A1530
Title: Testing sparse cointegration using static conditional ecm and dynamic tvp-vecm-sv with postsparsification on stock data
Authors: Przemyslaw Jasko - Krakow University of Economics (Poland) [presenting]
Abstract: The dynamics of the cointegrated stochastic processes of prices in the stock market can serve as a basis for establishing quantitative trading strategies, exploring error correction properties such as pairs trading or statistical arbitrage. Taking into consideration the properties of the stochastic processes generating prices and returns in the stock markets, cointegration testing becomes quite challenging. These properties include the stochastic volatility of the innovation process and the possible time variation of the cointegration relationship (with potential subperiods during which it vanishes). To infer about cointegration, we first consider frequentist statistical model of sparse cointegration, the CECM (Conditional ECM). This model represents the dynamics of stock (log) prices, assuming possible time constant cointegration relationship. Noting that the parameters of the models could change over time, we also consider time-varying cointegration Bayesian TVP-VECM-SV model with shrinkage priors and SAVS (Signal Adaptive Variable Selector) postsparsification. The latter enables us to simultaneously determine which stock price processes are cointegrated and whether this cointegration is constant or time-varying. We also briefly present MCMC procedures for the Bayesian estimation of the considered model.The dataset for the empirical research encompasses 21 time series of closing prices of the Warsaw Stock Exchange's index WIG20, and its 20 constituent stocks.
