A bagging algorithm for the imputation of missing values in time series

Classical time series analysis methods are not readily applicable to the series with missing observations. To deal with the missingness in time series, the common approach is to use imputation techniques to fill in the gaps and get a regularly spaced series. However, this approach has several drawbacks such as information and time bias, relationship causality, and not being suitable for the series with a high missingness rate. Instead of directly imputing the missing values, we propose a bagging algorithm to improve on the accuracy of imputation methods utilizing block bootstrap methods and marked point processes. We consider non-overlapping, moving, and circular block bootstrap methods along with amplitude modulated series and integer valued sequences. Imputation methods considered for bagging are Stineman and linear interpolations, Kalman filters, and weighted moving average. Imputation accuracy of the proposed algorithm is investigated by nearly 3000 yearly, quarterly, and monthly time series from different sectors under the "missing completely random" and "missing at random" missingness mechanisms. The results of the numerical study show that the proposed algorithm improved the accuracy of the considered imputation methods at most of the instances for different missingness rates and frequencies under both missingness mechanisms. (C) 2019 Elsevier Ltd. All rights reserved.