Financial Regulation / NL AnalyticsSecurities and Futures Commission2026-08-18

Hong Kong SFC: Natural-Language Market Analysis Across Five Systems

Supervisory questions cut across systems never designed to answer them together. A POC on MOI pre-computes derived indicators and routes a plain-language question — in Chinese or English — to the right SQL, pandas, news-join, MongoDB or CCASS workflow.

Securities and Futures Commission

The Securities and Futures Commission is the independent statutory body regulating Hong Kong's securities and futures markets, established in 1989, setting and enforcing market rules and safeguarding investors.

5 systems
Joined by one question
ZH + EN
Natural-language input
Pre-computed
Derived indicators
Text + charts
Answer format

The challenge

Supervisory questions cut across systems that were never designed to answer them together. Index and stock price history sits in SAS; market capitalization and industry classification in Oracle; company financials as JSON in MongoDB; broker shareholding movement only on the HKEX CCASS website. A question such as “which stocks traded at three times their 30-day average volume ahead of material news” spans two systems plus a news feed, and technical signals like ten consecutive closes above the 50-day moving average need rolling computation over very large price histories. Every question became a specification for a data team, so analysis ran at engineering speed and follow-ups were dropped.

The solution

The POC built a financial data analyst on OmniFabric. Derived indicators-50-day moving average, 30-day average volume, consecutive days above the average-are pre-computed T+1 in the source database, and OmniFabric connects to that result store alongside Oracle and MongoDB, with a webhook to external RPA for CCASS web data. A router understands the question, asks for a time range when one is missing, classifies intent and dispatches to the matching workflow: SQL for market and industry queries, a code interpreter running pandas rolling calculations for technical signals, an HTTP node joining news events to volume baselines, MongoDB aggregation for revenue growth. Answers return as text, chart and table, in Chinese or English.

The outcome

Index-drop volume, industry market-cap decline, moving-average streaks, pre-news volume spikes, CCASS shareholding movement and multi-year revenue growth are all asked in plain language-across SAS, Oracle, MongoDB and the HKEX website.

Every answer exposes the SQL it executed, and any pre-computed field shows its derivation, so a finding can be checked and reproduced rather than taken on trust.

“Change it to a bar chart”, “only the top 10”, “add data labels” re-render from the previous result without re-running the query-line, bar, pie, combination, heatmap and candlestick.

Solution Architecture

Data sources
  • Index and stock price history (SAS)
  • Market cap and industry classification
  • News and announcements
  • CCASS shareholding and trading data
MatrixOne Intelligence
  • Pre-computed derived indicators
  • Intent detection and clarification
  • Routing to SQL / pandas / MongoDB / CCASS
  • Cross-source joins and validation
  • Text and chart generation
Business applications
  • One question across five systems
  • Volume and market-cap movement
  • Moving-average streaks, pre-news spikes
  • Shareholding shifts and revenue growth

Related solutions and products