CMGM: A novel cross-market assets and multi-market modeling graph neural networks for financial market forecasting leveraging market states dependencies

The use of artificial intelligence (AI) in different financial services, such as financial technology (FinTech), is uprooting conventional ways and bringing novel alternatives. The latest trends in stock price forecasting are the use of Graph Neural Networks (GNN). However, these methods are still overlooked when modelling intricate dependencies of stock prices across multiple asset classes, including cryptocurrencies, commodities, bonds, and foreign exchange. Secondly, in graph learning, the correlations are overlooked to accumulate the impact of different financial conditions such as volatility trends, skewness/Kurtosis, and dynamic time-series correlations among different markets. To address such challenges, this research proposes a novel model called CMGM (Cross-Market Graph Modelling). It aims to model relationships between stocks within the same market and across different markets using specialized graph layers. The proposed CMGM designed the super and sub-graphs by leveraging the market state dependencies and highlights the benefits of bringing interconnected graphs with graph-based architectures for multi-market simulation. Such market-state dependencies are investigated around different factors using standard correlation, volatility-adjusted, skewness/kurtosis adjusted, as well as dynamic correlations that evolved over time. The proposed CMGM model is evaluated on U.S. stocks (S&P 500), commodities, forex, U.S. bonds, and cryptocurrencies. The findings of the research indicate that proposed CMGM models show good results over baseline methods, as well as showing improvements in multi-market simulation by achieving the lowest MAE and MSE errors of 0.01148 and 0.00026, respectively.