On the run-length of the structural change in time series data

Abstract

The movement of data changes in time series often cannot be seen as a single model throughout the time the serial data is recorded. The occurrence of model structure changes often must be accommodated in time series data modeling. This paper aims to study the Markov Switching models in capturing the structural changes the closing price stocks data of three companies (PT. Indofood Sukses Makmur Tbk. (INDF.JK), PT. Indofood CBP Sukses Makmur Tbk. (ICBP.JK), and PT. Mustika Ratu Tbk. (MRAT.JK)) from the member and not members of LQ45, estimating the run-length for each model structure, and forecasting the one-step-ahead of stocks. The parameters in the Markov Switching models are estimated using the Expectation Maximization (EM). The result shows that the three companies had more than one structural model. The best stock for investment is IDNF.JK share which have the longest Average Run Length (ARL) that provides the bigger probability of the forecasting regime and the stock values.