Technical Implementation of Equity Penetration: Graph Modeling, Algorithms, and Engineering Practices

In enterprise risk management, compliance due diligence, and relationship analysis, “equity penetration” is a fundamental technical capability. Starting from a target enterprise, it traces up the investment chain layer by layer, piercing through complex corporate structures to ultimately identify the natural person shareholders or actual controllers behind it. It also provides precise indirect shareholding percentages and complete paths. While it may seem like a simple recursive action of “finding shareholders, then finding their shareholders,” engineering it involves data modeling, graph algorithms, data quality governance, and explainability. This article systematically breaks down the complete technical implementation path of equity penetration.

1. Clarifying Requirements: Questions to Answer

Before writing code, business requirements must be broken down into computable problems. There are generally four categories:

1. Identifying Ultimate Beneficial Owners (UBOs): The industry commonly uses a threshold of cumulative shareholding percentage (e.g., 25%) for natural persons as the screening line for beneficial owners, supplemented by rules for executives and financial officers.
2. Calculating Indirect Shareholding Percentages: A target enterprise may be held indirectly through multiple layers of subsidiaries. It is necessary to multiply the shareholding ratios along each path and then sum the ratios of multiple paths leading to the same entity.
3. Detecting Anomalous Structures: Examples include circular shareholding (A holds shares in B, and B indirectly holds shares in A), suspected proxy holdings, or nesting levels that significantly exceed reasonable ranges. These are often critical risk signals.
4. Outputting Explainable Penetration Paths: Compliance scenarios require auditable conclusions. The system must provide a complete evidence chain of “who—through which chain—holds how much,” rather than just returning a single number.

These four requirements dictate the subsequent model and algorithm design: the computation must be both accurate and explainable.

2. Data Modeling: Representing Equity Relationships as Graphs

An equity network is naturally a graph structure: nodes are enterprises (legal entities) and natural persons, and edges are shareholding relationships. A minimally viable shareholding edge table typically includes the following fields:

• Shareholder identifier (enterprise or natural person)
• Invested enterprise identifier
• Shareholding ratio (recommended to be stored uniformly as a decimal for easy path product calculation)
• Capital contribution method and amount
• Data timestamp (registration date / change date)
• Data source and confidence level

There are two common routes for storage selection. The first is a relational database, utilizing recursive CTEs for hierarchical expansion, which is suitable for scenarios with a manageable scale and fixed query patterns. The second is a graph database, which natively supports multi-hop traversals and path queries, making it suitable for scenarios with large network scales, indeterminate depths, and complex graph algorithms requiring cycle detection. Neither is absolutely superior; the choice depends on data volume and query complexity. In practice, organizations often build the entire pipeline using general cloud platform capabilities—for example, using a graph database to carry the relationship network, object storage to save raw corporate registration materials, message queues to asynchronously process batch penetration tasks, logging services to沉淀 audit records, and monitoring/alerting to ensure task stability. Cloud platforms like Baidu AI Cloud provide these types of general-purpose components.

3. Core Algorithms: Multi-Layer Penetration and Ratio Calculation

3.1 Traversal Strategy

Using the target enterprise as the starting point, perform a graph traversal (either Breadth-First Search or Depth-First Search) to expand shareholders layer by layer. The traversal must terminate under three conditions:

• Reaching a natural person node: Equity penetration cannot continue beyond a natural person.
• Cumulative ratio falls below the pruning threshold: When a single path’s contribution is too small (e.g., less than 0.1%), its impact on the total is negligible. Early pruning significantly reduces computation.
• Reaching the maximum penetration depth: This prevents extremely long chains from degrading performance and controls the complexity of the results.

3.2 Ratio Calculation

The general definition of an indirect shareholding ratio is: for each path between the target enterprise and a specific shareholder, multiply the shareholding ratios of each segment along the path to obtain that path’s contribution, then sum all paths. The formula is:

Total Shareholding of a Shareholder = Σ (Product of ratios along each path)

Note that the same shareholder may hold shares in the target enterprise through multiple intermediate chains. These must be merged and summed, rather than just taking the maximum path, otherwise the shareholding will be systematically underestimated.

3.3 Cycle Handling

Circular shareholding will cause naive recursion to fall into an infinite loop. In engineering, three methods are typically combined: maintain a visited set for each traversal path to avoid re-entering the same node; set a global maximum depth as a fallback; and memoize the visited “enterprise—cumulative ratio” states to skip redundant expansions.

