Skip to Main Content
Time series forecasting attempts to predict future values of time series. Its work is based on studying previously observed values. A heteroskedastic time series features variable and unpredictable measures of dispersion. This uncertainty in statistical distribution parameters imposes a serious challenge to the forecasting models. There have been many attempts to identify the heteroskedasticity in time series. However, all these attempts rely on hypothesis testing and do not quantify the amount of heteroskedasticity in the examined time series. On the other hand, quantifying heteroskedasticity does provide extra information about the behavior of the time series. Studying this behavior will improve forecasting of behavioral dependent time series such as stock market data. This paper introduces a novel heteroskedasticity index based on variance of localized variances.