Econometrics 2: Comprehensive Lecture and Review Notes

0.0(0)
Studied by 0 people
call kaiCall Kai
learnLearn
examPractice Test
spaced repetitionSpaced Repetition
heart puzzleMatch
flashcardsFlashcards
GameKnowt Play
Card Sorting

1/29

encourage image

There's no tags or description

Looks like no tags are added yet.

30 Terms

1
New cards

Strict Exogeneity

The condition that the regression disturbance has a conditional mean of zero given the entire covariate matrix across all individuals and time periods: E(Ui∣X)=0E(U_i \mid X) = 0, or equivalently E(Yi∣X)=E(Yi∣Xi)=Xi′βE(Y_i \mid X) = E(Y_i \mid X_i) = X_i'\beta. It is the core requirement for the finite-sample unbiasedness of OLS.

2
New cards

Unbiasedness vs. Consistency

Unbiasedness is a finite-sample property stating that the mean of the sampling distribution equals the true parameter across repeated samples of fixed size NN (E(β^)=βE(\hat{\beta}) = \beta). Consistency is an asymptotic property stating that the estimator converges in probability to the true parameter as sample size approaches infinity (plimN→∞β^=β\text{plim}_{N \to \infty} \hat{\beta} = \beta). Neither property implies the other.

3
New cards

Downwards Bias of OLS in AR(1) Models

In stationary autoregressive time series (Yt=β0+β1Yt−1+UtY_t = \beta_0 + \beta_1 Y_{t-1} + U_t), conditioning on the entire covariate matrix XX includes future lags where YtY_t appears as an explanatory variable (Xt+1=(1,Yt)′X_{t+1} = (1, Y_t)'). Thus, E(Yt∣X)=Yt≠Xt′βE(Y_t \mid X) = Y_t \neq X_t'\beta, violating strict exogeneity and inducing finite-sample downward bias.

4
New cards

Role of N1/2N^{1/2} in OLS Asymptotic Normality

The scaling factor N1/2=NN^{1/2} = \sqrt{N} balances the rate of convergence of the standardized sum 1N∑i=1NXiUi\frac{1}{\sqrt{N}} \sum_{i=1}^N X_i U_i. Without N\sqrt{N}, the difference (β^−β)(\hat{\beta} - \beta) collapses to a degenerate point mass at zero; scaling by NN causes the variance to diverge to infinity. The exponent 1/21/2 stabilizes the variance to yield a non-degenerate Gaussian limit distribution.

5
New cards

Heteroskedasticity-Consistent (HC) Covariance Matrix

The sandwich variance estimator V^HC=(X′X)−1(∑i=1NU^i2XiXi′)(X′X)−1\hat{\mathcal{V}}_{HC} = (X'X)^{-1} \left( \sum_{i=1}^N \hat{U}_i^2 X_i X_i' \right) (X'X)^{-1} where U^i=Yi−Xi′β^\hat{U}_i = Y_i - X_i'\hat{\beta}. It yields standard errors that are asymptotically valid under both conditional homoskedasticity and arbitrary forms of conditional heteroskedasticity without changing the coefficient estimates.

6
New cards

Instrument Relevance Condition

The requirement that excluded instruments are correlated with the endogenous regressors after partialling out all included exogenous covariates. Formally, the matrix of population regression coefficients from the first-stage equation must have full column rank: rank(ΣZX)=K\text{rank}(\Sigma_{ZX}) = K. It is assessed via a joint first-stage FF or Wald test on the excluded instruments.

7
New cards

2SLS HC Residual Computation Rule

When constructing the heteroskedasticity-consistent variance-covariance matrix for 2SLS, residuals must be computed using the actual structural explanatory variables: U^i=Yi−Xi′β^2SLS\hat{U}_i = Y_i - X_i'\hat{\beta}_{2SLS}. Using first-stage fitted values X^i\hat{X}_i yields invalid residuals and inconsistent standard errors because the actual structural disturbance is Ui=Yi−Xi′βU_i = Y_i - X_i'\beta.

