Risk Modeling GuideSAFE

🔬 A curated collection of 23,000+ agent skills for empirical research across 8 social science disciplines. | 精选 23,000+ AI Agent 技能库,覆盖8大社会科学学科的实证研究。CoPaper.AI 20分钟完成一篇可复现的规范实证论文,并支持用户上传 Skills。-- Maintained by CoPaper.AI from Stanford REAP.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: risk-modeling-guide
description: "Financial risk modeling including VaR, stress testing, and credit risk"
metadata:
  openclaw:
    emoji: "📉"
    category: "domains"
    subcategory: "finance"
    keywords: ["risk-modeling", "var", "stress-testing", "credit-risk", "monte-carlo", "basel"]
    source: "wentor"
---

# Risk Modeling Guide

A skill for quantitative financial risk modeling, covering Value at Risk, Expected Shortfall, credit risk, stress testing, and Monte Carlo simulation methods. Essential for financial engineering research and regulatory risk analysis.

## Market Risk: Value at Risk

### VaR Methodologies

| Method | Description | Pros | Cons |
|--------|-------------|------|------|
| Historical simulation | Replay past returns | No distributional assumption | Assumes past repeats |
| Variance-covariance | Assume normal returns | Fast, analytical | Underestimates tail risk |
| Monte Carlo simulation | Simulate from fitted model | Flexible distributions | Computationally expensive |
| Filtered historical simulation | GARCH + historical innovations | Captures volatility clustering | More complex |

### Implementation

```python
import numpy as np
import pandas as pd
from scipy.stats import norm, t as t_dist

def historical_var(returns: np.ndarray, confidence: float = 0.99,
                    horizon_days: int = 1) -> dict:
    """
    Compute Value at Risk using historical simulation.
    returns: array of daily log returns
    confidence: confidence level (e.g., 0.99 for 99% VaR)
    horizon_days: risk horizon in days
    """
    # Scale returns to horizon
    if horizon_days > 1:
        # Rolling sum for overlapping returns
        scaled_returns = pd.Series(returns).rolling(horizon_days).sum().dropna().values
    else:
        scaled_returns = returns

    alpha = 1 - confidence
    var = -np.percentile(scaled_returns, alpha * 100)
    es = -np.mean(scaled_returns[scaled_returns <= -var])

    return {
        "VaR": round(var, 6),
        "Expected_Shortfall": round(es, 6),
        "confidence": confidence,
        "horizon_days": horizon_days,
        "n_observations": len(scaled_returns),
    }

def parametric_var(returns: np.ndarray, confidence: float = 0.99,
                    distribution: str = "normal") -> dict:
    """
    Parametric VaR assuming normal or Student-t distribution.
    """
    mu = np.mean(returns)
    sigma = np.std(returns, ddof=1)

    if distribution == "normal":
        z = norm.ppf(1 - confidence)
        var = -(mu + sigma * z)
        # Analytical ES for normal
        es = -mu + sigma * norm.pdf(norm.ppf(1 - confidence)) / (1 - confidence)
    elif distribution == "student-t":
        # Fit Student-t
        df, loc, scale = t_dist.fit(returns)
        z = t_dist.ppf(1 - confidence, df)
        var = -(loc + scale * z)
        # ES for Student-t
        t_pdf = t_dist.pdf(t_dist.ppf(1 - confidence, df), df)
        es = -loc + scale * (t_pdf / (1 - confidence)) * ((df + z**2) / (df - 1))
    else:
        raise ValueError(f"Unknown distribution: {distribution}")

    return {
        "VaR": round(var, 6),
        "Expected_Shortfall": round(es, 6),
        "distribution": distribution,
        "mean": round(mu, 6),
        "std": round(sigma, 6),
    }
```

### Monte Carlo VaR

```python
def monte_carlo_var(returns: np.ndarray, n_simulations: int = 100000,
                     confidence: float = 0.99,
                     horizon_days: int = 10) -> dict:
    """
    Monte Carlo VaR using GBM (Geometric Brownian Motion).
    """
    mu = np.mean(returns)
    sigma = np.std(returns, ddof=1)

    # Simulate daily returns for the horizon
    rng = np.random.default_rng(42)
    simulated = rng.normal(
        mu * horizon_days,
        sigma * np.sqrt(horizon_days),
        size=n_simulations,
    )

    alpha = 1 - confidence
    var = -np.percentile(simulated, alpha * 100)
    es = -np.mean(simulated[simulated <= -var])

    return {
        "VaR": round(var, 6),
        "Expected_Shortfall": round(es, 6),
        "n_simulations": n_simulations,
        "confidence": confidence,
        "horizon_days": horizon_days,
    }
```

## Credit Risk Modeling

### Probability of Default Estimation

```python
from sklearn.linear_model import LogisticRegression

def build_pd_model(features: pd.DataFrame,
                    default_flag: pd.Series) -> dict:
    """
    Build a Probability of Default (PD) model using logistic regression.
    Common features: debt-to-income, credit utilization, payment history,
    employment length, loan amount.
    """
    model = LogisticRegression(max_iter=1000, class_weight="balanced")
    model.fit(features, default_flag)

    # Coefficient interpretation
    coef_df = pd.DataFrame({
        "feature": features.columns,
        "coefficient": model.coef_[0],
        "odds_ratio": np.exp(model.coef_[0]),
    }).sort_values("coefficient", ascending=False)

    # Model discrimination
    from sklearn.metrics import roc_auc_score
    pred_proba = model.predict_proba(features)[:, 1]
    auc = roc_auc_score(default_flag, pred_proba)

    return {
        "auc": round(auc, 4),
        "coefficients": coef_df.to_dict("records"),
        "intercept": round(model.intercept_[0], 4),
    }
```

### Loss Given Default and EAD

```python
def compute_expected_loss(pd_score: float, lgd: float,
                           ead: float) -> dict:
    """
    Compute Expected Loss = PD x LGD x EAD.
    pd_score: probability of default (0-1)
    lgd: loss given default (0-1, fraction of exposure lost)
    ead: exposure at default (dollar amount)
    """
    el = pd_score * lgd * ead
    return {
        "PD": pd_score,
        "LGD": lgd,
        "EAD": ead,
        "Expected_Loss": round(el, 2),
        "Unexpected_Loss_99": round(el * 2.33 * np.sqrt(pd_score * (1 - pd_score)), 2),
    }
```

## Stress Testing

### Scenario-Based Stress Tests

```python
def run_stress_test(portfolio_returns: pd.DataFrame,
                    

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.