Trading using Hidden Markov Models during COVID-19 turbulences

Abstract Obtaining higher than market returns is a difficult goal to achieve, especially in times of turbulence such as the COVID-19 crisis, which tested the resilience of many models and algorithms. We used a Hidden Markov Models (HMM) methodology based on monthly data (DAX returns, VSTOXX index Germany’s industrial production and Germany’s annual inflation rate) to calibrate a trading strategy in order to obtain higher returns than a buy-and-hold strategy for the DAX index., following Talla (2013) and Nguyen and Nguyen (2015). The stock selection was based on 26 stocks from DAX’s composition, which had enough data for this study, aiming to select the 15 best performing. The training period was January 2000 - December 2015, and the out-of-sample January 2016 - August 2021, including the period of high turbulence generated by COVID-19. Fitting the best model revealed that the following regimes are the most suitable: two regimes for DAX returns, two regimes for VSTOXX and three regimes for the inflation rate and for the industrial production, while the posterior transition probabilities were event-depending on the training sample. Furthermore, portfolios built using HMM strategy outperformed the DAX index for the out-of-sample period, both in terms of annualized returns and risk-adjusted returns. The results were in line with expectations and what other researchers like Talla (2013), Nguyen and Nguyen (2015) and Varenius (2020) found out. We managed to highlight that a strategy calibrated based on HMM methodology works well even in periods of extreme volatility such as the one generated in 2020 by COVID-19 pandemic.