Parameter estimation and empirical analysis of fractional O-U process
based on three machine learning algorithms
- Yicun Li,
- Yuanyang Teng
Abstract
Abstract Fractional O-U process is a very classical stochastic process
which is used to describe the time series of financial volatility. Three
parameters need to be estimated in this process, and the estimation
method based on discrete observations can be realized by machine
learning optimization algorithm. In this study, the parameter estimation
method of fractional O-U process is briefly described, and three
optimization algorithms, Newton method, quasi Newton method and genetic
algorithm, are used to estimate the parameters. The comparison shows
that genetic algorithm is relatively accurate and efficient. Finally,
the minute data of stock index futures are estimated based on fractional
O-U process. The results show that the estimation of theta and Hurst
index is relatively accurate, and the estimation error of volatility is
large.