Order Book Based Market Simulation Using a State Machine

Abstract

Abstract Stock market trading has evolved into an electronic market, where trading happens according to automated algorithms. Trading algorithm developers are developing more competitive algorithms for tomorrow which can beat yesterdays and traders are trying to survive in this competitive market. Both trading algorithm developers and traders may benefit from a model which predicts the market impact caused from a specific limit order. This research studies the market impact of limit orders using a state machine with stable, upward or downward states by observing order book and trade data. An order book representative factor named, State Factor has been introduced in order to represent an order book instance as a single floating point value. All information in a single order book view has been taken into consideration when introducing this State Factor. Market state identification is done using trade data. As the machine learning approach, a Hidden Markov Model has been used for prediction. Model training is done using observed order book data and derived states from trade data. The trained model predicts the posterior probabilities of the market being in stable, upward or downward states for a given limit order or a limit order sequence. Continuous learning approach has been used to train and test the model. Model evaluation is done by comparing the real probability values with the predicted posterior probabilities. The model gives promising posterior probability results with at most 4% deviation. Also, the model predicts the state sequence which is likely to occur in a particular trading day, with an accuracy between 53 – 60 %. The results of the modelling and analysis are presented such that they can be used as an aid to future work. Keywords: Market impact, Hidden Markov Model, State Machine, Order Book