Обучение по макроэкономическому прогнозированию и ИИ в Гонконге

Обзор программы

Ministries of finance, central banks, planning commissions and statistics offices all depend on forecasts. Budget revenue estimates, debt sustainability analysis, monetary policy decisions and investment plans start from a view of growth, inflation, the external position and public finances. When that view is wrong or late, the cost shows up as budget revisions, emergency measures and lost credibility.

Machine learning and large language models now give forecasting teams new options: high-frequency and text data, faster nowcasts and automated checks. The evidence, however, is more nuanced than the marketing. A December 2025 IMF working paper comparing nowcasting methods found that well-specified bridge and dynamic factor models generally beat machine-learning algorithms for GDP, that linear methods such as Lasso and Elastic Net were the strongest of the machine-learning group, and that complex non-linear models tend to overfit short GDP series.

This five-day programme in Hong Kong is built around that practical balance. Participants start from the consistent macroeconomic framework used in IMF Financial Programming and Policies training, which links the real, fiscal, external and monetary sectors, and then learn where machine learning, alternative data and generative AI add measurable value, and where they do not.

Hong Kong is a fitting venue. The Hong Kong Monetary Authority (HKMA) and Cyberport run a GenAI Sandbox for banks, the BIS Innovation Hub has operated a centre in the city since 2019, and the Census and Statistics Department publishes advance GDP estimates soon after each quarter ends. In 2026 the Government raised its full-year growth forecast from 2.5–3.5% to 3.5–4.5% after a strong first half, a live example of forecast revision that participants will analyse.

The programme is applied throughout. Participants work in Python and spreadsheets with real public data, build a small nowcasting model, test it honestly against simple benchmarks, and design the governance needed to use AI tools responsibly in an official forecasting unit.

Программа укрепляет институциональный потенциал в следующих областях

• Building consistent macroeconomic frameworks across the real, fiscal, external and monetary sectors
• Producing baseline projections and alternative scenarios
• Nowcasting GDP and inflation with mixed-frequency data
• Applying bridge equations and dynamic factor models
• Using regularised regression (Lasso, Ridge, Elastic Net) for data-rich forecasting
• Testing tree-based and ensemble methods such as random forests and gradient boosting
• Understanding where neural networks and foundation models help and where they overfit
• Turning text, news and survey responses into indicators with language models
• Evaluating forecasts out of sample against simple benchmarks
• Explaining model results with interpretability tools
• Documenting models, data lineage and assumptions
• Managing data governance, privacy and vendor dependence
• Using macro-structural models such as the World Bank MFMod framework
• Integrating forecasting into budget and policy calendars
• Communicating uncertainty, risks and revisions to decision makers
• Setting up responsible AI governance with human oversight
• Planning skills, tools and team structures for a modern forecasting unit

Throughout the week, Risalat facilitators work alongside participants as partners, adapting examples to the economies and institutions represented in the room. Institutions planning wider digital change can also use Risalat’s free Digital Readiness Assessment before or after the programme.

The Growing Importance of Macroeconomic Forecasting and AI

Forecasting institutions are under review worldwide. In April 2024 the Bank of England published Ben Bernanke’s review of its forecasting for monetary policy, which made 12 recommendations grouped around three themes: modernising data, software and model infrastructure; handling risk, uncertainty and structural change; and communicating the outlook more clearly. The same three themes apply to most finance ministries and central banks.

At the same time, the Bank for International Settlements devoted a chapter of its 2024 Annual Economic Report to artificial intelligence and central banks. It describes how language models can convert unstructured text into real-time data for nowcasting, how payment and other granular data can sharpen the picture of consumption and output, and why data governance is a precondition. It also lists the risks: hallucinations, weak reasoning, cyber threats, limited explainability and reliance on a few model providers.

For practitioners the message is clear. AI is now part of the forecasting toolkit, but it must sit inside a disciplined framework with sound accounting identities, honest evaluation and clear accountability.

