Final Project
Applied Econ 440.614: Macroeconometrics
Spring, 2025
Your final project will have you use the tools we’ve developed to analyze the sources of macroeconomic fluctuations. In short, you will choose a macro dataset, and you will use that data to identify at least 2 interpretable shocks and illustrate their effects on the economy. You are allowed to work in groups, of up to five people. I am teaching two sections of this class, and I have no objection if you want to work with people who are registered the other section.
1 Expectations
1. You should select a dataset that we have not studied in class. Also, if an existing paper uses the same econometric methods on the same dataset, then you should find a new dataset. The data we’ve looked at in class has been mostly post-war U.S. macroeconomic data. Consequently, one way to proceed is to use data from another country, applying the methods that we may have already seen used on U.S. data. Make sure that any dataset you select is sufficiently long to generate usable results. (For monthly or quarterly data, I recommend at least 25 or 30 years. Keep in mind, you may lose observations from differencing or omitting outliers, such as pandemic-era observations.) You’re not forbidden from using U.S. data, but you can’t recycle something that we did in class, or do something that is virtually the same as what appears in an existing paper.
2. Besides analyzing your shocks of interest, include a discussion of the data itself. Explain briefly what each series measures, and where the data come from. Summarize whether the key variables of interest are pro-cyclical or counter-cyclical. (To isolate the business-cycle variation in the data, you may benefit from using the detrending methods we discussed in class.)
3. With your dataset, use macroeconometric methods to identify at least two types of structural shocks. Examples include shocks to monetary policy, fiscal policy, or technology. Notice how each of those examples is an interpretable economic force, not just regression residual or a forecasting error. The primary tools that empirical macroeconomists have used to identify such shocks are structural vector autoregressions (VARs), and you’re welcome to use the the structural VAR identification schemes that we discussed in class. If you’re feeling more entrepreneurial, you’re also welcome to explore alternative methods, such as local projections, vector error correction models, or DSGE models. Usually, you’ll have to use construct two separate models to identify the two shocks. Even if you only use VAR methods, the VAR that you use to estimate, say, a monetary shock will probably be different from the VAR that you use to estimate a technology shock. No matter what tools you use, you should explain how you can identify the shocks of interest using the data at your disposal.
4. Explain your specification choices. There are many decisions that go into building a suitable model. For example, which variables do you include? Do you transform any variables (e.g., taking logs)? Do the data enter in levels or differences? How many lags to you include? Do you include any exogenous variables (e.g., linear time trends, seasonal indicators, dummy variables for outliers)? These kinds of decisions may seem minor, but they sometimes matter for the results.
5. Your results should convey the role of the identified shock in the economy that you are studying. This assignment does not come with a checklist of model outputs to generate. Instead, you have to exercise your discretion to determine which model outputs are interesting and informative. That being said, there are several model outputs that are often worth investigating: impulse response functions, forecast-error variance decompositions, historical shock realizations, or historical decompositions. The output of your model does not speak for itself, so comment on your results. For example, do these shocks seem important for the economy that you’re studying? Do the results align with what you would expect from economic theory?
2 Deliverables
1. Proposal (10%): By Wednesday, April 23, send me an e-mail telling me who is in your group, what data you plan to use, and a preliminary indication of what kinds of shocks you want to identify. If you intend to use a programming language other than Matlab for the econometric analysis, you also need to mention that. The proposal does not have to be anything long or complicated. For the data, tell me what variables you have from what source(s), the frequency of the data (e.g., monthly or quarter), and time sample. When you submit your proposal, you don’t need to have finalized your choice of shocks or your choice of models. However, it needs to be clear that the data you’ve found will allow you to identify at least two shocks. For example, if you want to study tax shocks in a country, you need to make sure you can find data on taxes for that country. If your proposal does not seem viable for some reason, then I will ask you to revise and resubmit it.
2. Report (50%). Your final submission is due Wednesday, April 7, at noon. Please e-mail your report directly to me, with all members of your group CCed. This document should be a single PDF that summarizes what you did, and what you found. Like a research paper, it should be organized in sections. Include an introduction with a brief summary of what you’ll be studying. Include a summary of the data, a discussion of identification, an explanation of your choices in setting up your statistical models, and the results themselves. Cite sources appropriately, using one of the major bibliographic formats.
3. Replication Package (40%). The replication package should be submitted in the same e-mail as your written report, due Wednesday, April 7, at noon. This should contain all data and code necessary to produce all results in your report. For this component of the project, half of your grade will be for reproducibility, and half of it will be for correctness. I should be able to put all of your materials into a directory on my own computer, press “run,” and have your results generated in front me. In addition to the code and data, the replication package should also contain a “read me” file that explains what each file does. You should also include a data appendix that explains where you got each data series, so that someone could reassemble your dataset from scratch, if needed. The code itself should be well documented, so that I can read it and understand what each component is doing. The Matlab code that I have posted on Canvas gives you an idea of how much in-line documentation is appropriate.