2020 Volume 50 Issue 1 Pages 191-204
Time series are sampled at various frequencies.In the classical literature of multivariate time series analysis, target variables are aggregated to the common lowest frequency. The temporal aggregation causes the loss of information and consequently lowers the accuracy of statistical inference. To address this issue, a new approach called Mixed Data Sampling (MIDAS) has been developed rapidly. MIDAS allows target variables to have different sampling frequencies. Recently, the MIDAS approach was adopted to vector autoregression and Granger causality tests. This article reviews these advances, and analyzes the dynamic interdependence between monthly inflation and quarterly economic growth in the United States. Significant Granger causality from economic growth to inflation is detected. The significant causality vanishes when inflation is aggregated to the quarterly level, suggesting spurious non-causality. These empirical results highlight the practical use of MIDAS.