Çizge sinir ağı tabanlı zamansal derin öğrenme modelleri ile finansal varlık fiyat tahmini

Multiple financial asset price changes over time can be characterized by their complex nature and the interdependence between them. More traditional forecasting approaches may overlook the interdependencies among these assets, since they may not fully consider the spatial-temporal dependencies between them. Graph Neural Networks (GNNs) have emerged as a powerful tool for modeling complex relational dependencies across various data types, recently taking more attention for their remarkable performance in areas such as social network analysis and traffic forecasting. The high capability of GNNs is mainly due to their permutation-invariance, and local connectivity. However, their application in asset price prediction remains relatively unexplored. Here, we investigate GNNs' effectiveness in forecasting multiple financial asset prices jointly, specifically focusing on Foreign Exchange (Forex) and cryptocurrency markets. We employ two spatio-temporal GNN frameworks: MTGNN and StemGNN where both are recognized for their state-of-the-art performance in forecasting multivariate time series. Both models are uniquely capable of transforming time series data into graphs, and capturing both spatial and temporal dependencies. Both methods significantly outperform LSTM in predicting financial asset prices, especially in highly volatile markets such as cryptocurrencies. However, the performance difference between these GNN-based multivariate approaches and LSTM is less obvious for the Forex market which is less volatile than cryptocurrencies. The ability to model interdependencies between multiple assets both temporally and spatially is the main reason StemGNN and MTGNN outperform LSTM. Through a series of experiments and backtesting strategies, we assess the predictive power and profitability of these models in portfolio construction. Our code can be found at https://github.com/seferlab/temporal_gnn.