Finman Robustness

Robustness Strategy (finman-robustness)

When to trigger

• The paper has a wall of robustness tables and you cannot say which threat each one answers
• A coauthor wants to add "a few more specifications to be safe" before submission
• A referee asked whether the result survives an alternative explanation, not just an alternative specification
• You are unsure which checks belong in the main text vs. the internet appendix

The FM robustness philosophy (this is the distinctive part)

Most journals reward more robustness tables; FM explicitly does not. The editors state they put less weight on trivial robustness tests and more on new ideas and whether the result is interesting and credible. This inverts the usual instinct: at FM, a robustness section that mechanically re-runs the regression with five winsorization thresholds and three control sets signals weak ideas hiding behind volume. The skill here is subtraction, then precision — cut the busywork, then run the small number of checks that each defuse a named alternative explanation a smart referee would actually raise. One check that kills the leading confounder is worth more than ten that vary nothing important.

Threat-targeted robustness, not specification spraying

Build the robustness set by listing the alternative explanations a referee could offer, then pairing each with the one check that addresses it. Everything not on that list is a candidate for the appendix or the cut.

Curation sequence

1. List the alternative explanations explicitly — these are the only legitimate drivers of robustness work.
2. Map one check per threat. If two checks address the same threat, keep the stronger one.
3. Demote or delete the rest. Winsorization sweeps, control-set permutations that change nothing, and SE-flavor tables go to the internet appendix or out.
4. Lead with the most threatening alternative. The first robustness exhibit should answer the objection most likely to sink the paper.
5. Report stability, not a starred wall. Show the point estimate barely moves across the threat-targeted checks; that is the persuasive object, not significance stars.

What "less weight on trivial robustness" does NOT mean

Under-weighting trivial robustness is not a license to skip the checks that matter. FM still expects:

• Correct inference — the right clustering / standard errors are non-negotiable, not optional robustness.
• A defused leading alternative — the single most plausible confound must be addressed; this is identification, not busywork.
• Honest sensitivity — an Oster-style bound or a pre-trend plot is cheap and decisive, the opposite of trivial. The distinction FM draws is between checks that change what a reader believes (keep, lead with) and checks that only signal diligence (winsorization sweeps, SE menus, control churn — cut or appendix). When in doubt, ask: "If this check came out differently, would my conclusion change?" If no, it is trivial.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map: execution-with-mcp. Financial Management is empirical corporate finance + asset pricing; corporate-causal chain (DiD/IV/RDD) plus the factor-zoo haircut for cross-sectional pricing.

• Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg.
• OVB sensitivity: oster_delta / sensemakr.
• Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
• Re-fit off one handle: audit_result(result_id) lists missing checks + the exact suggest_function for each.
• Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Decisive checks in the body, exhaustive battery in the appendix. JF execution walkthrough.