FIN 3232 Financial Econometrics

Lesson 7 – Panel Data Regression

Tharindu Ediriwickrama

1
 Panel Data
• In panel data, the same cross-sectional units is
surveyed over time.
– Pooled OLS model
– The fixed effects model
– The random effects model

2
3
 Pooled OLS model
• The data will be simply pooled together.
• Ignore the heterogeneity among cross
sections and different time points.
• Ex: Data of six south Asian countries over last
20 years pooled together.

4
 Estimating the Uniqueness of Each
Individual Unit
• One of the strengths of panel data is that it
permits analysis of important economic
questions that can’t be addressed using data that
are exclusively cross sectional or time series.
• By utilizing repeated information on the
individual entities being investigated, you can
control for the effects of some missing or
unobserved variables.

5
 Estimating the Uniqueness of Each
Individual Unit
• An observable variable can be something like
age, education, or anything that’s typically
identified in surveys.
• An unobservable variable can be an
individual’s work ethic, natural ability, or any
information that’s not easily obtained when
data is collected.

6
 Estimating the Uniqueness of Each
Individual Unit
Yit = β0 + β1Xit + β2wit + εit

where,
i = 1,…, n represents the cross-sectional unit beginning
with the first individual unit (1) and proceeding to the
last (n), t = 1,…, T captures the time period in which the
subject is observed beginning with the first time period
(1) and proceeding to the last (T), X is an observable
independent variable, and w is an unobservable
independent variable. 7
 Estimating the Uniqueness of Each
Individual Unit
• The danger with combining panel data and
OLS estimation is that you may end up with
results containing heterogeneity bias.
• This bias occurs if you ignore characteristics
that are unique to your cross-sectional units
(relegate those things to the error term) and
they’re correlated with any of your
independent variables

8
In Figure 17-1, the pooled OLS estimate (line D) results in an overestimate of the
impact of X on Y, as illustrated by the parallel lines A, B, and C.

9
In Figure 17-2, the pooled OLS estimate (line D) results in an underestimate of the
impact of X on Y, as illustrated by the parallel lines A, B, and C

OLS estimate produce a flatter slope by ignoring individual fixed effects 10

In Figure 17-3, the pooled OLS results (line D) generate a negative estimated impact of X on Y
when, as illustrated by the parallel lines A, B, and C, the effect is actually positive.

11
 Estimating the Uniqueness of Each
Individual Unit
• The existence of unobservable factors that
consistently impact your outcome of interest
(Y variable) is likely with panel data, which
means you need to consider using one of
three estimation methods.
– First Difference (FD) Transformation
– Dummy Variable (DV) Regression
– The Fixed Effects (FE) Estimator

12
 First difference (FD) transformation
• With panel data, you can deal with
unobservable variables by applying a first
difference (FD) to the data.
• To transform the data into an FD, you subtract
the previous value of a variable from the
current value of that variable for a particular
cross-sectional unit and repeat the process for
all variables in the analysis.

13
 First difference (FD) transformation
• After you perform an FD transformation, you can
estimate the model using OLS with all the first-
differenced data.
• Doing so eliminates (differences out) any fixed
effects associated with the cross-sectional units,
even if those characteristics aren’t observable.
• Repeated observations for the same entities allow
you to get rid of the effect of unobservable factors
only if those characteristics are constant over time
for each entity. 14
 Dummy variable (DV) regression
• If you have panel data, the simplest approach
in estimating your model is to pool all the
years of data and apply ordinary least squares
(OLS) so that you’re essentially ignoring the
panel nature of the data.

15
 Dummy variable (DV) regression
Yit = β0 + β1Xit + vit
• where vit = wi + εit. The vit term is known as the
composite error because it contains individual
fixed effects and an idiosyncratic error. The
individual fixed effects are unobservable factors
associated with the individual subjects, whereas
the idiosyncratic error represents a truly random
element associated with a particular subject at a
specific point in time.

16
 Dummy variable (DV) regression
• One way to account for individual fixed effects
is by using the dummy variable (DV)
regression.
• You apply this approach by including dummy
variables in your model for each cross-
sectional unit, making it a straightforward
extension to the basic use of dummy
variables.

17
 Dummy variable (DV) regression
• If your data contains a large number of
individuals (cross-sectional units), which is
quite common with panel data, then the DV
approach can be computationally burdensome
(even for a computer) and impractical.

