A1522
Title: Random dynamical systems in a statistical arbitrage strategy on the stock market
Authors: Przemyslaw Jasko - Krakow University of Economics (Poland) [presenting]
Abstract: A statistical arbitrage is a long-short, market-neutral trading strategy. The first aim of the work is to mathematically establish structures of random dynamical systems and their generators (stochastic difference equations) representing the movement of prices of assets, such that these structures enable the pursuit of a statistical arbitrage strategy based on the modeled dynamics of the prices of related stocks. The second aim is to empirically find multivariate stochastic processes of related stock prices, forming a random dynamical system whose properties allow us to pursue a statistical arbitrage strategy based on it. The first aim is realized based on the theory of random dynamical models. As tools to find related processes of asset prices, the following tests are used: Breitungs tests for linear cointegration, RCC tests for nonlinear cointegration, and rank test for monotonic cointegration. Then, Bayesian model is presented for time-varying cointegration TVP-VECM-SV with shrinkage priors, and SAVS post-sparsification. The dataset includes 21 log-prices time series of WIG20 stock index and its constituents. For selected subsets of assets selected by cointegration tests, TVP-VECM-SV models are built with shrinkage priors and post-sparsification of the cointegration matrix. Such model structure enables the simultaneous test if cointegration is present, and when true, it is time-varying (with possible time subperiods in which cointegration disappears) or time-constant.
