Summary of the external review of the CNB’s modelling framework

Abstract:

In January 2024, the Governor and Board of the Czech National Bank requested that we should prepare an independent review of the analytical framework used for policy analysis and forecasting for monetary policy-making at the Czech National Bank, covering the set of core and satellite models as an integrated framework. They also encouraged observations on the way it is transformed to monetary policy recommendations and decisions. Our Review evaluated in detail the capabilities of the New Keynesian DSGE-type core model and satellite models, considered the current forecasting process and the underlying assumptions and expert judgments, and assessed how the models’ findings and forecasts are communicated to the Board for monetary policy decision-making. We spent three days at the CNB, for meetings with the monetary policy group (MPG) and the macroprudential policy group, and all Board members and their advisers. Further issues were discussed in online meetings and extensive email correspondence. We were provided with detailed documentation, not all in the public domain. The modelling staff are clearly sophisticated and technically competent, with access to a modern technical infrastructure of IT systems, data management and software at the CNB. The CNB also has an impressive record in the regular application of macroprudential tools to manage the credit cycle, as documented in their Financial Stability Reports. Our interviews suggested that four reasons had motivated the call for the external reviews. There had never previously been an external review, and it was timely. The inflation surge in 2022–23 and the 12-month ahead mis-forecast in October 2021, one of the largest in Europe, had signalled a need to re-examine the models generating the forecast. A third reason related to the model derived trajectory in the second half of 2022, which had implied raising the repo rate to 11%, which the Board wisely rejected. A fourth reason was unhappiness by most of the new Board after February 2023 with aspects of the decision-making process involving the model and model-generated forecasts. This concerned the usefulness of the models; “group think” and focusing on a narrow category of model; a lack of transparency about the models, assumptions and “expert judgments”; and limited flexibility for discussion and the simulation of a range of scenarios. Our recommendations are listed below and motivated in this article, with more detail in the Review itself.