Beyond the Smile: A Hybrid Convolutional VAE for Crypto Volatility Surfaces
Researchers have developed a hybrid convolutional variational autoencoder (VAE) designed to predict cryptocurrency volatility surfaces. This model, trained on extensive data from Binance Options for Bitcoin and Ethereum between May and October 2023, significantly reduces prediction errors compared to traditional methods. The VAE demonstrates superior performance, particularly in scenarios with missing data, and identifies specific market events like the August 17, 2023 flash crash as periods of elevated error. AI
IMPACT This research could lead to more accurate financial forecasting tools for cryptocurrencies.