The Research On Limit Order Book High-frequency Dynamic Modeling And Optimal Execution

Chinese stock market has undergone considerable development in recent years,with the scale of trading market value gradually expanding,and the listing system and trading system gradually improving.Compared with other financial markets,Chinese stock market has the characteristics of a large number of retail traders and low equity concentration,which provides opportunities and challenges for investors,market organizers and regulators.Based on high-quality limit order book data that details all order information within each trading day.We try to dig out effective information from the data with high-dimensional,multi-attribute,and non-equal time intervals characteristic.By clarifying the information transmission mechanism of the financial market,we attempt to explore the microstructure of the market and the characteristics of traders' behavior,which of importance for trader to optimize trading strategies and control risks,and for market organizers to strengthen supervision and improve transaction mechanism.We take the dynamic evolution of the limit order book as an entry point,reconstruct the dynamically evolving limit order book through high-quality order data,and construct models that combine the states of the limit order book and the real-time order flow.Based on the analysis of the interaction between the trader's order and the limit order book,we quantified the price impacts caused by the trading orders,discussed price formation process and market liquidity,hence provide a reference for traders to monitor market conditions,make short-term forecasts and optimize transaction execution.Each chapter discusses the interactive relationship between the limit order book order states and price formation process,and the information transmission mechanism from different time dimensions and perspectives.The research objectives are gradually focused and the sampling frequency is gradually increased.From the overall market state,to the limit order book network of stock portfolio,to the agglomeration and asymmetric characteristics of the order flow of sample stocks,we gradually refine the discussion on price impacts,price discovery process and information transmission mechanism.Chapter 3 discusses the regularity and relevance of the market conditions of Chinese stock market in various intraday periods based on market microstructure features,and applies the intraday status of the stock market to optimizing transaction execution strategies.First,we aggregate the stock sample data into a set reflecting the overall market status at different periods of the trading day;Second,Agglomerative Super-Paramagnetic Clustering is applied to analyze the inter structure of the dataset,and market state signature vectors are extracted and further used in online market state clustering.Third,based on the Almgren-Chriss optimal trade execution framework,we build an optimal transaction execution reinforcement learning model,in which market state signature vectors are included as input variables.This optimal trade execution problem is solved with the help of Deep Determinism Policy Gradient(DDPG).Empirical results show that the stock market exhibits obvious characteristics of intraday effects,and the characteristics of market conditions in the same or similar periods of each trading day have certain similarities.The market state signature vectors derived from clustering results can reduce the data dimension while retaining effective information,effectively improving the performance of the trading strategies.Chapter 4 uses the high-dimensional vector autoregressive model to model the high-dimensional,high-frequency limit order book sample data,and builds the limit order book network to quantify the self-impact and cross impact of stock prices and limit order book depth.First,we adopt the pre-average method and the capacity synchronization method to deal with the price-limit market microstructure problem and the non-equal interval time problem in high-frequency financial data,and obtain samples of the median price variable of the sample stocks and the best three-tier quotation volume variable data.Second,we use the post-double-select Lasso method and the bootstrap method to implement Granger causality test of high-dimensional time series and calculate sample generalized forecast error variance decomposition,respectively,and then build a limit order book network.Empirical results quantified the self-impact and cross-impact of prices and orders,verified the symmetry of information between buyers and sellers,and discovered the volatility factors that affect the performance of sample stocks in the limit order book network.Chapter 5 discusses the abnormal state changes of the market before and after the order burst from the vertical time dimension and the imbalance characteristics of the order order distribution from the horizontal price dimension,and then discusses the price shock and price discovery process.Based on the Hasbrouk price shock model,this chapter includes four different levels of measurement indicators: price changes,volume direction,multi-level order imbalance,and order burst,comprehensively discusses various influencing factors in the price formation process,and expands and analyzes the underlying factors.Mechanism and behavior characteristics of traders.Empirical results quantitatively analyze the effect of multi-level flow of information and order bursts on price formation,and explore the information breakdown of liquidity suppliers and liquidity demanders,and adverse selection caused by information asymmetry.Chapter 6 explores the incentive effect of limit order book events based on the state-dependent Hawkes process,focusing on the feedback between the state of the limit order book and the behavior of traders,and further analyzes the mechanism of price shocks.First,we constructed the limit order book state set based on the market microstructure characteristics,and constructed the limit order book event set based on the impact of the commission order on the limit order book.Secondly,we analyzed the self-incentive and mutual incentive effects of various limit order book events under different market conditions through the state-dependent Hawkes process set by different models.Empirical results show that traders in Chinese stock market have obvious agglomeration characteristics.The self-impacts of limit order book events are stronger than the cross-impacts,and there are significant differences under different market conditions.Dividing event types through aggressiveness can better reflect the impact of events and the information learning behavior of traders.Chapter 7 summarizes the research in the previous chapters from the perspective of application,based on the modeling of the evolution of the limit order book and the analysis of price shocks in the above chapters,and incorporates various market microstructure variables into reinforcement learning to optimize transaction execution from the perspective of model prediction The model is solved by the DDPG algorithm.This chapter constructs an effective and optimized transaction execution model and verifies the application value of the microstructure variables of the limit order book.In summary,we focuse on the dynamic evolution of the limit order book.First,we carried out dynamic evolution modeling from different angles and time dimensions,mined effective market microstructure information from data with complex data structure,and verified its interpretation and prediction capabilities.Secondly,based on the dynamic evolution modeling of the limit order book,we have identified and quantitatively analyzed the microstructure characteristics of Chinese stock market from the perspective of high-frequency financial measurement,such as the interactive impact relationship between stocks,the intraday effects of the market,traders herding behaviors and possible violations.Then,we analyze the limit order book from the perspective of traders' continuous bidding platform.Through the discussion of its dynamic evolution characteristics,the price discovery process and market information transmission mechanism have been more clarified,and we have enriched the research on the market microstructure under the framework of "observation-modelingapplication".The research has theoretical and practical value for traders to optimize trading strategies and for market organizers and regulators to improve trading mechanism.