algo-risk-credit

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Credit Scoring Model

信用评分模型

Overview

概述

Credit scoring models predict the probability of default (PD) from borrower characteristics using logistic regression or gradient boosting. Output: a score (300-850 range) or PD (0-1). Used for loan approval, pricing, and portfolio risk management.

信用评分模型利用logistic regression或gradient boosting算法，基于借款人特征预测违约概率（PD）。输出结果为300-850区间的评分或0-1区间的PD，用于贷款审批、定价及信贷组合风险管理。

When to Use

适用场景

Trigger conditions:

Building a scorecard for loan/credit approval decisions
Predicting default probability for risk-based pricing
Evaluating existing credit models for discriminatory power

When NOT to use:

For corporate bankruptcy prediction (use Altman Z-Score)
For market risk measurement (use VaR)

触发条件：

为贷款/信贷审批决策构建评分卡
基于风险定价预测违约概率
评估现有信用模型的区分能力

不适用场景：

企业破产预测（请使用Altman Z-Score）
市场风险计量（请使用VaR）

Algorithm

算法

IRON LAW: A Credit Model Must Discriminate AND Be Calibrated
Discrimination (AUC): correctly ranking good vs bad borrowers.
Calibration: predicted PD matches actual default rates.
A model with AUC=0.85 but predicted PD 2x actual default rate will
cause systematic over/under-pricing. Need BOTH properties.

IRON LAW: A Credit Model Must Discriminate AND Be Calibrated
Discrimination (AUC): correctly ranking good vs bad borrowers.
Calibration: predicted PD matches actual default rates.
A model with AUC=0.85 but predicted PD 2x actual default rate will
cause systematic over/under-pricing. Need BOTH properties.

Phase 1: Input Validation

阶段1：输入验证

Collect: borrower features (income, debt ratio, credit history length, delinquency count, utilization), outcome variable (default within 12-24 months). Handle: missing values, class imbalance (typically 2-5% default rate). Gate: Sufficient defaults (300+ events), features available at decision time.

收集：借款人特征（收入、债务比率、信用历史时长、逾期次数、信贷利用率）、结果变量（12-24个月内是否违约）。处理：缺失值、类别不平衡（通常违约率为2-5%）。 准入要求： 违约样本充足（300+个事件），决策时可获取相关特征。

Phase 2: Core Algorithm

阶段2：核心算法

Feature engineering: WOE (Weight of Evidence) binning for logistic regression, or direct encoding for GBDT
Train model: logistic regression (interpretable, regulatory-preferred) or GBDT (higher accuracy)
Calibrate: Platt scaling on holdout, ensure predicted PD matches actual default rate by decile
Convert to score: Score = offset + factor × log(odds), scaled to 300-850 range

特征工程：针对logistic regression使用WOE（Weight of Evidence）分箱，针对GBDT使用直接编码
模型训练：选择logistic regression（可解释性强，受监管偏好）或GBDT（精度更高）
模型校准：在验证集上使用Platt scaling，确保分位区间内预测PD与实际违约率匹配
转换为评分：评分 = 偏移量 + 系数 × log(优势比)，缩放至300-850区间

Phase 3: Verification

阶段3：验证

Evaluate: AUC (>0.70 acceptable, >0.80 good), KS statistic, Gini coefficient. Population stability index (PSI) for monitoring drift. Gate: AUC > 0.70, calibration acceptable, no discriminatory bias in protected attributes.

评估指标：AUC（>0.70为可接受，>0.80为良好）、KS statistic、Gini coefficient。使用Population Stability Index（PSI）监控数据漂移。 准入要求： AUC>0.70，校准效果合格，对受保护属性无歧视性偏差。

Phase 4: Output

阶段4：输出

Return score, PD, and key risk drivers.

返回评分、PD及关键风险驱动因素。

Output Format

输出格式

json

{
  "score": 680,
  "pd": 0.035,
  "risk_grade": "B",
  "top_risk_factors": [{"factor": "high_utilization", "impact": -45}, {"factor": "short_history", "impact": -30}],
  "metadata": {"model": "logistic_regression", "auc": 0.78, "vintage": "2024-Q3"}
}

json

{
  "score": 680,
  "pd": 0.035,
  "risk_grade": "B",
  "top_risk_factors": [{"factor": "high_utilization", "impact": -45}, {"factor": "short_history", "impact": -30}],
  "metadata": {"model": "logistic_regression", "auc": 0.78, "vintage": "2024-Q3"}
}

Examples

示例

Sample I/O

输入输出样例

Input: Borrower: income=$60K, DTI=35%, 5yr credit history, 0 delinquencies, 60% utilization Expected: Score ~680, PD ~3.5%, Grade B (some risk from high utilization)

输入： 借款人：收入6万美元，DTI=35%，5年信用历史，0次逾期，信贷利用率60% 预期输出： 评分约680，PD约3.5%，风险等级B（高信贷利用率带来一定风险）

Edge Cases

边缘案例

Input	Expected	Why
No credit history (thin file)	High uncertainty, default to conservative	Insufficient data for scoring
All features identical	Same score regardless of outcome	Model can't differentiate — need more features
Major economy shift	PSI > 0.25, model needs recalibration	Population has shifted from training distribution

输入	预期输出	原因
无信用历史（薄文件）	不确定性高，默认采用保守评估	评分数据不足
所有特征完全相同	无论结果如何评分一致	模型无法区分，需补充更多特征
经济环境重大变动	PSI>0.25，模型需重新校准	样本分布与训练集差异过大

Gotchas

注意事项

Reject inference: Training data only includes approved applicants. Rejected applicants' outcomes are unknown, creating selection bias. Use reject inference techniques.
Fair lending: Models must not discriminate by protected attributes (race, gender, age). Even proxy variables (zip code ≈ race) can create disparate impact. Test with fairness metrics.
Through-the-door vs on-the-books: TTD samples include all applicants; OTB only approved ones. Model purpose determines which sample to use.
Vintage analysis: Default rates vary by economic conditions. A 2019-trained model may not predict well in a recession. Track model performance by vintage.
Regulatory requirements: Financial regulators (Basel, OCC, FDIC) have specific requirements for model validation, documentation, and fair lending testing.

拒绝推断：训练数据仅包含获批申请人，被拒申请人的违约结果未知，会导致选择偏差。需使用拒绝推断技术。
公平借贷：模型不得因受保护属性（种族、性别、年龄）产生歧视。即使是代理变量（如邮政编码≈种族）也可能造成差异化影响，需使用公平性指标测试。
申请样本 vs 在贷样本：TTD样本包含所有申请人；OTB样本仅包含获批申请人。需根据模型用途选择对应样本。
** vintange分析**：违约率随经济环境变化。2019年训练的模型在衰退期预测效果可能不佳，需按vintage跟踪模型性能。
监管要求：金融监管机构（巴塞尔委员会、OCC、FDIC）对模型验证、文档记录及公平借贷测试有明确要求。

References

参考资料

For WOE binning methodology, see
```
references/woe-binning.md
```
For reject inference techniques, see
```
references/reject-inference.md
```

WOE分箱方法请参考
```
references/woe-binning.md
```
拒绝推断技术请参考
```
references/reject-inference.md
```