quant-tearsheet

name	quant-tearsheet
description	This skill should be used when the user asks to "create a tearsheet", "generate a PDF report", "strategy summary", "backtest report", "performance tearsheet", "factor tearsheet", or when producing single-page or multi-page PDF summaries of quantitative results.
version	1.0.0

Quantitative Tearsheet Standards

PDF tearsheets are the final deliverable — dense, polished, print-ready summaries that distill complex analysis into scannable documents. Built on the ML4T visual identity established in quant-viz.

Tearsheet Philosophy

A tearsheet is not a report with charts. It is a designed artifact where layout, hierarchy, and density are as important as the data itself. Think Bloomberg terminal printout meets editorial design.

Design Principles

1. Information density over whitespace: every square centimeter earns its place
2. Scannable in 30 seconds: key metrics visible without reading text
3. Self-contained: no external context needed to understand the story
4. Print-first: grayscale compatible, A4/Letter safe margins, 150+ dpi

Tearsheet Types

Strategy Tearsheet (Single Page)

The canonical one-pager for strategy evaluation:

+------------------------------------------------------------------+
|  STRATEGY NAME                              Period: 2015-2024     |
|  Subtitle / description                     Generated: 2026-03-02 |
+------------------+-------------------+---------------------------+
|  KEY METRICS     |  RISK METRICS     |  ROLLING PERFORMANCE      |
|  ─────────────── |  ─────────────── |  [Sparkline area chart]    |
|  Ann. Return  12%|  Max DD     -18% |                            |
|  Ann. Vol     15%|  DD Duration 4mo |                            |
|  Sharpe      0.80|  Calmar     0.67 |                            |
|  Sortino     1.12|  Tail Ratio 1.24 |                            |
|  Win Rate     54%|  VaR 95%   -2.1% |                            |
+------------------+-------------------+---------------------------+
|  CUMULATIVE RETURNS                                               |
|  [Line chart: strategy vs benchmark, regime shading]              |
|                                                                   |
+------------------------------------------------------------------+
|  DRAWDOWN                          |  MONTHLY RETURNS HEATMAP    |
|  [Fill chart, annotate max DD]     |  [Year x Month grid]        |
|                                    |  Jan Feb Mar ... Dec  Year  |
|                                    |  1.2 -0.3 2.1 ...    12.4  |
+------------------------------------+-----------------------------+
|  RETURN DISTRIBUTION               |  ROLLING SHARPE (252d)      |
|  [Histogram + VaR lines]           |  [Line + zero reference]    |
+------------------------------------+-----------------------------+
|  Footer: Data source | Methodology notes | Disclaimers            |
+------------------------------------------------------------------+

Factor Tearsheet

For alpha factor evaluation:

+------------------------------------------------------------------+
|  FACTOR: momentum_12m_1m               Universe: SP500            |
+------------------+-------------------+---------------------------+
|  IC SUMMARY      |  TURNOVER         |  IC TERM STRUCTURE        |
|  Mean IC   0.035 |  Monthly    42%   |  [Heatmap: horizon x lag] |
|  IC IR     0.45  |  Annual    180%   |                           |
|  Hit Rate   58%  |  Avg Hold   2.3mo |                           |
+------------------+-------------------+---------------------------+
|  QUINTILE RETURNS (annualized)                                    |
|  [Bar chart: Q1 through Q5, long-short spread annotated]          |
+------------------------------------------------------------------+
|  CUMULATIVE FACTOR RETURNS          |  IC TIME SERIES             |
|  [Line: long-short, long, short]    |  [Line + 12mo rolling avg]  |
+------------------------------------+-----------------------------+
|  SECTOR EXPOSURE                    |  FACTOR CORRELATION MATRIX  |
|  [Stacked bar by quintile]          |  [Heatmap, masked upper]    |
+------------------------------------+-----------------------------+

Multi-Strategy Comparison

Side-by-side strategy comparison:

