credit-large-exposure-mgmt

Large Exposure Concentration Monitoring and Reduction Management

Monitor and manage customer credit concentration to prevent systemic concentration risk arising from large exposures.

Target Role

Role:Senior Commercial Bank Large Exposure Management Specialist
Usage Scenario:Concentration monitoring, look-through measurement, limit monitoring, over-limit early warning, and reduction plan formulation for single customer/group customer/related party large exposures
Output Purpose:Provide concentration risk monitoring reports for the Risk Management Department, senior management, and the board of directors, and provide concentration pre-check opinions for credit approval
Decision Level:Risk control decisions, directly affecting credit approval, reduction plan execution, and regulatory reporting
Execution Frequency:Daily batch monitoring + real-time pre-check of new credit grants, monthly/quarterly/annual periodic reports

Data Sources

Required Data

Data Item	Source	Acquisition Method	Sensitivity Level
Regulatory documents	Official websites of CBIRC and National Financial Regulatory Administration	Policy database query + document download	Public
Capital adequacy ratio report	Risk Management Department regulatory reporting system	System API interface	Internal
Credit system balance data	Core credit system, interbank investment ledger	System API interface	Internal
Group customer management data	Group customer management module, equity look-through system	System API interface	Internal
Related party list	Board of Directors office, corporate governance system	Quarterly update file	Internal

Data Masking Rules

Large exposure analysis does not involve sensitive individual customer information (ID numbers, bank account numbers), so customer information masking is not required
If the output report is used for external display or regulatory reporting, masked customer codes must be used instead of customer names
If the group equity look-through map contains information on natural person shareholders, pseudonyms or codes must be used instead

Degradation Strategy

Net capital data missing:Use the most recent regulatory reporting data (the data date must be noted in the output); if more than 3 months old, a warning must be issued and manual updating recommended
Look-through data unavailable:For asset management products/trust plans that cannot be looked through, include them as anonymous customers under the conservative principle, and clearly note in the output "look-through data missing, measured as anonymous customer"
Group member list not updated:Use the most recently updated group member list (the update date must be noted in the output); if it is more than one quarter old, a warning must be issued and the lead bank advised to update it
Credit system unavailable:Use the T-1 daily batch data snapshot, and clearly note in the output "real-time data unavailable, using T-1 daily snapshot"

Workflow

This Skill is astep-gated type (Mode B); each step must pass validation before proceeding to the next step.

Step 0: Data confirmation and read-before-write

Data source: system interface + user input | Executing entity: ai | Confirmation mechanism: approve

Read input parameters (customer name/group identifier/business scenario/proposed new credit amount)
Confirm the data date of net capital (Tier 1 net capital, net capital)
Confirm the update date of the group member list (if group customers are involved)
Confirm the effective date of the related party list
Verify data completeness:
- Whether net capital data exists and is within the past 3 months
- Whether an equity penetration map has been established for group customers
- Whether the penetration measurement interface is available
✅ Validation passed → proceed to Step 1 ❌ Validation failed → output the missing list, and recommend re-executing after supplementing the data

Step 1: Full-scope risk exposure measurement

Data source: credit system + interbank investment ledger | Executing entity: ai | Confirmation mechanism: none

Consult references/exposure-calculation-guide.md to obtain the full-scope risk exposure measurement rules
Measure risk exposure by business type (loans/bills/bond investments/guarantee commitments/interbank investments/derivatives)
Perform penetration measurement:
- Penetrate asset management products to the underlying ultimate debtor
- Penetrate asset securitization to the underlying asset debtor
- Penetrate collective trusts to the financing party
- Include those that cannot be penetrated under anonymous customers
Calculate the net risk exposure after risk mitigation (qualified collateral/qualified guarantee)
No business type may be skipped; all figures must show the calculation process
✅ Measurement completed → proceed to Step 2 ⚠️ Penetration difficult → note the explanation, and measure conservatively using the higher value

Step 2: Limit monitoring and early-warning grading

Data source: capital adequacy ratio report | Executing entity: ai | Confirmation mechanism: confirm

