École Nationale Supérieure de
Techniques Avancées

Abstract

The behaviours of financial market often change abruptly from steady low-volatility regimes to high volatility regimes driven by panic. The problem of market regimes detection has been of great interest and the most frequently used models have been hidden Markov models. Recently, machine learning and especially clustering algorithms appear to be effecient too. This project focuses on comparing hidden markov models and clustering algorithms specifically K-mean and Gaussian mixture models. A last approach based on correlation matrices will be examine. First, the models will be compared using real financial data (SP500 index). Then, several tests with synthetic data are conducted to find the best approach.