# Economics 102B Applied Project

5 7

?> œ # 
3œ!

)3 C>3 
4œ"

04 ?>4  %>

(1)

where ?> is the unemployment rate, expressed as the deviation from the “natural” rate of _ unemployment, at time >, _> is the percentage deviation of ]> from “potential” GDP, ] > , so that C C> œ "!! ‚ Ðlog ]>  log ] > Ñ, %> is an error term and # and )3 , 3 œ "ß á ß 5 and 04 , 4 œ "ß á ß 7 are parameters to be estimated. The data set "appliedprojectdata.dta", available on CourseWork, contains data on ?> , ]> , and some other variables over the period 1950:Q3 to 2003:Q3 using obvious mnemonics.1 (a) Consider first the case of 5 œ ! with 04 œ ! a 4, so that we have ?> œ #  )C>  %> where ) œ )0 for reasons of notational parsimony. What, if any, sign to you expect ) to have? Assuming that 5 œ ! and 04 œ ! a 4, would OLS yield consistent estimators of the parameters # and )? If yes, prove so being careful to state and justify your assumptions. If no, determine the size and sign of the inconsistency of the OLS estimator of ). (b) Construction of the C> time series requires estimation of “potential” GDP. One way to do this _ _ is to model ] > as an exponential trend so that log ] > œ -!  -" > where > is a time trend. A little _ C> algebra applied to the definition of C> gives log ] > œ log ]>  "!! so we can write log ]> œ -!  -" >  C> "!! (2)

showing that we can calculate the output gap, C> , as 100 times the residuals from the regression of log ]> on a constant and a linear time trend. However, some economists believe that there was a break in the GDP trend in the early 1970s so we ought to test for this possibility before calculating the output gap. To this end, define " H> œ œ 0 if >   "*(!:Q1 otherwise

1The

first two commands in your log file should be clear and use appliedprojecttwodata.

so that the trend-break version of the model in equation (2) can be written C> . "!!

log ]> œ -!  9! H>  -" >  9" H> > 

(3)

Estimate equation (3) and test the hypothesis that there is no trend break in log ]> .2 If you reject this hypothesis use 100 times the residuals from this regression as your estimate of C> . If you do not reject this hypothesis, estimate equation (2) and use 100 times the residuals from that regression as your estimate of C> . _ (c) Provide a time-series graph showing log ]> and your estimate of log ] > . (d) Provide a time-series graph showing your estimate of C> . (e) Provide a scatter plot of ?> against C> . (f) Using OLS, estimate the parameters of ?> œ #  )C>  %> Report your results in the usual way. Does your estimate of ) have the expected sign? State and interpret the V # of the regression. Test the hypothesis that ) œ !. (g) The variables oil> and def> are, respectively, the relative price of oil (\$ per barrel/GDP deflator) and Federal defense spending as a fraction of GDP. Explain why these variables might be suitable instruments for C> . (h) Explain why C>" and C>2 might also be suitable instruments for C> . (i) Estimate the “first-stage” regression of C> on oil> , def> , C>" and C>2 . Comment on the results. (j) Perform a Hausman-Wu test for the endogeneity of C> in the model in part (f) using the instruments oil> , def> , C>" and C>2 . (k) Using 2SLS, with oil> , def> , C>" and C>2 as instruments, estimate the parameters of ?> œ #  )C>  %> Report your results in the usual way. Compare them to the results in part (f). Does your estimate of ) have the expected sign? Test the hypothesis that ) œ !.

this, and the other hypothesis tests in this project, you should specify the null and alternative hypotheses, the decision rule, the formula for the test statistic, before clearly stating the outcome of the test.

2For

(l) Test for first-order serial correlation of the error term in the model in part (k). (m) Consider now the version of equation (1) with 5 œ ! as above but without the assumption 04 œ ! a 4, viz.,
7

?> œ #  )C> 
4œ"

04 ?>4  %>

(4)

To determine 7, set 7 œ " so that the model becomes ?> œ #  )C>  0" ?>"  %> and estimate this model by 2SLS, with ?>" , oil> , def> , C>" and C>2 as instruments. If you can reject the hypothesis that 0" œ ! repeat the process for 7 œ #. Keep going until you find the first 7 such that you cannot reject the hypothesis that 07 œ ! and then set 7 equal to this value minus unity, that is, to the largest value for which you can reject the hypothesis that 07 œ !. (n) Test for first-order serial correlation of the error term in your preferred model in part (m) and compare the outcome with that in part (l). (o) Reestimate your preferred model from part (m) by 2SLS, with ?>" , á , ?>7 , oil> , def> , C>" and C>2 as instruments as in part (m), using the robust standard errors option. Do you think that heteroskedasticity is an issue here? There is no need to perform any explicit test for heteroskedasticity. (p) Perform a Hausman-Wu test for the endogeneity of C> in your preferred model from part (m). (q) Compute and plot the path of ?> following a temporary one unit shock to C> in your preferred model from part (m). HINT: Can you reject L! :# œ !?