+------------------------------------------------------------------+
|  STRATEGY COMPARISON                        Period: 2015-2024     |
+------------------------------------------------------------------+
|  METRICS TABLE                                                    |
|  ┌──────────┬────────┬────────┬────────┬────────┐                |
|  │ Metric   │ Strat A│ Strat B│ Strat C│ Bench. │                |
|  ├──────────┼────────┼────────┼────────┼────────┤                |
|  │ Return   │  12.3% │   9.8% │  15.1% │   8.2% │  ← best bold |
|  │ Vol      │  14.5% │  11.2% │  18.3% │  16.0% │                |
|  │ Sharpe   │  0.85  │  0.88  │  0.82  │  0.51  │  ← best bold |
|  │ Max DD   │ -18.2% │ -12.5% │ -24.1% │ -33.9% │                |
|  └──────────┴────────┴────────┴────────┴────────┘                |
+------------------------------------------------------------------+
|  CUMULATIVE RETURNS (all strategies overlaid)                     |
+------------------------------------------------------------------+
|  ROLLING 1Y SHARPE                  |  DRAWDOWN COMPARISON        |
+------------------------------------+-----------------------------+
|  CORRELATION MATRIX                 |  MONTHLY RETURN SCATTER     |
|  [Strategy x Strategy]              |  [Strat A vs Strat B]       |
+------------------------------------+-----------------------------+

Implementation

Page Setup

import matplotlib.pyplot as plt
from matplotlib.gridspec import GridSpec
from ml4t.style import COLORS, ml4t_palette, GRAYSCALE

# A4 landscape for single-page tearsheets
fig = plt.figure(figsize=(16.5, 11.7), dpi=150)
fig.set_facecolor('white')

# GridSpec for precise layout control
gs = GridSpec(4, 3, figure=fig,
             hspace=0.35, wspace=0.3,
             left=0.05, right=0.95,
             top=0.92, bottom=0.05)

# Header spans full width
ax_header = fig.add_subplot(gs[0, :])
ax_header.axis('off')

# Metric panels
ax_cumret = fig.add_subplot(gs[1, :])          # Full width
ax_dd     = fig.add_subplot(gs[2, :2])          # 2/3 width
ax_monthly= fig.add_subplot(gs[2, 2])           # 1/3 width
ax_dist   = fig.add_subplot(gs[3, 0])           # 1/3 width
ax_sharpe = fig.add_subplot(gs[3, 1])           # 1/3 width
ax_table  = fig.add_subplot(gs[3, 2])           # 1/3 width

Key Metrics Panel

def render_metrics_panel(ax, metrics: dict, title: str = "KEY METRICS"):
    """Render a metrics panel with label-value pairs.

    Args:
        metrics: {"Ann. Return": "12.3%", "Sharpe": "0.85", ...}
    """
    ax.axis('off')
    ax.set_xlim(0, 1)
    ax.set_ylim(0, 1)

    # Title bar
    ax.text(0.5, 0.95, title, transform=ax.transAxes,
            fontsize=10, fontweight='bold', ha='center', va='top',
            color=COLORS['blue'])
    ax.axhline(y=0.90, xmin=0.05, xmax=0.95,
               color=COLORS['amber'], linewidth=2)

    # Metrics rows
    y = 0.82
    for label, value in metrics.items():
        ax.text(0.08, y, label, fontsize=9, color=COLORS['neutral'],
                va='center')
        ax.text(0.92, y, str(value), fontsize=9, fontweight='semibold',
                color=COLORS['blue'], ha='right', va='center',
                fontfamily='monospace')
        y -= 0.12

Monthly Returns Heatmap

def render_monthly_heatmap(ax, returns: pd.Series):
    """Render year x month returns heatmap."""
    monthly = returns.resample('M').sum()
    table = monthly.groupby([monthly.index.year, monthly.index.month]).first().unstack()

    # Use diverging colormap centered at 0
    vmax = max(abs(table.min().min()), abs(table.max().max()))
    im = ax.imshow(table.values, cmap='RdYlGn', vmin=-vmax, vmax=vmax,
                   aspect='auto')

    # Annotate cells with values
    for i in range(table.shape[0]):
        for j in range(table.shape[1]):
            val = table.iloc[i, j]
            if pd.notna(val):
                color = 'white' if abs(val) > vmax * 0.6 else COLORS['neutral']
                ax.text(j, i, f"{val:.1%}", ha='center', va='center',
                       fontsize=7, color=color, fontfamily='monospace')

    ax.set_xticks(range(12))
    ax.set_xticklabels(['J','F','M','A','M','J','J','A','S','O','N','D'], fontsize=8)
    ax.set_yticks(range(len(table.index)))
    ax.set_yticklabels(table.index, fontsize=8)
    ax.set_title("Monthly Returns", fontsize=10, fontweight='semibold',
                 color=COLORS['blue'])

