This program models multivariate time series (MTS) relationships at automatically adjusted temporal scales. We developed a segmented regression algorithm in the frequency domain as well as the lag optimization procedure. We refer to the proposed method as multivariate adaptive temporal scale regression (MATSR).
The MATSR is easy to interpret, and effective in 1) model generation at different temporal scales; 2) variable selection; and 3) lag optimization as illustrated in simulation tests. The usefulness of MATSR is also demonstrated by an empirical application to the daily exchange rate from eight foreign countries.