The Czech National Bank has published a working paper on inflation forecasting in the Czech Republic that develops a quantile regression forest model and reports that a time-varying weighting scheme using information from the full predictive distribution produces better point forecasts than standard mean and median approaches and than conventional linear benchmark models. The research also models individual inflation subcomponents to help identify the drivers of future inflation and related risks. It adds a Shapley-value decomposition to improve interpretability and adapts the model’s predictors to the characteristics of a small open economy.
Czech National Bank 2026-04-01
Czech National Bank publishes research showing quantile regression forest improves Czech inflation forecasts
The Czech National Bank has published a working paper on inflation forecasting in the Czech Republic that develops a quantile regression forest model, finding that a time-varying weighting scheme using the full predictive distribution improves point forecasts relative to standard benchmarks. The research also models individual inflation subcomponents to identify drivers of future inflation and risks, incorporates a Shapley-value decomposition to enhance interpretability, and tailors predictors to the features of a small open economy.