Introduction

Prep: Statistical Concepts

The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Chapter 2 Panel Data

Advanced Financial Empirical Research

Lu Liu

1/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Introduction

Time series data A collection of observations on a single

entity such as a household, a firm and a country
over several time periods.
Cross-sectional data A collection of observations for multiple
entities at a single time point.

2/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Introduction

Panel data/ longitudinal data The pooling of observations on

a cross-section of entities such as households,
firms and countries over several time periods.
I Embodies information across both time and
space.
I Follows the same entities over time.

3/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Introduction

Repeated cross-section Observations that are not on the

same entities measured over time.

4/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Introduction
Basic econometric setup of panel data (Pooled OLS)

yit = α + βxit + uit , t = 1, . . . , T , i = 1, ..., N (1)

I β is a 1 × k vector of coefficients for the explanatory

variables,
I xit is a k × 1 vector of observations on the explanatory
variables,
I T is the number of time periods,
I N is the number of entities.
The OLS estimator of β is named the pooled OLS estimator.
5/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Introduction
Implement the regression
     
yi=1,t=1 xi=1,t=1 ui=1,t=1
 yi=1,t=2   xi=1,t=2   ui=1,t=2 
     
 ..   ..   .. 

 . 


 .  
  . 

 yi=1,t=T   xi=1,t=T   ui=1,t=T 
     
 .. 
= α + β
 .. +
  .. 

 . 


 . 
 
 . 

 yi=N,t=1   xi=N,t=1   ui=N,t=1 
     
 yi=N,t=2   xi=N,t=2   ui=N,t=2 
     
 ..   ..   .. 
 .   .   . 
yi=N,t=T xi=N,t=T ui=N,t=T

6/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Advantages of Panel Data I

1. Panel data give more variability and less collinearity

among the variables. The additional variation introduced
by combining cross-sectional data with time series data
can help to mitigate problems of multicollinearity that
plague time series models.

7/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Advantages of Panel Data II

Example: Baltagi and Levin (1992)

I The consumption of cigarette in the US is modelled as a
function of lagged consumption, price and income.
I Problem: in the aggregate time series for the US, there is
high collinearity between price and income.
I Solution: consider cigarette demand with a panel across
American states.
I Multicollinearity is less likely with a panel since the
cross-section dimension adds more informative data on
price and income, and hence increase variability in the
data.

8/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Advantages of Panel Data III

2. Panel data has the ability to control for all time-invariant

variables or cross-sectional invariant variables, whose
omission could bias the estimates in a typical cross-section
study or a time-series study.

9/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Advantages of Panel Data IV

Example: Baltagi and Levin (1992)
I The demand for cigarettes may be affected by variables
other than the variables of interest, i.e. price and income.
I State-invariant variables
I National advertising on TV that does not vary across states
I Time-invariant Variables
I Distribution of religious population (e.g. Utah, which has a
high percentage of Mormon population, has a per capita
sales of cigarette less than one half of the national average).
I Some cross-section-invariant and time-invariant variables
are difficult to measure or hard to obtain, but omitting them
from the regression may lead to bias in estimation.

10/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Advantages of Panel Data V

3. Panel data are not only able to model why individual

units behave differently (cross-sectional dimension)
but also why a given unit behaves differently at
different time periods (time dimension). This allows a
researcher to identify certain parameters or questions that
cannot be addressed using pure cross-sectional data or
pure time-series data.

11/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Advantages of Panel Data VI

Example: Consider a situation in which the average

consumption level rises with 2% from one year to another.
I It might be interpreted as 2% increase for all individuals.
I It might imply an increase of 4% for one half of the
individuals and no change for the other half (or another
combination).
To discriminate between these two models, we need to
utilize multiple-year consumption data on an individual
level, which is panel data.

12/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Panel Data Models

I Fixed effects model

I Random effects model

13/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Prep: Statistical Concepts I

In statistical models, true values of parameters are unknown.
The "estimator" of a parameter is itself a random variable,
which has certain distribution. A particular realization of the
estimator is an "estimate".
I The bias of an estimator is the difference between this
estimator’s expected value (given the sampling distribution)
and the true value of the parameter.
An estimator b is unbiased, if it is on average equal to the
true value β of the parameter. This is formally formulated
as

E{b} = β. (2)

14/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Prep: Statistical Concepts II

I Consistency is a large sample property and says that if

we obtain more and more observations, the probability that
our estimator is away from the true value β becomes
smaller and smaller.
The formal formulation of a consistent or asymptotically
consistent estimator is

lim P{|b − β| > δ} = 0 for all δ > 0 (3)

n→∞

We usually refer to this as ’the probability limit of b is β’ or

’b converges in probability of β’.

15/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Prep: Statistical Concepts III

I The efficiency of an estimator refers to how much

information it extracts about the parameter of interest from
the sample. The smaller the standard error is, the more
efficient the estimator.

16/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Fixed Effects Model I

Decompose the residual in equation (1) , uit , into an individual
specific effect, µi , and the remainder disturbance, vit

yit = α + βxit + µi + vit . (4)

µi is named individual- or entity-fixed effects

I µi encapsulates all the individual heterogeneity that are
time-invariant but vary across entities. - e.g., a person’s
gender, the sector that a firm operates in, the country
where a firm has its headquarters.
I In the (Baltagi and Levin, 1992)’s cigarette demand study,
µi controls for all the time-invariant variables such as the
distribution of religious population.
17/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Fixed Effects Model II

I Restriction in this model: µi = 0 for one arbitrary entity i to

avoid perfect multicollinearity between the individual-fixed
effect terms and the intercept term.

18/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Least Squares Dummy Variable Estimator

Eq(4) can be estimated using dummy variables

yit = α + βxit + µ2 D2i + µ3 D3i + · · · + µN DNi + vit . (5)

I Dji is a dummy variable that takes value one for all

observations on the jth entity (e.g. the jth firm) in the
sample and zero otherwise.
I Because of having the intercept term α, D1i is put to zero
in order to avoid the "dummy variable trap". Putting a
dummy for another entity to zero will do the same job.
I The parameters can be estimated by OLS. The implied
estimator for β is the least squares dummy variable
(LSDV) estimator.
19/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Least Squares Dummy Variable Estimator

An alternative econometric presentation of the fixed effects

model:

yit = βxit + µ1 D1i + µ2 D2i + µ3 D3i + · · · + µN DNi + vit (6)

The intercept term α is suppressed but dummy variables for all

the entities are included.

20/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

To estimate the LSDV estimators, we must estimate N + k

parameters in total in the fixed effects model. It is challenging to
estimate these many parameters when N is large.
I Let’s transform the data to get rid of the fixed-effect dummy
variables so as to make the estimation easier.
I The transformation, known as the within transformation,
involves subtracting the time-mean of each entity away
from the values of the variable.

21/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

Define:
1 PT
I ȳi = T t=1 yit : the time-mean of y for entity i.
1 PT
I x̄i = T t=1 xit : the time-mean of x for entity i.
Eq(4) averaged over time:

ȳi = α + β x̄i + µi + v̄i . (7)

Subtracting eq(7) from eq(4)

yit − ȳi = β(xit − x̄i ) + vit − v̄i (8)

µi and α are dropped out after the transformation!

22/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

Write eq(13) as

ÿit = β ẍit + v̈it (9)

where the double dots above the variables denote the

demeaned values.
I The implied estimator for β after the within transformation
is referred to as the fixed-effects estimator or within
estimator.
I Regression (9) can be estimated using OLS on the pooled
sample of demeaned data.
I µ̂i and α can be recovered from eq(7) after the estimation.
As µ1 = 0, α = ȳi=1 − β̂ x̄i=1 .
23/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

Do we get a larger number of degrees of freedom from within

transformation since we only estimate k parameters instead of
k + N? No.
I For every one of the N individuals for which we estimated
the mean, we lose a degree of freedom.
I Hence, the number of degrees of freedom that must be
used in estimating standard errors in an unbiased way
and when conducting hypothesis tests is NT − N − k .
I The LSDV estimator and the within estimator have identical
estimated values and standard errors.

24/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

Computation of standard errors

Var(β̂|x) = σ 2 (x 0 x)−1 (10)

q
ˆ β̂j |x) = s2 (x 0 x)−1
s.e.( jj (11)

I For a regression with k explanatory variables and an

intercept without
P fixed effects, s2 is the residual sums of
squares (i.e. 2
vi ) divided by NT − k − 1.
I With (N − 1) fixed effects (dummies), the s2 is the residual
sums of squares divided by NT − N − k . Standard errors
increase as the degree of freedom decreases.

25/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

Testing for fixed effects (H0 : µ2 = µ3 · · · = µN = 0)
(RRSS − URSS)/(N − 1)
F0 = ∼ FN−1,NT −N−k (12)
URSS/(NT − N − k )

I RRSS (the restricted residual sum of squares): the

residual sum of squares (RSS) of OLS on the pooled
model without fixed effects.
I URSS (the unrestricted residual sum of squares): the RSS
of the LSDV regression.
I (N − 1) is the difference in the number of parameters
between the restricted and the unrestricted models.
I NT − N − k is the degree of freedom of the unrestricted
model.
26/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

The Within Transformation Estimator

The LSDV estimator and the within estimator is
I consistent if E[(xit − x̄i )(vit − v̄i )] = 0 (i.e. E[ẍit v̈it ] = 0 ) .
Because of the averages, the condition implies that x is
strictly exogenous: E[vit |xis ] = 0, s = 0, 1, . . . , T .
I inefficient if the F-test for fixed effects is not rejected →
Fixed effects are redundant and thus pooled OLS
estimator is preferred).

yit − ȳi = β(xit − x̄i ) + vit − v̄i (13)

27/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Time-fixed Effects Model I

yit = α + βxit + λt + vit (14)

λt captures all of the variables that vary over time but are
constant cross-sectionally.
I business cycle
I regulatory environment
I tax rate
their effect may be the same for everyone in the economy but
vary over time.

