Volatility forecasting with garch models and recurrent neural networks

Abstract

The three main ways to estimate future volatilities include the implied volatility of option prices, time-series volatility models, and neural network models. This project investigates whether there are economically meaningful differences between those approaches. Seminal time-series models like the GARCH, as well as recurrent neural network models like the LSTM are investigated to forecast volatilities. An eventual informational advantage over the market’s expectation of future volatility in the form of implied volatility is sought after. Through trading strategies involving options, as well as investment vehicles that emulate the VIX, it is attempted to trade volatility in a profitable way.