econometric-research-writing

name

econometric-research-writing

description

End-to-end econometric analysis and economics/management paper-writing workflow. Use for data-grounded research grilling or mock defense, stress-testing research designs and identification strategies, dataset profiling, model selection (panels, IV, GMM, DiD, time-series, volatility, regimes), literature/citation verification, academic prose polishing, and publication-style Word DOCX reports with native OMML equations and three-line tables. Trigger when users ask to be grilled, questioned, challenged, or walked through one or more evidence-backed research decisions before implementation.

Econometric Research Writing

Overview

Use this skill to turn a research question, dataset, assignment, draft, result table, or literature need into a rigorous econometric analysis and a standard economics/management paper-style Word document. The skill combines method selection, empirical workflow, modular econometric method references, field-specific writing style, top-journal table/figure conventions, literature/citation integrity, and DOCX/OMML generation and validation.

Routing

1. For topic, identification, model choice, or method fit, read references/method-selection.md.
2. For research grilling, mock defense, data-grounded questioning, knowledge-adaptive explanation, or stress-testing a proposed empirical design before implementation, read references/research-grilling.md.
3. For single-series, multivariate time-series, nonstationarity, VAR, threshold, kernel, GARCH/ARCH, Markov-switching, or time-varying coefficient methods, read references/time-series-methods.md.
4. For static FE/RE, Mundlak/Hausman, GLS variance components, dynamic panels, heterogeneous slopes, panel trends, binary panels, panel VAR, spatial panels, or time-varying panels, read references/panel-methods.md.
5. For IV, dynamic-panel IV/GMM, panel treatment effects, synthetic counterfactuals, DiD/event-study logic, or causal claim boundaries, read references/iv-causal-methods.md.
6. For RDD, fuzzy RDD, propensity score matching, IPW, doubly robust weighting, PSM-DiD, or IPW-DiD, read references/rdd-matching-methods.md.
7. For dataset intake, variable profiling, agent-decided semantic roles, model-ready checks, executable descriptive statistics, baseline regression, clustered inference, or event-study output, read references/data-analysis-workflow.md.
8. For regression workflow, robustness, tables, or reproducibility, read references/empirical-workflow.md.
9. For table/figure style, regression tables, descriptive tables, event-study figures, coefficient plots, visual QA, or top-journal layout conventions, read references/tables-figures-style.md.
10. For paper search, literature review, source provenance, citation honesty, reference formatting, data/code citations, or Word superscript citations, read references/literature-citation-workflow.md.
11. For abstract, introduction, empirical strategy, results, robustness, conclusion, or style polishing, read references/econ-writing-style.md.
12. For Word output, formulas, rendering, superscript citation markers, or document QA, read references/docx-workflow.md.
13. For deterministic dataset profiling, use scripts/profile_econ_dataset.py; after the agent writes and reviews roles.json, use scripts/run_empirical_analysis.py for descriptive tables, baseline estimates, clustered standard errors, robustness results, design diagnostics, and event-study figures; for reference verification and enriched bibliography metadata, use scripts/verify_references.py; for DOCX generation, use scripts/build_paper_docx.py; for OMML helpers, use scripts/omml_math.py; for structural QA, use scripts/check_docx_integrity.py.

Default Workflow

When research grilling is active, use read-only evidence gathering and the questioning protocol in references/research-grilling.md. Suspend formal estimation, manuscript edits, final artifact generation, and repository changes until the user explicitly opens the implementation gate.

