Final Project Objectives

Table of Contents

For your final project, your group will select one of two tracks. Both tracks are weighted equally and graded on the same rubric: clarity of question, quality of code, and quality of deliverables.

Project Tracks

Feature Track A: Open Data Project Track B: Paper Reproduction
Focus Design and execute an original empirical project using public data. Duplicate the results of a prior study using the same data and procedures.
Requirements Pose a clear economic question and include an econometric baseline vs. ML comparison. Re-code the analysis in your own environment (computational reproducibility).
Deliverable An original analysis with a discussion on what worked and what didn’t. Regression tables and figures presented alongside the original paper’s results.

Use of AI and Code Standards

You may use AI assistants (Claude, ChatGPT, Copilot, etc.) for this project. However, all AI-assisted code must meet the following strictly enforced standards:

  • Modular Code: No function may be longer than 10 lines of code.
  • Professionalism: No emojis anywhere in code, comments, or notebook text.
  • Documentation: Every code cell must be preceded by a markdown cell explaining the “what” and “why.”
  • Reproducibility: The notebook must run start-to-finish with no manual intervention (“Run All”).
  • Data Handling: Any data not pulled from a public API at runtime must be included in your submission .zip file with a clear file path.
  • Disclosure: Briefly disclose AI use in an appendix (tools used, tasks performed, and verification methods). Include any agent.md or skill.md files used.

Timeline & Milestones

Note: Late submissions are not accepted at any stage. After group sign-up closes, discussion sections become optional office hours.

  • April 17: Group sign-up closes (Groups of 3–5).
  • April 19: Finalized topic selection.
  • April 24: Checkpoint 1 Due. (Short PDF: roles, track, topic, and repository screenshots).
  • May 1: Checkpoint 2 Due. (Google Form: report on at least 3 successful/attempted results).
  • May 6: Final Project Due.
  • May 8: Peer Review Due.

Project and Group Sign Up

Good collaboration and leadership will be needed for a successful reproduction. We ask that you organize yourself in groups of 3-5. Groups may have members from different discussion sections. We encourage you to use Ed as a place to look for team members.

We have included a group sign up sheet. Please fill out your full name and Berkeley email accordingly under the Group Signup subsheet. All groups should be finalized and we will close the sheets on April 15th at 11:59 pm.

You also have to decide a group leader who will be main contact for new members and other logistics within the group. Please only reach out to the group leader for any inquiries about joining the group.

There is also a group status drop down that only the group rep should change. When the drop down is set to open, you’re allowing new members to join by adding their names under your group directly. When you set the group status to closed, no new members can join unless an exception has been agreed upon.

Under the Project Signup subsheet, please specify the track and project your group will work on by putting the group leader’s name accordingly. There is a limit of 15 groups per project.

Deliverables and Submissions

Checkpoint 1

This will be a separate Gradescope assignment that will be due earlier. There is no need to submit any code; a write-up of your progress so far (preferably in LaTeX) will suffice.

Submission Guidelines:

  • Everything must be submitted in a single PDF.
  • Only the group representative should submit.
  • Important: Include all group members in the Gradescope submission.

Required Contents:

  • Group Composition: Names and assigned responsibilities.
  • Chosen Paper: Citation/Title of the selected paper.
  • Project Organization: Screenshots of the code organization (including any relevant data, notebooks, or scripts).
  • LaTeX Setup: Screenshots of the LaTeX project in Overleaf following the I4R template.
  • Progress Report: A description of what you have attempted to do so far.
  • Status Summary: A summary of what is going well and what you are struggling with.
  • Staff Support: A note on how course staff may help you succeed.

Checkpoint 2

One person from your group must fill out the designated Google Form for this checkpoint.

  • Requirement:
    • Track A: Showcase summary statistic tables, initial model setup/results, or any relevant EDA results. There is no strict number requirement.
    • Track B: While the form provides space to discuss up to 5 results, you are only required to discuss 3.