Header Block

def render_header(fig, name, subtitle, period, benchmark=None):
    """Render tearsheet header with title and metadata."""
    # Strategy name — large, deep blue
    fig.text(0.05, 0.97, name, fontsize=18, fontweight='bold',
             color=COLORS['blue'], va='top')

    # Subtitle
    fig.text(0.05, 0.94, subtitle, fontsize=10,
             color=COLORS['neutral'], va='top')

    # Period and date — right aligned
    fig.text(0.95, 0.97, f"Period: {period}", fontsize=9,
             color=COLORS['neutral'], ha='right', va='top')
    fig.text(0.95, 0.945, f"Generated: {pd.Timestamp.now():%Y-%m-%d}",
             fontsize=8, color=COLORS['silver_muted'], ha='right', va='top')

    if benchmark:
        fig.text(0.95, 0.92, f"Benchmark: {benchmark}", fontsize=8,
                 color=COLORS['neutral'], ha='right', va='top')

    # Amber rule line under header
    line = plt.Line2D([0.05, 0.95], [0.915, 0.915],
                      transform=fig.transFigure,
                      color=COLORS['amber'], linewidth=2)
    fig.add_artist(line)

Footer Block

def render_footer(fig, data_source, notes=None, disclaimer=None):
    """Render tearsheet footer."""
    footer_y = 0.02
    fig.text(0.05, footer_y, f"Data: {data_source}", fontsize=7,
             color=COLORS['silver_muted'])
    if notes:
        fig.text(0.50, footer_y, notes, fontsize=7,
                 color=COLORS['silver_muted'], ha='center')
    if disclaimer:
        fig.text(0.95, footer_y, disclaimer, fontsize=6,
                 color=COLORS['silver_muted'], ha='right', style='italic')

PDF Output

from matplotlib.backends.backend_pdf import PdfPages

# Single tearsheet
fig.savefig("strategy_tearsheet.pdf", dpi=150, bbox_inches='tight',
            facecolor='white', edgecolor='none')

# Multi-page report
with PdfPages("full_report.pdf") as pdf:
    pdf.savefig(strategy_fig, dpi=150, facecolor='white')
    pdf.savefig(factor_fig, dpi=150, facecolor='white')
    pdf.savefig(comparison_fig, dpi=150, facecolor='white')

    # Metadata
    d = pdf.infodict()
    d['Title'] = 'Strategy Performance Report'
    d['Author'] = 'ML4T Framework'
    d['CreationDate'] = pd.Timestamp.now()

Visual Standards for Tearsheets

Typography Hierarchy

Element	Size	Weight	Color
Strategy name	18pt	Bold	blue
Section header	10pt	Bold	blue
Metric label	9pt	Regular	neutral
Metric value	9pt	Semibold, mono	blue
Axis labels	8-9pt	Regular	neutral
Cell values	7pt	Regular, mono	varies
Footer	7pt	Regular	silver_muted

Conditional Formatting

• Positive returns: COLORS['positive'] (#10b981)
• Negative returns: COLORS['negative'] (#ef4444)
• Best-in-class metric: bold + COLORS['blue']
• Below-threshold metric: COLORS['negative'] + italic
• Neutral / benchmark: COLORS['neutral']

Density Guidelines

• Strategy tearsheet: 15-20 data elements on one page
• Factor tearsheet: 12-16 data elements
• Comparison: metrics table + 4 chart panels
• Never more than 6 chart panels per page
• Metric tables: max 8 rows, keep them scannable

Print Checklist

• All text readable at 100% zoom on A4
• No text smaller than 7pt
• Grayscale test passes (markers + linestyles distinguish series)
• Margins safe for binding (left margin 15mm+)
• No colored backgrounds that waste toner
• Amber accent line visible in grayscale (~65% gray)
• PDF metadata populated (title, author, date)

---

A tearsheet is a persuasion document. The data makes the argument; the design makes it effortless to follow.