Homework 1: CRSP, the CAPM, and Fama-French

Homework 1: CRSP, the CAPM, and Fama-French#

  • Launched: week 1. Due: Tuesday, October 20, at the start of the week 4 session. Also posted on Canvas.

  • Aim to finish by Tuesday, October 13, the week 3 session. You get three weeks, but HW 2 launches in week 3 and is itself due two weeks after that. The third week here is slack, not the plan: spend it and you are carrying two assignments at once.

  • Link to Assignment: a private copy is created for you in the hw-finm-32800 organization. The link to yours is on Canvas. Your copy is created from your answer to the GitHub username survey on Canvas, so fill that out first. You can open your copy once you accept the invitation that GitHub emails you.

  • Goes with: weeks 1 and 2

Overview#

One assignment, two papers, one data family. The first half rebuilds the CRSP market portfolio and the S&P 500 from their constituents. The second half merges CRSP with Compustat and replicates the Fama-French three-factor model. Together they tell one story: the CAPM says the market portfolio is the only priced risk, and Fama and French (1993) show what the market portfolio leaves unexplained and what fixes it. Week 1 teaches the first half and launches the assignment; week 2 teaches the second half.

Everything runs as a pipeline: a pinned environment, automated WRDS pulls, and unit tests that compare your series to published references. The way to work is to run pytest, read the failing test, and write the code that passes it. Run doit first so that the data is pulled before the tests run. Do not edit the test files.

The assignment has five parts, A to E, and each has a guide, listed above. Parts A to D are graded by the tests and are worth 34 points in total, one per test. Part E has nothing to fill in and no points. Every push runs the suite on the course runner and posts your score to the repository’s Actions tab. Search the repository for TODO to find the blanks.

You will need the wrds package and a .pgpass file so that the pulls can authenticate to WRDS without prompting; see the section “Creating a .pgpass file” in Connecting to the WRDS Platform With Python.

Before you start (not graded): background videos#

CRSP and Compustat in WRDS:

Fama-French factor construction, if you want to see it done in SAS first:

Part A (graded, 3 pts): GitHub Skills#

Complete three tutorials from the GitHub Skills page, using public repositories in your own GitHub account, and record the links in src/github_skills.py.

Part B (graded, 4 pts): the CRSP market index#

The CAPM’s market portfolio is, in practice, a value-weighted index of all stocks. You will replicate CRSP’s value-weighted and equal-weighted market indices from the monthly stock file. Complete src/calc_CRSP_indices.py.

Part C (graded, 15 pts): the S&P 500#

You will reconstruct the S&P 500 from its constituents two ways and see why only one of them is a tradable strategy. Complete the membership query in src/pull_SP500_constituents.py and the index calculations in src/calc_SP500_index.py.

Part D (graded, 12 pts): constructing the Fama-French factors#

You will merge CRSP with Compustat, replicate the size and value factors of Fama and French (1993), and reuse the same pipeline for a second sort, on corporate investment.

Fill in the code marked TODO. The blanks span the full pipeline, from query to automation:

  • SQL queries (src/pull_CRSP_Compustat.py): complete the WHERE clauses of the queries that pull Compustat fundamentals and the CRSP-Compustat link table. These are tested locally against a small in-memory database, without a WRDS connection, so you can iterate quickly. You must complete them before the automated download will work.

  • Portfolio construction (src/shared_portfolio_sorts.py and src/calc_Fama_French_1993.py): compute book equity, apply the common-stock and exchange screens, compute firm-level market equity, compute the NYSE breakpoints, and construct SMB and HML from the six size/book-to-market portfolios. The docstrings state what each function must produce; consult Fama and French (1993) and the CRSP/Compustat documentation linked in the code.

  • The investment sort (src/calc_inv_portfolios.py): the investment measure (asset growth), its NYSE breakpoints, and the portfolio assignment, as in the Fama-French five-factor model (Fama and French, 2015). It reuses the book equity, market equity, and universe screens from above, which is the point: a well-structured pipeline makes the second sort much cheaper than the first.

  • Pipeline automation (dodo.py): complete the PyDoit task definitions task_calc_Fama_French_1993 and task_calc_inv_portfolios so that doit runs these stages with the correct file dependencies and targets. See What is a task runner? and https://pydoit.org/tasks.html.

The reference factors and sorted portfolios are downloaded automatically from the Kenneth French data library. The unit tests check that the series your code builds from raw CRSP and Compustat data statistically match them. The unit tests are the specification: when unsure what a function should produce, read its test in src/test_*.py.

Part E (not graded): testing the CAPM and the three-factor model#

There is nothing to fill in. Once Part D works, doit runs src/calc_CAPM_and_FF3.py on your portfolios. The guide shows what they are for: whether the market is the tangency portfolio from HW 0, which alphas the CAPM leaves behind, and how many of them SMB and HML absorb.

For the theory behind it, without derivations, read From Mean-Variance to the CAPM and From the CAPM to Multifactor Models.

Contents