Detection of structural breaks in electricity consumption data: a comparative analysis of MLE and bayesian approach in a state space model / Nurul Najwa Jatarona

Structural break is an important issue in electricity consumption data. Previous studies concluded that if structural breaks or changes are ignored, inference and forecasting would lead to unreasonable consequences. In performing structural breaks analysis, a few issues need to be catered carefully by the researchers and one of the most crucial issues is to identify whether the series contain a break. Many unit root tests have been introduced by researchers over the decade in determining the existence of break, such as Augmented Dickey Fuller test, Phillip-Perron test, KPSS test, and many more. Researches concerning structural break or structural change most likely need to deal with the problem of the identification of the numbers of structural breaks in the data series and mostly in identification of single, two breaks or more than two structural breaks. Therefore, the correct procedure used is very important in order to ensure the reliability of the results. The main objectives of this thesis are first to model Malaysia electricity consumption data using state space models based on the Maximum Likelihood Estimation (MLE) and Bayesian Approach and secondly, to detect the structural break based on the proposed models. Data series starting from 1999 to 2012 of electricity consumption from residential, commercial, transport and industrial sector in Malaysia were used. Unit root tests were performed as a confirmatory test using the augmented Dickey-Fuller (ADF), Phillips-Perron (PP), the Kwiatkowski-PhillipsSchmidt-Shin (KPSS) and Zivot and Andrews (ZA) test. Then, the state space models using both the MLE and Bayesian methods were proposed, followed by Kalman Filtering, smoothing and forecasting. The results show that in detecting structural break, auxiliary residuals were used as they are potential in detecting both outliers and structural breaks simultaneously and able to distinguish between them. The MLE approach produces very confusing results when it identifies structural breaks as outliers. While the Bayesian approach is capable to detect structural breaks and is able to distinguish between outliers and structural breaks. This research hopefully aid in providing a positive direction in identifying structural breaks in electricity data from a set of more efficient and effective results.