Организациям все больше нужны специалисты, способные:

• Building macro-fiscal baselines that respect accounting and behavioural relationships
• Combining monthly, quarterly and daily indicators in one nowcast
• Choosing between econometric and machine-learning methods on evidence
• Preparing and cleaning large, messy datasets reproducibly
• Extracting sentiment and topic indicators from text
• Running out-of-sample tests and forecast-error decompositions
• Explaining model drivers to non-technical policy makers
• Designing scenarios and fan charts around a central forecast
• Documenting models so others can audit and maintain them
• Assessing data privacy, licensing and vendor risks
• Applying human oversight to AI-generated outputs
• Linking forecasts to budget, debt and monetary decisions

Strong macroeconomic forecasting and AI systems help organisations:

• Set more credible revenue and expenditure estimates
• Detect turning points earlier
• Reduce the gap between advance and final estimates
• Respond faster to shocks with scenario analysis
• Make forecast revisions transparent and explainable
• Retain institutional knowledge in documented code
• Use scarce analyst time on judgement, not data handling
• Manage model and AI risks in line with supervisory expectations
• Build trust with ministers, boards, markets and the public

The programme helps participants decide which of these improvements matter most for their own institution, and how to phase them realistically.

Who Should Attend This Macroeconomic Forecasting and AI Training?

Программа предназначена для:

• Macroeconomists in ministries of finance and economy
• Macro-fiscal and fiscal policy unit staff
• Revenue forecasting specialists
• Central bank economists and research staff
• Monetary policy and financial stability analysts
• National accounts and price statistics staff in statistics offices
• Planning commission and national development agency analysts
• Debt management office analysts
• Data scientists working on economic data in public institutions
• Economists in development banks and donor-funded programmes
• Chief economists and analysts in commercial banks
• Treasury and strategy teams in large companies and state-owned enterprises
• Risk managers using macroeconomic scenarios
• Heads of research and forecasting units
• Policy advisers who commission or use forecasts
• Academics and researchers supporting official forecasting

Participants should be comfortable with basic statistics and spreadsheets. Prior Python experience helps but is not required: guided notebooks are provided, and teams with mixed skills are welcome.

Macroeconomic Forecasting and AI Training Objectives

Upon successful completion of this Macroeconomic Forecasting and AI Training, participants will be able to:

• Build a consistent four-sector macroeconomic baseline
• Explain the strengths and limits of econometric, machine-learning and structural models
• Construct a GDP nowcast using bridge equations and a dynamic factor model
• Apply regularised regression to large indicator sets
• Test ensemble and neural network models without overfitting
• Create text-based indicators with language models and validate them
• Evaluate forecasts out of sample against naive benchmarks
• Interpret model drivers with interpretability tools
• Prepare scenarios and communicate uncertainty clearly
• Document models, data and assumptions for audit and handover
• Draft a responsible AI governance note for a forecasting unit
• Plan a realistic upgrade roadmap for their institution

Each objective is linked to a workshop output, so participants leave with working files and notes they can reuse.

Содержание и структура программы

Macroeconomic Forecasting and AI Training Methodology

The programme is practical and evidence-based. Short expert inputs are followed by guided exercises on real public data, so every concept is tested immediately.

Программа сочетает:

• Short technical presentations with worked examples
• Guided Python notebooks and spreadsheet templates
• Real data from statistics offices, central banks and international organisations
• Model comparison exercises with shared results tables
• Case studies on recent forecast revisions and reviews
• Group work in mixed teams of economists and data specialists
• Peer review of models and code
• Role play on presenting forecasts to senior decision makers
• Discussion of participants’ own institutional challenges
• Governance drafting exercises
• Daily reflection and action-planning sessions
• A capstone presentation with structured feedback

Facilitators adjust the technical depth to the group. Participants with less coding experience follow prepared notebooks, while more advanced participants extend the models. All materials remain available after the programme.

