Financial Time Series Forecasting Using Prophet

Abstract

Forecasting the financial time series had been a difficult endeavor for both academia and businesses. Advances of the financial time series forecasting had moved from traditional techniques to automated and intelligent techniques that based on machine learning and deep learning. However, many methods of automatic forecasting have been tailored to the specific nature of the time series. As such, a recently introduced Prophet model, which is based on time series decomposition, is adopted with variants of its input parameters and applied to six different financial time series data sets obtained from Standard & Poor’s 500 index (SP500), Dow Jones Industrial Average index (DJIA), China Securities Index (CSI300), Malaysia’s stock market of Kuala Lumpur Composite Index (KLCI), Hong Kong Hang Seng 300 index (HS300) and Tokyo’s stock market of Nihon Keizai Shinbun index (Nikkei). The results of the time series forecasting show that the Prophet model is competitive in modeling the actual market movement by simply adopting appropriate parameters where the measure of Mean Absolute Percentage Errors (MAPE) was 6% at most. In addition, the errors of the forecasting result are also comparable to the results of much more complex forecasting models from the literature.