Forecasting accuracy of Holt-Winters Exponential Smoothing: evidence from New Zealand.

Abstract

Financial time series is volatile, dynamic, nonlinear, nonparametric, and chaotic. Accurate forecasting of stock market prices and indices is always challenging and complex endeavour in time series analysis. Accurate predictions of stock market price movements could bring benefits to different types of investors and other stakeholders to make the right trading strategies.

Adopting a technical analysis perspective, this study examines the predictive power of Holt-Winters Exponential Smoothing (HWES) methodology by testing the models on the New Zealand stock market (S&P/NZX50) Index. Daily time-series data ranging from January 2009 to December 2017 are used in this study. The forecasting performance of the investigated models is evaluated using the root mean square error (RMSE], mean absolute error (MAE) and mean absolute percentage error (MAPE).

Employing HWES on the undifferenced S&P/NZX50 Index (model 1) and HWES on the differenced S&P/NZX50 Index (model 2) we find that model 1 is the superior predictive algorithm for the experimental dataset. When the tested models are evaluated overtime of the sample period we find the supportive evidence to our original findings. The evaluated HWES models could be employed effectively to predict the time series of other stock markets or the same index for diverse periods (windows) if substantiate algorithm training is carried out.