Embedding is a technique for converting text, images, or other forms of data into vector representations. OpenAI's Embedding API utilizes extensive text training to convert any text into multi-dimensional vectors, making similar text closer together in vector space. These vectors can be used for searching, clustering, recommendations, anomaly detection, diversity measurement, and classification.

MacroMicro utilizes OpenAI's Embedding API to calculate the consistency level of each Federal Reserve statement by converting them into vectors. The consistency level algorithm is 1 - vector distance, where distance is calculated using cosine similarity. The x-axis and y-axis represent the dates of the statements, and the color represents the level of consistency between two statements, with red (higher value) indicating higher consistency.

By analyzing the consistency heatmap, it is possible to determine whether the Federal Reserve's stance has changed. For example, during the period from March 2020 to the end of 2021, the consistency level between statements was high, coinciding with a period of loosened monetary policy in response to the COVID-19 pandemic. However, after the Federal Reserve began its interest rate hiking cycle in 2022, the consistency between statements during the pandemic period noticeably declined.