

Study with the several resources on Docsity
Earn points by helping other students or get them with a premium plan
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
A university economics homework focusing on estimating modified versions of the carr, markusen, and maskus (cmm) model of fdi determinants using the usfdimaster database. Instructions, results of a log-log base specification regression, and questions related to endogeneity concerns, fixed-effects estimator, first-difference estimator, and random-effects estimator.
Typology: Assignments
1 / 2
This page cannot be seen from the preview
Don't miss anything!


University of Oregon Department of Economics
ECN 607 – Econometrics II: Applied Microeconometrics Prof. Bruce Blonigen Due Date: Thursday, Feb. 5
For this homework, we will continue to use the USFDIMaster database and be estimating modified versions of the Carr, Markusen, and Maskus (CMM) model of FDI determinants using data on U.S. outbound FDI activities. We’ll begin with the following log-log base specification from homework 1:
reg lrpos lsumgdp lgdpdifsq lskdiff lht_tcost lhm_tcost lht_beri ldistance if outbound==0;
which yields the following results:
Source | SS df MS Number of obs = 892 -------------+------------------------------ F( 7, 884) = 74. Model | 1447.03983 7 206.719976 Prob > F = 0. Residual | 2441.38939 884 2.7617527 R-squared = 0. -------------+------------------------------ Adj R-squared = 0. Total | 3888.42922 891 4.3641181 Root MSE = 1.
lrpos | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------- lsumgdp | 3.729948 .8214597 4.54 0.000 2.117709 5. lgdpdifsq | -.669287 .3119789 -2.15 0.032 -1.281593 -. lskdiff | -.0924125 .1253975 -0.74 0.461 -.3385241. lht_tcost | -.0681695 .0361272 -1.89 0.059 -.1390745. lhm_tcost | 1.062487 3.970349 0.27 0.789 -6.729923 8. lht_beri | -2.67275 .2543142 -10.51 0.000 -3.17188 -2.
ldistance | -1.065663 .1085663 -9.82 0.000 -1.278741 -. _cons | 1.42479 28.47055 0.05 0.960 -54.45296 57.
1) Endogeneity Concerns.
A) The variable lht_beri is a measure of the “costs of FDI” and is an index of how a country in a given year is perceived in terms of its business climate, legal protection of foreign assets, etc. Provide an explanation for why this variable may be endogeneous. B) The USFDIMaster dataset contains variables connected with bilateral tax treaties between countries that determine tax rates on repatriated income from foreign affiliates. One of these is “ageeff”, the number of years that a country has had a bilateral tax treaty with the U.S., where “0” indicates that no treaty exists with the U.S. Explain how well you think this variable meets the criteria for an instrumental variable. Run 2SLS using this variable as an instrument and calculate
a Hausman test for endogeneity. Does the 2SLS coefficient estimate for “lht_beri” differ from the OLS estimate of this coefficient in the way you would expect given you explanation for the endogeneity of “lht_beri”? Explain. Is this a “weak” instrument? Explain. C) An alternative instrument for “lht_beri” is “cnum”, the number assigned to each country in the dataset largely based on alphabetical order. Explain how well you think this variable meets the criteria for an instrumental variable. Run 2SLS using this variable as an instrument and calculate a Hausman test for endogeneity. Which instrument do you prefer and why? Is this a “weak” instrument? Explain.
2) Fixed-Effects Estimator
A) The outbound sample we are running has both cross-sectional variation and time variation. “Cnum” indexes countries, the cross-sectional units, while “year” indexes time periods. Why is it an unbalanced sample? Explain why it may be important to control for unobserved cross- sectional heterogeneity in the context of this FDI model. B) Can one control for such unobserved cross-sectional heterogeneity using Stata’s “areg” command? If so, what type of estimator would it be? C) Find and use a Stata command to generate the fixed effects estimator. Are the fixed-effects statistically significant? Compare the fixed-effects coefficients with the OLS pooled coefficients. Why does Stata drop “ldistance” out of the fixed-effects specification? D) Use the “by”, “collapse” and “merge” commands to transform your data so that you can run the fixed effects estimator with a simple OLS command. Run that OLS on transformed data and verify that you get the same coefficient vector.
3) First-Difference Estimator
A) Transform your data into first-difference form. First differences can be generated in Stata using the command “gen xdif=x[_n]-x[_n-1]” after you have properly sorted the data by N then by T. Small “n” in the command is really referring to time periods. Be very careful to do this by cross-sectional unit, using the “by” command. Run the first-difference estimator and compare results with the fixed-effects estimator. Are they the same? Why or why not? Explain. B) Now estimate the first-differenced data using a fixed-effects estimator. Explain how would one interpret the estimated fixed-effects in this context? Are the fixed-effects statistically significant?
4) Random-Effects Estimator
A) Estimate the outbound FDI data specification with a random-effects estimator. Compare the coefficient estimates with the fixed-effects estimators’ coefficients. Why isn’t “ldistance” dropped from this specification? B) Run a Hausman test of the fixed-effects estimator to the random-effects estimator. Given the test, which estimates do you prefer and why?