Final Submission

Final submission for both tracks consists of three components:

  1. The Report: 10–15 pages including graphs and tables.
    • Track A: Submit as .md or .tex AND a PDF.
    • Track B: Submit as .tex AND a PDF.
  2. Analysis Notebook: A PDF export of your Jupyter Notebook(s) submitted to Gradescope.
  3. Reproducibility Link: A zip file with runnable notebook with organized data subfolders.
  4. Peer Evaluation: A Google Form regarding group performance.

Report Requirements

Feature Track A: Open Data Track B: Paper Reproduction
Reproducibility (10 pts) Whether Colab or Zip file runs start to finish. Whether Colab or Zip file runs start to finish.
Introduction (5 pts) Define a specific policy/business challenge and its significance. Summarize the original paper’s contribution and causal mechanism.
Track Specific (10 pts) Detailed description of where the data is accessed or made. Address the main findings and conclusions of the paper; state if it can be replicated.
Data (20 pts) Detail source (APIs, scraping), cleaning methods, and summary stats. EDA and Describe. Detail access to original data vs. your own. Provide summary stats.
Model (20 pts) Justify method choice. Show the logic and variable selection. State the original model and exact equations authors used.
Results (15 pts) Trends, regression tables, and coefficient explanations. Replicate original graphs/tables as closely as possible.
Discussion (15 pts) Data hurdles and limitations that could skew findings. Reproduction hurdles and whether full replication was achieved.
AI Usage (5 pts) Document help with code optimization or data cleaning. Document help with code translation or LaTeX formatting.

Enrollment and Topic Selection

  • Starter Lists: Track A topics and Track B papers are posted on Ed. You may propose your own by Checkpoint 1.
  • Enrollment Cap: A maximum of 15 groups may sign up for any single topic. Topics close once they reach this limit.
  • Peer Review: Groups will exchange submissions via email. Group leaders must submit a review form verifying if the code runs and evaluating the quality of the work.

Track A: Starter Topics

Each topic below is a starting point, not a script—your group is expected to refine the question, justify your data choices, and make real modeling decisions. All projects must include both an econometric baseline and a machine learning comparison, and must satisfy the code standards in the main spec.

Enrollment Cap: A maximum of 15 groups may sign up for each topic. Sign-ups are first-come, first-served via the sign-up sheet. Once a topic hits 15 groups, it closes; you must pick another or move to Track B (Reproductions).

1. Sovereign Credit Rating Reproduction

Build a model that predicts S&P or Moody’s sovereign credit ratings from publicly available macro indicators, following the spirit of Cantor and Packer (1996). Pull country-level data from World Bank WDI and IMF WEO (GDP per capita, growth, inflation, external debt, fiscal balance, default history). Scrape current ratings from the Wikipedia “List of countries by credit rating” page. Baseline: ordered logit or probit. ML comparison: ordinal random forest or gradient boosting. Discuss which countries the model misses and why.

2. GDP Nowcasting from Mixed-Frequency Indicators

Nowcast current-quarter US real GDP growth using monthly and weekly indicators released before the BEA advance estimate. Pull data from FRED via fredapi: industrial production, retail sales, nonfarm payrolls, ISM, initial claims, and similar series. Target: BEA quarterly GDP. Baseline: bridge equation or simple monthly-to-quarterly aggregation with OLS. ML comparison: dynamic factor model or a simple recurrent network. Benchmark your nowcasts against the Atlanta Fed’s GDPNow.

3. State Unemployment Insurance Claims Forecasting

Forecast weekly initial unemployment insurance claims for three to five US states using FRED’s state-level claims series (e.g., CAICLAIMS, TXICLAIMS, NYICLAIMS, FLICLAIMS, MIICLAIMS), accessed via fredapi with a free API key. Augment with state-level Google Trends data for unemployment-related search terms, pulled via the pytrends package and cached to CSV for reproducibility. Baseline: ARIMA or OLS with lagged claims and seasonal dummies. ML comparison: gradient boosting (XGBoost or LightGBM) or a small LSTM. Groups must explicitly address how they handle the COVID structural break, and should pick states with contrasting labor market structures (e.g., a tech-heavy state, a manufacturing state, a tourism-dependent state) to compare model performance across contexts.

