Financial institutions obtain enormous amounts of data about user\ntransactions and money transfers, which can be considered as a large graph\ndynamically changing in time. In this work, we focus on the task of predicting\nnew interactions in the network of bank clients and treat it as a link\nprediction problem. We propose a new graph neural network model, which uses not\nonly the topological structure of the network but rich time-series data\navailable for the graph nodes and edges. We evaluate the developed method using\nthe data provided by a large European bank for several years. The proposed\nmodel outperforms the existing approaches, including other neural network\nmodels, with a significant gap in ROC AUC score on link prediction problem and\nalso allows to improve the quality of credit scoring.\n