Topic: Advanced Panel Data Regression Techniques
Panel Data Structure:
● Observations on multiple entities (individuals, firms, countries) over multiple time
periods.
● Notation:
○ yity_{it}yit: Dependent variable for entity iii at time ttt.
○ XitX_{it}Xit: Vector of independent variables.
Fixed Effects Model (FEM):
● Controls for unobserved heterogeneity by allowing each entity to have its own
intercept.
● Model Specification: yit=αi+β′Xit+ϵity_{it} = \alpha_i + \beta' X_{it} +
\epsilon_{it}yit=αi+β′Xit+ϵit
● Estimation Methods:
○ Least Squares Dummy Variable (LSDV):
■ Include dummy variables for each entity.
■ Issue: May cause degrees of freedom loss with many entities.
○ Within Transformation:
■ Subtract the mean over time for each entity:
yit−yˉi=β′(Xit−Xˉi)+(ϵit−ϵˉi)y_{it} - \bar{y}_i = \beta' (X_{it} - \bar{X}_i) +
(\epsilon_{it} - \bar{\epsilon}_i)yit−yˉi=β′(Xit−Xˉi)+(ϵit−ϵˉi)
Random Effects Model (REM):
● Assumes individual-specific effects are random variables uncorrelated with the
regressors.
● Model Specification: yit=α+β′Xit+μi+ϵity_{it} = \alpha + \beta' X_{it} + \mu_i +
\epsilon_{it}yit=α+β′Xit+μi+ϵit
○ μi\mu_iμi: Random individual effect.
● Estimation:
○ Generalized Least Squares (GLS).
○ Variance Components:
■ Need to estimate the variance of μi\mu_iμiand ϵit\epsilon_{it}ϵit.
Hausman Test:
● Tests the null hypothesis that the individual effects are uncorrelated with the
regressors.
● Procedure:
○ Estimate both FEM and REM.
○ Compute the test statistic:
H=(β^RE−β^FE)′[Var(β^FE)−Var(β^RE)]−1(β^RE−β^FE)H = (\hat{\beta}_{RE}
- \hat{\beta}_{FE})' [Var(\hat{\beta}_{FE}) - Var(\hat{\beta}_{RE})]^{-1}
(\hat{\beta}_{RE} -
\hat{\beta}_{FE})H=(β^RE−β^FE)′[Var(β^FE)−Var(β^RE)]−1(β^RE−β^FE)
○ Decision Rule:
■ If HHH is significant, reject REM in favor of FEM.
Dynamic Panel Data Models:
● Include lagged dependent variables to capture persistence.
● Model Specification: yit=γyi,t−1+β′Xit+αi+ϵity_{it} = \gamma y_{i,t-1} + \beta' X_{it}
+ \alpha_i + \epsilon_{it}yit=γyi,t−1+β′Xit+αi+ϵit
● Challenges:
○ The lagged dependent variable is correlated with the error term due to the
fixed effects.
● Estimation Techniques:
○ Arellano-Bond Estimator:
■ Difference the equation to eliminate αi\alpha_iαi.
■ Use yi,t−2y_{i,t-2}yi,t−2and earlier as instruments for Δyi,t−1\Delta
y_{i,t-1}Δyi,t−1.
○ System GMM (Blundell-Bond):
■ Combines equations in levels and first differences.
Advanced Topics:
● Cross-Sectional Dependence:
○ When entities are interrelated (e.g., countries in a trade network).
○ Tests:
■ Pesaran's CD test for cross-sectional dependence.
● Heteroskedasticity and Autocorrelation:
○ Standard errors may need to be adjusted.
○ Robust Standard Errors:
■ Clustered at the entity level.
Non-Stationary Panel Data:
● Unit Root Tests:
○ Im-Pesaran-Shin (IPS) test.
○ Levin-Lin-Chu (LLC) test.
● Cointegration:
○ Pedroni's test for panel cointegration.
Example Study:
● Objective: Analyze the impact of R&D expenditure on firm productivity.
● Model:
ln(Productivityit)=αi+β1ln(R&Dit)+β2ln(Capitalit)+β3ln(Laborit)+ϵit\ln(Productivity_
{it}) = \alpha_i + \beta_1 \ln(R\&D_{it}) + \beta_2 \ln(Capital_{it}) + \beta_3
\ln(Labor_{it}) +
\epsilon_{it}ln(Productivityit)=αi+β1ln(R&Dit)+β2ln(Capitalit)+β3ln(Laborit)+ϵit
● Estimation:
○ Use FEM to control for firm-specific characteristics.
○ Test for serial correlation in errors.
Practical Tips:
● Software Commands:
○ Stata: xtreg, xtabond, xtserial.
○ R: plm package functions plm(), phtest().
Assignments:
● Empirical Project:
○ Use a panel dataset to estimate the determinants of economic growth.
○ Apply fixed and random effects models and conduct the Hausman test.
● Reading:
○ Baltagi's "Econometric Analysis of Panel Data," Chapters 4-6.
● Exercises:
○ Simulate panel data and practice estimating models under different
assumptions.
●
