Abstract:
An analysis of time series and forecasting is usually inaccurate, invalid and unreliable when there are missing data. The objective of this research is to compare missing value estimation methods and to propose a new method by combining individual methods so called combined missing value estimation methods. The proposed method combines three individual missing value estimation methods by using weighted average. There are three methods for evaluating weighted averages which are Equally Weight (EW), Least Absolute Value (LAV) and Minimum Variance (MV). The Monte Carlo simulation technique is used in the R program with 50,000 trajectory samples, where sample sizes are 30, 50 and 100, standard deviations of error are 1, 2 , 5 and 10, missing percentages are 5%, 10%, 15% and 20% and correlation levels of independent
variables are 0, 0.2 and 0.5. The Mean Absolute Percentage Error is used as a criterion for comparison of missing value estimation methods. The proper method for cross sectional data is the Mean method. For time series data, the Mean method is suitable for three cases; 1) standard deviation of the error is 5 and the sample size is 30 2) standard deviation of the error is 10 and the sample size is 30 3) standard deviation of the error is 10 and the sample size is 50, otherwise, the Ordinary Least Square Method (OLS) is the best estimation method. The weighted of combined estimation method by least absolute value is superior to other methods for all cases studies.