1. Clarify the research claim: outcome, treatment/regressor, unit, time, sample, and target interpretation.
2. Audit the data structure: cross-section, time series, panel, balanced/unbalanced panel, binary outcome, nonstationary series, or high-dimensional nonlinear setting.
3. Inspect both variable names and observed data, then use the research question and any codebook to build a variable dictionary and causal/mechanism map: outcome, treatment/main regressor, controls, mechanisms, moderators, instruments, fixed effects, clustering unit, and bad controls. The agent must make and record these semantic decisions; never promote deterministic column-name hints into analysis roles.
4. Choose one primary econometric family and one optional supporting robustness family. Do not mix methods without a clear identification or modeling reason.
5. Define the baseline equation, assumptions, estimator, diagnostics, and threat model.
6. For a standard linear or event-study specification, write the reviewed roles to JSON and run scripts/run_empirical_analysis.py to generate focused descriptive statistics, baseline estimates, clustered or heteroskedasticity-robust inference, robustness specifications, VIF/rank/condition-number diagnostics, and event-study outputs when event time is defined. Then run or request any additional method-matched checks.
7. If external literature is used, build a claim-to-source ledger, verify paper metadata, and never include fabricated or unsupported citations.
8. Write the paper in economics/management style: claim first, method bounded by assumptions, results tied to tables/figures, limitations explicit.
9. Format empirical tables and figures in top-journal style: three-line tables by default, self-contained notes, source notes, readable figures, and text callouts explaining what to look for.
10. If the deliverable is Word, generate a .docx using the script/template, insert Word-native OMML formulas, use Word superscript runs for numeric citation markers when required, render pages, inspect layout, then run structural integrity checks.
11. Classify files as final, formal archive, or temporary; remove scratch scripts/renders unless they are intentional reproducibility materials.

Method Boundaries

• Treat Granger causality as predictive precedence, not structural causality.
• Treat fixed effects as controls for time-invariant heterogeneity, not a full endogeneity solution.
• Treat IV/dynamic panel estimates as assumption-dependent; state instrument relevance and exclusion restrictions.
• Treat nonparametric/time-varying coefficient models as flexible heterogeneity or evolving association unless a separate identification design exists.
• Treat cointegration as a stable long-run relation, not causal proof.

DOCX Generation

Use assets/econ-paper-template.docx as the long-term Word style template when a template is needed. For net-new papers from structured content, prefer:

python3 scripts/build_paper_docx.py input.json output.docx --template assets/econ-paper-template.docx
python3 scripts/check_docx_integrity.py output.docx
python3 scripts/render_validate_docx.py output.docx --output-dir .qa/econ_docx_render

Rendered PNG/PDF files are temporary QA artifacts unless the user asks to keep them.

Validation

Before final delivery:

1. Confirm the final file path and authority status.
2. Confirm formulas are Word-native OMML when formulas matter.
3. Confirm expected headings, tables, figures/media, references, and page count when applicable.
4. Confirm citations are verified, source-backed, and formatted in the requested style; if superscript markers are expected, confirm they are true superscript runs.
5. Render and visually inspect the .docx; if rendering is unavailable, state that visual QA could not be completed.
6. Remove temporary files or explicitly mark them as temporary.

References And Assets

• references/method-selection.md: choose models from research question and data structure.
• references/research-grilling.md: evidence-first, knowledge-adaptive questioning and the pre-implementation confirmation gate.
• references/time-series-methods.md: AR, unit-root, threshold, kernel, cointegration, Granger, VAR, TV-VAR, and nonlinear time-series methods.
• references/panel-methods.md: FE/RE, balanced/unbalanced panels, heterogeneous slopes, dynamic panels, panel nonstationarity, binary panels, and time-varying panels.
• references/iv-causal-methods.md: IV/2SLS, dynamic-panel IV, DiD/event studies, and causal claim boundaries.
• references/rdd-matching-methods.md: RDD, fuzzy RDD, propensity score matching, IPW, and combined matching/weighting DiD designs.
• references/data-analysis-workflow.md: econometric dataset intake, variable semantics, quality checks, mechanism maps, and model-readiness rules.
• references/empirical-workflow.md: end-to-end econometric analysis and robustness workflow.
• references/tables-figures-style.md: economics/management top-journal table and figure conventions.
• references/literature-citation-workflow.md: literature search, citation integrity, reference verification, and Word citation markers.
• references/econ-writing-style.md: economics/management academic writing style and section templates.
• references/docx-workflow.md: Word, OMML, render QA, and file-handling rules.
• assets/econ-paper-template.docx: reusable standard paper template.