctd
The Federal Open Market Committee (FOMC) meets 8 times a year.
– If the FOMC believes the economy is growing too fast and inflation
is too high, the Committee may increase interest rates (more
precisely the effective federal funds rate) to temper economic activity.
– In the opposite scenario they may set a lower federal funds rate
target to spur greater economic activity.
In any case: to take appropriate actions, they need accurate
predictions concerning the rate of price inflation.
Based on econometric techniques (time series analysis).
David Preinerstorfer Introduction to Econometrics Week 1 18 / 38
Questions we try to answer with econometric methods Prediction
Q4 ctd
Inflation, consumer prices for the United States
Effective Federal Funds Rate
15.0
12.5
10.0
7.5
Percent
5.0
2.5
0.0
-2.5
1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Sources: Board of Governors, World Bank
[Link] [Link]/g/eUJ0
Data: [Link]
Very useful for working with US time series.
David Preinerstorfer Introduction to Econometrics Week 1 19 / 38
Questions we try to answer with econometric methods Summary
From these examples we learn:
Appropriate answers to economic questions often need to be
numerically precise, i.e., quantitative.
Economic theory often provides an idea about the nature of the
answers, a qualitative statement, but the actual effect must be
learned with data.
Qualitative statement: cigarette consumption should decrease
if taxes are increased.
Quantitative statement: what is the price elasticity of demand.
Knowing the latter lets us decide by how much one should in-
crease taxes.
In this particular example theory provides a functional (mono-
tonicity) relationship between the variables of interest.
David Preinerstorfer Introduction to Econometrics Week 1 20 / 38
Questions we try to answer with econometric methods Summary
CAUSALITY
David Preinerstorfer Introduction to Econometrics Week 1 21 / 38
Causality
Causality
Most of the questions above were concerned with “causal” effects:
A specific action (treatment) leads to a specific, measurable
outcome.
Examples: touching a hot item ⇒ getting burned.
Learning causal relationships (rules) from data is a topic that goes
well beyond the content of this course.
We will only cover the most basic/important tools.
Lots of ongoing research in this field!
David Preinerstorfer Introduction to Econometrics Week 1 22 / 38
Causality Fundamental problem
Potential outcome approach
Main problem (in a nutshell): Formally, the effect of a binary
treatment T on an outcome y for unit i can be defined as:
yi1 − yi0 ,
where yi1 and yi0 are the potential outcomes under treatment and no
treatment, respectively.
But: we can only observe either yi1 or yi0 !
The unobserved is called the counterfactual outcome – what would
have happend if we had assigned the unit to the other group.
So-called “fundamental problem of causal inference”.
A very interesting article is: Rubin, D.B. (1974). Estimating causal effects of treatments in randomized and nonrandomized
studies. Journal of Educational Psychology, 86 (5), 688-701.
David Preinerstorfer Introduction to Econometrics Week 1 23 / 38
Causality Average treatment effects
Dealing with this fundamental problem
Statistical approach:
Shift the focus from individual treatment effects – which can not be
observed – to “average treatment effects” in a certain population.
“What is the average causal effect of a treatment in a population.”
To answer such questions one can use randomized controlled
experiments.
David Preinerstorfer Introduction to Econometrics Week 1 24 / 38
Causality RCEs
Randomized controlled experiments
We have a sample of N subjects from a population.
They are assigned to:
– Control group (CG) that receives no treatment.
– Treatment group (TG) that receives treatment.
How should we assign them to the groups?
Randomized assignment of subjects to CG and TG.
Why randomized assignment?
Eliminate systematic relationship between assignment and out-
come.
Outcomes of CG and TG are then compared statistically.
David Preinerstorfer Introduction to Econometrics Week 1 25 / 38
Causality RCEs
Classical example type:
Effect of a certain type of fertilizer on crop yield of a plant.
Treatment group: gets a certain amount a of the fertilizer.
Control group: does not get any fertilizer.
Potential confounders: Sunlight, field aspects, ...
⇒ Which plot is fertilized is randomly assigned.
At the end of the season the crop yield is measured, the average
difference in yield is the effect of the fertilizer on crop yield.
If implemented on a large enough scale ⇒ gives a good estimate of
causal effect of fertilizer on outcome.
