Skip to content

Section A: Geometry Overview

Section A visualizes the 2D distribution of MFE vs MAE to reveal the overall shape of trade outcomes.

What It Does

  • Creates scatter plots showing every trade as a point in (MFE, MAE) space
  • Generates marginal distributions (histograms + KDE) for MFE and MAE separately
  • Calculates summary metrics (median, percentiles, win rate, correlation)
  • Supports outlier removal for cleaner visualizations

Key Questions Answered

  • What is the overall shape of outcomes?
  • Is there asymmetry suggesting an edge?
  • Are there obvious data quality issues (outliers)?
  • Is the signal "tight-stop friendly" or does it need room?

API Usage

from signal_analyzer import analyze, AnalysisConfig

# Run Section A
config = AnalysisConfig(H=20, sections=['A'])
result = analyze(df, sig_col='sig', config=config)

# Access metrics
metrics = result.section_a['long']['metrics']
print(f"Median MFE: {metrics['median_mfe']:.2f}%")
print(f"Median MAE: {metrics['median_mae']:.2f}%")
print(f"Win Rate: {metrics['win_rate']*100:.1f}%")
print(f"MFE 95th: {metrics['mfe_p95']:.2f}%")
print(f"MAE 5th: {metrics['mae_p5']:.2f}%")

Output Structure

The section_a result contains:

result.section_a = {
    'long': {
        'scatter': {
            'raw': {'mfe': np.array, 'mae': np.array},
            'trimmed': {'mfe': np.array, 'mae': np.array},
        },
        'marginals': {
            'mfe_bins': np.array,
            'mfe_counts': np.array,
            'mfe_kde_x': np.array,
            'mfe_kde_y': np.array,
            # ... same for MAE
        },
        'metrics': {
            'n_trades': int,
            'median_mfe': float,
            'median_mae': float,
            'win_rate': float,
            'mfe_p95': float,
            'mae_p5': float,
            'mae_mfe_corr': float,
        }
    },
    'short': { ... }  # Same structure
}

Generated Plots

Scatter Plot

Scatter Plot

What to look for:

  • Dense cluster near (0, 0): Weak follow-through, consider filtering signals
  • Right tail in MFE: Edge comes from subset of winners
  • Large negative MAE tail: Survival risk, need wide stops
  • Wedge shape: Good asymmetry (winners bigger than losers)
  • Diagonal correlation: MFE and MAE scale together (regime-mixing?)

Marginal Distributions

Marginal Distributions

What to look for:

  • MFE distribution: Right-skewed = few big winners carry the edge
  • MAE distribution: Left-skewed = fat tail of large losers
  • Bimodal distributions: Multiple trade archetypes (see Section F)
  • Tight distributions: Consistent behavior, predictable outcomes

Interpreting Metrics

Median MFE/MAE

The typical achievable profit and typical max drawdown. Use these as baseline targets/stops.

Example: 

Median MFE: 2.5%
Median MAE: -1.2%
→ Target ~2.5% profit, expect ~1.2% drawdown on typical trade.

Win Rate

Percentage of trades with MFE > 0. Note: This is MFE-based (best achievable), not final P&L.

Example: 

Win Rate: 65%
→ 65% of trades show some profit potential within horizon H.

Percentiles

MFE 95th percentile = best 5% of trades. MAE 5th percentile = worst 5% of trades.

Example: 

MFE 95th: 8.5%
MAE 5th: -4.2%
→ Best trades can reach 8.5%. Worst drawdowns hit -4.2%.

Correlation

Correlation between MFE and MAE. High positive correlation suggests regime-mixing or volatility dependence.

Example: 

MAE/MFE Corr: 0.75
→ Strong positive correlation. Big winners also have big drawdowns. Consider Section E (vol normalization).

Configuration Options

Outlier Removal

Control trimming with trim_method and trim_k:

# IQR method (default)
config = AnalysisConfig(trim_method='iqr', trim_k=1.5)

# More permissive (keep more outliers)
config = AnalysisConfig(trim_method='iqr', trim_k=3.0)

# Percentile method
config = AnalysisConfig(trim_method='percentile', trim_k=5)  # Remove top/bottom 5%

# No trimming
config = AnalysisConfig(trim_method=None)

When to use:

  • Default (IQR, k=1.5): Most cases, removes extreme outliers
  • Permissive (k=3.0): When you have clean data or want to keep tail events
  • Percentile: When you know exact percentage to trim
  • None: When every trade matters (e.g., small sample size)

Decision Guide

Based on your Section A output:

Observation Action
Dense cluster at (0,0) Signal has weak follow-through. Consider filtering or shorter horizon.
Wedge shape (wide MFE, narrow MAE) Good asymmetry. Proceed to Section B for stop sizing.
High MAE/MFE correlation Regime-mixing or vol-dependent. Run Section E.
Bimodal distributions Multiple trade types. Run Section F for clustering.
Win rate <50% Edge comes from few big winners. Need patient exits.
Win rate >70% High hit rate. Use smaller targets, wider stops.

Example Interpretation

Long Trades (n=1247):
  Median MFE:  1.85%
  Median MAE: -0.92%
  Win Rate:    68.5%
  MFE 95th:    5.42%
  MAE 5th:    -3.18%
  Correlation: 0.31

Interpretation:

  • Positive edge: Median MFE (1.85%) > |Median MAE| (0.92%). Good asymmetry.
  • High win rate: 68.5% hit rate suggests mean-reversion or tight signal.
  • Reasonable tails: 95th MFE at 5.42% shows upside. 5th MAE at -3.18% is manageable.
  • Low correlation: 0.31 suggests consistent behavior across regimes.

Next steps:

  • Section B: Find optimal stop size (likely around 1-2% based on MAE)
  • Section C: Check if trailing stops work (high win rate suggests MFE-first)
  • Section D: Validate TP=2%, SL=1% feasibility

API Reference