estimate-var

Estimates VAR/SVAR/VECM models with diagnostics, identification and impulse responses.

SKILL.md

GitHub
<!-- Generated from .claude/skills/estimate-var/SKILL.md by tools/sync_agents.py. Edit the source, then run the script. -->

/estimate-var — Estimate VAR

Category: Estimation · Primary agent: var-vecm-specialist

Estimates VAR/SVAR/VECM models with diagnostics, identification and impulse responses.

Arguments: whatever the user wrote after the command name.

Purpose

Estimate a VAR, structural VAR or VECM with the specification choices made explicitly, diagnostics reported, identification stated and tested, and impulse responses shown with honest uncertainty bands. The output is a set of results whose conclusions survive an alternative identification or specification.

Use it when dynamic interactions among a few macro variables matter, structural shocks must be identified from data, or cointegration relations are of interest.

Do not use it when the goal is forecasting many variables (/estimate-bvar), when responses at long horizons with external instruments are needed (/local-projections), or when the shock series is the deliverable (/identify-shocks).

Before you start

  1. Check required inputs. For each input below record FOUND or ABSENT (never 'assumed'):
    • Data sheet
    • Identification scheme If a required input is ABSENT, that part is NOT ASSESSED — NO DATA: do not estimate or invent it; name what is missing and which skill produces it. If all required inputs are absent, stop and say so.
  2. Check confidentiality. Identify the data classification level of the inputs (.claude/docs/confidentiality-and-ai-use.md). Do not proceed with restricted or confidential data in a way the policy does not allow.
  3. Collaboration. Follow .claude/docs/collaboration-protocol.md: ask questions first, present options, get approval before writing files, and update production/session-state/active.md after each approved section.

Agents

  • var-vecm-specialist — primary agent; coordinates the work and integrates results
  • shock-identification-specialist — Identifies structural shocks
  • data-quality-seasonal-specialist — Ensures data quality

Each agent is defined in .claude/agents/<name>.md. If your tool runs subagents, run these agents as subagents, in parallel where the steps are independent; otherwise read each agent's file and apply its instructions in turn. Subagents only write to the paths named in your instructions.

Key design decisions (present options, let the user choose)

Decision Typical options What it changes
Specification levels, differences, VECM Long-run information and inference
Lag length information criteria, residual tests Dynamics and degrees of freedom
Identification recursive, sign restrictions, long-run, external instruments, heteroskedasticity What the shocks mean
Inference bootstrap, Bayesian, robust bands for set identification Honest uncertainty
Sample full, post-break, excluding outliers Stability

Workflow

  1. Test integration order and cointegration; choose lag length and deterministic terms
  2. Estimate the reduced form; run stability and residual diagnostics
    • Roots, serial correlation, heteroskedasticity, normality; parameter stability tests
  3. Apply identification and compute IRFs, FEVD with bands
    • State the economic justification for each restriction
  4. Run robustness under an alternative identification or specification

Checkpoints with the user

  • After specification: confirm variables, sample and deterministic terms
  • After identification: confirm restrictions are defensible

Common pitfalls

  • Recursive orderings without justification
  • Sign restrictions reported as point identification
  • Ignoring structural breaks
  • Over-reading forecast-error variance decompositions

Output

docs/models/<name>/var-results.md; code and settings in models/econometrics/<name>/

The results document reports specification tests, diagnostics, identification, IRFs and FEVD with bands, and robustness; code and settings allow exact rerun. Every run saves its results to output/<run-id>/ in the model folder, with a run log (.claude/rules/model-code.md).

Gate / verdict

  • Stability and residual tests reported
  • Robustness to identification shown
  • Identification restrictions are justified economically
  • Set-identified results are reported as sets

Suggested next steps

  • /validate-model
  • /scenario-analysis