David Preinerstorfer Introduction to Econometrics Week 1 26 / 38
Causality RCEs
Back to Q1 – class size and achievement
1980s state of Tennessee financed large randomized controlled
experiment – Project STAR (about 12 Mio dollars)
Thousands of students (kindergarten – grade 3) were randomly
assigned (within schools) to
– small classes (13-17 students) with one teacher.
– regular classes (22-25 students) with one teacher.
– regular classes (22-25 students) with one teacher and a full-time
teachers aide
Teachers were also randomly assigned to the classes they would teach.
Results suggest that smaller classes lead to better learning outcomes.
See also the description at: [Link]
David Preinerstorfer Introduction to Econometrics Week 1 27 / 38
Causality RCEs
Difficulties
Costs
Ethical problems
Duration
Difficult administration (e.g., students switch classes or school)
General problems in experimental settings:
Hawthorne effect.
Generalizability (across “people” and “situations”).
David Preinerstorfer Introduction to Econometrics Week 1 28 / 38
Causality RCEs
Internal and external validity
Most difficult: in an experiment one needs to balance:
– control the situation so that no extraneous variables are influencing
results (internal validity ).
– ensure that the results can be generalized to realistic setups
(external validity ).
David Preinerstorfer Introduction to Econometrics Week 1 29 / 38
Causality Working Definition of Causality
One definition of causality
causal effect = effect as measured in an
“ideal randomized controlled experiment”.
The only systematic reason for differences in outcomes for different
“treatment groups” is the treatment itself.
David Preinerstorfer Introduction to Econometrics Week 1 30 / 38
Causality Non-experimental data
Observational data
Financial, practical and ethical problems ⇒ experimental data in
economics is rare.
Instead one often needs to resort to non-experimental data, i.e.,
so-called observational data.
Surveys (telephone interviews), administrative records, ...
Lots of possible confounders ⇒ extremely difficult (often impossible)
to learn causal relationships from observational data.
David Preinerstorfer Introduction to Econometrics Week 1 31 / 38
Causality Non-experimental data
ctd
“Treatments” are not assigned randomly.
Difficult to distinguish between treatment effect and effect of other
potentially unobserved confounding variables.
Typically one can only learn about relationships (correlations)
between variables of interest;
Sometimes an econometric analysis based on observational data can
convincingly address the effect of confounders, so that it strongly
suggests causal relationships.
David Preinerstorfer Introduction to Econometrics Week 1 32 / 38
Data types
Classes of data types – terminology
Cross sectional data – multiple entities observed at a single time
point.
Time series data – single entity observed at multiple time points.
Panel data (longitudinal data) – multiple entities, each being
observed at multiple time points.
David Preinerstorfer Introduction to Econometrics Week 1 33 / 38
Summary
Summary
– Economic decisions require knowledge of quantitative relationships.
– Typically, causal effects are of interest.
– Ideally, causal effects tested in randomized controlled experiment.
– But: Experimental data is rare in economics.
– Observational data: difficult to draw conclusions concerning causal
effects; many potential confounders to consider.
– Given observational data, econometric methods can help to assess
the plausibility of potential causal effects.
David Preinerstorfer Introduction to Econometrics Week 1 34 / 38
Summary
REGRESSION MODELS
David Preinerstorfer Introduction to Econometrics Week 1 35 / 38
Summary
Main focus of this course
Our focus is on various types of
Regression models.
In a nutshell regression models are statistical models of the form
y = g (x1 , . . . , xk ) + u,
where y is a variable of interest, which is modelled as a function of
explanatory variables x1 , . . . , xk plus an “error term” u.
In this course we mainly focus on the case where g is linear.
Since g is typically unknown – only qualitative properties are provided
by theory, e.g., linearity – it needs to be estimated.
David Preinerstorfer Introduction to Econometrics Week 1 36 / 38
Summary
Some stylistic economic examples:
A consumption function:
C = β1 + β2 I + u,
where C is consumption and I is disposable income.
The variable u represents other variables besides disposable income
that influence consumption.
A wage equation:
log(W ) = β1 + β2 S + β3 T + β4 E + u
where W is the wage rate, S is the education, T is the years on the
current job, and E is the experience in the labor force.
David Preinerstorfer Introduction to Econometrics Week 1 37 / 38
“Homework”
Reading until next week
Homework: revise material from probability and statistics!
In particular, basic properties of the expectation, variance and
covariance of random variables.
For example: revise course material, or go through Chapter 2 of SW.
David Preinerstorfer Introduction to Econometrics Week 1 38 / 38