This study aims at comparing several deep learning-based forecasters in the
task of volatility prediction using multivariate data, proceeding from simpler
or shallower to deeper and more complex models and compare them to the naive
prediction and variations of classical GARCH models. Specifically, the
volatility of five assets (i.e., S\&P500, NASDAQ100, gold, silver, and oil) was
predicted with the GARCH models, Multi-Layer Perceptrons, recurrent neural
networks, Temporal Convolutional Networks, and the Temporal Fusion Transformer.
In most cases the Temporal Fusion Transformer followed by variants of Temporal
Convolutional Network outperformed classical approaches and shallow networks.
These experiments were repeated, and the difference between competing models
was shown to be statistically significant, therefore encouraging their use in
practice.

该研究旨在比较基于深度学习的多变量数据预测模型在波动率预测方面的表现，结果表明 Temporal Fusion Transformer 优于传统方法和浅层网络，因此鼓励将其应用到实践中。