IMPLEMENTATION OF CROSS-VALIDATION ON HANG SENG INDEX FORECASTING USING HOLT’S EXPONENTIAL SMOOTHING AND AUTO-ARIMA METHOD

Abstract

This study applies a rolling window cross-validation to evaluate the multi-step forecasts instead of using the traditional single split for Hang Sheng Index (HSI) forecasting. The forecasting methods discussed in this study are Holt's Exponential Smoothing and auto ARIMA, chosen because of their ability to model trend data as in the daily HSI. This research aims to evaluate up to five step forecast values obtained by the two forecasting methods built in the training data with rolling window cross-validation. In the experiment, each of the 21 auto ARIMA and Holt's models was constructed from 84 observations (as in-sample data) obtained from the rolling window cross-validation. The one to five step forecast values of daily HSI are then calculated using those models, and the accuracy of each forecast value is evaluated based on Mean Absolute Percentage Error (MAPE). The results show that the Auto ARIMA model produces a lower MAPE value than Holt's model, namely 2.9196%, 4.6553%, 6.4012%, 8.3083%, and 10.3781%, respectively, for one to five steps ahead. Therefore, auto ARIMA is more recommended for forecasting HSI values up to five steps ahead than Holt's method.