jcf-tables-figures - SKILL.md Agent Skill

name: jcf-tables-figures description: Use when building the tables and figures for a Journal of Corporate Finance (JCF) empirical paper — summary statistics, main regression tables with fixed effects and clustering disclosed, event-study/CAR plots, and self-contained notes. It shapes the exhibits; pair with jcf-data-analysis for the underlying estimates.

Tables & Figures (jcf-tables-figures)

When to trigger

Designing the summary-statistics, main-result, and robustness tables
Building event-study or CAR figures with confidence bands
Writing self-contained table/figure notes a reviewer can read without the text

Table conventions (empirical corporate finance)

Summary statistics: N, mean, SD, key percentiles; flag winsorizing and units. Match the sample to the regression sample.
Main regression tables: coefficients with standard errors (or t-stats) clearly labeled; report the fixed effects included, the clustering level, N, and within/adjusted R². Show economic magnitude (e.g., a 1-SD change) near the headline coefficient.
Robustness tables: vary one thing per panel/column (definition, subsample, FE, estimator) so the reader sees what moves the result.
Use a consistent decimal precision; do not hide the dependent variable's scaling.

Figure conventions

Event-study / CAR plots: plot coefficients (or cumulative abnormal returns) with confidence bands; mark the event date; show pre-event leads to support parallel trends.
Binned scatters for nonlinearity; avoid 3D, gradients, and chartjunk.
Vector output (PDF/EPS) for print resolution.

Self-contained notes (required)

Each exhibit's note states: sample and period, variable definitions or a pointer, fixed effects, clustering, winsorizing, and significance markers. A referee should not need the body text to read the table.

Formatting and policy

"Your paper, your way" at first submission means exhibit styling need only be consistent; full Elsevier styling comes at revision.
Tables/figures count toward the paper's exhibits but JCF states no fixed length ceiling (待核实) — still, keep the set tight and non-redundant.

Exhibit architecture (calibration, hedged)

Accepted empirical JCF papers tend to follow a recognizable arc — treat the counts as calibration, not rules, and confirm against the journal's current author guidelines:

Table 1: sample construction and/or summary statistics on the regression sample.
Table 2: main fixed-effects estimates, sparse controls first, saturated last.
Tables 3–4: identification exhibits — event-study dynamics, pre-trends, first stage, or density tests.
Tables 5+: mechanism splits and robustness, one varied dimension per panel.
Figure 1 is frequently the event-study or RDD plot — many JCF readers judge identification from this figure alone.
Six to nine main exhibits is a common footprint; overflow robustness belongs in an appendix or internet appendix numbered IA.1, IA.2, and cited by number in the text.

Worked headline-table sketch

Illustrative skeleton for a governance-shock paper (numbers invented):

                       (1)        (2)        (3)
Treated x Post        0.014**    0.015**    0.013**
                     (0.006)    (0.006)    (0.006)
Firm FE                 Y          Y          Y
Year FE                 Y          —          —
Industry x Year FE      —          Y          Y
Controls                —          —          Y
N                    21,043     21,043     20,877
Within R²             0.08       0.11       0.12
Magnitude: 1 SD of treatment exposure ≈ 12% of mean investment

The magnitude line beneath the table body is the JCF habit worth copying — it answers "is this economically big?" before the referee asks.

Exhibit pushback and the fix

"I cannot tell what is clustered." → Put FE and clustering rows in every regression table, not in a global footnote.
"The event-study figure has no bands." → Re-plot with 95% CIs and at least four pre-period leads visible.
"Summary stats do not match the regression N." → Recompute Table 1 on the estimation sample; reconcile any drop in a sample-construction panel.
"Too many tables." → Merge robustness variants into panels; push the remainder to the internet appendix and cite IA numbers in the text.

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers. Full map: execution-with-mcp. JCF is corporate finance — endogeneity of corporate policies is the central threat; foreground IV/DiD identification.

Tables: etable (multi-model) or did_summary_to_latex straight from the result_id.
Figures: plot_from_result / enhanced_event_study_plot / event_study_table — axis units and the SE/clustering note baked in.
Every note names the estimator + clustering and states the magnitude in interpretable units.

See a full fitted-result → exhibit chain in the JF execution walkthrough.

Anti-patterns

A regression table that hides the FE and clustering structure.
"Stars only" with no economic magnitude.
Event-study plots without confidence bands or pre-trend leads.
Summary stats computed on a different sample than the regressions.

Output

【Tables】headline FE/cluster/N disclosed? [Y/N]; magnitude shown? [Y/N]
【Figures】event/CAR with CIs + leads? [Y/N]; vector output? [Y/N]
【Notes】each exhibit self-contained? [Y/N]