Title: Reversed order monitoring CUSUM test with factor structure
Authors: Shou-Yung Yin - National Taipei University (Taiwan) [presenting]
Chang-Ching Lin - National Cheng Kung University (Taiwan)
Wen-Jen Tsay - Academia Sinica (Taiwan)
Abstract: The reversed order monitoring cusum (ROM) type test with the factor structure is considered. Despite detecting the latest break, we allow the unobserved factors as predictors. We then show that after an appropriate transformation of recursively estimated factors, these estimated factors would not change the asymptotic property of the Brownian motion under regular conditions with suitable ratio of $N$ to $T$. Monte Carlo simulations show that by using these transformed factors, there is almost no size distortion and the power is promising when we extend the evaluating time period with structure change. We also use the simulation to compare the proposed procedure with the fixed windows approach, and the results reveal that the ROM approach, in general, dominates the fixed window method. We then apply the ROM method to predict monthly growth rate of the U.S. industrial production (IP) and real personal income less transfers (RPI). The results support that the use of the ROM method can improve the out-of-sample forecast as compared to the usual fixed window approach.