Authors: Dragana Radojicic, Friedrich Hubalek, Thorsten Rheinlände, Simeon Kredateus and Nina Radojicic
Abstract: This study aims to explore the dynamics of the Limit Order Book by introducing a stochastic model for the limit order book in discrete time and space, characterized by a symmetric random walk. Special emphasis is placed on order avalanches, defined as sequences of order executions with periods of no trade events not exceeding ε > 0. The primary focus is on analyzing the distribution of order avalanche lengths. Further, to assess the informativeness of order book data, a machine learning model based on the Gated Recurrent Unit (GRU) neural network is proposed. The customized for reconstructing order book data with aim to extract features of interest from the market data is introduced. Additionally, technical indicators are incorporated into the dataset, which was previously aggregated and formatted to meet the requirements of this research. Additionally, the quantified results demonstrate significant performance improvement achieved by selecting features based on the proposed feature selection methods.
Abstract: This study aims to explore the dynamics of the Limit Order Book by introducing a stochastic model for the limit order book in discrete time and space, characterized by a symmetric random walk. Special emphasis is placed on order avalanches, defined as sequences of order executions with periods of no trade events not exceeding ε > 0. The primary focus is on analyzing the distribution of order avalanche lengths. Further, to assess the informativeness of order book data, a machine learning model based on the Gated Recurrent Unit (GRU) neural network is proposed. The customized for reconstructing order book data with aim to extract features of interest from the market data is introduced. Additionally, technical indicators are incorporated into the dataset, which was previously aggregated and formatted to meet the requirements of this research. Additionally, the quantified results demonstrate significant performance improvement achieved by selecting features based on the proposed feature selection methods.
Link to work: