AI-Based Forecasting in Sweden — the Riksbank's Machine-Learning Models for GDP and Inflation Prediction

Sveriges Riksbank, Sweden's central bank, has relied on dynamic factor models and other classical time-series econometrics within its Monetary Policy Department to forecast GDP growth and inflation ahead of interest-rate decisions.

An April 2025 Staff Memo set out to test whether machine-learning approaches could improve on these traditional forecasting benchmarks, as a step toward more automated forecasting processes that could free analyst time for judgement and communication.

Riksbank researchers (Ard den Reijer, Pär Stockhammar, David Vestin, Davide Bucci Vincenzo and Xin Zhang) systematically compared random-forest and neural-network models against the Bank's traditional time-series benchmarks in backtesting for GDP and inflation forecasting, publishing the full methodology and author list openly on riksbank.se.

The ML-based models better captured non-linear relationships and delivered measurably better forecast accuracy than the traditional benchmarks for both GDP and inflation in backtesting, though no operational adoption in interest-rate decisions has been claimed.

Because Riksbank Staff Memos are explicitly research publications 'free of policy conclusions,' this is evidence-generation ahead of possible institutional adoption rather than a claim that AI is already driving monetary-policy decisions; open publication of the methodology allows external replication and scrutiny.