Publication-Ready Figures

For this repo, the production plotting toolkit is fintools.figures. Read docs/ai/figures.md first, then choose the plotting path deliberately:

• Use native fintools.figures helpers for repo work, Word proof packs, validation checks, and dataframe-to-figure suites.
• Use style="fins" for the house publication style and style="ft" for FT-style output.
• Use the legacy skill-local finance.mplstyle / figutils.py assets only when the user explicitly wants the older standalone helper style or needs a portable snippet outside the package workflow.

To recreate the FT validation gallery:

bash
1python tools/figure_examples.py --style ft --docx --output results/figures

To recreate the house-style gallery:

bash
1python tools/figure_examples.py --style fins --docx --output results/figures

Generated PNG/PDF/DOCX/caption files belong under ignored results/figures/ paths. Do not commit maintainer proof packs or local gallery outputs.

Apply these conventions whenever creating figures. The goal: every figure Claude produces is publication-ready by default — no manual cleanup needed.

Legacy Helper Quick Start

Prefer fintools.figures for this repo. The helper assets below remain available for standalone or explicitly requested legacy publication-style plots.

Copy finance.mplstyle from this skill directory into the project, then:

python
1import matplotlib.pyplot as plt
2plt.style.use('path/to/finance.mplstyle')

Or apply inline (no file needed):

python
1import matplotlib.pyplot as plt
2plt.rcParams.update({
3    'font.family': 'serif',
4    'font.serif': ['Times New Roman', 'STIXGeneral', 'DejaVu Serif'],
5    'mathtext.fontset': 'stix',
6    'font.size': 9,
7    'axes.labelsize': 9,
8    'xtick.labelsize': 8,
9    'ytick.labelsize': 8,
10    'legend.fontsize': 8,
11    'axes.linewidth': 0.6,
12    'axes.spines.top': False,
13    'axes.spines.right': False,
14    'lines.linewidth': 1.2,
15    'xtick.direction': 'out',
16    'ytick.direction': 'out',
17    'legend.frameon': False,
18    'figure.dpi': 150,
19    'savefig.dpi': 600,
20    'savefig.format': 'pdf',
21    'pdf.fonttype': 42,
22})

Default Aesthetic

• Fonts: Times New Roman / STIX (serif, no LaTeX dependency)
• Spines: Bottom + left only (no top/right)
• Grid: Off by default
• Ticks: Outward, 8pt labels
• Colors: Okabe-Ito colorblind-safe palette (blue first)
• Export: PDF vector, 600 DPI, fonts embedded (type 42)

Color Palettes

Default cycle (Okabe-Ito, colorblind-safe):

python
1PALETTE = ['#377EB8', '#E41A1C', '#4DAF4A', '#984EA3',
2           '#FF7F00', '#A65628', '#F781BF', '#999999']

Two-series (long vs short, treatment vs control):

python
1BLUE_RED = ['#377EB8', '#E41A1C']

Grayscale-safe (for guaranteed print clarity):

python
1GRAYSCALE = ['#000000', '#555555', '#999999', '#CCCCCC']
2# Combine with linestyles: '-', '--', ':', '-.'

Sequential/diverging colormaps: Use viridis or cividis (colorblind-safe) for heatmaps. Use RdBu_r for diverging (correlation matrices).

Figure Sizing

Aspect ratio: Default to golden ratio (width / 1.618). Use square for heatmaps, wide (width / 2.0) for time series.

python
1def set_size(width='single', ratio='golden'):
2    widths = {'single': 3.5, 'onehalf': 5.25, 'double': 7.0, 'slide': 10.0}
3    ratios = {'golden': 1.618, 'square': 1.0, 'wide': 2.0}
4    w = widths.get(width, width)
5    r = ratios.get(ratio, ratio)
6    return (w, w / r)

Common Figure Types in Empirical Finance

Time Series

python
1fig, ax = plt.subplots(figsize=set_size('double', 'wide'))
2ax.plot(dates, values)
3ax.set_xlabel(''); ax.set_ylabel('Return (%)')

