Chapter Four

Introduction to Panel Data

Econometrics II

May 17, 2024

Panel Data

I In panel data, the same units are observed over certain periods of
time.⇒ so we have cross-sectional and time series dimension.
I If all the cross-sectional units have the same number of time series
observations, the panel is known as balanced, if not it is unbalanced.
I If the number of cross-sectional subjects, N, is greater than the
number of time periods, T ⇒ short panel
I If T is greater than N ⇒ long panel

Econometrics II Chapter Four May 17, 2024 2 / 26

Example of Panel dataset

Econometrics II Chapter Four May 17, 2024 3 / 26

Total number of observation = N ∗ T
I In econometrics, panel data models are formally specified as:
yi t = α + βxi t + ui t (1)
i = 1. . . n companies and t = 1. . . T time periods.

Econometrics II Chapter Four May 17, 2024 4 / 26

Importance of Panel Data

I can take heterogeniety explicitly into account by allowing for subject -

specific variables
I more informative, more variability, less collinearity, more degrees of
freedom
I allows for the study of the dynamics of change
I by structuring the model in appropriate way, we can remove the
impact of certian forms of omitted variable bias

Econometrics II Chapter Four May 17, 2024 5 / 26

Estimation of model (1) depends on the assumptions that we make about
the intercept (α0 ), the slope coefficient (α1 ) and the error term (ui t ).
Several possible assumptions can be made in order to estimate model (1):
1 Assume that the intercept and slope coefficients are constant across
time and firms and that the error term captures differences over time
and over firms.
2 The slope coefficient is constant but the intercept varies over firms.
3 The slope coefficient is constant but the intercept varies over firms
and over time.
4 All coefficients (intercept and slope) vary over firms.
5 The intercept as well as the slope vary over firms and time.

Econometrics II Chapter Four May 17, 2024 6 / 26

Estimations of Panel Data

1. Pooled Regression

yi t = α + βxi t + ui t (2)

I the simplest way is to estimate a single, pooled regression on all the

observations together
I but pooling the data assumes that there is no heterogenity, i.e, the
same relationship holds for all data

Econometrics II Chapter Four May 17, 2024 7 / 26

2. Fixed Effects Model

yi t = α + βxi t + µi + ui t (3)

I µi encapsulates all of the variables that affect yi t cross-sectionally,

but do not vary overtime.
I for example, the sector that a firm operates in, a person’s gender, or a
country where a bank has its head quarter, etc
I thus we capture the heterogeniety in µi by allowing for different
intercepts for each cross-sectional unit.

Econometrics II Chapter Four May 17, 2024 8 / 26

3. Least Square Dummy Variable (LSDV) model

I The FE model could be estimated using LSDV approach.

yi t = βxi t + µ1 D1i + µ2 D2i + µ3 D3i + ...... + µN DNi + ui t (4)

D1i is dummy variable (=1) for all observations on the first entity in
the sample and zero otherwise
D2i is dummy variable (=1) for all observations on the second entity
in the sample and zero otherwise, etc.

Econometrics II Chapter Four May 17, 2024 9 / 26

3. Least Square Dummy Variable (LSDV) model....

yi t = βxi t + µ1 D1i + µ2 D2i + µ3 D3i + ...... + µN DNi + ui t

I LSDV can be seen just as the standard regression model and thus,
can be estimated by OLS
I the model above has N + k parameters to estimate
I to avoid estimating too many dummy variable parameters, a
transformation, known as the within transformation is used.

Econometrics II Chapter Four May 17, 2024 10 / 26

The Within Transformation

I substracting the time mean of each entity away from the values of the
variable
I define the time-mean of the observations for the cross-sectional unit i
as
T
1 X
y¯i = yi t (5)
T t=1
I similarly calculate the mean of all explanatory variables
I subtract the time-mean from each variable to obtain a regression
containing a demeaned variables only
I such regression does not require an intercept term since the
dependent variable will have a zero mean by construction.

Econometrics II Chapter Four May 17, 2024 11 / 26

Within Transformation...

I the model containing the demeaned variables is

yi t − ȳi = β(xi t − x̄i + ui t − ūi (6)

I we could write this as

ÿi t = βẍi t + üi t (7)
where double dots above the variables denotes the demeaned values
I this model can be estimated using OLS, but we need to make degrees
of freedom correction.

Econometrics II Chapter Four May 17, 2024 12 / 26

The Between Estimator

I alternative to demeaning, run a cross-sectional regression on the

time-averaged values of the variables

ȳi = α + βx̄i + ūi (8)

1 PT 1 PT 1 PT
where y¯i = T t=1 yi t , x¯i = T t=1 xi t , ūi = T t=1 ui t

I an advantage of the between estimator over the within estimator is

that the process of averaging is likely to reduce the effects of
measurement error in the variables on the estimation

Econometrics II Chapter Four May 17, 2024 13 / 26

4. First Difference

I a further possibility is that the FD operator could be applied

I the model explians the change in yi t rather than its level
I any variable that do not change overtime will again cancel out
I FD and the within group transformation will produce identical
estimates when there are only two periods.

