Neural networks for time series leverage deep learning architectures to model and predict sequential data by capturing temporal dependencies and patterns. These models, including recurrent and convolutional networks, excel at handling complex, non-linear relationships in time series data, often outperforming traditional statistical methods.