The European Central Bank (ECB) has published a working paper presenting Deep-π*, a neural network model that links persistent inflation to observable economic conditions rather than unexplained statistical shocks. Using 122 macroeconomic indicators and US consumer price inflation from February 1970 to December 2025, it estimates both the persistent component of current inflation and a trend designed to predict inflation over the following year. The results identify labor market conditions as the largest contributor to the current trend’s rise after the pandemic and show forecasting gains against selected statistical benchmarks. Labor market indicators accounted for 41% to 52% of the current trend’s increase from January 2019 to its peak. Real activity, employment, and money and credit were the main contributors to the rise in the trend used to predict future inflation, although their relative importance differed between network designs, reflecting model uncertainty. Energy and commodity price movements contributed little to the estimated trends. These decompositions are statistical attributions, not identified causal effects. In estimates using the full sample, the trend used to predict future inflation peaked in March 2021, eight to 12 months before the current trend. Monthly forecasting tests from December 2020 to November 2023 used successive samples from a revised dataset, rather than historical data vintages. Forecasts were more accurate than those of a standard unobserved components model with stochastic volatility and a random walk, and broadly comparable with an autoregression. In early 2021, the model predicted inflation close to 4% over the following year, compared with professional forecasters’ 2%, against an eventual outcome of 7.3%. Professional forecasters anticipated the subsequent decline better. The model initially overestimated inflation persistence because the relationship between labor market tightness and future inflation weakened in 2022.
ECB publishes research on neural network model to estimate US trend inflation and explain its drivers
The European Central Bank has published research on a neural network model that estimates US trend inflation from 122 macroeconomic indicators and explains their statistical contributions. Labor market indicators accounted for 41% to 52% of the estimated rise in the persistent component of current inflation after the pandemic. Tests using revised data produced better forecasts than standard trend inflation and random walk models, but professional forecasters anticipated the subsequent decline in inflation earlier.