Resources
Code and research-workflow configurations from my own work, free to reuse. Dates mark first release.
Research-assistant kits
Two setups for using an AI agent as a research assistant in applied econometrics. The Claude Code kit gives the agent more autonomy; the local kit keeps a smaller model on one approved step at a time.
2026 September
Local LLM research starter
A configuration for running a local model as a research assistant on well-defined tasks, with data kept on the workstation. A single AGENTS.md sets the rules the agent follows each session:
- Role. The PI decides; the assistant computes and never invents data, variables, or results.
- Context discipline. At most three files per task; datasets are summarized, never opened directly.
- One step at a time. Plan, execute, report, and stop for approval.
- Reference notes. Nine notes in
refs/(econometrics in R and Julia, verification, replication, proofreading, and others), read only when needed.
Setup I use: Ollama with Qwen3.8 27B and OpenCode, on a workstation with 128 GB of RAM and an RTX 5090.
- Runs on
- Ollama Qwen3.8 27B OpenCode
- Languages
- R Julia Stata LaTeX
- Shell
- Windows PowerShell
- Status
- Active
- License
- MIT
2026 February
Claude Code research starter kit
A Claude Code setup for applied econometrics projects in R, Stata, Julia, and LaTeX. Copy it into a new project, paste the starter prompt, and Claude adapts the configuration and works plan-first.
- 16 skills. Workflow commands (
/data-analysis,/review-paper,/lit-review, and others) and econometrics references for R, Julia, causal inference, and Stata (reghdfe,csdid,rdrobust, and others). - 4 review agents. Domain review, proofreading, R code review, and verification.
- Rules, hooks, and quality gates for planning, logging, replication, and code conventions.
Hooks are written for PowerShell; the README covers macOS and Linux.
- Runs on
- Claude Code
- Languages
- R Stata Julia LaTeX
- Shell
- PowerShell (bash notes included)
- Status
- Active
- License
- MIT
Code and software
Replication code and a Stata command from my papers.
2023 August
Spatial autoregressive stochastic frontier with fixed effects: crime under-reporting in Bogotá
R replication notebook for Chancí, Kumbhakar, and Sandoval, Crime under-reporting in Bogotá: a spatial panel model with fixed effects, Empirical Economics 66(5), 2024. A spatial autoregressive stochastic frontier panel with fixed effects, where the one-sided error measures under-reporting, estimated for six offences across Bogotá’s police quadrants. Reproduces Tables 1–3 and 5.
2019 March
chks: nonlinear index and zero-inefficiency stochastic frontier models
Stata command for nonlinear-index models (CES or Cobb–Douglas), estimated by NLS, as a stochastic frontier, or as a zero-inefficiency stochastic frontier (ML or EM). Developed for a 2019 working paper with Kumbhakar and Sandoval.
net install chks, from("https://luischanci.github.io/chks/") replace
help chks- Language
- Stata
- Version
- 1.1
- Status
- Provided as is; not maintained