As noted in the syllabus, an econometrics project is required for Econ 107 and is worth 10 percent of the course grade. A good project can also be the basis for a senior capstone (Econ 199, required for BA Econ and BA Quantitative Econ majors).

Project deadlines for spring 2010

• Proposal due Friday, March 12. This should be one typed page describing your topic, the equation you plan to estimate (write it out, defining dependent variables and regressors), and the data sources you plan to use.
• First draft due Friday, Apr 16 at turnitin.com. Format must be as described below.
• In-class oral presentations April 9, 16, and 30. Each student's presentation will last about 5 minutes. Each presenter must bring a paper copy of tables for the projector. Listeners should be prepared to ask questions and to make helpful suggestions.
• Final draft due Friday May 7 at turnitin.com. The final draft should respond to questions and suggestions of listeners and written comments from the instructor. Projects will be graded according to my scoring rubric.

Project steps

(1) Choose a topic. Consult your old economics textbooks for ideas. It is OK to estimate an equation similar to those in the textbook or your computer exercises, but with new data. It is even OK to estimate an equation outside economics, but talk to me first.

(2) Gather data. You will probably need at least 50 observations to get reasonably precise estimates, even with a small number of regressors. Always use government data sources whenever available. For cross-section data on states or cities, a good source is the Census's Statistical Abstract of the United States (in PDF format at www.census.gov/compendia/statab/). For time-series macroeconomic data, a good source is the St. Louis Fed's FRED database (in a variety of formats at research.stlouisfed.org/fred2/), or the Economic Report of the President appendix tables (in Excel format at www.gpoaccess.gov/eop/tables09.html). More sources are available at www.drake.edu/cbpa/econ/data.html

(3) Estimate the equations. Think carefully about what should be included in the equation and what functional form makes the most sense (causes on right side, effect on left side). For example, a demand equation normally shows quantity on the left side, and price, income, and possibly the prices of related goods on the right side; and is usually best estimated in constant-elasticity form--that is, with all variables in logarithms. For another example, a Keynesian consumption function normally shows real consumption on the left side, and real income on the right side. Always estimate several alternative equations for comparison.

(4) Test for heteroskedasticity or serial correlation. For cross-sectional data, check for heteroskedasticity related to population using a Breusch-Pagan test. If homoskedasticity is rejected, use weighted least squares. For time-series data, test for serial correlation using Durbin's alternative test or a Breusch-Godfrey test. If zero serial correlation is rejected, quasi-difference the data (if ρ is less than 0.8) or just difference the data (if &rho is greater than 0.8).

(5) Write up your results. Begin by creating the tables. Then write your narrative using the following REQUIRED FORMAT.

• Cover page: Title, your name, your email address, and the date.
• Body: 3-4 pages. Use the following section headings. Each section should be about a paragraph.
  1. Motivation: What is the underlying economic relationship you are trying to estimate? Why is it interesting?
  2. Data sources: What are the definitions of your variables? Where did you get the data? Describe the means and correlations of the variables.
  3. Specification: Write out the equation(s) to be estimated. Motivate your choice of functional form (linear, log-linear, dummy variables, etc.) Based on economic theory, what signs do you expect the coefficients to have (positive or negative)?
  4. Results: Discuss the coefficient estimates. Do the signs of the coefficients make sense? Are the coefficient estimates significantly different from zero at 5 percent? Use numerical examples to illustrate what the values of the coefficients mean. ("If X increases by 5 units, then Y increases by ...") Indicate which equation you prefer. Highlight any interesting results.
  5. Tests: Report test statistics and p-values for tests for either heteroskedasticity or serial correlation, as appropriate. If you reject homoskedasticity, report estimates using weighted least squares. If you reject no serial correlation, report estimates after differencing or quasi-differencing.
  6. Conclusions: Summarize your project in a paragraph.
• Tables: You should have at least three tables, namely
  1. Descriptive statistics of the raw data (sample means, standard deviations, minima, and maxima).
  2. Sample correlations of the raw data.
  3. Regression results, including coefficient estimates, standard errors, R², adjusted R², and sample size. Do NOT simply print out the Excel regression output page (wrong format and too ugly!). Estimates for each version of the equation should be in a separate column.
  See example of required table format.
• Bibliography of data sources: Always give the author or government agency that published the data. If the data are from a book or periodical (even if you actually got them over the web) cite the source as a book or periodical. Put the web address in parentheses like this: <http://stats.bls.gov:80/eag/eag.us.htm>. If the data are only available on the web, give the author or agency, title of web page, web address, and the date you downloaded the data.

[end of project instructions]