Azme et al. / Malaysian Journal of Fundamental and Applied Sciences Vol. 14, No. 3 (2018) 382-385 382 A robust vector autoregressive model for forecasting economic growth in Malaysia Azme Khamis * , Nur Azreen Abdul Razak, Mohd Asrul Affendi Abdullah Faculty of Applied Science and Technology, University of Tun Hussein Onn Malaysia, 86400 Parit Raja, Batu Pahat, Johor, Malaysia. * Corresponding author: azme@uthm.edu.my Article history Submitted 20 February 2018 Revised 15 March 2018 Accepted 16 April 2018 Published Online 3 September 2018 Graphical abstract Abstract Economic indicator measures how solid or strong an economy of a country is. Basically, economic growth can be measured by using the economic indicators as they give an account of the quality or shortcoming of an economy. Vector Auto-regressive (VAR) method is commonly useful in forecasting the economic growth involving a bounteous of economic indicators. However, problems arise when its parameters are estimated using least square method which is very sensitive to the outliers existence. Thus, the aim of this study is to propose the best method in dealing with the outliers data so that the forecasting result is not biased. Data used in this study are the economic indicators monthly basis starting from January 1998 to January 2016. Two methods are considered, which are filtering technique via least median square (LMS), least trimmed square (LTS), least quartile difference (LQD) and imputation technique via mean and median. Using the mean absolute percentage error (MAPE) as the forecasting performance measure, this study concludes that Robust VAR with LQD filtering is a more appropriate model compare to others model. Keywords: Forecasting, economic growth, robust, imputation, filtering © 2018 Penerbit UTM Press. All rights reserved INTRODUCTION Vector autoregression (VAR) is a stochastic process model used to capture the linear interdependencies among multiple time series data. The VAR models generalize the univariate autoregressive model (AR model) by allowing for more than one evolving variable, where it described at least two factors where the dependent factors are found as a lag on the correct hand side of the equation. Also, VAR is a tools that take control of both the changing and interdependent relationships of the numerous variables (Onwukwe, 2014). VAR model is a very popular tool in multiple time series analysis. Its parameters are usually estimated by the least squares procedure which is very sensitive to the presence of errors in data known as outliers. If outliers were present, the estimation results would become unreliable (Jonas, 2009). Outliers in forecasting are data points that are not considered to be part of the overall pattern of demand and it can affects the forecasting accuracy. Since outliers can affect and skew forecast accuracy, it can be useful to exclude them from overall forecasting calculations to improve forecast accuracy (Anscombe (1960), Chen (1993) and Yu (2014)). In order to overcome this problem, an alternative method have been developed that are not so easily affected by outliers that is filtering and imputation technique. Filtering technique is done by employing robust regression methods (Shumacker et. al., 2002). Robust regression is an essential method for analysing data that are contaminated with outliers. It can be employed to detect outliers and to provide resistant results in the presence of outliers (Almonem, 2015). The robust methods are Least Median of Squares (LMS), Least Trimmed Squares (LTS) and Least Quartile Difference (LQD) (Berk (1990), Birkes & Dodge (1993). Meanwhile, imputation is a common approach for dealing with missing values in statistical databases. The imputer fills in missing values with draws from predictive models estimated from the observed data such as mean and median, resulting in completed versions of the database (Fan & Jerome, 2017). An economic indicator such as exchange rate and interest rate acts as the important indicator in the analysis of economic performance of a country and the predictions of future performance as stated by Agalega & Antwi (2013). Economic growth is defined as the increase in the capacity of the economy, where it can be measured by using an economic indicators as they move markets and measure how vigorous an economy of a country is. Besides that, they also can quantify particular divisions of an economy, for example, the lodging or retail division, or they provide quantification or estimations of an economy in general, such as unemployment or Gross Domestic Product (GDP) (Gaspar et. al., 2017). Thus, this study aims to propose and investigate the performance of the robust VAR models as an alternative tool in forecasting the economic growth in Malaysia in a specific divisions such as currency in circulation, exchange rate, external reserve and reserve money. EXPERIMENTAL Materials Data that were used in this study are the economic indicators consist of Currency in Circulation (CIC), Exchange Rate (EXC), External Reserve (EXT) and Reserve Money (RM) that were monthly basis starting from January 1998 to January 2016, which a sample of 217 observations is available. The first 171 values of actual data that is January 1998 to December 2011 is used in the estimation period to help RESEARCH ARTICLE