Ожидаемые результаты

Upon completion of this Macroeconomic Forecasting and AI Training, participants will be able to:

• Produce a consistent macroeconomic baseline and scenario
• Run and maintain a working GDP nowcast
• Select forecasting methods based on tested performance
• Use language models safely to build text-based indicators
• Explain model results to non-specialists
• Document models for audit, review and handover
• Apply a responsible AI policy within their unit
• Present forecasts and revisions with clear uncertainty ranges
• Lead a phased upgrade of forecasting tools and skills

These outcomes support better budget and policy decisions and stronger trust in published forecasts.

Institutional Application of Macroeconomic Forecasting and AI Training

Программа помогает организациям укрепить:

• Macro-fiscal frameworks underpinning annual budgets
• Revenue forecasting and tax policy analysis
• Monetary policy forecasting and communication
• Nowcasting between official data releases
• Data management and reproducible pipelines
• Model risk management and validation
• Responsible use of generative AI
• Collaboration between economists and data scientists
• Forecast evaluation and accountability
• Scenario planning for shocks
• Coordination between finance ministries, central banks and statistics offices
• Skills planning and succession in research units

Участники возвращаются с практическими инструментами для улучшения:

• A baseline consistency checklist
• A nowcasting notebook template
• A model comparison and forecast error tracker
• A text indicator validation guide
• A model documentation template
• An AI use policy template for forecasting units
• A forecast communication and revision note template
• A phased upgrade roadmap

Institutions wishing to map skills before nominating staff can use Risalat’s free Training Needs Assessment, and can browse related courses across Risalat’s programme categories.

Региональная актуальность и применение

Азия

Asian monetary authorities are active in this field. The HKMA’s second GenAI Sandbox cohort, announced in October 2025, brought together 27 use cases from 20 banks and 14 technology partners, with trials from early 2026. The BIS Innovation Hub Hong Kong Centre, set up in 2019, works on central bank digital currency, open finance and green finance. Participants from Asia can compare their own approaches directly.

Европа

The 2024 Bernanke review of Bank of England forecasting has influenced debate across European institutions on data infrastructure, scenario use and communication. Its 12 recommendations give participants a structured checklist for reviewing their own forecasting process.

Африка

Many African finance ministries work with macro-structural models from the World Bank’s MFMod family, which covers 184 country models and has been the Bank’s main forecasting tool since 2015. The open-source ModelFlow package in Python makes these models usable without proprietary software, which lowers cost for smaller units.

Ближний Восток

Economies exposed to commodity prices need fast, scenario-based forecasting and clear communication of revisions. The programme’s nowcasting and scenario sessions apply directly to revenue-dependent budgets.

Latin America and the Pacific

Small and commodity-dependent economies often have short GDP series, which is exactly where the IMF evidence warns against complex non-linear models. Participants learn to choose parsimonious methods suited to limited data.

Across all regions, the common thread is disciplined use of new tools inside a consistent framework, with clear human accountability.

Профиль принимающей страны – Гонконг

Hong Kong is an international financial centre with a dense community of central bankers, regulators, banks, technology firms and universities. Its compact size, efficient public transport and wide choice of hotels make it a practical venue for an intensive five-day programme.

For this topic the city offers live reference points. The Census and Statistics Department publishes advance quarterly GDP estimates, and the Government reviews its full-year forecast as data arrive: real GDP grew 3.5% in 2025, the February 2026 Budget forecast 2.5–3.5% growth for 2026, and in August 2026 the forecast was raised to 3.5–4.5% after first-half growth of 5.1%. In October 2024 the Government issued a policy statement on responsible AI in the financial market, taking a dual-track approach that promotes adoption while managing risks, and the HKMA and Cyberport run the GenAI Sandbox on Cyberport’s AI supercomputing platform.

Hong Kong International Airport has extensive direct connections to Asia, Europe, the Middle East and Africa. Many nationalities can visit Hong Kong without a visa for short stays, while others need a visa in advance; participants should check current requirements with the Hong Kong Immigration Department. Risalat provides complimentary visa support and invitation letters where needed.

Macroeconomic Forecasting and AI Training Deliverables

Участники получают полный пакет поддержки до, во время и после программы.

