From ectheory-skills
Designs Monte Carlo simulations and numerical illustrations for Econometric Theory papers, showing finite-sample behavior tracks asymptotics.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ectheory-skills:ectheory-data-analysisThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
- Your theorem is proved and you need simulations showing it bites in finite samples
ET is theorem-proof first; numerical work is evidence that the asymptotics are useful, not the contribution itself. Two distinct, optional components:
At a theorem-proof venue the referee treats simulations as a stress test of whether the limit approximation is useful, not as the result. The first checks:
| Referee check | Passes for ET | Triggers a revision |
|---|---|---|
| DGP vs assumptions | Spans the boundary (near-unit-root, weak ID, growing dim) | One interior DGP that flatters |
| Metric vs claim | Size and power for a test; coverage, bias, RMSE for an estimator | Size only, or RMSE without coverage |
| Sample sizes | A grid of n that makes the rate visible | A single n hiding slow convergence |
| Honesty | Breakdown region reported | Uniformly green tables, no failure regime |
A Monte Carlo that never visits the regime where the proof's delicate step lives is desk-reject-adjacent.
For a refinement that reduces the error in rejection probability of a t-test from order n^(-1/2) to n^(-1) under local-to-unity asymptotics, report the design:
# Monte Carlo skeleton for the refinement illustration
seed = 20260610 # fixed and reported
reps = 50000 # per cell
n_grid = [50,100,200,400,800]
c_grid = [0,-5,-10,-20] # local-to-unity drift c, root rho = 1 + c/n
# size under H0 (first-order vs refined); power under local alt theta0 + h/sqrt(n)
# show ERP=|size-0.05| decays faster for the refined test; flag breakdown at large |c|
The fixes: "rate without distribution theory" → upstream (route ectheory-identification-strategy), since a
Monte Carlo cannot supply a missing limiting law; "no finite-sample evidence" → add the boundary-spanning
design; "simulations avoid the hard regime" → extend the c-grid into the regime where the proof's delicate
step operates. The ET structure is theorem → proof → simulation; confirm Supplement conventions against the
author guidelines.
【Components】Monte Carlo / empirical illustration / both
【DGP coverage】boundary cases included? [Y/N]
【Metrics】size / power / coverage / bias / RMSE
【n grid】reveals rate? [Y/N]
【Benchmark】existing method compared? [Y/N]
【Reproducibility】seeds + reps + DGP specified? [Y/N]
【Next step】ectheory-tables-figures
npx claudepluginhub brycewang-stanford/awesome-journal-skills --plugin ectheory-skillsDesigns and audits Monte Carlo simulation evidence for Econometrica manuscripts, covering finite-sample performance, regularity-condition stress tests, and degenerate cases.
Guides design and audit of Monte Carlo simulations, empirical applications, and estimator comparisons for The Econometrics Journal, focusing on reproducibility and theoretical alignment.
Guides Monte Carlo simulation design and empirical illustration for Journal of Econometrics submissions, covering size/power, DGP stress tests, and computational hygiene.