Predictive Analysis for Financial Volatility

Abstract

In this thesis we take a closer look at the generalized autoregressive con- ditional heteroscedasticity (GARCH) model, which models the variance of the error term in particular stochastic time series. The standard procedure for estimating the model parameters is the maximum likelihood (ML) estima- tion method. We instead use machine learning to estimate parameters of a GARCH(1,1) model for a specific dataset and find that a machine learning model is able to find the ML estimator when it exists. We conclude that ma- chine learning is a good alternative to the ML estimation method and that it has the potential to better predict data.