2 of 17

A Network Analysis of Licensed Professionals and Institutions

Why Hong Kong Financial Services Sector?

Hong Kong’s financial services sector is a vital pillar of its economy, contributing 23% to GDP and employing over 277,000 professionals. It plays a crucial role as a leading global financial hub
The Securities and Futures Commission (SFC) regulates this sector, maintaining a public register of licensed professionals and firms since 2003, providing a detailed and structured view of its evolution.

Why Study this Network?

Traditional economic indicators like GDP and employment lag behind real-time market developments. Network analysis provides a more granular understanding of key dynamics such as firm creation, longevity, and career trajectories.
Through complex network analysis, we can uncover the intricate relationships between professionals and firms, revealing the structure and health of Hong Kong's financial ecosystem.

What Makes This Study Unique?

21-year dataset enriched with demographic and organizational insights using large language models (LLMs), enabling a deeper understanding of nationality, gender, and firm classifications.
This analysis explores how individual behaviors and relationships shape broader financial patterns, providing fresh insights into market trends and workforce dynamics.

Implications:

Findings from this study can inform recruitment strategies, policy-making, and risk management, providing value for regulatory bodies and financial institutions.�

3 of 17

The study analyzes the public register of licensed professionals and institutions by the Securities and Futures Commission (SFC) of Hong Kong through complex network analysis
A dataset spanning 21 years offers insights into the evolving social network of licensed professionals and firms in Hong Kong's financial sector
Large language models have been leveraged to classify firms and infer the likely nationality and gender of employees based on names, enriching the dataset with demographic and organizational information.
Preliminary findings reveal important structural features of Hong Kong's financial landscape, providing new insights into its dynamics
The structured dataset will be released to support further research in network analysis, informing strategies in recruitment, risk management, and policymaking in the financial industry

General Overview of the Conference Paper

4 of 17

Why Focus on Networks and Create New Datasets?

Networks are generic mathematical framework
Model complex relationships effectively
Networks are simple at local scale (2 nodes connected by an edge) and complex at global scale.
Scalable and flexible modelling
Integrate diverse data types

Why Networks?

Ready datasets exist, but they represent different domains not directly related to our research.
This dataset has never been examined by academic researchers, offering new insights to the field.
Enhanced control allows us to tweak the dataset, such as enriching it with LLMs.
The dataset can assist Register Authorities in monitoring activities and specializations within the financial sector.
It can help predict future trends in professional movements and enable authorities to create customized incentive plans to attract specialized talent and activities.
Additionally, the dataset can assist HR departments in companies with recruitment efforts.

Rational and Potential benefits of this dataset

5 of 17

Dataset Description

Origin of the Dataset:

Derived from Hong Kong’s public register under the Securities and Futures Ordinance (SFO), covering licensed individuals, corporations, and financial institutions since April 2003.

Includes recent data on Virtual Asset Service Providers (VASPs) licensed under the Anti-Money Laundering and Counter-Terrorist Financing Ordinance (AMLO) as of June 2023. (Crypto firms are now regulated)

Dataset acquisition and overview:

Compiled via systematic web scraping from the Hong Kong SFC public register, which isn't directly downloadable

The final dataset includes 519,860 rows and 12 columns, covering all licensed persons and registered institutions.

License types: Dealing in Securities, Advising on Securities, Advising on Corporate Finance, Asset Management, Providing Automated Trading Services, Dealing in Futures Contract, Leveraged Foreign Exchange Trading, Advising on Futures Contracts, Securities Margin Financing, Providing Credit Rating Services,

6 of 17

The Dataset before LLMs enrichment

Describing the Dataset columns:

Column Name	Description
effectiveDate	Start date of the license or regulated activity
endDate	Termination or expiration date of the license or activity
fullname	Full legal name of the license holder (given and family names)
sfcid	Unique ID assigned by the SFC to identify each licensee
lcRole	RE: Representative authorized to carry out regulated activities under supervision. RO: Responsible Officer, authorized to supervise regulated activities
prinCeName	Official English name of the firm employing the licensee.

