Skip to Main Content
Time series prediction or forecasting is an important area of research in various fields of science and engineering. Predictability refers to the degree that a correct forecasting of a time series can be made. Prediction can be erroneous. So it is important to know the predictability of the time series before going for prediction. In this paper six new predictability metrics have been presented. The proposed predictability metrics are evaluated by using single-step-ahead and multistep-ahead forecasting by nearest neighbor method. The variation of predictability metrics with signal to noise ratio is determined. The proposed metrics found stable. The predictability metrics for three benchmark time series is calculated and compared with already proposed metrics in literatures. It is found that proposed metrics are more useful for analyzing the time series.