8
New cards

Over-Identifying Restrictions (OIR / Sargan / Hansen J) Test

A diagnostic test available when the number of excluded instruments exceeds endogenous regressors (M>KM > K). It evaluates H0:E(ZiUi)=0H_0: E(Z_i U_i) = \mathbf{0} by regressing structural residuals U^i\hat{U}_i on all instruments ZiZ_i (J=NRU^2→dχM−K2J = N R^2_{\hat{U}} \stackrel{d}{\to} \chi^2_{M-K}). Failure to reject supports over-identifying restrictions but does not prove exogeneity of the baseline just-identified model.

9
New cards

Regression-Based (Wu-Hausman) Endogeneity Test

A test for regressor endogeneity conducted by estimating the first-stage regression, extracting the residual vector V^i\hat{V}_i, and including it as an additional explanatory variable in the structural regression estimated via OLS: Yi=Xi′β+δV^i+εiY_i = X_i'\beta + \delta \hat{V}_i + \varepsilon_i. A rejection of H0:δ=0H_0: \delta = 0 indicates endogeneity, requiring 2SLS over OLS.

10
New cards

Indirect Least Squares (ILS)

A method to recover structural parameters in just-identified simultaneous equations systems by estimating the reduced-form equations via OLS and algebraically solving for the structural parameters as ratios and combinations of reduced-form coefficients (e.g., α^12=π^12/π^22\hat{\alpha}_{12} = \hat{\pi}_{12} / \hat{\pi}_{22}).

11
New cards

Cluster-Robust (HAC) Panel Covariance Matrix

A panel variance estimator V^cluster=(X′X)−1(∑i=1NXi′U^iU^i′Xi)(X′X)−1\hat{\mathcal{V}}_{cluster} = (X'X)^{-1} \left( \sum_{i=1}^N X_i' \hat{U}_i \hat{U}_i' X_i \right) (X'X)^{-1} that groups all TT residuals for individual ii into a vector U^i\hat{U}_i. The cross-product matrix U^iU^i′\hat{U}_i \hat{U}_i' accounts for arbitrary heteroskedasticity across individuals and arbitrary serial correlation (autocorrelation) over time within each cross-sectional unit.

12
New cards

Long-Run Propensity (LRP) in FDL Models

In a Finite Distributed Lag model of order qq (Yit=β0+∑j=0qδjXi,t−j+UitY_{it} = \beta_0 + \sum_{j=0}^q \delta_j X_{i,t-j} + U_{it}), the Long-Run Propensity or long-run multiplier measures the total cumulative change in the outcome resulting from a permanent, one-unit increase in XX, given by LRP=∑j=0qδjLRP = \sum_{j=0}^q \delta_j.

13
New cards

First-Differences (FD) Estimator

An estimator that subtracts period (t−1)(t-1) observations from period tt: ΔYit=ΔXit′β1+ΔVit\Delta Y_{it} = \Delta X_{it}'\beta_1 + \Delta V_{it}. It completely eliminates unobserved, time-invariant individual heterogeneity αi\alpha_i, but drops all time-invariant observed regressors (ΔZi=0\Delta Z_i = \mathbf{0}) and reduces the effective sample size from NTNT to N(T−1)N(T-1).

14
New cards

Fixed Effects (Within) Estimator

An estimator obtained by demeaning all variables by their entity-specific time averages: (Yit−Yˉi)=(Xit−Xˉi)′β+(Vit−Vˉi)(Y_{it} - \bar{Y}_i) = (X_{it} - \bar{X}_i)'\beta + (V_{it} - \bar{V}_i). It is algebraically identical to the Least Squares Dummy Variable (LSDV) estimator and relies exclusively on within-unit variation over time to identify β\beta, conditioning out individual intercepts αi\alpha_i.

