1) Use data from PRMINWAGE.DTA, which is described in Examples 10.3, 10.9, and12.2. Use the –tsset- command to declare –year- as a time variable. Run the OLSregression of the log of the Puerto Rican employment/population ratio (lprepop)against:

? log of the effective minimum wage (lmincov)

? log of US GNP (lusgnp)

? log of PR GNP (lprgnp)

? a time trend term (year)

a) Interpret these results

b) Use equation 12.14 to test for serial correlation in the error terms of thisequation. What do you conclude?

c) Use Durbin’s alternative estimator (In Stata, it is –estat durbinalt-) to testfor serial correlation. (Note: you must re-run the regression in a) just prior torunning -estat durbinalt-). What do you conclude?

d) Run a Dickey Fuller test, with a trend term, to determine if –lprepop- has aunit root. Interpret the results.

e) Run another Dickey Fuller test, with a trend term, to determine if –lmincov- hasunit root. Interpret the results; what does this tell us about the results in a)?

f) Now rerun the regression in a) but this time using first differences of allvariables. Drop the time trend (“year”), since differencing takes care of lineartrends. Interpret the results for –lmincov- and compare them to a). Which resultsare more likely to be correct?

g) Use Durbin’s alternative test to test for serial correlation in the error termsof the differenced equation that you just ran. Do we have a problem with serialcorrelation? Explain.

h) Whether or not you see serial correlation, run the regression in differencesagain, using Newey-West standard errors. (In Stata the command is –newey-.)Specify that you want –newey- to consider 2 lags of the error term, using thelag(2) option. Does the use of robust standard errors change your conclusions aboutthe effects of the minimum wage?

i) Now use FGLS Cochran-Orcutt method to run the model (still using firstdifferences for all variables). (In Stata, this is the –prais- command with the,corc option.) Does this change your conclusions?