09/05/2024 | Press release | Distributed by Public on 09/05/2024 08:23
Similar results have been obtained by broadening the Fed's research to include other large language models (), countries and time horizons. have generally been at least as good as professional forecasters at projecting inflation since 2018, with 2023 being the only year that US and Canadian forecasters outperformed AI platforms in this area.
But AI isn't without its limitations. In examining historical forecasting performance, it's very difficult to prevent from "cheating" by taking into account information that wasn't available when professional forecasters issued their projections. Further, some of the platforms' technical characteristics can cause them to just naively repeat prior forecasts.
All told, should be considered a useful tool in the economic forecaster's toolbox, especially since it incorporates different information than traditional tools. However, until an analysis is untaken which properly controls for all of 's current limitations, it isn't yet time to replace the human with the machine.