Notes1

ECO341K Introduction to Econometrics

() Lecture 1 1 / 31

Lecture 1 - Outline

Course Outline

Course objectives

Textbook

Assessment

() Lecture 1 2 / 31

Course Objectives

The aim of the course is to help you develop a working knowledge ofeconometrics and its applications to real-world economic data.

The course will cover a range of topics:

− Simple Regression− Multiple Regression− estimation, inference− extensions

− Learn a specialised econometric software package

By the end of session you will be able to:

=⇒ read and understand most analyses performed by econometricians=⇒ conduct your own empirical research.

() Lecture 1 3 / 31

Textbooks:

The required text is:

J.M. Wooldridge (2008) Introductory Econometrics: A ModernApproach. 4th Edition

3rd edition is fine

A useful companion book:

J.M. Wooldridge (2008) Student Solution Manual for IntroductoryEconometrics available electronically through the text website.

() Lecture 1 4 / 31

Assessment

1. Weekly assignments (drop lowest) 10%2. Two Mid-Term Exams 50%3. Final Exam 40%

() Lecture 1 5 / 31

Course Website (Blackboard)

This site will contain:

Lecture handouts (syllabus, etc)

Lecture notes for each class

Homework assignments, including data sets

Homework solutions

Data sets and STATA logs for in class examples

Special announcements (sent via email also)

() Lecture 1 6 / 31

Topics covered – rough timeline

Topic Classes Approx.

Introduction 2

Simple Linear Regression 3

Multiple Linear Regression 5

Small Sample Inference 2

Large Sample Inference 1

Further Issues including Dummies 4

Time Series 2

Panel Data 2

Qualitative Response 2

Endogenous Regressors and IV 2

() Lecture 1 7 / 31

Software – STATA

It is crucial that you have access to STATA and that you do theempirical exercises

Access Options:

1 Timeshare through the web from anywhere2 Purchase through STATA gradplan – see syllabus – about $1003 Labs in Burdine?

What does STATA look like? Lets see!

() Lecture 1 8 / 31

Lecture 2 - Outline

The Nature of Econometrics

What is Econometrics ?

The Structure of Economic Data

Causality ‘Ceteris Paribus’ and correlation

() Lecture 1 9 / 31

What is Econometrics ?

Econometrics concerns the use of statistical methods in:

estimating economic relationships

testing economic theory

evaluating government and business policy.

forecasting and prediction

() Lecture 1 10 / 31

Applications

Econometrics has many wider applications, for example

1 the effect of class size or spending on student performance

2 the effect of education on wages

3 testing for discrimination in labour and credit markets

4 the effect of minimum wages on unemployment

5 the effect of CEO compensation on firm performance

6 the effect of govt policies on inflation and economic growth

Common feature: Econometrics deals with nonexperimental datadrawn from observing economic events (the data are not collectedthrough controlled experiments in a laboratory).

() Lecture 1 11 / 31

Conducting Empirical Economic Analysis

Econometrics is used in every branch of applied economics

Empirical Analysis uses data to test the predictions of a theory orestimate a relationship.

An empirical analysis generally consists of

1 An economic model - which may be formally developed (e.g. derivationof consumer demand equations from a model of utility maximisation)or based on intuitive reasoning.

2 An econometric model - which requires specifying the nature of therelationship between variables

() Lecture 1 12 / 31

Conducting Empirical Economic Analysis

Example of an econometric model – a multiple regression model:wage = β0 + β1.educ +β2.exper + υ

where:wage = hourly wage rateeduc = years of educationexper = years of employmentβ0, β1, β2 = parameters which describe the direction and strength of therelationship between the wage and the factors which determine itυ = error term (or ‘disturbance term’) which contains unobservable factors(innate ability, job characteristics,...)

() Lecture 1 13 / 31

Conducting Empirical Economic Analysis

With this model a range of hypotheses can be stated in terms of theunknown parameters (β0, β1, β2).

Empirical analysis requires data, and econometric methods are used toestimate the parameters of the model and to formally test hypothesesof interest. The model can also be used to make predictions.

Methods need to take into account the structure of the data

4 main data structures: cross-sectional data, time-series data, pooledcross sections, panel data

() Lecture 1 14 / 31

The Structure of Economic Data

A. Cross-Sectional Data

sample of individuals, households, firms, countries or other unitstaken at a point in time (“snapshot”)

usually obtained by random sampling from the population (and thesample is “representative”)

cross-sectional data are widely used in economics and other socialsciences. Very common in applied micro such as labor economics,public economics, industrial organization, health economics

=⇒ this is the main data structure we will focus on

if randomly sampled, order of observations is unimportant

() Lecture 1 15 / 31

Cross-Sectional Data

Examples of a cross-sectional data set:(a) Data set on wages and other personal characteristics

() Lecture 1 16 / 31

Cross-Sectional Data

(b) Data set on economic growth and country characteristics

() Lecture 1 17 / 31

Time Series Data

B. Time Series Data

observations on a variable (or set of variables) over time

E.g. stock prices, cpi, gdp, crime rates.

