Wild Bootstrap Tests for IV Regression
read more
Citations
Fast and wild: Bootstrap inference in Stata using boottest:
Assessing the Incidence and Efficiency of a Prominent Place Based Policy
Assessing the Incidence and Efficiency of a Prominent Place Based Policy
The Returns to College Admission for Academically Marginal Students
Are Some Degrees Worth More than Others? Evidence from college admission cutoffs in Chile
References
Numerical recipes in C
Numerical Recipes in C: The Art of Scientific Computing
Bootstrap Methods and Their Application
Instrumental variables regression with weak instruments
Numerical Recipes 3rd Edition: The Art of Scientific Computing
Related Papers (5)
Frequently Asked Questions (15)
Q2. What have the authors stated for future works in "Wild bootstrap tests for iv regression" ?
The authors also show how to apply the RE and WRE bootstraps to models with two or more endogenous variables on the right-hand side, but their performance in this context remains a topic for future research.
Q3. Why does the pairs bootstrap test not impose the null hypothesis?
Because the pairs bootstrap DGP does not impose the null hypothesis, the bootstrap t statistics must be computed ast(β̂∗j , β̂) = β̂∗j − β̂ se(β̂∗j ) .
Q4. Why is it desirable to impose a null hypothesis on the bootstrap DGP?
This is because imposing a (true) restriction makes estimation more efficient, and using more efficient estimates in the bootstrap DGP should reduce the error in rejection probability (ERP) associated with the bootstrap test.
Q5. What is the importance of the DGP used to generate the bootstrap samples?
The choice of the DGP used to generate the bootstrap samples is critical, and it can dramatically affect the properties of bootstrap tests.
Q6. What are the parameters that influence the performance of the tests?
In the context of the DGP given by (27) and (28), there are only four parameters that influence the finite-sample performance of the tests, whether asymptotic or bootstrap.
Q7. What is the smallest measure of how strong the instruments are?
One of several possible measures of how strong the instruments are is the concentration parameter, which can be written asa2 ≡ 1 σ22 π⊤W⊤MZWπ. (18)– 6 –Evidently, the concentration parameter is large when the ratio of the error variance in the reduced-form equation to the variance explained by the part of the instruments that is orthogonal to the exogenous variables in the structural equation is small.
Q8. how to construct confidence intervals by inverting bootstrap tests?
In addition, the authors discuss how to construct confidence intervals by inverting bootstrap tests based on bootstrap DGPs that impose the null hypothesis, such as the RE and WRE bootstraps.
Q9. How can the authors expect a bootstrap confidence interval to work?
It can be expected to work well whenever the rejection frequencies for tests at level α based on the relevant bootstrap method are in fact close to α.1.
Q10. Why do the authors believe that symmetric bootstrap tests are more attractive in the context of t?
The authors believe that these are more attractive in the context of t statistics than tests based on the symmetric P value (7), because IV estimates can be severely biased when the instruments are weak.
Q11. What is the striking result in Figure 6?
The most striking result in Figure 6 is that using RE, the bootstrap method which does not allow for heteroskedasticity, along with any of the test statistics that require homoskedasticity (ts, AR, and K) often leads to severe overrejection.
Q12. What is the rescaling of the REC bootstrap?
The biascorrected estimator can be used in a modified version of the RE bootstrap, called the REC bootstrap by Davidson and MacKinnon.
Q13. What is the way to proceed with the test statistics?
For test statistics that are always positive, such as the AR and K statistics that will be discussed in the next section, the authors can use (7) without taking absolute values, and this is really the only sensible way to proceed.
Q14. What is the difference between the equal-tail tests and the symmetric ones?
the equal-tail tests seem to perform better than the symmetric ones, and they are less sensitive to the values of ρ, which further justifies their choice to focus on them.
Q15. What is the way to construct a bootstrap confidence interval?
The authors now explain precisely how to construct such an interval with nominal coverage 1− α.– 19 –The method the authors propose can be used with any bootstrap DGP that imposes the null hypothesis, including the RE and WRE bootstraps.