Measure twice before you cut: differences in Furfine-type algorithm implementations

Abstract

Our study focuses on the practical implementation aspects of “Furfine-type” algorithms used to identify money market loans from payments data. We use two different versions of Furfine-type algorithms applied to data from TARGET2, the Eurosystem’s large-value payment system. We illustrate the magnitude and development over time of the differences between the implementations and study the explanations depending on the technical specifications of the algorithms. Paramount are the applied range of plausible rates and the identification of zero-rate loans. Consequently, the monetary policy environment and market conditions have an influence on the differences in the resulting data sets. These even affect aggregate indicators including interest rate dispersion, especially in times of stress, such as during the sovereign debt crisis and, to a lesser extent, the global pandemic. We find that an environment with methodological plurality can reduce overall uncertainty by using cross-checks between different data sources. The results strengthen the foundation for applying Furfine-type algorithms and are useful as guidance for their implementation and calibration in the euro area and beyond.