15
New cards

Incidental Parameters Problem

In short panels where N→∞N \to \infty with fixed TT, estimating individual intercepts αi\alpha_i directly (as in LSDV) requires estimating NN separate parameters. Because each αi\alpha_i is estimated using only TT observations, α^i\hat{\alpha}_i is unbiased but inconsistent, even though the within slope estimator β^FE\hat{\beta}_{FE} is consistent.

16
New cards

Identification Failure in Two-Way Fixed Effects (TWFE)

Double-demeaning in TWFE models (Yit∗=Yit−Yˉi⋅−Yˉ⋅t+Yˉ⋅⋅Y_{it}^* = Y_{it} - \bar{Y}_{i\cdot} - \bar{Y}_{\cdot t} + \bar{Y}_{\cdot\cdot}) eliminates any regressor that can be decomposed additively into an individual component and a time component (Xit=ai+btX_{it} = a_i + b_t). Examples include Ageit=t−BirthYeariAge_{it} = t - \text{BirthYear}_i and potential experience, which become exactly zero after transformation.

17
New cards

Nickell Bias

The severe downward finite-sample bias of OLS applied to first-differenced dynamic panel models (ΔYit=ρΔYi,t−1+ΔVit\Delta Y_{it} = \rho \Delta Y_{i,t-1} + \Delta V_{it}). Because Yi,t−1Y_{i,t-1} contains the disturbance shock Vi,t−1V_{i,t-1}, the differenced regressor is correlated with the differenced error: Cov(ΔYi,t−1,ΔVit)=−Var(Vi,t−1)≠0\text{Cov}(\Delta Y_{i,t-1}, \Delta V_{it}) = -\text{Var}(V_{i,t-1}) \neq 0.

18
New cards

Anderson-Hsiao and Arellano-Bond Estimators

Instrumental variable solutions for dynamic panels. Anderson-Hsiao instruments the differenced lag ΔYi,t−1\Delta Y_{i,t-1} using lagged levels (Yi,t−2Y_{i,t-2}) or lagged differences (ΔYi,t−2\Delta Y_{i,t-2}). Arellano-Bond generalizes this via GMM, utilizing all available linear orthogonality conditions from deeper historical lags (Yi1,…,Yi,t−2Y_{i1}, \dots, Y_{i,t-2}) as time progresses.

19
New cards

Optimality of the Conditional Expectation Function (CEF)

Under expected quadratic loss (E[(Y−m(X))2]E[(Y - m(X))^2]), the CEF m(X)=E(Y∣X)m(X) = E(Y \mid X) is the Minimum Mean Squared Error (MMSE) predictor over all possible functions. Population OLS is the Best Linear Predictor (BLP), representing the MMSE approximation constrained to linear functions X′bX'b.

20
New cards

Decomposition of MSPE

Mean Squared Prediction Error on out-of-sample data decomposes mathematically into three additive terms: MSPE=σU2+[Bias(Y^i)]2+Var(Y^i)MSPE = \sigma_U^2 + [\text{Bias}(\hat{Y}_i)]^2 + \text{Var}(\hat{Y}_i), representing irreducible random noise, squared model approximation bias, and parameter estimation variance, respectively.

21
New cards

Conditional Mean CI vs. Individual Prediction Interval

A 95% confidence interval for the conditional mean (C′β^±1.96⋅se(C′β^)C'\hat{\beta} \pm 1.96 \cdot se(C'\hat{\beta})) captures only parameter estimation uncertainty. A 95% prediction interval for an individual (C′β^±1.96⋅se(C′β^)2+RMSE^2C'\hat{\beta} \pm 1.96 \cdot \sqrt{se(C'\hat{\beta})^2 + \widehat{RMSE}^2}) also incorporates the irreducible variance of the idiosyncratic individual shock U0U_0, making it substantially wider.