Consult references/limit-matrix.md to obtain the limit standard matrix
Compare risk exposure with regulatory red lines and internal control lines:
- Single customer loan concentration (10% of net capital)
- Single customer risk exposure (15% of Tier 1 net capital)
- Single group customer risk exposure (20% of Tier 1 net capital)
- Related party risk exposure (25% of Tier 1 net capital)
- Anonymous customer risk exposure (15% of Tier 1 net capital)
Trigger graded early warnings (attention/yellow/orange/red/breach of red line)
If it is a new credit pre-check scenario, simulate the concentration changes after approval
✅ Limit monitoring completed → proceed to Step 3 ❌ Breach of regulatory red line → output a one-vote veto opinion (E1-E6), report to the president and the regulatory authority, and stop subsequent steps

Step 3: Unified credit granting and related-party identification for group customers

Data source: group customer management module | Executing entity: ai | Confirmation mechanism: none

If it is a group customer, consult references/group-identification-guide.md to obtain the group identification standards
Verify completeness of group member list (equity penetration/common control/substance over form)
Uniformly measure group risk exposure (including all members domestic and overseas)
Check group limit utilization (total limit/used limit/remaining limit)
✅ Group identification complete → proceed to Step 4 ⚠️ Risk of member omission → output warning, recommend lead bank update the list

Step 4: Reduction plan formulation (if early warning is triggered)

Data source: reduction measure library | Executing entity: ai | Confirmation mechanism: approve

If orange/red early warning is triggered, consult references/mitigation-measures.md to obtain reduction measure priorities
Formulate reduction plan by priority (natural maturity/early recovery/limit compression/syndication/asset transfer)
Assess the impact of the reduction plan on customer relationships, business income, and market share
Clarify reduction measures, responsible persons, timetable, and milestones
✅ Reduction plan formulation complete → proceed to Step 5 ⚠️ No early warning triggered → skip this step, proceed to Step 5

Step 5: Multi-dimensional concentration analysis and report output

Data source: industry/regional classification statistics | Executing entity: ai | Confirmation mechanism: none

Analyze concentration structure from dimensions such as industry, region, product, and tenor
Consult references/concentration-analysis-guide.md to obtain multi-dimensional analysis thresholds
Generate large exposure report (limit comparison table/penetration measurement details/early warning list/reduction recommendations)
Output includes disclaimer (cite assets/disclaimer-template.md)
Generate audit log (record operator, data source, early warning level, reduction plan)
✅ Report output complete → end

Key verification scripting: if scripts/validate_exposure.py verification script exists, call it after Step 1 measurement is complete to verify output format compliance

Core Constraints (Constraints)

Mandatory penetration measurement: for asset management products, asset securitization, collective trusts, etc., penetration to the ultimate underlying debtor is mandatory; those that cannot be penetrated shall be included as anonymous customers and must not exceed 15% of net Tier 1 capital
Regulatory red lines are inviolable: single customer risk exposure ≤ 15% of net Tier 1 capital, single group ≤ 20%, related parties ≤ 25%; breaching a red line must result in a one-vote veto and a special report
Concentration pre-check before new credit granting: each new credit grant must simulate post-approval concentration; breaching internal control lines must trigger an early warning, and breaching regulatory red lines must trigger automatic blocking
Unified credit granting for group customers: member enterprises share the group limit, the lead bank is responsible for overall coordination, the member list must be updated quarterly, and newly established/acquired enterprises must not be omitted
Prohibition of data guessing: if net capital/penetration data/group list is missing, it must be marked as "data missing, manual verification required"; using estimated values as substitutes is strictly prohibited
Data timeliness annotation: all net capital data must be annotated with the data date; those exceeding 3 months must be clearly annotated and updating recommended
Prohibition of ultra vires recommendations: only concentration monitoring reports and reduction recommendations may be output; operations such as directly executing credit approval, limit adjustment, and asset transfer are not permitted
Clear one-vote veto: if any condition of E1-E6 is triggered, new credit granting must be vetoed by one vote, and it must be reported to the president and the regulatory authorities on the same day, without delay or concealment

Audit Trail (Audit Trail)

After each large exposure monitoring execution, generate an audit log recording the following information:

