To get the most out of this class, you will need to do more than just read the text, do the homework, and study for the exams. You need to get your hands “dirty”, so to speak, by building an econometric model, so as to gain a deeper understanding of the issues involved in the “art” of econometrics. You are thus expected to write an econometric research paper with another student or two on a topic of your own choosing (as long as it is about economics, and employing methods learned in this class). This will involve:
1) identifying a problem that can be analyzed with econometric techniques,
2) formulation a model with testable hypothesis,
3) collecting data,
4) estimating your model using the techniques covered in this class, and
5) interpreting the results.
While published papers might not go to as much details as I am asking you to do here, these extra requirements are to demonstrate (to me) how much you have learned in this class. You should thus understand the reasons behind this outline, and follow it as much as you can. Points might be taken otherwise. (See the summary at the end of this outline.)
Your main paper should not be more than 6 pages long, and no more than 10 pages including all graphs, tables and data appendices. You might have an extra cover sheet with your names, title of your paper, etc., and number all pages except the cover page.
1) Introduction / Review of the Literature: (1/2 to 1 page)
• In this section you introduce your topic, state why others might be interested in the topic, or your motivation for studying it.
• You should then summarize, in a few paragraphs, what other researchers using econometrics methods have found on this topic (using econometric techniques), or a topic that is very closely related to yours, following author (year) or author (year, p.xxx) style. Be sure to provide references for them at the end of your paper. You must have at least one paper from a peer-reviewed journal.
• Finally, state your own hypothesis.
I can’t stress enough how important this review of literature is for your paper. To identify a problem that can be analyzed with econometric technique, you really need to carefully study what other scholars have done. It will give you ideas about what are the issues involved in the area that you are interested. In other words, don’t reinvent the wheel. You are not likely to be the first person whom ever thought about this problem. Find out what others have done first, before getting too deeply into your own study.
You might use http://scholar.google.com/ to search for papers on a topic of your interest, or look into the following list of possible sources that might give you ideas and/or data.
Research Paper in Economics http://www.repec.org,
Econlit http://www.econlit.org,Jstor http://www.jstor.org,
ProQuest http://www.umi.com/proquest,SSRN http://www.ssrn.com/ern,
NBER http://www.nber.org,IMF http://www.imf.org,
WTO http://www.wto.org/, World Bank http://www.worldbank.org/
2) Model and Data: (1 to 2 pages)
• This is where you write down your model using the format shown for all equations as in your textbook. Use the subscripts and Greek letters as indicated. Markdown the expected sign of each variable under its name. Use informative names for your variables instead of X1 to Xk. Don’t be lazy, and make your reader a detective! For example, for cross sectional models (only if you can’t make it into a panel data model) use
log(wagei)=b0 + b1educi + b2experi + b3femali +ui
where wage is the hourly wage in dollars,
educ is the year of schooling a worker has,
exper is the years of working experience a worker has, and
female is an indicator variable if the person is a female, 0 otherwise.
and for time series models, use
Dlog(GDPt) =b0 + b1Dintt + b2Dlog(M2t) + b3Dlog(FDIt) + ut.
where GDPt is the Gross Domestic Production (million dollars) at time t,
intt is the interest rate (3 month Treasurer rate) at time t,
M2t is the money supply (million dollars) for time t, and
FDIt is the Foreign Direct Investment (million dollars) at time t.
Or, for panel data mode use
log(wageit)=b0 + b1educit + b2experit + b3kidsit +ai +τt +uit
where wageit is the hourly wage in dollars for worker i at time t,
educit is the year of schooling a worker i has at time t,
experit is the year of work experience a worker i has at time t,
kidsit is the number of kids a worker i has at time t.
• You should put variable definition, their unit of measurements in a table for easy reference.
• You then explain the underlying economic theory behind your variable selection and your hypothesized signs of the coefficients if you did not indicate under your equation.
• You should pay close attention to the unit of measurement of your variables and normalize them so as to be compatible with each other. For time series data, be sure to deflate them properly. For cross sectional data, be sure to think about the correct functional forms for the variables, and see if you want to match gross of one variable with gross, average, or percentage, etc. of another variable, so that the variables will be generally compatible (and will not be affected by entity merging of splitting), to reduce heteroskedasticity, as well as for the ease of interpretation.
• You should have at least one nonlinear functional form, and justify your linear or nonlinear functional forms by a few scatter plots (included in the appendix).
• Next, you discuss the data sources for your model (you must use the latest available data, and no dataset from any textbook is allowed for this paper). It should have 30 or more data points (over 100 for time series) and 3 or more independent variables (lags or powers of a variable don’t count), and from more then one data source (or data table). Furthermore, if your model has a time series component you should keep as much data as the data series is available. Otherwise, you must justify why you are using only a selected section of the time series data.
• If you collected the data yourself, be sure to document the data collection methodology carefully.
• You should explain all data transformation, smoothing or any adjustment performed in this section.
3) Analysis and Results: (1 to 2 pages)
• You should then present your regression results in standard format, as shown below. You should include estimated coefficients with associated (robust) standard errors (use ‘*’, ‘**’, or ‘***’ to indicate significance of 10%, 5% or 1% level respectively), R2, 2, F-statistics, D.W -statistics (for time series only), as well as the sample size n. They should be independent of the software used. For example, using HTV dataset resulted in:
=0.174+ 0.116educ+0.034exper+ 0.0083motheduc+ 0.021fatheduc
(.196) (.0105)*** (.00688)*** (.0086) (.0063)***
n=1230, R2=0.19, 2=0.188, SSR=.534, F=72.2
Do not report the results in EViews, STAT, SAS, R, or SPSS output format here (instead, you must include the original output without editing in the appendix, including the STAT or R commends, or SAS log).
