The Bank for International Settlements (BIS) is collaborating with the European Central Bank (ECB) and Deutsche Bundesbank to investigate the potential of generative AI in enhancing inflation forecasting techniques.
Known as Project Spectrum, this initiative focuses on extracting actionable insights from consumer price data, which is often characterized by its raw, high-volume, and unstructured nature. Given the complexity of this data, extensive automation is necessary to effectively glean meaningful results.
The project aims to leverage generative AI to automatically categorize billions of product descriptions and price data points, thereby refining the accuracy of inflation nowcasting. Initial experiments will involve mapping data from the ECB’s Daily Price Dataset onto globally recognized household expenditure categories, such as those defined by the United Nations’ Classification of Individual Consumption According to Purpose (COICOP).
The BIS indicates that the algorithms and insights generated through this project could have applications that extend beyond Europe. The training dataset encompasses information from global retail chains and multiple European languages, enabling large language models (LLMs) and embedding technologies to yield findings that can be adapted even for languages not present in the original training set.