John Barrios · Yale School of Management Claude Code for Accounting Research

General Setup

Everything on this page happens before Module 1. The course moves fast because the environment work is done in advance — a laptop that can’t run Claude Code on the first morning costs you the whole first lab. Budget 45–90 minutes, more if you’re on Windows. Work top to bottom; each step ends with a check that tells you it worked.

NoteWhat you need by Module 1 (summary)

A laptop you have admin rights on (macOS or Linux natively; Windows via WSL2). A working local Python — plus R or Stata if that’s your language. Claude Code installed and authenticated on a Claude Pro or Max subscription (or an API key). git installed and a GitHub account. And a WRDS account with Duo multi-factor enrolled — the account by Module 1, a working ssh wrds login by end of Module 2 (that checkpoint is a hard gate for Module 4).

1. Operating system

macOS and Linux work natively — nothing special to do. Windows requires WSL2 (Windows Subsystem for Linux): Claude Code runs inside the Linux environment, not native Windows. Install Ubuntu via wsl --install in an administrator PowerShell, reboot, and do everything else on this page inside the Ubuntu terminal. The most common Windows failure across the course is mixing the two worlds — a Python installed in Windows is invisible from WSL and vice versa. Pick WSL and stay there.

Check: you can open a terminal (Terminal.app, your Linux terminal, or the Ubuntu/WSL window) and echo hello prints hello.

2. Local runtime: Python (and R or Stata)

Claude Code writes and runs code using whatever is installed on your machine — it does not bundle a Python or R of its own. The course’s examples are Python; a Stata bridge is taught in Module 4, and R users can follow along with everything.

  • Python 3.10+: python3 --version should print 3.10 or higher. If not: macOS brew install python; Ubuntu/WSL sudo apt update && sudo apt install python3 python3-pip python3-venv.
  • Install the handful of packages the labs lean on: pip3 install pandas matplotlib duckdb pyarrow requests (add --break-system-packages on newer Ubuntu if pip refuses).
  • R users: confirm Rscript --version runs. Stata users: confirm you can launch Stata from your machine and know the path to the executable — Module 4’s .do/.dta handoff needs it.

Check: python3 -c "import pandas; print('ok')" prints ok.

3. Claude Code — install and authenticate

You need either a Claude Pro or Max subscription or an API key. Pro (roughly $20/month at the time of writing — check current pricing) is sufficient for this course; Max tiers buy more usage headroom but are not required. API-key billing is pay-per-token and also works fine.

Install (either method):

# npm (requires Node 18+; macOS: brew install node / Ubuntu: sudo apt install nodejs npm)
npm install -g @anthropic-ai/claude-code

# or the standalone installer
curl -fsSL https://claude.ai/install.sh | bash

Then, from any folder, run claude. On first launch it walks you through login — choose your subscription account (or configure the API key when prompted). Grant the terminal access it asks for.

Check (the one that matters): cd into a fresh empty folder, run claude, and ask it: “Create a file called hello.txt containing one sentence about accounting research, then read it back to me.” If it creates the file, reads it back, and the file exists when you ls — you are ready for Module 1.

WarningCommon failure: it installs but won’t launch

Three usual suspects, in order: (1) Node too old — node --version should be 18+; (2) npm’s global bin directory not on your PATH — the installer prints the path to add; (3) on Windows, you installed in PowerShell instead of inside WSL. If you burn more than 30 minutes here, stop and bring the error message to Module 1’s lab — Part A has time and staff for exactly this, and the AI-research-setup guide below has step-by-step walkthroughs for both macOS and Windows.

4. git and GitHub

git is the course’s safety net (Module 1) and its collaboration layer (Module 5). Install: macOS xcode-select --install (or brew install git); Ubuntu/WSL sudo apt install git. Create a free GitHub account if you don’t have one, and set your identity:

git config --global user.name "Your Name"
git config --global user.email "you@university.edu"

Check: git --version prints a version; you can log in to github.com.

5. WRDS account and Duo — start this early

Module 4 runs on your own WRDS account; the course never distributes WRDS data or shares logins. If you don’t have an account, request one now through your institution’s WRDS representative — approval can take days, which is why this is a pre-course item, not a Module 3 one.

Two stages, two deadlines: your account exists and Duo multi-factor is enrolled by Module 1 (WRDS Account → Settings → Two-Factor); and you can complete an SSH login — ssh your_username@wrds-cloud.wharton.upenn.edu, approve the Duo push, see a WRDS prompt, type exit — by end of Module 2. Module 2 closes with a checkpoint where you submit proof of that login or flag a problem; students without working access by Module 4 use the public-data fallback track (same skills, public SEC data), so nobody is stranded — but the real thing is better.

Check: the SSH login above succeeds end-to-end.