Below is a schematic implementation using a recursive CTE (SQL pseudo-code):

WITH RECURSIVE chain AS (
    SELECT :root AS company_id, shareholder_id, ratio,
           ratio AS acc_ratio, 1 AS depth,
           ARRAY[shareholder_id] AS path
    FROM equity WHERE company_id = :root
  UNION ALL
    SELECT c.company_id, e.shareholder_id, e.ratio,
           c.acc_ratio * e.ratio, c.depth + 1,
           c.path || e.shareholder_id
    FROM equity e
    JOIN chain c ON e.company_id = c.shareholder_id
    WHERE c.depth < 6                                 -- Depth fallback
      AND NOT e.shareholder_id = ANY(c.path)          -- Cycle protection
      AND c.acc_ratio * e.ratio >= 0.001              -- Threshold pruning
)
SELECT shareholder_id, SUM(acc_ratio) AS total_ratio
FROM chain
GROUP BY shareholder_id
ORDER BY total_ratio DESC;

The same logic expressed in Python is as follows:

from collections import defaultdict
def penetrate(edges, root, max_depth=6, min_ratio=0.001):
    """edges: {company_id: [(shareholder_id, ratio), ...]}"""
    total = defaultdict(float)
    stack = [(root, 1.0, 0, frozenset())]
    while stack:
        company, acc, depth, path = stack.pop()
        if depth >= max_depth:
            continue
        for sh, r in edges.get(company, []):
            new_acc = acc * r
            if new_acc < min_ratio:
                continue                      # Pruning
            total[sh] += new_acc              # Multi-path merge
            if sh in edges and sh not in path:
                stack.append((sh, new_acc, depth + 1, path | {sh}))
    return dict(total)

These two code snippets are skeletal demonstrations. A real system would also need to incorporate processing logic for dimensions such as executive roles, equity freeze statuses, and change history.

4. From Shareholding Ratio to Actual Controller

The shareholding ratio is only the starting point. Identifying the actual controller typically requires overlaying the following general rules:

• Relative Control Assessment: In the absence of a single shareholder holding the majority, compare the gap between the largest shareholder and others, combined with governance information such as board seats, to make a comprehensive judgment.
• Concerted Action Consolidation: Treat multiple shareholders with concerted action agreements as a single entity when calculating voting rights.
• Agreements and Special Arrangements: Voting rights entrustment, or the arrangement of general partners in limited partnerships, may grant actual control to a party with a lower shareholding ratio.
• Executive Affiliations: Cross-appointments of legal representatives, directors, supervisors, and senior executives can serve as auxiliary identification signals.

A mature system will consolidate these rules into a configurable rule engine, allowing compliance personnel to adjust thresholds and weights according to regulatory requirements, rather than hardcoding the logic into the traversal code. This ensures rule adjustments take effect without requiring software deployments.

5. Engineering Implementation: Data, Performance, and Explainability

Data quality determines the upper limit. Corporate data universally suffers from missing values, delays, and inconsistent definitions. For instance, when shareholding ratios are missing, they must be inferred from capital contribution amounts and registered capital. Multi-source data cross-validation, retaining source and confidence fields, and establishing data change subscription mechanisms are the foundations for ensuring trustworthy penetration results.

Timestamp Consistency. Equity changes continuously. Penetration reports must be anchored to a unified data snapshot timestamp; otherwise, mixing current-month data upstream with last-year data downstream will lead to incorrect conclusions. When historical retrospection is needed, shareholding relationships must also be versioned along the time dimension.

Performance Optimization. For super-large enterprise groups, full-graph expansion is highly expensive. Common techniques include threshold pruning, on-demand lazy loading of subgraphs, caching results for hotspot subgraphs, and batch task asynchronization (submitting tasks to be distributed via message queues, with callback notifications upon completion).

Explainability. Every conclusion should be able to replay the complete path and intermediate calculation process for manual review. The penetration tasks themselves must record audit logs to meet compliance traceability requirements. Penetration results lacking explainability have virtually no usability in compliance scenarios.

6. Common Pitfalls

1. Penetrating solely based on shareholding ratios while ignoring voting rights and special agreements, leading to misjudgment of the actual controller.
2. Ignoring the time dimension, using a current snapshot to answer questions about a historical point in time.
3. Failing to set limits on depth and thresholds, causing performance issues on massive equity networks.
4. Returning only numbers without paths, making conclusions unauditable and leaving the business side afraid to trust them.

Conclusion

The essence of equity penetration is performing pruned, cycle-protected, and explainable multi-hop traversals with weighted calculations on a continuously evolving graph. The algorithms themselves are not mysterious; the real barrier lies in data quality and timestamp governance. By clearly modeling equity relationships first, making penetration rules configurable, and logging the calculation process, you can build an equity penetration analysis service that is both accurate and capable of withstanding compliance scrutiny.