До программы

• Предварительная оценка участников
• Macroeconomic Forecasting and AI Readiness Assessment
• Macroeconomic Forecasting and AI Resource Pack
• Материалы для предварительного чтения
• Бесплатная визовая поддержка и консультации
• Встреча в аэропорту и приветственное сопровождение
• Проживание с завтраком
• Приветственный ужин для нетворкинга

Во время программы

• Комплект участника Risalat (сумка, бутылка, кружка, футболка, кепка, обложка для паспорта, ручки, блокнот, магнит и брелок)
• Comprehensive Macroeconomic Forecasting and AI Training Manual
• Baseline Consistency Checklist
• Nowcasting Notebook Template (Python)
• Model Comparison and Forecast Error Tracker
• Text Indicator Validation Guide
• Model Documentation Template
• AI Use Policy Template for Forecasting Units
• Практические семинары и симуляции
• Групповые дискуссии и обмен опытом
• Ежедневный обед и кофе-брейки
• Полный доступ к инфраструктуре места проведения
• Мини-планшет с цифровыми учебными материалами
• Выездной визит (при наличии)
• Развлекательная экскурсия по городу

После программы

• Сертификат о прохождении
• Награда Risalat за участие в программе
• Справочные материалы после обучения
• Code Repository with Workshop Notebooks
• Forecasting Unit Upgrade Roadmap Template
• Поддержка и консультации после обучения
• Отчет по программе

Central banks, ministries, statistics offices and financial institutions can also request a tailored proposal for an in-house version built around their own data, models and policy questions.

Common Challenges in Macroeconomic Forecasting and AI, And How This Training Helps

How can we tell whether machine learning actually improves our forecasts?

Day Three runs a structured horse race between econometric, regularised and tree-based models using rolling out-of-sample tests, so the decision rests on measured errors rather than claims.

How do we nowcast when official GDP data arrive late and are revised?

Day Two covers bridge equations, mixed-frequency methods and dynamic factor models, including how to handle ragged data edges and revisions, with a Hong Kong case study.

How can we avoid overfitting when our data series are short?

Day Three explains why complex models overfit short samples, shows time-series cross-validation, and helps participants choose parsimonious methods suited to their data.

How should staff use generative AI tools without creating risk?

Day Four reviews hallucination, privacy and reproducibility risks and ends with a practical AI use policy that each participant adapts for their unit.

How do we explain model results to ministers and boards?

Day Four introduces interpretability tools, and Day Five covers fan charts, scenarios and clear explanations of forecast revisions.

How can we keep the macroeconomic picture consistent while adding new tools?

Day One builds a four-sector baseline with accounting checks, and the capstone on Day Five requires every new model to feed back into that framework.

How do we plan the skills and systems a modern forecasting unit needs?

Day Five sets out roles, tools and phased investments, and the capstone produces an upgrade roadmap that participants take back to their institutions.

Часто задаваемые вопросы

Do I need to know Python to attend?

No. Guided notebooks and spreadsheet templates are provided, and facilitators adjust the technical depth. Basic statistics and spreadsheet skills are enough to take part fully.

Does the programme claim that AI beats traditional forecasting models?

No. It follows the evidence: recent IMF research found that bridge and dynamic factor models generally beat machine-learning methods for GDP nowcasting on short series, while linear methods such as Lasso and Elastic Net can help when data are rich. Participants test this themselves.

What data will we work with?

Public data from statistics offices, central banks and international organisations, including Hong Kong's published GDP series. In-house versions can use your institution's own data.

Is generative AI covered?

Yes. Day Four covers text-based indicators, analyst support uses, risks such as hallucination and data leakage, and a practical AI use policy for forecasting units.

Can the programme be delivered for one institution?

Yes. Risalat can deliver an in-house version focused on your own models, data and policy calendar. Use the request proposal form on the Risalat website.

Все программы в ГонконгеMore Technology programmes

Другие учебные курсы в Гонконге · Все программы в Азии · Полный календарь обучения