• Generally, you should only present your final model or models, and briefly describe how you arrived at this, leaving the detail of modeling steps to the appendix if need be. If you do need to compare results from different regressions, it is best to present then in a tabular form, and be sure not to let your table run over multiple pages. If you must, it should have new headings to indicate such.
• It is important that you present the regression results in laymen’s terms. You will need to discuss the following points in your narrative:
a) State the significance of each of the independent variables in your model using t-test, or p-value (only if you did not mark the standard error with appropriate asterisks ‘*’ as in the example above), and relating your findings to the hypothesis stated in section 2 above.
b) Report any statistical test conducted for the variable and/or model selection, such as F-test, serial correlation test, unit root test, and the test for heteroskedasticity, etc. This is one of the most important sections of your report, for it is where you demonstrate your mastery of the statistical concepts learned in the class; demonstrate your ability to apply those concepts in actual situations. However, don’t conduct all tests you can find, just to show that you can do them. Conducting irrelevant tests exposes your lack of understanding! And it might event herm your grade!
c) Interpret the meaning of the transformed variables, if you have any.
d) If you use intercept dummies or slop dummies, justify their usage.
e) Interpret the meaning of the coefficient estimates in laymen’s terms and be specific about what is being held constant (for I want to see how much do you understand the results from your own model).
f) Could there be bias in estimated coefficients due to omitted variables? What variables might have been omitted and what are the likely effects on the estimated coefficients?
g) Are the model assumptions justified, such as E(u|X)=0, or E(uit|Xit)=0?
h) If your model is a cross-sectional one, you should test for the presence of heteroskedasticity. If it is found to be significant, you should discuss its consequence and state any corrective measures that you’ve taken.
i) If your model is a time series one, you should test and correct for the problem of nonstationarity, serial correlation, etc.
4) Conclusions: (1/2 page) You should summarize your findings here, such as
• What conclusions can you make based on your study.
• How is it compared to those you’ve found in your review of literature?
• Does your analysis support your initial hypotheses?
• Based on your model, what policy recommendation could you make, if any?
• What suggestion you might have for future research?
5) Bibliography: List all sources that you used in your project. If the source is in Chinese, only translate the authors’ names into English. It includes references you made in the review section of your paper, and any other written sources that you have consulted. List items alphabetically by the last name of the author or the name of the organization that produced the item (not its web address). List each item’s name (year), “Title of Article” /Title of a book, Title of journal or magazine volume, page number /city of publication: publisher. If you use Internet as your source, it should not be the only source.
6) Appendices: (1-2 pages) You should include in the appendices the following (some of the items asked here is to help your instructor to understand your model, not to be included for publication):
a) Clear indication of the data source, so that others can replicate your results without difficulty.
b) A table of the first 5 and last 5 observation from your dataset used for regression with clear labels for each column, even for panel data.
c) A table of summary descriptive statistics with clear labeling for each variable and its unit of measurement. (It might help you detect some data errors)
d) A few informative scatter plots of some of the key non-dummy independent variables verses the dependent variable used in your model. For panel data, you either plot the combined data, or plot only for a particular year.
e) The key regression outputs from EViews or any other software you used.
f) Any other model run or tests you’ve conducted.
Note: You will not be graded on how will your model performs. Instead, you will be evaluated by how well you use and explain the statistical test employed and how well you convey and articulate your ideas. Outstanding work in review of literature, data collection, using advanced models learned in this class, or thoughtful statistical tests might lead to extra points. Using color in your printout adds no points to your grade in most cases, and any distracting graphics might even harm your grade.
Remember your reader (that is me) don’t have a lot of time to read and grade your paper, so don’t make your reader a detective in finding out the nature of your data, the variables you’ve used (give meaningful and easy to understand names, for example, use GDP/Pop rather than GDPPC, or log(GDP) rather than LGDP), and what is the aim of your paper, etc.
Grading Criterion (Tentative):
Introduction / Review of the Literature (10%)
Introduction to your project / Motivation (how clear is the goal of your project) Literature review (how is your work compare to others)
Model and Data (45%)
List of relevant variables (whether or not you can find them)
Data collection (effort)
Variable format (proper functional forms)
Analysis and Results (35%)
Presentation and interpretation of your regression result(s)
Stat tests (Heteroskedasticity, serial correlation, F-test, etc.)
Analysis of omitted variable bias
Style and Presentation (5%)
Summary of major point deductions if this instruction was not followed:
Plagiarism is 100 points off!
Missing review of literature is 20 points off!
Using out dated data is up to 20 points off.
Independent variables less then 3 is 10 points off for each missing variable.
Observations less then 30 is 1 point off for each &lt;30, with a maximum of 5 points off.
Report the regression results in original software output format (including data sample, descriptive statistics and scatter plots, etc.) within the main report is 5 points off.
Missing unedited original output in the appendix is 5 points off.
Use stepwise regression is 5 points off.
Dragging your reader through all of your intermediate regression results is 5 points off.
Missing a section on interpret the meaning of the coefficient estimates is 5 points off.
Page numbers &gt;10 is 1 point off for each page, with a maximum of 5 points off.
- Among other benefits, we guarantee:
Essays written from scratch – 100% original,
Competitive prices and excellent quality,
24/7 customer support,
Priority on customer’s privacy,
Unlimited free revisions upon request, and
Plagiarism free work.
Providing Quality University Papers , written from scratch,
delivered on time, at affordable rates!