Moving Holidays’ Effects on the Malaysian Peak Daily Load Fadhilah Abd. Razak 1 , Amir Hisham Hashim 2 , Izham Zainal Abidin 3 and Mahendran Shitan 4 1 Department of Science & Mathematics, College of Engineering, Universiti Tenaga Nasional. Email: fadhilah@uniten.edu.my 2,3 Department of Electrical Engineering, College of Engineering, Universiti Tenaga Nasional. Email: amir@uniten.edu.my, izham@uniten.edu.my 4 Department of Mathematics, Faculty of Science, Universiti Putra Malaysia. Email: mahen@fsas.upm.edu.my Abstract—Malaysia’s yearly steady growth in electricity consumption as a result of fast development in various sectors of the Malaysian economy have increased the need to have a more robust, reliable and accurate load forecasting for short -, medium-, or long-term. A reliable method for short term load forecasting is crucial to any decision maker in a power utility company. Many studies have been made to improve the forecasting accuracy using various methods. The forecasting errors for the holiday seasons are known to be higher than those for weekends. This paper aims to determine which model would be a better model to estimate the holiday effects and therefore give a better forecasting accuracy for the peak daily load in Malaysia. Some of the holiday effects in Malaysia are from Eid ul-Fitr, Christmas, Independence Day and Chinese New Year. The seasonal ARIMA (SARIMA) and Dynamic Regression (DR) or Transfer function modelling are considered. Furthermore, the final selection of the models depends on the Mean Absolute Percentage Error (MAPE) and others such as the sample autocorrelation function (ACF), the sample partial autocorrelation function (PACF) and a bias-corrected version of the Akaike’s information criterion (AICC) statistic. The Dynamic Regression (DR) model recorded 2.22% as the lowest MAPE value for the 2004 New Year’s Eve and 2.39% for the seven days ahead forecasting. And therefore, DR model is the most appropriate model to be considered for forecasting any public holidays in Malaysia. Keywords—ARMA; SARIMA; Transfer Function; Dynamic Regression; MAPE I. INTRODUCTION Malaysia’s electricity consumption which has been increasing steadily in the past decade is expected to increase by 4% to 5% from an estimated 104 terawatt hours in 2009 to over 125 terawatt hours by 2014. The increase is mainly contributed by strong demands from the industrial and residential sectors. In addition, Malaysia’s population growth over the past decade, together with the improved and changing lifestyle has also contributed to the increase in electricity consumption. Since the availability of electricity plays a very significant role in driving the socio-economic growth of any country, many studies have been carried out over the past four decades on load forecasting using a wide variety of methods to improve - its accuracy. Load forecasting either for short, medium and long-term is important for planning and operational decision due to long and heavy investment required for building any power plant [1]. Short term load forecasting (STLF) which is usually from one hour to one week, is concerned with forecast of hourly and daily peak system load, and daily or weekly system energy. In addition, with privatization of the energy industry, a reliable and robust forecasting method is required by the utility company to maximize profit through efficient operations, and at the same time satisfying future customers’ needs with the most cost effective manner [2]. An under forecasted load demand will result in shortage of supply which will affect both economy and utility’s company reputation negatively. In contrast, over forecasted load demand will result in over supply and unnecessary extra cost that will reduce utility company profit. The aforementioned major inaccuracies in load forecasting not only are costly to any utilities companies, but in some instances will result in major blackouts. This incident had happened in Malaysia, whereby a few states had experienced major blackouts in January 2005. Forecasting the electricity load from one day to one week ahead for the Spanish system operator was discussed in [3]. The Spanish system’s daily model was computed for forecasting the daily load up to ten days ahead was derived by starting from an ARIMA (auto- regressive integrated moving average) model with dummy variables to capture the influence of special days which include holidays. The issue of identifying and eliminating deterministic seasonality due to holidays tied to the lunar calendar as well as analyzing the consequences of ignoring them by utilizing methods in the time and the frequency domains for Turkey was addressed in [4]. The result had shown that the variables become relatively more predictable and the estimation results based on these variables will be more reliable. It was found out that adding holiday regressors can effectively control the impact of moving holidays and improve the seasonal decomposition of the selected series in Taiwan [5]. This study is focused on Malaysia due the country’s uniqueness of having both fixed such as Christmas and