Question: Suppose a Ph.D. student has a sample (Yi ,Xi ,Zi : i 1, ...,n) and estimates by OLS the equation Y Z X0e
Suppose a Ph.D. student has a sample (Yi ,Xi ,Zi : i Æ 1, ...,n) and estimates by OLS the equation Y Æ Z®Å X0¯Åe where ® is the coefficient of interest. She is interested in testing H0 : ® Æ 0 against H1 : ® 6Æ 0. She obtains b®
Æ 2.0 with standard error s(b®) Æ 1.0 so the value of the t-ratio for H0 is T Æ b®/s(b®) Æ 2.0. To assess significance, the student decides to use the bootstrap. She uses the following algorithm 1. Samples (Y ¤
i ,X¤
i ,Z¤
i ) randomly from the observations. (Random sampling with replacement).
Creates a randomsample with n observations.
2. On this pseudo-sample, estimates the equation Y ¤
i
Æ Z¤
i ®ÅX¤0 i ¯Åe¤
i by OLS and computes standard errors, including s(b®
¤). The t-ratio for H0, T ¤ Æb®
¤/s(b®
¤) is computed and stored.
3. This is repeated B Æ 10,000 times.
4. The 0.95th empirical quantile q¤
.95
Æ 3.5 of the bootstrap absolute t-ratios jT ¤j is computed.
5. The student notes that while jT j Æ 2 È 1.96 (and thus an asymptotic 5% size test rejects H0), jT j Æ
2 Ç q¤
.95
Æ 3.5 and thus the bootstrap test does not reject H0. As the bootstrap is more reliable, the student concludes that H0 cannot be rejected in favor of H1.
Question: Do you agreewith the student’smethod and reasoning? Do you see an error in her method?
Step by Step Solution
There are 3 Steps involved in it
Get step-by-step solutions from verified subject matter experts
