International Journal of Energy and Statistics Vol. 4, No. 1 (2016) 1650005 (6 pages) c  Institute for International Energy Studies DOI: 10.1142/S2335680416500058 Missing value imputation in time series using Singular Spectrum Analysis Rahim Mahmoudvand ∗,§ and Paulo Canas Rodrigues †,‡,¶ ∗ Department of Statistics Bu-Ali Sina University, Hamedan, Iran † Department of Statistics, Federal University of Bahia Salvador, BA, Brazil ‡ CAST – University of Tampere, Tampere, Finland § R.mahmoudvand@basu.ac.ir ¶ paulocanas@gmail.com Received 12 February 2016 Revised 1 March 2016 Accepted 5 March 2016 Published 31 March 2016 This paper introduces a new algorithm for gap filling in univariate time series by using SSA. In this algorithm, the data before the missing values and the data after the missing values (in reverse order) are treated as two separate time series. Then using the recurrent SSA forecasting algorithm, two estimations of the missing values are obtained, one from the data before the missing values and one from the data after the missing values. Finally, using bootstrap resampling and a given weighting scheme, based on sample variances, these two estimates are combined to produce a unique estimation for missing values. Keywords : Bootstrap; missing values; Singular Spectrum Analysis; forecasting. Nomenclature SSA : Singular Spectrum Analysis. SSA - R : Recurrent SSA. SSA - V : Vector SSA. SVD : Singular Value Decomposition. RIM : Recurrent Imputation Method. NIORDC : The National Iranian Oil Refining & Distribution Company. 1. Introduction Singular Spectrum Analysis is a tool for time series analysis that can be effectively applied for different purposes, see for example [1–4]. As with other applications of SSA, several techniques were proposed to impute missing values in time series. 1650005-1