ETF Markets’ Prediction & Assets Management Platform Using Probabilistic Autoregressive Recurrent Networks

Abstract

The significance of macroeconomic policy changes on ETF markets and financial markets cannot be disre-garded. This study endeavors to predict the future trend of these markets by incorporating a group of selected economic indicators sourced from various ETF markets and utilizing probabilistic autoregressive recurrent net-works (DeepAR). The choice of economic indicators was made based on the advice of a domain expert and the results of correlation estimation. These indicators were then divided into two categories: "US" indicators, which depict the impact of US policies such as the federal reserve fund rate and quantitative easing on the global markets, and "region-specific" indicators. The findings of the study indicate that the inclusion of "US" indicators enhances the prediction accuracy and that the DeepAR model outperforms the LSTM and GRU models. Fur-thermore, a web platform has been developed to apply the DeepAR models, which enables the user to predict the trend of an ETF ticker for the next 15 time-steps using the most recent data. The platform also possesses the capability to automatically generate fresh datasets from corresponding RESTful API sources in case the current data becomes outdated.