The chronological ordering of observations is important

=⇒ observations cannot be assumed to be independent over time, mosteconomic time series are (strongly) related to their recent histories

=⇒ econometric model needs to take this into account

Data frequency is important, due to seasonal patterns (e.g. daily,weekly, monthly, quarterly, annual)

() Lecture 1 18 / 31

Time Series Data

Example: data on minimum wages

() Lecture 1 19 / 31

Pooled Cross Sections

C. Pooled Cross Sections

some data sets have both cross-sectional and time series properties.

E.g. 2 cross-sectional family surveys in US

- one in 2000 recording income, expenditure, family size,...- a new random sample in 2005, with same questions=⇒ pool them to increase sample size=⇒ no family is in the sample for the 2 years

() Lecture 1 20 / 31

Pooled Cross Sections

Pooled cross-sections can be an effective way to analyse govt policies(e.g. look at economic relationships before and after the policy wasintroduced)

Pooled cross sections are also very useful for studying group dynamicsover time (e.g. how are average wages evolving for the group whoentered the labour market during the last recession; what determineschanges in median house prices in specific areas of US)

Can analyse like a standard cross-section, though need to allow forchanges in variables over time

() Lecture 1 21 / 31

Pooled Cross Sections

Example: Two years of house prices

() Lecture 1 22 / 31

Panel Data

D. Panel (or Longitudinal) DataConsists of a time series for each cross-sectional unit

=⇒ follow the same individuals / firms etc. over time

Example: crime statistics at the city level – to study things like effectof law enforcement or economic conditions on crime

() Lecture 1 23 / 31

Panel Data

panel data has some important advantages over other data structures(we can control for certain types of unobserved characteristics, andcan study lags in behaviour).

some important questions cannot be answered without panel data

=⇒ e.g. studying dynamics behaviour of individual units

briefly consider simple panel data methods

() Lecture 1 24 / 31

Causal Effects, Ceteris Paribus and Correlation

Causality and Ceteris Paribus in Econometric Analysis

In most tests of economic theory, and for evaluating policy, the goal isto infer a causal effect of one variable on another

Most propositions in economics are ‘ceteris paribus’ by nature

Example: the responsiveness of the demand for coffee to price -holding all other factors constant (such as income, prices of othergoods). If these other factors are not constant, we cannot determinethe casual effect of a price change on quantity demanded

not feasible to literally hold ‘all else equal’ .... but have enough otherfactors been held constant to infer causality ?

properly applied, econometric methods can simulate a ceteris paribusexperiment

(⇒) economic theory and econometrics together can help us uncovercausal effects.

() Lecture 1 25 / 31

Causality and Correlation

We may recall (hopefully!) from Prob/Stats ECO329 the concept ofcorrelation and covariance

Measures of linear association between 2 variables

Example: Education and Wages.

Do people with higher levels of education tend to have higher wages?Do people with higher wages have more education?

Correlation is a measure of this assocation

Let r be the correlation between wage for person i - say yi andeducation xi

() Lecture 1 26 / 31

Recall (?????) that,

r =∑n

i=1(xi − x)(yi − y)√∑n

i=1(xi − x)2√

∑ni=1(yi − y)2

r > 0 means that large y are associated with large x

r < 0 means that large y are associated with small x

r = 0 means no linear association

could be nonlinearcould be no association at all

() Lecture 1 27 / 31

Keep in Mind

One cannot conclude causation by simply looking at correlation

Note r is symmetric in x and y so:

does x cause ydoes y cause x

Even if one thought it went a particular direction there may be othermitigating factors that need to be taken into account

() Lecture 1 28 / 31

Examples:

Wages and Education are correlated (as we will see)

Which direction is plausible and why?

Other factors?

() Lecture 1 29 / 31

Examples:

SportsIn watching football I often hear/see statements that are of the form:

“When team x (insert your favorite) runs the ball more than 30 times theywin 80% of the games but when they run less than 30 times they only win

30% of the games.”

What the heck does this mean? Clearly it looks like there is acorrelation between number of running plays and the chance ofwinning.

But is it a causal effect?

If it were causal then it would mean the coach could just make surehe runs the ball at least 30 times (regardless) and will win more often

Is this how it works?

() Lecture 1 30 / 31

Examples:

In the popular press there are many instances of people trying to infer acausal relationship between variables based simply on correlations betweentwo variables.

Try and listen for examples of this.

Now lets play with some data!

1 Wage data – relation between education and wages

2 Test score data – relation between class size and standardized testscores

() Lecture 1 31 / 31