Comparison of algorithmic trading using the homogeneous and non-homogeneous Markov chain analysis
Comparison of algorithmic trading using the homogeneous and non-homogeneous Markov chain analysis

Author(s): Pavla Říhová, Milan Svoboda
Subject(s): Business Economy / Management, Methodology and research technology, Financial Markets
Published by: Masarykova univerzita nakladatelství
Keywords: algorithmic trading; Markov chain analysis; share price prediction;
Summary/Abstract: This empirical study deals with stochastic modelling of a short-term share price development. We use Markov chain analysis (MCA) to predict the share price development. When defining a state space we assume that the share price moves in three types of trends: primary, secondary and minor. The subject of our interest is a minor trend, which usually lasts for several days. During this trend the share price accumulates a certain profit or loss in relation to the price at the beginning of the trend. The state space is defined by the amount of the accumulated profit or loss. The aim of this study is to compare two approaches to modelling the state space. In the first approach, we assume homogeneous Markov chains, i.e. approximately the same volatility, and MCA is performed with unvarying state space. In the second approach, we assume non-homogeneous Markov chains, i.e. a changing volatility, and MCA is performed with varying state space. We create trading strategies for automatic generation of buying and selling orders based on these models. Three business systems have been created for each approach. The profitability of each business system is calculated and compared. The study was performed using historical daily prices (opening and closing) of CEZ shares from the beginning of 2006 until the end of 2016. This study has proved that trading models with varying state space, on the average, outperform trading models with unvarying state space.

• Details
• Contents