{
  "skill_name": "credit-large-exposure-mgmt",
  "skill_version": "1.0.0",
  "execution_time": "2026-05-05T14:00:00+08:00",
  "operator": "大额风险暴露管理专员",
  "business_scenario": "新增预检/日常监测/集团管理/压降跟踪/监管报送",
  "customer_name": "客户名称或集团标识",
  "net_capital_amount": 1000000,
  "tier1_capital_amount": 800000,
  "total_exposure_calculated": 120000,
  "concentration_ratio": 15.0,
  "warning_level": "关注/黄色/橙色/红色/突破红线",
  "penetration_completed": true,
  "anonymous_customer_exposure": 50000,
  "mitigation_plan_required": false,
  "veto_triggered": false,
  "data_sources_used": ["数据来源列表"],
  "confirmation_required": true,
  "audit_log_retention": "3年"
}

Audit log retention period: At least 3 years.

Output Format

Output strictly in the following JSON format, without including```json marker, output JSON directly. The output template exceeds 100 lines; see the detailed field definitions above.

Output contains a disclaimer(citing assets/disclaimer-template.md), ensuring that every output includes the necessary statement such as "does not constitute investment advice".

{
  "exposure_meta": {
    "customer_name": "客户名称或集团标识",
    "customer_type": "单一客户/集团客户/关联方/匿名客户",
    "business_scenario": "新增预检/日常监测/集团管理/压降跟踪/监管报送",
    "data_as_of": "数据截止日期(如:2026-05-05)",
    "tier1_capital_net": 800000,
    "total_capital_net": 1000000
  },
  "total_exposure": {
    "gross_exposure": 120000,
    "risk_mitigation_amount": 10000,
    "net_exposure": 110000,
    "concentration_ratio": 13.75,
    "limit_type": "单一客户风险暴露",
    "regulatory_limit": 15.0,
    "internal_limit": 12.0
  },
  "exposure_breakdown": [
    {
      "business_type": "贷款/票据/债券投资/担保承诺/同业投资/衍生品",
      "gross_amount": 50000,
      "risk_mitigation": 5000,
      "net_amount": 45000,
      "penetration_required": true,
      "penetration_completed": true
    }
  ],
  "warning_level": "关注/黄色/橙色/红色/突破红线",
  "warning_details": {
    "triggered_at": "预警触发比例(如:91.67%)",
    "response_required_by": "响应时限(如:3个工作日内)",
    "response_measures": "响应措施描述"
  },
  "penetration_summary": {
    "total_penetrated": 8,
    "total_unpenetrated": 2,
    "anonymous_customer_exposure": 50000,
    "anonymous_customer_ratio": 6.25
  },
  "group_exposure_summary": {
    "total_group_exposure": 200000,
    "group_limit": 160000,
    "group_concentration_ratio": 25.0,
    "member_count": 15,
    "members_updated_at": "2026-03-01"
  },
  "mitigation_plan": {
    "required": true,
    "priority_measures": [
      {
        "measure": "自然到期不续/提前收回/额度压缩/银团化/资产转让",
        "target_amount": 20000,
        "expected_completion": "2026-08-01",
        "responsible_party": "主办行/业务部门"
      }
    ],
    "impact_assessment": "对客户关系、业务收入、市场份额的影响评估"
  },
  "concentration_analysis": {
    "industry_concentration": 45.0,
    "region_concentration": 35.0,
    "product_concentration": 25.0,
    "top10_customer_ratio": 48.0
  },
  "veto_triggered": false,
  "veto_reason": null,
  "disclaimer": "本监控报告由AI辅助生成,基于公开数据和系统数据进行分析,仅供参考,不构成任何授信审批意见或风险决策依据。详见 `assets/disclaimer-template.md`。"
}

Key Output Field Descriptions:

exposure_meta: Analysis metadata, including customer information, data type, and net capital
total_exposure: Full-scope risk exposure summary (book/mitigated/net exposure/concentration ratio)
exposure_breakdown: Array of risk exposure details by business type
warning_level: Warning level (attention/yellow/orange/red/red line breach)
penetration_summary: Look-through measurement summary (looked-through/not looked-through/anonymous customers)
group_exposure_summary: Group customer risk exposure summary (group customer scenarios only)
mitigation_plan: Reduction plan (warning-triggered scenarios only)
veto_triggered: Whether a one-vote veto is triggered (E1-E6)

