Midterm III

Due Wednesday, December 9, 2009 by Noon

Your termpaper should be a complete exercise, not a contrived one as in the previous assignment. You need to make independent specification decisions, determine whether the specifications are appropriate, and determine whether there are problems with the error term. For the last midterm, you need to repeat problem 4 of the previous assignment with the primary regression in your termpaper.

As in the previous assignment, I strongly encourage you to:

• Look over your literature review before deciding on your specification.

• Try to estimate as few regression runs as possible.

1. Each of you will have a different dependent variable and set of independent variables. So that I can understand your results provide a codebook of your variables and basic descriptive statistics of each variable.
2. Hypothesize the coefficients of the expected signs for all these variables in your regression equation. I only the hypothesized sizes, but in the termpaper you should economic content of each hypothesis.
3. Choose the best combination of explanatory variables for this model and run a regression. Print out your results.
4. Based on your regression results, comment on whether you believe the model suffers from irrelevant variables or omitted variables. How do you plan to address your comments?
5. Based on your regression results, comment on whether you believe the model suffers from multicollinearity. How do you plan to address your comments?
6. If you have cross-sectional data, run a Park-Test. If the test is strongly significant, divide all independent variable and the dependent variable by the proportionality factor. Print out your results. If it is not strongly significant, move to 7.
7. Run a White test, Breush-Pagen, Cook-Weisberg test, or the built in test in your econometrics program. Print out the test and report the results.
8. If you ran a weighted least squares in 6, skip to 9. If you did not run a weighted least squares, use the econometrics program's built-in correction for heteroskedasticity. Print out your results.
9. Determine the Durbin-Watson statistic and comment on potential problems based on your result.
10. If you have time-series data and the Durbin-Watson statistic is significant, re-run your regression with one of the built-in generalized least squares.