22
New cards

Jensen's Inequality in Log Retransformations

Because the exponential function is strictly convex, E(exp⁡(Ui)∣Xi)>exp⁡(E(Ui∣Xi))=exp⁡(0)=1E(\exp(U_i) \mid X_i) > \exp(E(U_i \mid X_i)) = \exp(0) = 1. Consequently, the naive retransformed predictor Y^i=exp⁡(Xi′β^)\hat{Y}_i = \exp(X_i'\hat{\beta}) systematically underestimates the true conditional expectation E(Yi∣Xi)E(Y_i \mid X_i), requiring a retransformation adjustment.

23
New cards

Duan's Smearing Estimator

A non-parametric retransformation adjustment that does not assume normality of disturbances. It estimates the multiplicative correction factor as the sample average of exponentiated residuals: a~0=1N∑i=1Nexp⁡(U^i)\tilde{a}_0 = \frac{1}{N} \sum_{i=1}^N \exp(\hat{U}_i), yielding level predictions Y^i=a~0exp⁡(Xi′β^)\hat{Y}_i = \tilde{a}_0 \exp(X_i'\hat{\beta}).

24
New cards

Ridge vs. LASSO Regularization

Ridge (L2L_2 penalty: λ∑bk2\lambda \sum b_k^2) shrinks coefficients continuously toward zero via closed-form solution (X′X+λI)−1X′Y(X'X + \lambda I)^{-1}X'Y, handling multicollinearity but never setting coefficients to zero. LASSO (L1L_1 penalty: λ∑∣bk∣\lambda \sum |b_k|) features a non-differentiable kink at zero, driving small coefficients exactly to zero to produce sparse models and automatic variable selection.

25
New cards

Selection of Shrinkage Tuning Parameter (\lambda)

λ\lambda cannot be selected by minimizing in-sample SSR because in-sample SSR is mechanically minimized at λ=0\lambda = 0 (OLS). It must be chosen by evaluating out-of-sample predictive performance over a grid of candidate values using MM-fold cross-validation to find the value that minimizes MSPE^CV(λ)\widehat{MSPE}_{CV}(\lambda).

26
New cards

Residual Autocorrelation Theorem

By the Law of Iterated Expectations, prediction errors from a correctly specified conditional expectation function cannot be autocorrelated at any lag: Cov(Ut,Ut−j)=E[Ut−jE(Ut∣Ft−1)]=0\text{Cov}(U_t, U_{t-j}) = E[U_{t-j} E(U_t \mid \mathcal{F}_{t-1})] = 0. Statistically significant autocorrelation in regression residuals indicates lag underspecification.

27
New cards

Akaike Information Criterion (AIC)

A model selection metric defined as AIC(p)=log⁡(SSRpn)+2(p+1)nAIC(p) = \log\left( \frac{SSR_p}{n} \right) + \frac{2(p+1)}{n}. It balances in-sample goodness-of-fit against a linear parameter penalty. Rather than targeting structural model-selection consistency, AIC targets predictive optimality by minimizing out-of-sample forecast error variance.

28
New cards

Multi-Step Forecasting Failure of ARDL Models

While an ARDL(p,q)ARDL(p, q) model can produce one-step-ahead forecasts using lagged covariates, it cannot generate iterated forecasts for horizons h>1h > 1 because it lacks structural equations describing the future evolution of the auxiliary predictors E(Xn+1∣Fn)E(X_{n+1} \mid \mathcal{F}_n). A Vector Autoregression (VAR) solves this by modeling all variables simultaneously.

29
New cards

Granger Causality

Variable XX Granger-causes variable YY if historical lags of XX contain statistically significant information that helps predict future values of YY beyond the information contained in past lags of YY alone. It evaluates predictive forecasting content via a joint Wald/F test, not structural or economic causality.

30
New cards

Empirical Size vs. Power in Monte Carlo Experiments

Empirical size is the rejection frequency of a test evaluated when the null hypothesis is true (b=β0b = \beta_0); an unbiased test should produce a rejection rate within sampling error of nominal size (α=0.05\alpha = 0.05). Empirical power is the rejection frequency when the null hypothesis is false (b≠β0b \neq \beta_0); for a consistent estimator, power converges toward 1.01.0 (100%100\%) as sample size grows.