A0640
Title: Regular and reverse Midastar models: Threshold autoregression with mixed frequency data
Authors: Kaiji Motegi - Kobe University (Japan) [presenting]
John Dennis - Institute for Defense Analyses (United States)
Seok Young Hong - Nanyang Technological University (Singapore)
Abstract: The aim is to propose Midastar, a novel extension of the threshold autoregression (TAR) to the Mixed Data Sampling (MIDAS) framework. In the regular Midastar, the target variable is observed less frequently than the threshold variable. In the reverse Midastar, the target variable is observed more frequently. These models accurately capture threshold effects, whereas standard TAR with temporally aggregated data can point to spurious non-threshold effects. The parameters are estimated via profiling, and the no-threshold-effect hypothesis is tested via wild bootstrap. Uniform consistency and asymptotic normality are established under much weaker conditions than in the literature. In particular, the challenges overcame arising from the potential lack of stationarity. Monte Carlo simulations and empirical applications demonstrate that the Midastar models are useful for modeling and predicting macroeconomic and financial indicators.
