In an increasingly digital environment where data and advanced analytics challenge traditional economic modeling, bank of England To better understand complex phenomena such as inflation, we apply a fusion of machine learning (ML) and economic theory.
Two working papers shared by the Bank of England are bringing the spotlight to this development.
Written by experienced research and industry experts, these studies envisage either tackling or tackling the opacity of ML's “black box” head-on.
By embedding economic principles into algorithms, they aim to be a more keen policy tool for central banks navigating the volatile post-pandemic economy.
The first paper laid a conceptual foundation, arguing that Raw ML's predictive capabilities often sacrifice interpretability and are not suited to policy scrutiny.
Buckmann and Potjagailo introduce structured ML techniques, particularly block auxiliary models (BAMS) and monotonic constraints that fit the theory, injecting economic logic without diluting performance.
Enforce the entire additive while allowing nonlinear interactions within the block based on predictors of BAMS partitions (e.g., grouping supply chain variables or labor market indicators) based on economic evidence.
This gives you a clear and attributable contribution. One block could reveal how oil prices spike nonlinear inflation, while another block separates wage pressure.
The monotonic constraints further align the model with theory, ensuring output to respect established relationships, such as those that increase monotonically with demand but drop with oversupply.
These methods applied to simulated actual datasets preserve the predictive edge of ML while providing a “economically meaningful story.”
Central banking operations have an impact. Policymakers can investigate “what-if” scenarios, such as inflation fallout from a trade war.
A second paper built on this foundation will work with a customized inflation model, the Blockwise Boost Inflation Model (BBIM).
Here, the trio unfolds a boosted tree (gradient boost ML variant) with a block-structured framework inspired by the hybrid Philips curve of the open economy.
Inflation is analyzed into supply and demand blocks, while monotony guarantees an intuitive link. The harsher labor market clearly highlights price pressures, but excess capacity cools them.
Adjusted with UK Consumer Price Index (CPI) data from 1997 to 2024, BBIM discovered nonlinear drivers that linear models missed.
The recent surge in inflation, which peaked at 11.1% in October 2022, has emerged as a supply shock cascade rather than a spiral of uniform wage prices.
“This model shows that the recent surge is driven primarily by global supply shocks transmitted through supply chains,” the author points out, identifying disruptions like the energy ripple effects of the Ukrainian war.
The labor market reveals the “L-shaped” Phillips curve. Slack broadly suppresses inflation during recession, but in the boom, tension tightens an asymmetric upward burst that is amplified by adopting frenzy.
The expectations of household inflation add another layer of nuance.
The short-term view shaking on headline spikes showed a sustained nonlinear effect, increasing trend inflation by 0.5-1 percentage points.
However, the long-term anchor has secured a 2% target for the bank, highlighting the resilience of reliability.
Out-of-sample forecasts from 2022 were up to 20% better than self-network benchmarks of unstructured ML and vectors with average absolute error.
These papers appear to show collectively a paradigm shift.
Traditional economy is highly transparent, but flap the nonlinearity of big data. Pure ML thrives in patterns, but it confuses decision makers.
The Bank of England hybrid is structured, boosted and economically connected, and praises them for making the most of both.
For inflation forecasts, the BBIM equips the Monetary Policy Committee with clever insights.
The broader impact extends to global finance.
As climate shocks and geopolitical fractures intensify, interpretable ML can even attempt to predict ripple effects in emerging markets or analyze the proposition of cryptographic deflation.
However, the challenges last long. Scaling these models to real-time data requires computational weight, and potentially bias is at risk due to excessive reliance on UK-specific blocks.
Ultimately, these works confirm the maturity of ML's economics.