• Use ax.axhline(0, color='grey', linewidth=0.5, zorder=0) for zero reference
• Add NBER recession bands with ax.axvspan(start, end, alpha=0.1, color='grey')

Cumulative Return / Wealth Paths

python
1cumret = (1 + returns).cumprod()
2ax.plot(cumret.index, cumret.values)
3ax.set_ylabel('Growth of $1')

• Start at 1.0 (or 100 for percentage scale)
• Log scale optional for long horizons: ax.set_yscale('log')

Decile Portfolio Bar Chart with Newey-West CIs

python
1# returns_df: DataFrame with columns 0..9 (portfolio return time series)
2plot_portfolio_bars(ax, returns_df, show_ls=True, ls_label='10-1')
3# Computes means, Newey-West SEs (lag = floor(T^0.25)), 95% CI error bars
4# Includes long-short bar with t-stat annotation

• Use plot_portfolio_bars for any portfolio sort figure — it handles NW SEs automatically
• Short leg (decile 1) colored red, long leg (decile 10) green, L-S bar purple
• t-stat annotated on the L-S bar

For simple bars without CIs (pre-computed means):

python
1plot_decile_bars(ax, means, highlight_extremes=True, spread_label=True)

Coefficient Plot (Forest Plot)

python
1ax.errorbar(coefs, range(len(coefs)), xerr=[coefs-ci_lo, ci_hi-coefs],
2            fmt='o', color='#377EB8', capsize=3, markersize=4)
3ax.axvline(0, color='grey', linewidth=0.5, linestyle='--')
4ax.set_yticks(range(len(names))); ax.set_yticklabels(names)

Event Study (CAR Plot)

python
1days = range(event_window[0], event_window[1] + 1)
2ax.plot(days, car, color='#377EB8')
3ax.fill_between(days, ci_lo, ci_hi, alpha=0.2, color='#377EB8')
4ax.axvline(0, color='grey', linewidth=0.5, linestyle='--')
5ax.axhline(0, color='grey', linewidth=0.5)
6ax.set_xlabel('Days Relative to Event'); ax.set_ylabel('CAR (%)')

Correlation Heatmap

python
1import seaborn as sns
2mask = np.triu(np.ones_like(corr, dtype=bool), k=1)
3sns.heatmap(corr, mask=mask, cmap='RdBu_r', center=0, vmin=-1, vmax=1,
4            annot=True, fmt='.2f', linewidths=0.5, ax=ax,
5            cbar_kws={'shrink': 0.8})

Multi-Panel Figures

python
1fig, axes = plt.subplots(1, 3, figsize=set_size('double', 'wide'))
2# Label panels
3for i, ax in enumerate(axes):
4    ax.text(-0.1, 1.05, f'({chr(97+i)})', transform=ax.transAxes,
5            fontsize=10, fontweight='bold', va='top')

Export Checklist

Before saving any figure:

1. Format: PDF (vector) for papers; PNG (300+ DPI) for slides/web
2. Font embedding: pdf.fonttype = 42 (already in style)
3. Bbox: bbox_inches='tight' to avoid clipped labels
4. DPI: 600 for publication, 150 for screen preview
5. Size: Match target journal column width — don't rescale in LaTeX/Word

python
1fig.savefig('figure.pdf', bbox_inches='tight', dpi=600)
2# Also save PNG for quick preview:
3fig.savefig('figure.png', bbox_inches='tight', dpi=150)

Journal-Specific Overrides

Do NOT

• Use rainbow colormaps (jet, hsv) — not perceptually uniform, not colorblind-safe
• Use 3D plots unless the data genuinely requires a third dimension
• Add chartjunk: unnecessary gridlines, borders, background colors
• Scale figures in LaTeX/Word — set correct size in matplotlib, include at 100%
• Use different fonts/sizes across figures in the same paper
• Use LaTeX escapes (\&, \%, \_) in matplotlib text (titles, labels, annotations) — matplotlib's default text engine renders backslashes literally. Write plain S&P 500, not S\&P 500. LaTeX escapes only work when plt.rcParams['text.usetex'] = True, which requires a full LaTeX installation and is NOT enabled by default in our style.