Hierarchical Model Combination for UK Inflation Forecasting: Integrating Econometric and Machine Learning Models

Hierarchical Model Combination for UK Inflation Forecasting: Integrating Econometric and Machine Learning Models

Economics

Machine Learning in Forecasting Inflation

Nitay Carmi

2025

Under the supervision of Professor Mark Steel.

This paper evaluates the forecasting performance of various econometric and machine learning approaches for predicting UK headline Consumer Price Index (CPI) inflation. Following the methodology of Stock and Watson (2007), I conduct pseudo-out-of-sample forecasts over multiple horizons (1, 2, and 4 quarters) using a comprehensive set of macroeconomic activity variables. My econometric models include Vector Autoregression (VAR), Vector Error Correction Models (VECM), and traditional benchmarks such as rolling IMA(1,1) and autoregressive models. Machine learning methods encompass Random Forest, XGBoost, Support Vector Regression, and Kernel Ridge Regression algorithms, implemented following the nonlinear framework of Coulombe et al. (2020). To address model uncertainty, I im- plement a hierarchical model combination framework using predictive performance- based weighting schemes that adapt through rolling re-estimation. Results indicate that while traditional econometric benchmarks remain competitive, particularly the rolling IMA(1,1) model, Random Forest demonstrates consistent and statistically significant improvements across all forecast horizons. The VECM approach shows horizon-dependent performance, improving from substantial underperformance at short horizons to near-parity at longer horizons. My findings contribute to under- standing inflation dynamics in the UK context and demonstrate the importance of algorithm selection in machine learning applications to macroeconomic forecasting.

Machine Learning, Data Science, Economics, Econometrics, Artificial Intelligence