Unsupervised Learning Of Markov-Switching Stochastic Volatility With Application To Market Data
Ivan Gorynin, SAMOVAR, Telecom SudParis, CNRS, Universite Paris-Saclay
Emmanuel Monfrini, SAMOVAR, Telecom SudParis, CNRS, Universite Paris-Saclay
Wojciech Pieczynski, SAMOVAR, Telecom SudParis, CNRS, Universite Paris-Saclay

Abstract:
We introduce a new method for estimating the regime-switching stochastic volatility models from the historical prices. It uses a specifically designed assumed density filter (ADF) to evaluate the quasi-likelihood function. Then an optimization algorithm maximizes the quasi-likelihood and provides the parameter estimates. The simulation experiments show the efficiency of our method. We then use it to analyze different market price histories for consistency with a regime-switching model.