28/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Time-fixed Effects Model II

The LSDV estimator for the time-fixed effects model

yit = α + βxit + λ2 T2 + λ3 T3 + ... · · · + λT TT + vit (15)

where Tj denotes a dummy variable that takes value 1 for time

period j and zero elsewhere, and so on.

29/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Time-fixed Effects Model III

Within-transformation of the time-fixed effects model

Define:
I ȳt = N1 N
P
i=1 yit : the cross-sectional mean of y for time
period t.
I x̄t = N1 N
P
i=1 xit : the cross-sectional mean of x for time
period t.
Subtracting the cross-sectional averages from each observation

yit − ȳt = β(xit − x̄t ) + vit − v̄t (16)

30/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Time-fixed Effects Model IV

Testing for time-fixed effects (H0 : λ2 = λ3 · · · = λT = 0)

(RRSS − URSS)/(T − 1)
F0 = ∼ FT −1,NT −T −k (17)
URSS/(NT − T − k )

31/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Two-Way Fixed Effects Model

yit = α + βxit + µi + λt + vit . (18)

The LSDV equivalent model with both cross-sectional and time

dummies

yit = βxit + µ1 D1 + µ2 D2 + µ3 D3 + · · · + µN DN
+ λ1 T1 + λ2 T2 + λ3 T3 + ... · · · + λT TT + vit . (19)

The number of parameters is now k + N + T .

32/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Two-Way Fixed Effects Model

Two-way within transformation:
1. Subtract the time-mean of the variables in eq(18). µi is
dropped.

yit − ȳi = β(xit − x̄i ) + (λt − λ̄t ) + (vit − v̄i ) (20)

written as

ÿit = β ẍit + λ̈t + v̈it (21)

where the double dots above the variables denote the

values after subtracting the time-mean.
2. Subtract the cross-sectional means of the variables in
eq(21). λ̈t will be dropped.
33/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Two-Way Fixed Effects Model

I An alternative to the two way transformation is to

implement within transformation only for the large data
dimension, and then estimate the regression on demeaned
variables with dummy variables for the small dimension.
I Microdata panels, panels with large N and small T , e.g.
panel data solicited from household surveys: first transform
the variables by subtracting time-mean to remove individual
fixed effects and then run a regression on the variables with
dummy variables for time periods.
I The opposite applies to macrodata panels, i.e.
panels with small N and large T , e.g. panel of observations
for countries over many time periods.

34/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Two-Way Fixed Effects Model

Testing for two-way fixed effect against

I entity-fixed effect model (H0 : λ2 = λ3 · · · = λT = 0)
I time-fixed effect model (H0 : µ2 = µ3 · · · = µN = 0)
I pooled ols model (H0 : λ2 = λ3 · · · = λT = 0 ;
µ2 = µ3 · · · = µN = 0)
RRSS and the number of degrees of freedom vary with H0 .

35/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Implementation in Stata
Consider an application of panel data on a test of capital asset
pricing model due to Fama and MacBeth (1973). The test
involves a two-step estimation procedure:
1. First, the market beta is estimated in separate time series
regression for each firm.
2. Second, for each time point t, a cross-sectional regression
of the excess returns on the estimated betas is conducted
Rit − Rft = λ0 + λm βit + uit (22)
where the dependent variable is the excess return of i at
each t and the independent variable is the estimated beta
for i.
If CAPM holds, λ0 =0 and λm should approximate the (time
average) equity market risk premium, Rm − Rf .
36/48
 Introduction
Prep: Statistical Concepts The Least Squares Dummy Variable Estimator
The Fixed Effects Model Within Transformation
The Random Effects Model Time-Fixed Effects Model
Fixed Effects or Random Effects? Two-Way Fixed Effects Model
References

Implementation in Eviews
Fama and MacBeth (1973) proposed estimating this second
stage regression separately for each time period, and then
taking the average of the parameter estimates to conduct
hypothesis tests. Regression (22) is a combination setting of
cross-section and time-series, therefore, we can achieve a
similar objective using a panel approach. We will use a sample
comprising the annual returns and estimated betas for eleven
years on 2500 UK firms provided by Brooks (2019):
https://www.cambridge.org/as/academic/
subjects/economics/finance/
introductory-econometrics-finance-4th-edition?
format=PB → ’Resources’ → ’General Resources’ → ’Excel
files’ → ’panelex.xls’.
37/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

The Random Effects Model

The loss of degrees of freedom in the fixed effects model can
be avoided if µi and λt are random. The entity-random effects
model is

yit = α + βxit + µi + vit , µi ∼ IID(0, σµ2 ); vit ∼ IID(0, σv2 ),

(23)

µi + vit is an error term consisting of two components:

I an individual specific component, which does not vary over
time,
I a remainder component, which is assumed to be
uncorrelated over time.

