Centram Centre for Applied Artificial Intelligence

AI ANALYTICS AND AUTOMATION COURSE FOR MONETARY POLICY MANAGEMENT

Course Overview

This course is designed for designed for Economists in Central Bank, Ministry of Finance or Government Consultants, it balances macro/econ theory, machine learning for time series and text, automation/MLOps, model governance and policy communication, with hands‑on labs and a capstone.

Course title

 AI Analytics and Automation for Monetary Policy Management

Course description

The course introduces methods and tools that combine AI, econometrics and automation to support monetary policy design, implementation and communication. Covers real‑time data ingestion and vintage handling, now casting and forecasting of macro aggregates, policy rule optimization and scenario analysis, text analytics for communications and sentiment, systemic risk indicators, stress testing, and deployment/monitoring of operational analytics. Emphasis on robustness, interpretability, reproducibility and governance in a policy context.

Core learning objectives

By course end students will be able to:

  1. – Identify and access relevant macro‑financial, payments and high‑frequency datasets and handle data vintages and revisions.
  2. Build nowcasts and forecasts for GDP, inflation, unemployment and financial stress using econometric and ML approaches.
  3. Apply NLP to central bank communications, news and market commentary to extract sentiment and narrative indicators.
  4. Design automation pipelines to produce reproducible, real‑time policy indicators and dashboards (MLOps).
  5. Evaluate model uncertainty, robustness, model risk and interpretability for policy use.
  6. Conduct scenario analysis and automated stress tests; implement simple policy optimization (rules, reinforcement learning prototypes).
  7. Communicate analytics and uncertainty clearly to policymakers and stakeholders; apply governance and ethical best practices.

Course Duration

The course duration is 2 weeks’ intensive psychical boot camp training PLUS 2 weeks of online study OR 3 weeks of intensive physical boot camp training whichever is convenient to the participant

Course Content

Policy context, data types and infrastructure

  1. Monetary policy objectives, transmission mechanisms, indicators policymakers use.
  2. Data types: macro aggregates, financial series, payments, market prices, survey data, real‑time indicators (Google Trends, mobility), text (FOMC minutes, speeches).
  3. Data infrastructure & tooling: Python, R, Jupyter, Git, SQL, cloud basics.
  4. Lab: Set up environment; ingest sample FRED series and one payments/market series; explore data vintages.

Data engineering for macro analytics and vintage handling

  1. – Data cleaning, seasonal adjustment, calendar effects, handling revisions and real‑time vintages, nowcasting data flow.
  2.  APIs & sources: FRED, ECB SDW, BIS, IMF, national statistical offices, data licensing issues.
  3. Lab: Build reproducible ETL that preserves vintages and creates real‑time datasets for nowcasting.

Classical time series & macro econometrics refresh

  1.  ARIMA, state‑space models, Kalman filter, VAR/VECM, identification for policy analysis.
  2. Model evaluation for forecasts (MSFE, CRPS) and density forecasts.
  3. Lab: Build a small VAR for inflation/unemployment/interest rates; impulse response interpretation.

Nowcasting and forecasting with ML

  1. Bridge models, dynamic factor models, mixed-frequency models, machine learning (gradient boosting, LSTM, temporal convolution, transformers for time series).
  2. Combining forecasts and forecast reconciliation.
  3. Lab: Nowcast GDP and headline inflation using mixed-frequency and ML approaches; evaluate and compare.



Probabilistic forecasting, uncertainty and evaluation

  1. Density forecasting, quantile regression, bootstrapping, Bayesian approaches, scoring rules.
  2. Communicating forecast uncertainty to policymakers.
  3. Lab: Produce probabilistic inflation forecasts and calibration diagnostics.

Regime shifts, change detection and stress indicators

  1. Structural breaks, regime switching (Markov‑switching), early warning indicators, financial stress indices and systemic risk measures.
  2. Lab: Detect regime change around crisis periods and build a simple financial stress index from market indicators.

Text analytics for central bank communications and market narratives

  1. NLP pipelines: tokenization, embeddings, topic modeling, sentiment and narrative detection; event detection from news and social media.
  2. Use cases: extracting forward guidance, measuring policy stance and credibility.
  3. Lab: Analyze central bank minutes/speeches to construct a policy‑stance index and relate to market moves.

