Use the data in BARIUM to answer this question.

(i) In Table 12.1 the reported standard errors for OLS are uniformly below those of the corresponding standard errors for GLS (Prais-Winsten). Explain why comparing the OLS and GLS standard errors is flawed.

(ii) Reestimate the equation represented by the column labeled “OLS” in Table 12.1 by OLS, but now find the Newey-West standard errors using a window g = 4 (four months). How does the Newey-West standard error on lchempi compare to the usual OLS standard error? How does it compare to the PW standard error? Make the same comparisons for the afdec6 variable.

(iii) Redo part (ii) now using a window g = 12. What happens to the standard errors on lchempi and afdec6 when the window increases from 4 to 12?