A modified cross-entropy method for detecting change-points in the Sri-Lankan stock market

Change-points or break-points in a sequence of observations occur because of sudden changes in the underlying process. These changes can be due to internal or external factors, or both. It is a well observed fact that economic time series heavily depend on external factors such as market information, apart from the inherited internal variations. Thus, they are open to have abrupt changes naturally in the process. We propose a procedure "mCE-ARMA", which utilizes Autoregressive-moving-average (ARMA) time series modelling and a modified version of the Cross-Entropy method to identify the change-points in time series observations. A multi-core architecture based on parallel implementation of the propose method is also considered. We observe that the parallel implementation has significantly reduced computation cost in which the processes are highly time consuming. The methodology is assessed in terms of segmentation performance through artificially generated data and a well known real data set in the literature. Finally, we use the propose procedure to identify potential change-point locations in the All Share Price Index (ASPI) of Sri Lanka from 2001 to 2012. Our results suggest that the proposed mCE-ARMA method is an effective way of segmenting time series data as compared to the competing methodologies.