Course: Advanced Panel Data Methods
Data Structures:
- Cross-sectional: information from several individuals at one period of
time
- Time-series: one individual over a period of time
- Panel data: a mix of both of the previous, repeated observations over
time, many indivduals over a period of time
Panel Data:
Number of observations: n x t (n: number of individuals; t: period of time)
_____
How do we control for individual effects between the different
individuals tested
Base line model is the regression equation
How can we control for temporal effect: dummy variables
Explanation:
1. uituit:
o This is the total error for individual ii at time tt. It's like a
"leftover" part of the model that can't be explained by the
variables.
2. γiγi:
o The individual effect. This captures factors that are specific
to each individual (ii) and do not change over time.
o Example: If the individuals are companies, this might
represent their unique culture or management style.
3. dtdt:
o The temporal effect. This represents factors that change
over time (tt) but are the same for all individuals.
o Example: A global economic trend like inflation or a pandemic
that affects everyone in the same time period.
4. vitvit:
o The error term. This is the truly random part of the error that
varies both across individuals and over time. It represents
unobserved factors that are unique to each individual and
each time point.
o Example: A sudden, unexpected event affecting one company
in one specific year.
- Behavioral models the only way is panel data
Models of Panel Data:
- Fixed effects
- Random effect
The difference is the individual effect
To choose which model, we can do a test to decide
The Hausman test: to choose which model is more appropriate in
our data set
- Endogeneity is impossible to avoid, explanatory related to
error term
- Covariance between the precious is different from zero
- We cannot use random effects model
- Here we looking the total error term not just the individual
effect
- In this case we will have bias reutls because we cannot
control
Fixed and random we assume that the covariance between
explanoty and error is 0, because individual effects are estimated
seperatrely
And in ramdon epxl and individual covar 0
- Hwen we have endogeity outr assumption doesn’t fulfill
because the covariance will be different from 0
Why do we have endogentiy?
- Omittd variable: we cannot include them because we cannot
measure or we don’t have enough data
- Although this varialbel is relevant in explaining our key
varialble
- If we cannot observe it, this information will be part of the
error term
- If this variable is correlated with the explanatory variable
we have a problem, if not it’s ok
- If business opoo explain firm value and correlated with
performance we have endogentiy check the slides
- Measurement error:
- We usually use proxies to measure variables and quantify
them, the difference between the proxy ad the truwvalue of
our dependent variable is a measurement error
3. reverse causality is a source of endogenity
- There is a double causality in the model the indpendent
explain the dependent and vice versa
Try to consider all varilable that might influence the Y
varilable
another is to lag the independent varialble here we are
forcing the direction of causality
