Claude Code for Economic Research

Introduction and Setup

ECONOMIST · MACROECONOMICS · QUANTITATIVE ECONOMICS

Claude Code for Economic Research

Sebastián Fernández Rivera · Research Assistant, LSE

APRIL 2026 · ISSUE 26.04

1

§ 1 · CHAPTER

Introduction to
Claude Code

What is Claude Code?

§ 1 · INTRODUCTION

An agentic coding tool that turns plain-English research design into running code, in your terminal or IDE.

Claude Code is developed by Anthropic and operates directly in your terminal, IDE, desktop app, and browser. It builds features from plain-English descriptions, drafts plans, writes the code, and verifies that the result matches what you asked for.

Key capabilities

  • Builds features from plain-English descriptions
  • Drafts plans, writes code, ensures functionality
  • Automates the tedious parts of an empirical pipeline

2

§ 2 · CHAPTER

Applications in
Economic Research

Research Workflow Support

Claude Code can carry the entire pipeline — from raw data to publication-ready outputs.

Complete data analysis pipeline

Cleaning, modelling, visualization, and reproducible documentation in one continuous, auditable trail.

  1. Data management. Navigate codebases and datasets.
  2. Statistical analysis. Run econometric models.
  3. Visualization. Generate publication-quality figures and tables.
  4. Documentation. Read and document code for replicability.

Econometric Modeling

Specialized econometric techniques live one prompt away.

Advanced statistical methods

As a coding tool, Claude Code can implement the modern econometric techniques essential for current empirical work.

Native support for R and Python. Stata and MATLAB via specialized MCP servers. The tool follows your conventions, not the other way around.

Example: Difference-in-Differences

Describe the research design in English; Claude assembles the code.

$ claude "Run a difference-in-differences regression
on the treatment effect of minimum wage policy using
panel data from 2010-2020, controlling for state and
year fixed effects"

Claude Code will load and clean the dataset, construct the right variables, run the regression with robust standard errors, and emit both regression tables and visualizations (Anthropic, 2026).

3

§ 3 · CHAPTER

Why It Matters

Why Adopt Claude Code

Five gains, one caveat.

  • Speed. Accelerate the tasks that consume your time.
  • Transparency. Every step documented and auditable.
  • Flexibility. Diverse cleaning + econometric methods.
  • Integration. Connects to institutional data sources.
  • Reproducibility. Built-in version control and documentation.

Caveat

Claude Code is an assistant, not a substitute for your expertise. Always review the code and results. Final responsibility for accuracy rests with you.

References

Anthropic. (2026). Claude Code Docs. https://code.claude.com/docs

Thank you.

SEBASFR.GITHUB.IO · APRIL 2026

SFR
/* SFR slide animations — letter-mask, axis-morph, clip-reveal, fade-rise, content-sweep, stamp. Mirrors the website's _animations.scss vocabulary and re-fires animations on slide change. */ (function () { document.documentElement.classList.add('js-motion'); // Detect Fraunces variation-axis support (Safari guard, see _animations.scss). try { if (CSS && CSS.supports && CSS.supports('font-variation-settings', '"opsz" 144, "SOFT" 100, "WONK" 1')) { document.querySelectorAll('.anim-axis-morph').forEach(el => el.setAttribute('data-fvs-ok', '1')); } } catch (e) { /* noop */ } // Split text inside .anim-letter-mask into per-character spans with --i index. function splitLetters(root) { root.querySelectorAll('.anim-letter-mask').forEach(host => { if (host.dataset.splitDone) return; host.dataset.splitDone = '1'; // Walk text nodes recursively so nested .hero-wordmark-first/.last keep structure. const walk = (node, counter) => { Array.from(node.childNodes).forEach(child => { if (child.nodeType === Node.TEXT_NODE) { const frag = document.createDocumentFragment(); for (const ch of child.nodeValue) { const span = document.createElement('span'); span.className = 'anim-char' + (ch === ' ' ? ' is-space' : ''); span.style.setProperty('--i', counter.i++); span.textContent = ch; frag.appendChild(span); } child.parentNode.replaceChild(frag, child); } else if (child.nodeType === Node.ELEMENT_NODE) { walk(child, counter); } }); }; walk(host, { i: 0 }); }); } // Re-trigger CSS animations on a slide by removing+reinserting nodes. const ANIM_SELECTOR = [ '.anim-letter-mask', '.anim-axis-morph', '.anim-clip-reveal', '.anim-fade-rise', '.anim-content-sweep', '.anim-stamp' ].join(','); function retrigger(slide) { if (!slide) return; slide.querySelectorAll(ANIM_SELECTOR).forEach(el => { // Force reflow to restart CSS animations. const animsBackup = el.style.animation; el.style.animation = 'none'; // Read offsetHeight to flush. // eslint-disable-next-line no-unused-expressions void el.offsetHeight; el.style.animation = animsBackup; }); } // Initial split + first-slide animation on Reveal ready. if (window.Reveal) { Reveal.on('ready', e => { splitLetters(document); retrigger(e.currentSlide); }); Reveal.on('slidechanged', e => { // Slides not yet visited may still need splitting (Quarto loads all up-front, // so this is mostly a safety net). splitLetters(e.currentSlide); retrigger(e.currentSlide); }); } else { document.addEventListener('DOMContentLoaded', () => splitLetters(document)); } })();