Granger causality is a statistical hypothesis test for determining whether one time series is useful in forecasting another. While traditional Granger causality tests are based on linear regression models, neural models can capture more complex, non-linear relationships in the data.