About This Event
“Know where the economy is today, not where last quarter's data says it was.”
A practical workshop on machine-learning techniques for macroeconomic nowcasting. You will prepare mixed-frequency and alternative data, build and tune random forests, gradient boosting and neural networks, compare them with dynamic factor benchmarks, and produce GDP and inflation nowcasts that policy committees can use between official releases.
What You'll Explore
A reproducible nowcasting pipeline in Python for GDP or inflation
A tested comparison of ML methods against factor-model benchmarks
Worked code for random forests, gradient boosting and LSTMs
An interpretability toolkit using SHAP values for model outputs
A nowcast briefing template with uncertainty bands
Who Should Attend
Open to all qualifying staff, particularly: Forecasting Economists, Research & Data Science Staff, Statisticians, Monetary Policy Analysts, Heads of Modelling Units.
Why This Course Matters
Timelier Insight
Official GDP often arrives months after the quarter ends, too late for many decisions. A reliable nowcast gives the committee a current reading of activity when it has to decide on rates.
More Data, Used Well
Machine-learning methods handle many indicators, including mobile money and tax data, at once. Used carefully, they extract signal from data that traditional models would have to leave out.
Tested Accuracy
Out-of-sample testing against simple benchmarks shows which method earns its complexity. That testing means you present nowcasts with a known track record, not an untested black box.
Programme
Day 1
Nowcasting foundations & mixed-frequency data
You will set out the nowcasting problem, handle ragged-edge and mixed-frequency data, and build bridge-equation and dynamic factor benchmarks in Python to measure later models against.
Day 2
Regularised regression & tree-based methods
You will apply LASSO and elastic net for variable selection, then build random forests and gradient-boosted trees, tuning them with time-series cross-validation.
Day 3
Neural networks for macroeconomic time series
You will train feed-forward and LSTM neural networks on macroeconomic series, and learn how to avoid overfitting when the sample of quarterly observations is short.
Day 4
Alternative data, evaluation & interpretability
You will add alternative data such as mobile money, tax receipts and search trends, evaluate models out of sample, and use SHAP values to explain what drives each nowcast.
Day 5
Building a nowcasting pipeline & policy briefing
You will assemble a reproducible nowcasting pipeline for GDP or inflation, combine model outputs, and present a nowcast briefing with uncertainty bands to a mock committee.
Standards & Faculty Benchmark
BIS Irving Fisher Committee (IFC) work
Central bank statistics community reports on big data and machine learning in official use.
OECD AI Principles
International principles for trustworthy, transparent and accountable use of AI systems.
IMF Data Standards (e-GDDS, SDDS)
IMF standards on the timeliness and periodicity of official data that nowcasts complement.
Central banks are expected to use machine learning transparently and explain its outputs. Aligning your work to IFC practice and the OECD AI Principles means your nowcasts can be explained to a committee and reviewed by auditors, not simply trusted.
Is This Right for You?
- ☑You produce forecasts or short-term economic assessments
- ☑Your institution has data but slow official GDP releases
- ☑You want ML methods you can test, explain and defend
Good to Know
Participants should have basic econometrics and some Python or R; code templates are provided so time goes on judgement, not syntax. You leave with a reproducible nowcasting pipeline and a tested comparison of methods on a case dataset.
The Bottom Line
Bring back a working nowcast your committee can read each month, with methods you can explain and a record you can show.
Recommended For
Open to all qualifying staff, particularly: Forecasting Economists, Research & Data Science Staff, Statisticians, Monetary Policy Analysts, Heads of Modelling Units.