Better forecasting with AI-based models
News, Staff memo AI-based models outperform traditional time series models in forecasting Swedish GDP and inflation. This is according to a survey conducted by economists at the Riksbank's Monetary Policy Department.
AI models offer significant potential to provide new insights and tools that are important for forecasting. The ability to capture non-linear relationships, adapt to new and large data sets, and automate forecasting processes presents a considerable advantage over traditional forecasting models.
In “AI-based forecasting in Sweden”, the authors examine how well AI models, especially random forests and neural networks, can forecast Swedish GDP and inflation. The forecasting performance of the AI models is compared with traditional benchmark models such as autoregressive models, as well as with commonly used forecasting models such as dynamic factor models.
The results show that the AI-based forecasting models improve the forecasts for Swedish GDP and inflation.
As AI models improve and become more integrated into the forecasting process, we can expect them to increasingly complement and, in many cases, continue to outperform traditional forecasting models.
Authors:, Ard Den Reijer, Pär Stockhammar and David Vestin, who work at the Monetary Policy Department, and Davide Bucci Vincenzo and Xin Zhang who previously worked at the Riksbank.
Staff Memo
A Staff Memo provides Riksbank staff members with the opportunity to publish advanced analyses of relevant issues. It is a staff publication, free of policy conclusions and individual standpoints on current policy issues. Publication is approved by the head of department concerned. The opinions expressed in Staff Memos are those of the authors and should not be regarded as the Riksbank’s standpoint.