18
 Fixed effects (FE) estimator
• The most common method of dealing with fixed
effects of cross-sectional units is known as the fixed
effects (FE) estimator.
• FE estimation is applied by time demeaning the data.
• In other words, you calculate the average value of a
variable over time for each cross-sectional unit and
subtract this mean from all observed values of a
given cross-sectional unit, repeating the procedure
for all the cross-sectional units.

19
 Fixed effects (FE) estimator
• For FE estimation, you must first specify the model as,

• Where, β1 is known as the fixed effects estimator (or

within estimator). The unobservable variable (w) has
been demeaned away because the values are assumed
constant over time. Finally, I place the ~ above the other
variables to note they’ve been transformed into their
time-demeaned versions (also called the within
transformation). 20
 Fixed effects (FE) estimator
• Where,

21
 Fixed effects (FE) estimator
• You may be tempted to calculate the degrees of
freedom with FE estimation using the
traditional (OLS) calculation (total number of
observations minus the number of estimated
parameters).
• Consequently, the correct formula for
calculating the degrees of freedom is nT – n – p,
where n is the number of cross-sectional units,
T is the number of time periods, and p is the
number of independent variables.
22
Increasing the Efficiency of Estimation with
Random Effects
• With panel data, the advantage of the RE
model over the FE model is more efficient
estimates of the regression parameters. The
RE technique doesn’t estimate the fixed
effects separately for each cross-sectional
unit, so you get fewer estimated parameters,
increased degrees of freedom, and smaller
standard errors.

23
 Increasing the Efficiency of Estimation with
Random Effects
• The error term in a RE model is known as the
composite error term because it combines two
components.
• This term was also used in the previous section (where
you learn about the fixed effects model), but the
random effects model requires that you pay more
attention to the specific components of the error term:
– The unobserved effects associated with each particular
cross-sectional unit (wi)
– A completely random element that isn’t associated with the
cross-sectional units (εit) 24
Increasing the Efficiency of Estimation with
Random Effects
• A critical assumption of the RE model is that
the unobserved individual effect (wi) isn’t
correlated with the independent variable(s);
Cov(Xit, wi) = 0.
• If the individual effect is correlated with the
independent variable(s), then the RE estimate
is biased.

25
 Increasing the Efficiency of Estimation with
Random Effects
• In an RE model, your independent variables can
include individual characteristics that don’t vary
over time (such as gender and race) because they
won’t be differenced away as they are in the FE
model.
• In addition, RE models are also likely to include
time-effect controls — added dummy variables for
each time period in which cross-sectional
observations were obtained.
26
 Testing Efficiency against Consistency with
the Hausman Test
• The RE model produces more efficient estimates
than the FE model.
• However, if individual fixed effects are correlated
with the independent variable(s), then the RE
estimates will be biased.
• In that case, the FE estimates would be preferred.
• The Hausman test checks the RE assumptions and
helps you decide between RE and FE estimation.

27
 Testing Efficiency against Consistency with
the Hausman Test
• A Hausman test examines differences in the
estimated parameters, and the result is used
to determine whether the RE and FE estimates
are significantly different.

28
 Unit Root Testing in Panel Data
• There are so many tests for this.
– Levi, Lin and Chu (LLC)
– Im, Pesaran, Shin W stat
– ADF Fisher Chi square
– PP Fisher Chi square
• Null and alternative hypotheses for above are as follows.
– Null: Panel data has unit root or non-stationary
– Alt: Panel data has no unit root and data is stationary.
• However, there is another test called Hadri Test.
– Null: Panel data has no unit root and data is stationary.
– Alt: Panel data has unit root or non-stationary

29
 Panel Data Regression
• 1st Step: Estimate a pooled regression model
with neither fixed nor random. But this pooled
model assumes intercepts are the same for
each firm and for each year.
• 2nd Step: Estimate a random effects model.
Select only random effect for firm and not
over time.
• 3rd Step: Conduct Hausman Test.
– If P < 0.05, Fixed Effect
– If P > 0.05, Random Effect 30
 Panel Data Regression
• 4th Step: It is worth determining whether fixed
effect are necessary or not using Redundant
Fixed Effects – Likelihood ratio test.
– If P < 0.05, pooled model is not appropriate
– If P > 0.05, Pooled model is appropriate

31
THANK YOU!

32