jbf-tables-figures - SKILL.md Agent Skill

name: jbf-tables-figures description: Use when preparing Journal of Banking & Finance (JBF) tables and figures — descriptive statistics, regression tables, event-study plots, mechanism and heterogeneity exhibits, appendix robustness inventories, and self-contained notes that report sample, fixed effects, clustering, and units.

Tables & Figures (jbf-tables-figures)

When to trigger

Main empirical results exist but the exhibits are hard to read
You need JBF-ready descriptive, regression, event-study, or robustness tables
Figure and table notes do not identify sample, fixed effects, clustering, or units

Exhibit architecture

Table 1: sample and summary statistics. Include variable definitions, units, winsorization, and observation counts.
Table 2: baseline results. Build from sparse to saturated specifications so readers can see identifying variation.
Table 3: identification checks. Pre-trends, placebo outcomes, alternative fixed effects, alternative clustering, or IV first stage.
Table 4: mechanisms and heterogeneity. Tie each split to banking/finance theory, not to fishing.
Appendix tables: robustness inventory. Alternative samples, definitions, windows, and estimation methods.

Figures that work for JBF

Event-study coefficient plots with confidence intervals and a clear omitted period
Sample-selection flow charts for complex merges
Binned scatter or marginal-effect plots for nonlinear mechanisms
Time-series plots for policy shocks, treatment timing, or market conditions

Table notes

Every regression table note should state:

Dependent variable and units
Sample and period
Fixed effects
Clustering level
Controls
Treatment or event-window definition
Stars or confidence-interval convention

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers. Full map: execution-with-mcp. JBF is empirical banking/finance — corporate/bank causal designs around regulation and shocks.

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

Ten nearly identical coefficient columns with no specification logic
Statistical significance without economic magnitude
Appendix robustness that is impossible to map back to the main claim
Figure axes with undefined units

JBF exhibit calibration

Accepted JBF empirical papers commonly carry roughly six to nine main-text exhibits — summary statistics, baseline, identification diagnostics, mechanism/heterogeneity, and one or two robustness pieces — with deeper batteries in an appendix (a stylized norm, not a journal rule). Event-study figures usually earn main-text space when the design is staggered adoption; pure robustness plots rarely do.

Worked magnitude conversion (illustrative)

Suppose the baseline coefficient on a post-LCR × exposure interaction is −0.014 with loans/assets as the outcome.

A one-standard-deviation exposure (0.6, illustrative) implies −0.014 × 0.6 = −0.84 percentage points of loans/assets.
Against a sample-mean loans/assets of 62%, that is a 1.4% relative decline; against mean quarterly loan growth of 1.1%, it is most of one quarter's growth.
Put the conversion in the results text and the benchmarks (mean, SD) in Table 1 so the table note can stay short.

Regression table skeleton

Table X: [Outcome] and [treatment], bank-quarter panel
               (1)       (2)       (3)       (4)
Treat x Post   b/se      b/se      b/se      b/se
Controls       No        Yes       Yes       Yes
Bank FE        Yes       Yes       Yes       Yes
Quarter FE     Yes       Yes       --        Yes
State x Qtr FE --        --        Yes       --
Cluster        Bank      Bank      State     State
N / R2         ...
Note: sample, period, units, winsorization, FE, clustering, star convention.

Columns must encode a specification logic (sparse → saturated → alternative FE and clustering), not a sensitivity dump.

Figure conventions for bank panels

Event-study plots: mark the omitted period, shade confidence intervals, and date both announcement and implementation when they differ.
Treatment-rollout plots: show adoption timing across states or countries for staggered designs so readers see the control pool.
Bunching or density plots around supervisory asset thresholds whenever a cutoff design is used.
Sample-attrition flowcharts when merging Call Reports with DealScan-style loan data; match the counts to the attrition table.

Referee exhibit complaints

"I cannot reconstruct N across columns." → report observations per column, not once per table.
"Stars without magnitudes." → add a standardized-effect or economic-magnitude row.
"Appendix Table A12 contradicts Table 3." → audit the specification map before resubmission; mismatches read as carelessness.

Exhibit economy

Keep the main exhibit set to: sample/table 1, baseline, identification check, mechanism, and one decisive robustness table. Move specification inventories to the appendix and make the appendix map back to the main claim so reviewers can audit without rereading the whole paper.

Output format

[Main exhibit claim] ...
[Table/Figure role] baseline / identification / mechanism / robustness
[Required note fields] sample, FE, clustering, units, controls
[Missing diagnostics] ...
[Next step] jbf-contribution-framing or jbf-writing-style