Equity Penetration: Identifying Affiliated Risks and Verifying Information

1. Core Problems Equity Penetration Must Solve

Equity penetration is not simply about listing multi-layer shareholder information. It must answer several business-relevant questions:

1. Who ultimately controls this company?
2. Are there hidden affiliations between this company and another?
3. Do complex structures exist, such as circular ownership, nominee holdings, or concert-party arrangements?
4. Under a multi-layer ownership structure, how is a given shareholder’s actual beneficial ratio calculated?

The answers to these questions often do not appear directly in business registration records. When the ownership structure exceeds three layers, or involves entities across provinces or even borders, manual layer-by-layer lookup is nearly infeasible. In engineering practice, equity penetration therefore typically combines data collection, graph computation, a rules engine, and human review.

From a technical perspective, equity penetration can be abstracted as a directed graph: nodes represent companies or natural persons, edges represent shareholding relationships, and edge weights are shareholding ratios. The penetration process starts from a target node and traces upward along edges until a termination condition is met. Termination conditions usually include: reaching a natural person, reaching a state-owned entity, reaching an overseas entity, hitting a preset depth limit, or falling below a shareholding ratio threshold.

But real-world ownership structures are far more complex than this graph. Circular ownership introduces cycles, concert-party agreements create control links between entities with no equity relationship, and special entities such as partnerships, funds, and trusts introduce additional penetration conventions. A usable equity penetration system must therefore add rule-based judgment and human review mechanisms beyond graph traversal.

2. Data Sources and Quality Constraints

Equity penetration results depend heavily on the completeness and timeliness of underlying data. Available data sources mainly include:

• Business registration information: shareholder names, subscribed capital contributions, contribution ratios, legal representatives, and executive information;
• Periodic reports of listed entities: top ten shareholders, descriptions of actual controllers, and related-party transaction disclosures;
• Judicial and administrative penalty information: useful for corroborating actual connections between related parties;
• Behavioral data such as intellectual property, bidding, and recruiting: used to uncover soft affiliation signals when equity relationships are unclear.

In multi-source data fusion, the biggest challenge is not data volume itself but entity normalization. The same company may appear under different names, abbreviations, former names, or with a missing unified social credit code across sources, preventing direct linkage. Natural persons and legal entities sharing the same name is also common; matching on name alone produces a large number of false associations.

In practice, engineering teams first build a master data model, using the unified social credit code, business registration number, or personal identity mapping information as the primary key, then apply cleaning, deduplication, and entity alignment to obtain a relatively clean entity repository. On top of that repository, an equity relationship edge table is built, recording the source, time, ratio, and confidence of each shareholding relationship.

Data timeliness is another key point. Business registration changes can occur at any time. If penetration calculations rely on a shareholder snapshot from months ago, the resulting actual-controller conclusion may already be invalid. For high-risk scenarios, it is advisable to attach a data update timestamp to penetration results and set shorter refresh cycles for entities that change frequently.

3. Engineering Implementation of Penetration Algorithms and Ratio Calculation

Once entity and equity relationship data are prepared, the core challenge of the penetration algorithm is path enumeration and ratio calculation.

The most direct approach is depth-first search or breadth-first search. Starting from the target company, read its direct shareholders, then recursively read those shareholders’ upstream shareholders until a termination condition is met. For most scenarios with shallow layers and no cycles, this traversal is sufficient.

However, when multiple paths lead to the same ultimate shareholder, ratios must be merged. For example, Company A holds 30% of Company C, Company B holds 40% of Company C, and an individual holds 50% of Company A and 50% of Company B. The individual’s combined shareholding ratio in Company C is:

30% × 50% + 40% × 50% = 15% + 20% = 35%

When paths are few, this calculation can be done by accumulating along each path, but in real scenarios the number of paths can grow exponentially. To control computation, common practices include:

• Setting a maximum penetration depth, for example 5 layers by default, beyond which paths are no longer expanded;
• Setting a minimum shareholding ratio threshold, for example paths below 0.1% are no longer traced upward;
• Caching already-visited nodes to avoid redundant computation;
• Using a traversal strategy with visit markers when cycles exist in the graph, to prevent infinite loops.

