Rfs Tables Figures

Tables & Figures (rfs-tables-figures)

When to trigger

• The main table has too many columns or unlabeled specifications
• Standard errors, significance stars, or clustering are unclear or inconsistent
• A figure is a default software plot rather than a publication-grade exhibit
• A reader cannot tell from a table alone what the regression is

What an RFS exhibit must do

Every main table or figure should be self-contained: a referee skimming exhibits should grasp the design, the result, and the inference without the text. RFS articles are typeset by Oxford University Press for the SFS, so build exhibits to OUP production specs (vector figures, clean two-column-friendly tables). One RFS-specific constraint shapes every exhibit: because RFS requires public release of all code as a condition of publication, each table and figure must be regenerable by the released scripts — a number a referee cannot reproduce from your code is a liability. The portfolio-sort and spanning exhibits in Hou, Xue, and Zhang (2015) "Digesting Anomalies" (RFS 28(3)) are a good template for decile monotonicity and long–short t-stats.

Table conventions

• One question per table. A main paper typically has a handful of focused tables, not a dozen sprawling ones; overflow goes to the Internet Appendix (rfs-internet-appendix).
• Self-contained caption: state the sample, period, unit of observation, estimator, and what is being shown.
• Coefficients with SE below in parentheses; state the SE type (e.g., "standard errors clustered by firm") in the notes — not only stars.
• Report significance honestly: stars are a convenience; the SE/t-stat is the substance. State the convention in the notes.
• Summary-statistics table: means, SD, and key percentiles for every variable used; a correlation matrix when relevant.
• Decimal alignment, consistent rounding, consistent variable names across all tables.
• Economic magnitude, not only statistical significance — report effect sizes in interpretable units.

Figure conventions

• Event-study / coefficient plots with confidence bands for DID and dynamic effects — strongly preferred over a pre/post dummy table.
• Binned scatter / RD plots for discontinuity designs.
• Label axes with units; legends self-explanatory; greyscale-legible (do not rely on color alone).
• Vector format (PDF/EPS) for print; consistent fonts and sizing.

Reporting conventions that signal craft to finance referees

• Report N and within/adjusted R² (or pseudo-R²) for every regression column.
• State which fixed effects are absorbed in each column (a "FE" panel of yes/no rows).
• For Fama–MacBeth, report the number of cross-sections and the time-series length.
• For portfolio sorts, show monotonicity across deciles and the long–short spread with t-stat.
• Keep one canonical ordering of variables across all tables so readers can scan rows.
• Put units in the caption (returns in %, dollars in millions) — never leave a number unitful.

Execution bridge (StatsPAI / Stata MCP)

Generate exhibits from the fitted result, not by retyping numbers. Full map: execution-with-mcp. RFS is finance top-3 (with JF, JFE) — corporate-causal chain for corporate papers, factor-zoo haircut for asset pricing.

• 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.

Checklist

• Each main table answers one question; the rest is in the IA
• Every table caption states sample, period, unit, estimator
• SEs reported below coefficients; SE type stated in notes (clustering/adjustment)
• Economic magnitudes reported, not just significance
• Summary-statistics table present for all variables
• DID/dynamic effects shown as an event-study figure with CIs
• Figures legible in greyscale; axes labeled with units; vector format (OUP production-ready)
• Every reported number is regenerable from the code RFS requires you to release
• Variable names, rounding, and notation consistent across all exhibits

Anti-patterns

• An 11-column main table where no one can tell what each column estimates.
• Stars with no statement of the standard-error type or clustering.
• Reporting t-stats but never the economic magnitude of the effect.
• A default Stata/R plot pasted in without axis labels or units.
• Color-only figures that become unreadable in print.
• Variable named x1 in one table and Leverage in another.

Output format

【Main exhibits】[table/figure list, one question each]
【SE reporting】type + clustering stated in every table note? yes/no
【Magnitude】economic effect sizes reported? yes/no
【Figures】event-study/RD plots with CIs? format?
【Moved to IA】[...]
【Next step】rfs-internet-appendix