Column Name	Description
prinCeNameChin	Official Chinese name of the firm.�
prinCeRef	Unique ID assigned by the SFC to each licensed firm.
regulatedActivity.status	Current status of the regulated activity: R: Registered/Active. A: Archived/Inactive.
regulatedActivity.actType	Numerical code for the type of regulated activity (e.g., 1: Dealing in Securities; 2: Dealing in Futures Contracts; 3: Leveraged Foreign Exchange Trading; etc.).
regulatedActivity.actDesc	Description of the regulated activity in English.
regulatedActivity.cactDesc	Corresponding description in Chinese

7 of 17

Exploratory Data Analysis

License Types and Professional Specialization

Most professionals in the SFC register typically hold four or fewer license types (30% hold two and another 30% hold one type)

High specialization in trading (dealing with securities), often paired with asset management or corporate finance licenses. (encompassing mutual funds, hedge funds, and asset management divisions of banks)

Another pair of “Dealing with Securities” paired with “Advising on Corporate Finance” (covering investment banking activities like M&A and IPOs)

8 of 17

Exploratory Data Analysis

9 of 17

Exploratory Data Analysis

10 of 17

Exploratory Data Analysis

Job Market Dynamics: License Creations and Terminations

2009 and 2020 saw nearly equal license creations and terminations due to the Global Financial Crisis and COVID-19 pandemic.

Stagnation in 2012-2013 aligned with real estate cooling measures, limiting financial sector growth.

In 2023, terminations outpaced creations, possibly indicating economic uncertainty or migration to emerging global hubs like Singapore and the UAE.

11 of 17

Data Enrichment using LLMs

Data Enrichment:

We have leveraged LLMs to extract features like country of origin, gender, and business classification, enabling detailed demographic and organizational analysis in Hong Kong's financial sector

Initial Findings:

66% of the 121,833 individuals are likely from China, while Hong Kong & UK account for 7% and 3%, respectively

Western expatriates are departing Hong Kong, while the presence of individuals from China has been steadily increasing, especially since 2014.

12 of 17

Data Enrichment using LLMs

13 of 17

Data Enrichment using LLMs – Gender Analysis (1/2)

Initial Findings:

Gender representation varies widely, with lower female participation among Western expatriates (10%) and higher among Asian nationalities (20%-40%).

This is influenced by cultural and societal factors.

The model was able to predict gender based on names with reasonable accuracy but faced difficulties with unisex which are more frequent in China.

14 of 17

Data Enrichment using LLMs – Gender Analysis (2/2)

15 of 17

Network Constructions – Firm-Firm Based on Shared Employees

Firms are nodes, connected by edges if they share employees over time
Graph captures dynamic employee movement between firms

Construction of a Temporal Graph:

Vertices: Firms with at least one active employee
Edges: Exist between firms that have shared employees within a specific time frame

Vertices and Edges:

Weights: Based on the number of shared employees
Normalized Weight: Adjusted for firm size and employee turnover

Edge Weights:

3,116 nodes (firms) and 95,516 edges (shared employees) as of Feb 15, 2024
Network shows scale-free structure: few firms act as hubs
Average length path: 2.5 Vs 1.95 for the random network suggesting the presence of distinct clusters.
Clustering Coefficient: 0.42 Vs 0.02 for the random graph. Indicating the formation of tightly-Knit groups

Key Findings:

Firm 3

16 of 17

Network Constructions – Employee-Employee Firm Based on Shared Employers

Employees are nodes, connected by edges if employees have worked together at the same firm.
Graph captures dynamic employee movement between firms

Construction of a Temporal Graph:

Vertices: Employee active at time t
Edges: Exist between employees if they have worked together at the same firm

Vertices and Edges:

Weights: Based on the number of days that have overlapped in their employment at any firm
Normalized Weight: Adjusted for the total duration of each employee’s career

Edge Weights:

121,488 nodes (active employees) and 48,728,078 edges (shared employees) as of Feb 15, 2024
Network shows scale-free structure: a few firms act as hubs
Average length path: 2.8 Vs 1.95 for the random network suggesting the presence of distinct clusters.
Clustering Coefficient: 0.7 Vs 0.02 for the random graph. Indicating the formation of tightly-Knit groups

Key Findings:

17 of 17

Conclusion

Explored Hong Kong’s Financial Ecosystem:

Network analysis of the SFC Public Register enriched with LLM classifications.

Key Structural Insights:

Identified heavy-tailed degree distribution and high clustering in firm-firm and employee-employee networks.

Contribution:

Released a structured dataset, providing a valuable resource for future financial network research.

Impact:

Lays the foundation for deeper understanding of financial ecosystems, possibly aiding in policy-making by regulators and risk management by companies.

Opportunities for Future Research:

Potential for predictive models to improve forecasting (e.g, economic variables, employee turnover, firm ceasing activities).

Extend analysis to global firm activities and cross-regional employee movements.