Order book dynamics: statistical analysis, modeling and forecasting.

Summary This thesis consists of two related parts, the first on the order book and the second on tick value effects. In the first part, we present our backlog modeling framework. The tail-reactive model is first introduced, in which we revise the traditional zero-intelligence approach by adding dependence on the order book state. An empirical study shows that this model is very realistic and reproduces many interesting microscopic features of the underlying asset such as the backlog distribution. We also show that it can be used as an efficient market simulator, allowing the evaluation of complex investment tactics. We then extend the tail-reactive model to a general Markovian framework. Ergodicity conditions are discussed in detail in this setting. In the second part of this thesis, we are interested in studying the role played by the tick value at two microscopic and macroscopic scales. First, an empirical study of the consequences of a change in tick value is performed using data from the Japanese 2014 tick size reduction pilot program. A prediction formula for the effects of a tick value change on transaction costs is derived. Then, a multi-agent model is introduced to explain the relationships between market volume, price dynamics, bid-ask spread, tick value and equilibrium order book state.