Comparing the performance of variable-breakdown and normal approaches in forecasting

Forecasting accuracy is a primary criterion in selecting appropriate methods of prediction. Even though there are various methods of forecasting, different data characteristics require different approaches to be able to predict with good accuracy. This paper introduces the variable-breakdown approach in forecasting and compares the approach with the normal approach. The variable-breakdown approach is a process where the population data set is divided into sub-populations before forecasting methods are applied to each group. The forecasting methods used for comparison were the method of average, exponential smoothing, and Box-Jenkins. For the purpose of this study, the Malaysia labour force data sets covering the period 1982 up to 2019 were obtained from the Department of Statistics Malaysia. This data set was divided into three age groups; 15-24 years old, 25-54 years old and 55-64 years old. The fitted values for each group with the lowest MSE were selected to generate the new population fitted value. The new fitted population value was then compared to the normal approach population fitted value. From the finding, it is found that the variable-breakdown approach gives a smaller MSE value of 14,268.9, compared to the normal approach, 30,171.1. Therefore, it can be concluded that the variable-breakdown approach can give better forecast accuracy for the Malaysia labour force data set.