38/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

The Random Effects Model

The OLS estimators for α and β from eq(23) are

I consistent,
I because µi and vit are mutually independent and
independent of xjs (for all entity j and time s).
I inefficient,
I because the composite error term µi + vit exhibits a
particular form of autocorrelation as µi is constant over
time. Consequently, routinely computed standard errors of
the OLS estimator are incorrect.

39/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

The Random Effects Model

Generalized Least Squares (GLS) transformation for a more
efficient estimator:
I Define the ’quasi-demeaned ’ data:
yit∗ = yit − θȳi (24)
xit∗ = xit − θx̄i (25)
ȳi and x̄i are time-mean of the observations on yit and xit ,
respectively.
I
σv
θ =1− q (26)
T σµ2 + σv2

σv2 is the variance of the observation error term, and σµ2 is

the variance of the entity-specific error term.
40/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

The Random Effects Model

The transformation ensures no autocorrelations in the error
terms of the transformed model:

yit∗ = α(1 − θ) + xit∗ β + εit (27)

where the error term εit is iid over individuals and time.
I Estimate σµ2 and σv2 in the first step.
I In the second step, estimate the transformed model using
OLS. The estimators are named as the random-effects
estimators.
Similarly to the time-fixed effects model and the two-way fixed
effects model, we can estimate a time-random effects model
and a two-way random effects model including random effects
for time periods.
41/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Fixed Effects or Random Effects?

Random effects model states that

E{yit |xit } = α + xit0 β, (28)

Fixed effects model states that

E{yit |xit } = α + xit0 β + µi . (29)

One may prefer the fixed effects estimator if some interest lies
in µi , which makes sense if the number of units is relatively
small and of a specific nature. The random effects are more
appropriate when the entities in the sample can be thought of
as having been randomly selected from the population.

42/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Fixed Effects or Random Effects?

I The fixed-effects (within) estimator cannot estimate

the effect of any time-invariant variable. These
time-invariant variables are wiped out by demeaning the
variables. Alternatively, time-invariant variables are
spanned by the individual dummies.
I The random-effects estimator does not wipe out the
time-invariant variables.

43/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Fixed Effects or Random Effects?

I Random effects estimators are inconsistent if µi are

correlated with the explanatory variables.
I Fixed effects estimators is consistent regardless of the
relationship between explanatory variables and µi .

44/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

Fixed Effects or Random Effects?

I The fixed effects estimator exploits the within dimension of

the data (differences within individuals) only. The between
dimension of the data is lost due to within transformation.
I The random-effects estimators use more of the variation in
the data (the between-individual variation). So, if
random-effects estimators are consistent (i.e. µi
uncorrelated with xit ), the random-effects estimators are
more efficient than the fixed-effects ones.

45/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

The Hausman Test I

The Hausman test (Hausman, 1978) H0 : xit and µi are

uncorrelated.
I The fixed-effects estimator is consistent under both H0 and
H1 .
I The random-effects estimator is consistent (and typically
efficient) under H0 only.
I A significant difference between the two estimators
indicates that H0 is unlikely to hold.

46/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

The Hausman Test II

Consider the difference vector β̂FE − β̂RE . The Hausman test

statistic is

ξH = (βˆFE − β̂RE )0 [V̂ {β̂}FE − V̂ {β̂RE }]−1 (β̂FE − β̂RE ) (30)

where V̂ s denote estimates of covariance matrices of the

estimators. The statistic ξH has an asymptotic Chi-squared
distribution with k degrees of freedom.

47/48
 Introduction
Prep: Statistical Concepts
The Fixed Effects Model
The Random Effects Model
Fixed Effects or Random Effects?
References

References

Baltagi, B. H. and Levin, D. (1992). Cigarette taxation: raising

revenues and reducing consumption. Structural Change and
Economic Dynamics, 3(2):321–335.
Brooks, C. (2019). Introductory econometrics for finance.
Cambridge university press.
Fama, E. F. and MacBeth, J. D. (1973). Risk, return, and
equilibrium: Empirical tests. The journal of political economy,
pages 607–636.
Hausman, J. A. (1978). Specification tests in econometrics.
Econometrica: Journal of the Econometric Society, pages
1251–1271.

48/48