GNN 방식을 활용한 실현변동성 예측 연구

Abstract

Risk means uncertainty in finance, and is an important in Portfolio selection, Asset allocation, and Risk management. Therefore, studying risk forecast methodologies is very crucial for us to solve problems in financial markets. Realized volatility is a measure of the actual change in stock returns over a specific period of time. In the past, studies using a linear framework have been mainly conducted for forecasting realized volatility. Nonlinear frameworks such as LSTM have outperformed the realized volatility prediction performance compared to linear frameworks, but studies on nonlinear frameworks are still lacking. In this study, we create a graph based on company similarity to forecast realized volatility, and propose a GNN approach using the graph. This approach not only performed better at forecasting the realized volatility of individual stocks than the existing algorithm, but also performed better than the existing algorithm in the return of the risk parity portfolio based on the predicted realized volatility.