Exploring Graph Neural Networks for Stock Market Predictions with\n Rolling Window Analysis

Abstract

Recently, there has been a surge of interest in the use of machine learning\nto help aid in the accurate predictions of financial markets. Despite the\nexciting advances in this cross-section of finance and AI, many of the current\napproaches are limited to using technical analysis to capture historical trends\nof each stock price and thus limited to certain experimental setups to obtain\ngood prediction results. On the other hand, professional investors additionally\nuse their rich knowledge of inter-market and inter-company relations to map the\nconnectivity of companies and events, and use this map to make better market\npredictions. For instance, they would predict the movement of a certain\ncompany's stock price based not only on its former stock price trends but also\non the performance of its suppliers or customers, the overall industry,\nmacroeconomic factors and trade policies. This paper investigates the\neffectiveness of work at the intersection of market predictions and graph\nneural networks, which hold the potential to mimic the ways in which investors\nmake decisions by incorporating company knowledge graphs directly into the\npredictive model. The main goal of this work is to test the validity of this\napproach across different markets and longer time horizons for backtesting\nusing rolling window analysis. In this work, we concentrate on the prediction\nof individual stock prices in the Japanese Nikkei 225 market over a period of\nroughly 20 years. For the knowledge graph, we use the Nikkei Value Search data,\nwhich is a rich dataset showing mainly supplier relations among Japanese and\nforeign companies. Our preliminary results show a 29.5% increase and a 2.2-fold\nincrease in the return ratio and Sharpe ratio, respectively, when compared to\nthe market benchmark, as well as a 6.32% increase and 1.3-fold increase,\nrespectively, compared to the baseline LSTM model.\n