Roy Yang

Summary

Data scientist and senior manager with over 15 years in banking and insurance, including more than nine years applying analytics, AI, cloud platforms and executive reporting to acquisition, risk management, liquidity and profitability.

I work in Python, SQL, Azure, GCP and Power BI on ML forecasting and customer segmentation, with front-office, risk and treasury teams.

Experience

1. Data scientist, Group Treasury

   2022 – present

   AIA Company Limited (seconded by Protiviti), Hong Kong

   • Redesigned and automated the Liquidity Assessment (LQA) reports, cutting manual reporting effort by about 90% and moving the cadence from monthly to daily. Insights go straight to the C-suite through Azure-backed pipelines and an interactive dashboard that combines Bloomberg and internal systems.
   • Led a strategic AI initiative to build a unified cash-flow forecasting platform across 18 Asia-Pacific markets, giving senior management one dashboard for balances and inflow and outflow forecasts. Integrated ML forecasting models that cut error (sMAPE) from about 90% to 40%. Managed the rollout, set up monitoring controls and automated the reporting.
   • Increased Treasury profit with an optimal cash model on the liquidity analytics platform. It estimates required operating cash and minimum balances, so significant surplus cash could be redeployed into investments for measurable annualised investment income.
   • Project lead for data analysis and system integration, working with several business units to put analytical strategies in place and build treasury risk management tools.

2. Business strategy manager, Global Corporate Banking

   2017 – 2022

   Bank of China (Hong Kong) Limited, Hong Kong

   • Reduced risk-weighted assets (RWA) by more than 20% in targeted corporate banking portfolios while delivering double-digit profit growth, through large-scale deep-dive analyses that combined ML and big-data techniques into quantitative models and business strategies.
   • Influenced strategic entry into 3 new emerging industries with industry-level research, market-trend analyses and competitor benchmarks for senior leadership.
   • Built a Power BI executive dashboard tracking income structure, product volume and key financial ratios across business lines. More than 90% of senior managers adopted it.
   • Built ML models for customer segmentation, campaign optimisation, product recommendation and client profiling. Forecast deposit and loan volumes, and income, by product, department and division.

3. Senior officer, Middle Office Personal Banking

   2013 – 2015

   Bank of China (UK) Limited, London

   • Managed projects to integrate and implement the core banking system and the internet and telephone banking systems, coordinating IT teams in Europe and China.
   • Ran project management tasks including user acceptance testing (UAT), risk assessment and budget forecasting.
   • Built a cross-selling database and used quantitative data analysis to find business opportunities in personal and corporate banking.
   • Monitored and improved system capabilities and staff competencies. Wrote and revised policies, procedures and terms and conditions to control risk and meet compliance standards.

4. Associate relationship manager, Premier Banking

   2010 – 2012

   HSBC, Hong Kong

   • Achieved profitable growth through financial and investment product sales to high-net-worth (HNW) clients.
   • Provided wealth management solutions tailored to each client's financial needs, consistently above sales targets.

5. Personal financial manager, Premium Banking

   2009 – 2010

   Standard Chartered Bank (Hong Kong), Hong Kong

   • Managed corporate and personal banking portfolios, advised on investment strategies, acquired clients and increased account profitability through targeted financial solutions.

6. Personal banker, Personal Banking Department

   2007 – 2009

   Citibank (Hong Kong) Limited, Hong Kong

   • Built and kept client relationships through tailored financial planning, acquired new accounts and drove revenue growth through targeted product cross-selling.

Projects at work

• Machine learning forecasting. Models that predict deposit and loan volumes by product type, department and client profile, and income by type (interest and fee) across products, departments and divisions. Used for budget planning and profitability.
• Client data projects. ML for customer segmentation, behaviour analysis and feature analysis, and models for sales and marketing campaign optimisation, product recommendations and client profiling.
• Liquidity assessment report. Remodelled the liquidity calculations for more accurate measurement, connected Bloomberg data and internal systems through Azure cloud and databases, and designed an interactive Power BI dashboard for real-time portfolio liquidity.
• Business analytics dashboard. An interactive dashboard for senior executives tracking income structure, product volume and financial ratios across dimensions, with insight into departmental performance, customer segments and financial metrics.
• Premium persistence model. An ML model of premium persistence using payment method, policyholder demographics and product features, which improved policy retention and profitability.

Personal projects

• AI trading swarm. A committee of AI analysts that reads point-in-time market data, cites every claim and hands its ideas to Strategy Orchestra to test, with a human approving each idea and every number reproduced before a strategy counts.

  Python, PostgreSQL, TradingAgents, LangGraph, LM Studio, FastAPI, React

• TSBench: multi-series forecasting benchmark. A benchmark harness that runs forecasting models from several libraries over 16 years of 15-minute bars for 75 US tickers, scores them against finance-specific baselines, and picks winners with a Model Confidence Set.

  Python, pandas, statsforecast, LightGBM, arch, neuralforecast, Chronos-Bolt

• Reinforcement-learning trading strategy. Two agents, PPO and A2C, choose which of 10 low-correlation ETFs to trade every 15 minutes, and how much. The pipeline runs end to end. The evaluation doesn't hold up yet, and this page shows why.

  Python, stable-baselines3, Gymnasium, PyTorch, TA-Lib