Econometrics II Chapter Four May 17, 2024 14 / 26

5.Time Fixed Effect Models

I a time-fixed effects model rather than an entity-fixed effects model

I if we think that the average value of yi t changes over time but not
cross-sectionally
I intercepts allowed to vary overtime but assumed to be the same
across entities at each given point in time

yi t = α + βxi t + λt + vi t (9)

where λt is a time varying intercept.

I it captures all of the variables that affecy y and that vary overtime,
but are constant cross-sectionally

Econometrics II Chapter Four May 17, 2024 15 / 26

Time Fixed Effect Models....

I e.g., the regulatory environment or tax rate changes may well

influence y, but in the same way for all firms.
I time-variation in the intercept terms can be allowed for exactly the
same way as with entity fixed effects (LSDV)

yi t = βxi t + λ1 D1t + λ2 D2t + λ3 D3t + .... + λT DTt + vi t (10)

where D1t a dummy variable(=1) for the first time period and zero
elsewhere), and so on

Econometrics II Chapter Four May 17, 2024 16 / 26

Time Fixed Effect Models....

yi t = βxi t + λ1 D1t + λ2 D2t + λ3 D3t + .... + λT DTt + vi t (11)

I dummy variables capture time variation rather than cross-sectional

variation
I to avoid estimating a model containing all T dummies, a within
transformation can be conducted to subtract away the cross-sectional
averages from each observations
I it is possible to allow for both entity fixed effects and time fixed
effects within the same model (two way error component model)

yi t = α + βxi t + λt + µi + vi t (12)

Econometrics II Chapter Four May 17, 2024 17 / 26

Random Effects Model

I aka the error component model

I different intercept terms for each entity and these intercepts are
constant over time
I the intercepts assumed to arise from a common intercept α, plus a
random variable εi (varies cross-sectionally but constant overtime)

yi t = α + βxi t + ωi t , ωi t = εi + vi t (13)

I εi measures the random deviation of each entity’s intercept term from

the "global" intercept term α

Econometrics II Chapter Four May 17, 2024 18 / 26

How the RE Model Works

yi t = α + βxi t + ωi t , ωi t = εi + vi t

I unlike FE, no dummy variables to capture the heterogeneity

(variation) in the cross-sectional dimensions
1 this occurs via the εi terms
I RE assumes that εi (the cross-sectional error term)
1 has zero mean
2 independent of vi t (the individual observation error term)
3 has constant variance
4 independent of the explanatory variables

Econometrics II Chapter Four May 17, 2024 19 / 26

How the RE Model Works.....

yi t = α + βxi t + ωi t , ωi t = εi + vi t

I OLS estimation
1 the parameters (α and β) are estimated consistently but inefficiently by
OLS
2 the conventional formulea need modification as a result of the
cross-correlation between error terms for a given cross-sectional unit at
a different points in time

Econometrics II Chapter Four May 17, 2024 20 / 26

How the RE Model Works....

yi t = α + βxi t + ωi t , ωi t = εi + vi t

I GLS estimation
1 generalised least squares (GLS) procedure is usually used
2 the transformation involved is to subtract a weighted mean of yi t
overtime
(i.e., part of the mean rather than the whole mean as was the case of
fixed effect estimation)

Econometrics II Chapter Four May 17, 2024 21 / 26

Fixed Effects vs Random Effects Model?

yi t = α + βxi t + ωi t , ωi t = εi + vi t

I RE model is appropraite
1 when the entities have been randomly selected from the population

RE is superior to FE

I will not remove the explanatory variables that do not vary over time
I fewer parameters to be estimated, saves degrees of freedom
I produce more efficient estimation than the FE approach

Econometrics II Chapter Four May 17, 2024 22 / 26

Fixed Effects vs Random Effects....

yi t = α + βxi t + ωi t , ωi t = εi + vi t

I RE is valid only when ωi t (the composite error term) is uncorrelated

with all of the explanatory variables

I this assumption is more strigent than the corresponding one in the FE

case, because with RE we require both εi and vi t tobe independent of
all of the xi t

Econometrics II Chapter Four May 17, 2024 23 / 26

Fixed Effects vs Random Effects....

yi t = α + βxi t + ωi t , ωi t = εi + vi t

I any unobserved omitted variables(that were allowed for by having

different intercepts for each entity) must be uncorrelated with the
included explanatory variables

I if they are uncorrelated, a random effects approach can be used,

otherwise the fixed effects model is preferable.

Econometrics II Chapter Four May 17, 2024 24 / 26

Fixed Effects vs Random Effects....

I Hausman test can be used to test for whether this assumptions is

valid for the RE estimator ⇒ whether εi is correlated to explanatory
variable (xi t ) or not.
H0 : Random effects are independent of explanatory variables
HA : H0 is not true
I if the assumption does not hold, the parameter estimates wil be
baised and inconsistent
I to see how this arises
1 suppose that we have only one explanatory variable x2i t that varies
positively with yi t , and also with the error term, ωi t
2 the estimator will ascribe all of the any increase in y to x when in
reality some of it arises from the error term, resulting in biased
coefficients.

Econometrics II Chapter Four May 17, 2024 25 / 26

 End of Chapter 4!

Econometrics II Chapter Four May 17, 2024 26 / 26