Advanced Econometrics
1. Abbreviations
FWL Frisch-Waugh-Lovell
GMM Generalized method of moments
IID Independent and identically distributed
IV Instrumental variables
LLN Law of large numbers
MM Method of moments
OLS Ordinary least squares
2SLS Two-stage least squares
2. Table of Contents
A. A review of the properties of LS estimators
1. Deriving the OLS estimator
a. OLS in scalar notation
b. OLS in matrix notation
2. Small sample properties of the OLS estimator
3. An introduction to asymptotic theory
a. Convergence in probability
b. Convergence in distribution
4. Beyond the Gauss-Markov assumptions
a. Stochastic regressors OLS
b. Dependent regressors OLS
c. Non-normal error terms OLS
d. Heteroskedasticity OLS/GLS
e. Autocorrelation OLS/GLS
f. Endogeneity GMM
B. Simulation methods
1. Motivation
2. Monte Carlo
3. Bootstrap
4. Jackknife
C. The Generalized Method of Moments
1. OLS as a Method of Moments Estimator
2. The Instrumental Variables Estimator
a. IV as a Method of Moments estimator
b. Extension: multivariate model
c. Asymptotic distribution of the IV estimator
d. Two-Stage Least Squares (2SLS) estimation
e. Choice of instruments
f. The problem of weak instruments
3. The Generalized IV Estimator
a. Identification
b. Estimating overidentified models
c. Choice of optimal weighting matrix
d. One-step GMM
e. 2SLS representation of the one-step GMM estimator
f. Two-step GMM
g. Overidentifying restrictions test
h. The problem of too many instruments
3. Case
Part 1: Theoretical Derivation
Step 1: Understand the Data Generating Process (DGP) and Model Misspecification
We start with the baseline regression model provided in the case: yt=ρyt−1+θwt+εt+Tαεt−1
This model includes:
- An autoregressive structure in yt
- An exogenous regressor wt∼i.i.d. N(0, σw2)
- A small misspecification via an MA(1)-like term that shrinks with sample size
The objective is to study the impact of this misspecification on the properties of the OLS
estimator.
Step 2: Derive the Asymptotic Distribution of the OLS Estimator
Using matrix notation and asymptotic theory, we derive the asymptotic distribution of the
OLS estimator β^=(ρ^,θ^)′.
Specifically, we analyze: VT(β^−β)→dN(Bias,Variance)
We compare this to the standard asymptotic distribution under correct specification (i.e.,
α=0) to understand how inference is affected by the model misspecification.
Part 2: OLS Estimation and Monte Carlo Simulation
Step 3: Implement the Data Generating Process in R
Use the script DataGeneratingProces.R to generate synthetic datasets under controlled
conditions.
Step 4: Estimate the Model Using OLS
Use the OLS() function from Estimators.R to estimate the parameters ρ\rho and θ\theta
from the simulated data.
Wrap this estimation inside a function compatible with the MonteCarlo() interface.
Step 5: Conduct Monte Carlo Simulations
Call the MonteCarlo() function to simulate the sampling distribution of the OLS estimator.
Use the custom DGP, with settings such as:
Sample sizes: T = 25, 50, 100, 500, 2500
Number of iterations: e.g., 1000+
Estimator: OLS
Evaluate the results in terms of:
Bias and standard error of the estimators
Jarque-Bera test for normality
Accuracy of inference using t-tests
Part 3: Inference and Bootstrap Correction
Step 6: Assess Inference Accuracy
Use the functions ttest() and BiasTest() from Inference.R to examine:
Whether the t-tests based on estimated SEs are valid
How the misspecification affects p-values and rejection rates
Step 7: Implement Bootstrap Inference
Use the functions BootStrap() or ParametricBootStrap() from Bootstrap.R to generate
bootstrapped distributions of the estimator.
Compare bootstrap-based inference to analytical inference and assess whether it improves
test size and power.
Part 4: GMM Estimation and Simulation
Step 8: Implement the GMM Estimator
Use GMM() from Estimators.R to estimate the model using valid instruments.
Typical choices include lagged values of wt, assuming they are uncorrelated with the
misspecification error.
Step 9: Simulate Properties of the GMM Estimator
Run MonteCarlo() using GMM as the estimator.
Evaluate and compare its performance to OLS in terms of:
Bias and standard errors
Robustness to misspecification
Correct size and power of inference tests
Part 5: Visualization and Interpretation
Step 10: Plot Sampling Distributions
Use PlotBetaDistribution() from Plot.R to visualize the empirical distribution of ρ^\hat{\rho}
and θ^\hat{\theta} across different sample sizes.
This helps assess asymptotic normality and convergence rates.
Step 11: Interpretation and Reporting
Write a discussion that:
Connects theoretical derivations to simulation results
Highlights how misspecification affects OLS inference
Discusses when and why GMM provides more reliable estimates
Evaluates the role of bootstrap methods for correcting inference