The pseudocode can be described as:

function calc_benefit(target, depth, visited):
    if depth > MAX_DEPTH:
        return {}
    if target in visited:
        return {}
    visited.add(target)
    result = {}
    for edge in get_upstream_edges(target):
        ratio = edge.ratio
        upper = edge.upper_node
        sub = calc_benefit(upper, depth + 1, visited)
        for person, sub_ratio in sub.items():
            result[person] += ratio * sub_ratio
        if is_natural_person(upper) or is_state_owned(upper):
            result[upper] += ratio
    visited.remove(target)
    return result

The logic above addresses simple shareholding chains, but in partnership and fund scenarios, the GP-LP relationship and profit distribution may not fully align with shareholding ratios. In such cases, penetrating solely by contribution ratio may fail to reflect the true control relationship. Engineering practice therefore typically calculates shareholding ratios and control relationships separately: the former focuses on economic benefit, the latter on voting rights and actual control.

4. Handling Circular Ownership and Complex Structures

Circular ownership is one of the more difficult patterns in equity penetration. For example, A holds 60% of B, B holds 50% of C, and C holds 20% of A. Direct recursive traversal would loop infinitely among the three nodes.

There are two main approaches to solving this. One is graph-theoretic cycle detection and decomposition: identify strongly connected components before traversal and handle nodes within cycles specially to avoid repeated traversal. The other is rule-based simplification: truncate when a cycle involves very low shareholding ratios, and when ratios within the cycle are high, flag it as a complex structure for human judgment.

From a business perspective, circular ownership often implies a deliberately designed ownership structure that may serve to evade regulation or facilitate internal benefit transfer. In an equity penetration report, circular structures should be listed separately and highlighted rather than silently absorbed by the algorithm.

Concert-party arrangements and voting rights delegation are another problem that cannot be resolved by shareholding ratios alone. Two shareholders with relatively low ratios may form a concert-party relationship through an agreement, jointly controlling the company. Such information usually comes from listed company disclosures, articles of association, or judicial judgments, and is poorly structured. In practice, a combination of rule matching and text mining can first automatically extract passages containing keywords such as “concert party,” “voting rights delegation,” or “actual control,” which then enter a human confirmation workflow.

5. Application of Penetration Results and Human Review

The final output of equity penetration generally includes:

• A list of actual controllers and ultimate beneficial paths;
• A complete ownership structure tree with shareholding ratios at each layer;
• Risk flags: circular ownership, frozen equity, frequent changes, multi-layer domestic and offshore nesting, etc.;
• A related-party list: used for subsequent related-party transaction checks, blacklist screening, or concentration analysis.

In supplier onboarding scenarios, equity penetration helps buyers detect whether different bidders are in fact controlled by the same person, thereby identifying bid-rigging and collusion risks. In credit and risk control scenarios, penetration results can be used for group customer identification, preventing a single actual controller from obtaining credit separately through multiple affiliated companies and causing uncontrolled exposure limits.

It must be emphasized, however, that automated penetration results cannot be used directly as the final basis for decisions. Missing data, low shareholding ratios, and contractual control can all cause automated results to deviate from the true control relationship. A complete equity penetration workflow should therefore include a human review step. The system automatically generates a draft report, flags all suspicious points and data gaps, and analysts confirm, supplement, or correct key paths in the interface before producing the formal version.

From an engineering implementation perspective, equity penetration is better suited as a review assistance tool than as a black-box system replacing human judgment. For development teams, the explainability of system output, path traceability, and display of data update times are often more important than merely pursuing deeper and more complete penetration layers. Only when users can clearly see the source and calculation process of every path does the penetration result carry real business value.