4. Intergenerational Mobility Across US Counties

Use the Opportunity Insights county-level mobility data to build a model predicting upward mobility from county characteristics: segregation, school quality, family structure, income inequality, and commuting patterns. Baseline: OLS with the variables Chetty and coauthors emphasize. ML comparison: random forest with feature importance analysis. The interesting question is not just predictive accuracy but whether ML uncovers interactions that the linear model misses, and what those interactions imply for policy.

5. Election Forecasting from Economic Fundamentals

Replicate and extend a fundamentals-based presidential election model along the lines of Abramowitz’s “Time for Change” or the Fair model, using FRED data on growth, inflation, and unemployment plus presidential approval from Gallup or 538’s archive. Baseline: OLS, the original specification. ML comparison: regularized regression (Lasso or Ridge) given the small sample. Use county level voting data for voting and for explanatory variables.

6. Local Housing Price Nowcasting

Nowcast monthly housing price changes for a chosen metro area using Zillow ZHVI as the target and a mix of FRED macro series, Census ACS demographics, and mortgage rate data as features. Baseline: OLS or ARIMA. ML comparison: gradient boosting or a small neural network. Pick at least three metros with different market dynamics (e.g., Bay Area, Austin, Cleveland) and compare what features matter where.

Track B: Paper Reproduction

For Track B, your group will attempt to replicate the results from a high-impact economics paper using Python. Each of the papers listed below received a 2025 or 2026 Best Paper Award from the American Economic Journal (AEJ).

Main Objective

Your objective is to reproduce the analysis using Python and the data science tools introduced in this course.

  • Code Originality: While you may reference the original replication archives (often in Stata or R), the Python code you submit must be your own work.
  • Environment: The analysis must be fully functional within a Google Colab environment.
  • Data Restrictions: While most papers include full replication archives, some may have restricted data due to privacy. You must identify these limitations early.

Approved Paper List

Applied Economics

  • Do Carbon Offsets Offset Carbon?
    • Journal: AEJ: Applied Economics, Vol. 17, No. 1 (January 2025)
  • Working Remotely? Selection, Treatment, and the Market for Remote Work
    • Journal: AEJ: Applied Economics, Vol. 16, No. 4 (October 2024)

Economic Policy

  • Employed in a SNAP? The Impact of Work Requirements on Program Participation and Labor Supply
    • Journal: AEJ: Economic Policy, Vol. 15, No. 1 (February 2023)
  • Adaptation and Adverse Selection in Markets for Natural Disaster Insurance
    • Journal: AEJ: Economic Policy, Vol. 14, No. 3 (August 2022)

Macroeconomics

  • Optimal Fiscal and Monetary Policy with Distorting Taxes
    • Journal: AEJ: Macroeconomics, Vol. 17, No. 2 (April 2025)
  • Earnings-Based Borrowing Constraints and Macroeconomic Fluctuations
    • Journal: AEJ: Macroeconomics, Vol. 15, No. 2 (April 2023)

Microeconomics

  • The Fake News Effect: Experimentally Identifying Motivated Reasoning Using Trust in News
    • Journal: AEJ: Microeconomics, Vol. 16, No. 2 (May 2024)
  • Competition in Pricing Algorithms
    • Journal: AEJ: Microeconomics, Vol. 15, No. 2 (May 2023)

Accessing Replication Archives

You can find the papers and their complete replication archives through the American Economic Association (AEA) website. Ensure you download the “Replication Package” associated with your chosen paper to access the original data and source code.

Back to Top

Accessibility Nondiscrimination

Copyright ©2026, Regents of the University of California and respective authors.

This site is built following the Berkeley Class Site template, which is generously based on the Just the Class, and Just the Docs templates.

View all course offerings