A0162
Title: Revealing cluster structures based on mixed sampling frequencies: Application to the state-level labor markets
Authors: Yeonwoo Rho - Michigan Technological University (United States) [presenting]
Yun Liu - Michigan Technological University (United States)
Hie Joo Ahn - Federal Reserve Board (United States)
Abstract: Mixed data sampling (MIDAS) models have drawn much attention among professional forecasters for their capability for a concise yet data-driven summary of information in frequently observed variables. While a parametric MIDAS model provides a parsimonious tool to summarize information in high-frequency data, one parametric form may not necessarily be appropriate for all cross-sectional subjects. A penalized regression approach is proposed that lets the data reveal their underlying cluster structure. To ease this clustering procedure, a simple yet flexible nonparametric MIDAS specification is proposed. The proposed clustering algorithm delivers reasonable clustering results, both in theory and in simulations, without requiring knowledge of the true group membership. An empirical application on the state-level labor markets in the United States is presented, clustering the states based on the response of the unemployment rate to the regional gross domestic product growth and weekly initial unemployment insurance claims. The mixed-frequency Okun's law relationship suggests that the state-level labor markets can be clustered into two groups composed of 11 states and the rest distinguished by the cyclical sensitivity of the unemployment rates and the changing predictability of weekly initial claims through the quarter.
