Everything but the Kitchen Sink: Building a Metadata Repository for Time Series Data at the Federal Reserve Board

In support of the Board's duty to conduct monetary policy for the United States, the research divisions at the Federal Reserve Board use a variety of time series data for both research and forecasting. The collection, maintenance, and upkeep of more than 50,000 time series from more than sixty sources in a central location are daunting tasks; documenting the metadata for the compilation and use of these data is even more so. We are currently building a comprehensive metadata repository that links three kinds of metadata about our time series: structural metadata, reference metadata, and operational metadata. Many of the pieces of this puzzle currently exist in an array of disparate formats: attributes in a proprietary database, HTML pages on a website, Word documents buried on a file server, etc. We are bringing these pieces of information together in a relational database setting to allow users to search and display relevant metadata for a particular series or economic concept. In addition, we are working to make the metadata entries in the repository time sensitive to accommodate the database of contemporaneous " real time " or " snapshot " time series that we are building for future research purposes. This paper will examine the different types of metadata gathered for time series data, and how this information is collected, stored, and made available to staff at the Federal Reserve Board. Metadata Types The concept of metadata or " data about data " raises questions concerning what users need to know about different pieces of information. We have identified three types of metadata that cover the information we gather. The first and most fundamental type is structural metadata. Structural metadata are usually a small set of concrete details that identify and define a particular time series; they can often be expressed as short strings of text that convey the information essential to the meaning of a number. For example, users would have difficulty interpreting the statement that output was 12487.1. What kind of output? How is it measured? When was it measured? These questions may seem basic, but their answers are integral to the correct interpretation of this string of digits as economic information. If a few short strings such as description, country, time period, and unit of measure were defined, and you were told that the gross domestic product of the United States for 2005 was 12,487.1 billion U.S. dollars, the …