Regime Detection using Machine learning

This paper focuses on Regime Detection in historical markets. It utilizes a Hidden Markov Model (hereinafter referred to as HMM) and Support Vector Machine (hereinafter referred to as SVM) to detect regimes in the iShares MSCI EAFE ETF adjusted close price time series from 2000 to today (chosen mainly due to its greater exposure to overseas mid- and large-cap companies), in order to enable the accurate prediction of the 24-hour trend of a market-on-open order in a domestic exchange. We observe that the HMM accurately predicts regimes in the historical time series very effectively and in relatively short timespans. It is also shown that the classified data after an unsupervised SVM clustering is further enhanced and the regimes further refined. The accuracy is the SVM clusters is heavily influenced by the parameter ν and γ of the Radial Basis kernel Function, as demonstrated is a series of experiments where each of the parameters is altered. The results of the classification can be used to develop a trading strategy for a market-on-open order that is informed by historical market fluctuations.