Multiple Choice

1. In a quasi-experiment:

A) quasi  differences  are  used,  i.e.,  instead  of  ∆Y   you  need  to  use  (Y¯ after  − λ × Y¯ before), where 0 < λ < 1.

B) the t-statistic is no longer normally distributed in large samples.

C) the causal effect has to be estimated through quasi maximum likelihood estimation.

D) randomness is introduced by  variations in individual circumstances that make it appear   as if the treatment is randomly assigned.

2. In the case of heterogeneous causal effects, the following is NOT true:

A) in the circumstances in which OLS would normally be consistent (when E(u_i|X_i) = 0), the OLS estimator continues to be consistent.

B) OLS estimation using heteroskedasticity-robust standard errors is identical to TSLS.

C) the OLS estimator is properly interpreted as a consistent estimator of the average causal effect in the population being studied.

D) the TSLS estimator in general is not a consistent estimator of the average causal effect      if an individual’s decision to receive treatment depends on the effectiveness of the treatment for that individual.

3. Threats to internal validity of quasi-experiments include:

A) failure of randomization.

B) failure to follow the treatment protocol.

C) attrition.

D) all of the above with some modifications from true randomized controlled experiments.

4. The MSPE for the standardized predictive regression model can be written as:

A) MSPE = σ² + E[(βˆ1 − β₁)X^OOS + . . . + (βˆk − β_k)X^OOS].

B) MSPE = E[(βˆ1 − β₁)X^OOS + . . . + (βˆk − β_k)X^OOS]².

1 k

C) MSPE = σ² + E[βˆ1X^OOS + . . . + βˆk X^OOS]².

u 1 k

D) MSPE = σ² + E[(βˆ1 − β₁)X^OOS + . . . + (βˆk − β_k)X^OOS]².

5. With panel data, the causal effect:

A) cannot be estimated since correlation does not imply causation.

B) is typically estimated using the probit regression model.

C) can be estimated using the “difference-in-differences” estimator.

D) can be estimated by looking at the difference between the treatment and the control group after the treatment has taken place.

6. In a sharp regression discontinuity design:

A) crossing the threshold influences receipt of the treatment but is not the sole determinant.

B) the population regression line must be linear above and below the threshold.

C) X_i will in general be correlated with u_i.

D) receipt of treatment is entirely determined by whether W exceeds the threshold.

7. The Lasso estimator minimizes the following penalized sum of squares:

A) SLasso(b; λLasso) = Σn

B) SLasso(b; λLasso) = Σn

C) SLasso(b; λLasso) = Σn

2 k

j=1

(Y_i − b₁X_1i − ... − b_kX_ki)² + λ_Lasso ^k

(Y_i − b₁X_1i − ... − b_kX_ki)².

b².

|b_j|.

D) S^Lasso(b; λ_Lasso) = Σn (Y_i − b₁X_1i − ... − b_kX_ki)² + λ_Lasso Σk (b₁X_1i + ... + b_kX_ki).

8. The ADL(p, q) model is represented by the following equation:

A) Y_t = β₀ + β_pY_t−p + δ_qX_t−q + u_t.

B) Y_t = β₀ + β₁Y_t−1 + . . . + β_pY_t−p + δ_q + u_t−q.

C) Y_t = β₀ + β₁Y_t−1 + . . . + β_pY_t−p + δ₀X_t + δ₁X_t−1 + . . . + δ_qX_t−q + u_t−q.

D) Y_t = β₀ + β₁Y_t−1 + . . . + β_pY_t−p + δ₁X_t−1 + . . . + δ_qX_t−q + u_t.

9. Stationarity means that the:

A) error terms are not correlated.

B) probability distribution of the time series variable does not change over time.

C) time series has a unit root.

D) forecasts remain within 1.96 standard deviation outside the sample period.

10. To choose the number of lags in either an autoregression or in a time series regression model with multiple predictors, you can use any of the following test statistics with the exception     of the:

A) Augmented Dickey-Fuller test.

B) Akaike Information Criterion.

C) Bayes Information Criterion.

D) F -statistic.