Scenario analysis, stress testing and macro-financial linkages

  1. Scenario design, reverse stress testing, counterfactuals, stress-test automation.
  2. Linking macro forecasts to balance sheet models and stress metrics.
  3. Lab: Construct automated scenario pipeline that propagates shocks to macro and banks’ indicators.

Policy optimization, rules and RL prototypes

  1. Policy rules (Taylor rule), optimal control basics, overview of reinforcement learning applied to policy (pros/cons), model calibration and constraints.
  2. Lab: Backtest simple policy rules; prototype RL agent in a simulated macro environment (toy model).

Automation, governance & MLOps for policy analytics

  1. Production pipelines for near‑real‑time indicators: scheduling, reproducibility, CI/CD, model registries, monitoring and retraining.
  2. Model risk management, documentation, versioning, data lineage and auditability.
  3. Tools: Airflow/Prefect, Docker, MLflow, Great Expectations, dbt.
  4. Lab: Implement an automated pipeline that updates a forecast and publishes to a dashboard.

Interpretability, model risk, ethics and communication

  1. Explainable AI tools (SHAP, partial dependence), robustness checks, back testing, avoiding overfitting.
  2. Ethical and governance issues: transparency, accountability, legal constraints, political economy of automated advice.
  3.  Lab: Build interpretability report and model risk checklist for a forecasting model intended for policy use.

Capstone presentations, policy exercises and course wrap-up

  1. Final project presentations: end‑to‑end solution (data → model(s) → automated pipeline → dashboard + governance documentation).
  2. Simulated policy meeting using analytics outputs; peer review and synthesis.

Practical labs & capstone project ideas

  1. Nowcasting dashboard for GDP and inflation with automated daily/weekly updates and uncertainty bands.
  2. NLP monitoring of central bank communications to detect shifts in forward guidance and produce an actionable policy‑stance index.
  3. Automated early‑warning system for banking sector stress using market and payments data.
  4. Backtest and evaluate alternative policy rules across historical vintages; build a reproducible comparison notebook.
  5. Scenario pipeline linking macro shocks to bank capital ratios and market risk exposure, with automated reporting.
  6. Prototype RL policy agent tested in a calibrated macro toy model and evaluated for robustness.

Datasets & data sources

  1. FRED, ECB SDW, BIS statistics, IMF IFS, national statistical offices (GDP, CPI, labor), central bank releases.
  2. Market data: yields, credit spreads, FX (note licensing for proprietary feeds).
  3. Real‑time/high‑frequency: payments data (where available), card transactions, money market rates, Google Trends, mobility data, news feeds (Refinitiv/Reuters if licensed).
  4. Text corpora: FOMC minutes, speeches, central bank press releases, ECB bulletin.
  5. Research datasets: Haver Analytics, Bank data repositories (subject to access).

Tools & libraries

  1. Python: pandas, statsmodels, pmdarima, arch, xgboost/lightGBM/CatBoost, scikit‑learn, PyTorch/TensorFlow/Keras.
  2. Time series & state‑space: statsmodels, pykalman, darts, Prophet, VARtools.
  3. Bayesian: PyMC, ArviZ, Stan (CmdStanPy).
  4. NLP: spaCy, Hugging Face Transformers, gensim, NLTK, BERTopic.
  5. Data engineering & MLOps: Airflow/Prefect, Docker, MLflow, dbt, Great Expectations, Kubernetes.
  6. Visualization/dashboards: Plotly/Dash, Streamlit, Power BI/Tableau.
  7. Reproducibility: Git, Docker, environment.yml, notebooks with tests.

Core readings & resources

  1. Giannone, Reichlin & Small (2008) — Nowcasting
  2. Stock & Watson — Nowcasting/forecasting papers
  3. Box & Jenkins — Time series foundations
  4. Guidotti et al. & Molnar (explainable AI) for interpretability
  5.  Central bank reports and working papers on data and nowcasting (ECB, Fed, Bank of England)

Papers on model risk, governance and AI ethics (BCBS, IMF, OECD guidance)