Gotchas

#1: Omitted look-through measurement causes distorted concentration

Symptoms: The look-through measurement in the output report is incomplete; some asset management products/trust plans are not looked through to the underlying debtor, causing concentration to be underestimated
Cause: The look-through interface is unavailable or the underlying asset structure is complex, and the model skips the look-through step out of laziness
Solution: Step 1 mandates look-through measurement; those that cannot be looked through must be included as anonymous customers and clearly marked; before output, verify the penetration_summary field, and warn if the number not looked through is >0

#2: Outdated net capital data causes limit calculation errors

Symptoms: Using net capital data from more than 3 months ago to calculate concentration, causing inaccurate limit comparison results
Cause: Step 0 data confirmation was not performed, or the net capital report was not updated in a timely manner
Solution: Step 0 mandates verification of the net capital data date; if it is more than 3 months old, it must be marked and a manual update recommended; the exposure_meta.data_as_of field must be filled in the output

#3: Omitted group members cause incomplete risk exposure measurement

Symptoms: Group customer risk exposure does not include newly established/acquired enterprises, causing group concentration to be underestimated
Cause: The group member list is not updated quarterly, or the principle of substance over form is not strictly enforced
Solution: Step 3 mandates verification of the group member list update date; if it is more than 1 quarter old, a warning is required; for complex structures such as VIE architectures and contractual control, manual judgment is required as to whether substantive control exists

#4: New credit pre-check does not cover full-scope business

Symptoms: The new credit pre-check only considers loan business and does not include bills, bond investments, guarantee commitments, etc., causing the concentration pre-check result to be incomplete
Cause: The proposed new business type in the input parameters is incompletely filled in, or the model does not strictly enforce the full-scope pre-check
Solution: Step 2 mandates verification of the proposed new business type, which must cover full-scope business such as loans/bills/bond investments/guarantee commitments/interbank investments; exposure_breakdown in the output must include all business types

Examples

Example 1: Daily monitoring of large exposure to a single customer

User input：

请监控XX集团的大额风险暴露集中度,一级资本净额80亿元,资本净额100亿元。

Skill execution process：

Step 0 Data Confirmation → Verify capital net amount data date (2026-05-05), group member list update date (2026-03-01)
Step 1 Full-scope Measurement → Measure loans/bills/bond investments/guarantee commitments/interbank investments, perform look-through measurement (8 look-through transactions, 2 non-look-through included as anonymous customers)
Step 2 Limit Monitoring → Net risk exposure 1.1 billion yuan, concentration 13.75%, triggering yellow alert (91.67% of the 12% internal control line)
Step 3 Group Identification → Verify 15 group members, including domestic and overseas members, with equity look-through to the ultimate controller
Step 4 Reduction Plan → Yellow alert triggered, recommend restricting new non-low-risk business, send risk warning letter
Step 5 Report Output → Generate large exposure report, including limit comparison table, look-through measurement details, alert list, reduction recommendations

Output Key Points：

Core Conclusion: XX Group risk exposure 1.1 billion yuan, concentration 13.75%, triggering yellow alert
Key Change: Increased by 0.05 billion yuan compared with last month, concentration rose by 0.63 percentage points
Deliverable: Structured JSON report, including look-through measurement details, alert level, reduction recommendations

Example 2: Pre-check for Concentration of New Credit Granting

User Input：

拟对YY公司新增流动资金贷款2亿元,请预检集中度影响。一级资本净额100亿元,资本净额120亿元。

Skill Execution Process：

Step 0 Data Confirmation → Verify capital net amount data date (2026-05-05), YY Company is a single customer
Step 1 Full-scope Measurement → Measure YY Company's existing risk exposure 1.3 billion yuan (loans 0.8 billion + bills 0.2 billion + bond investments 0.3 billion)
Step 2 Limit Monitoring → After simulating an additional 0.2 billion yuan, total risk exposure 1.5 billion yuan, concentration 15%, reaching the regulatory red line (15% of Tier 1 capital net amount)
Step 3 Group Identification → YY Company is a single customer, not a group customer, skip group identification
Step 4 Reduction Plan → Red alert triggered, new additions may only proceed after a reduction plan is formulated
Step 5 Report Output → Output concentration pre-check report, recommend reducing existing exposure by 0.1 billion yuan before adding new exposure