Outliers detection in state-space model using indicator saturation approach

Farid Zamani Che Rose, Mohd Tahir Ismail, Mohd Hanafi Tumin


Structural changes that occur due to outliers may reduce the accuracy of an estimated time series model, shifting the mean distribution and causing forecast failure. This study used general-to-specific approach to detect outliers via indicator saturation approach in the local level model framework. Focusing on impulse indicator saturation, performance recorded by the suggested approach was evaluated using Monte Carlo simulations. To tackle the issue of higher number of regressors compared to the number of observations, this research utilized the split-half approach algorithm. We found that the impulse indicator saturation performance relies heavily on the size of outlier, location of outlier and number of splits in the series examined. Detection of outliers using sequential and non-sequential algorithms is the most crucial issue in this study. The sequential searching algorithm was able to outperform the non-sequential searching algorithm in eliminating the non-significant indicators based on potency and gauge. The outliers captured using impulse indicator saturation in financial times stock exchange (FTSE) United States of America (USA) shariah index correspond to the financial crisis in 2008-2009.


additive outlier; general-to-specific; indicator saturation; local level; Monte Carlo simulation; state space;

Full Text:


DOI: http://doi.org/10.11591/ijeecs.v22.i3.pp1688-1696


  • There are currently no refbacks.

Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

shopify stats IJEECS visitor statistics