GNN Volatility Modeling
How volatility spreads between sectors
- 2.4%better than a GARCH(1,1) baseline (3.37 → 3.29 loss)
- 11market sectors modeled
I predicted long-horizon realized volatility for 11 market sectors, capturing how volatility spills over between them with a temporal graph neural network and a GRU encoder.
It beat a GARCH(1,1) baseline by 2.4% (loss 3.37 to 3.29) on a loss weighted by directional accuracy. Learned edge embeddings capture how the links between sectors change over time.