| Abstract [eng] |
Nowadays, many researches are made in ultra high frequency data series. Considering the data in time intervals as it arrives, helps to understand a variety of issues relating to trading process and market microstructure. The empirical analyses of such data present a number of new and unique statistical challenges. This work considers the dynamics of time intervals between transactions of one share in London Stock Exchange. The aim of this work is to make a statistically relevant model suitable for analysis of time intervals, and making the forecasts. First, some classical ACD-type models where tried to apply for a description of the data under consideration. After, a more sophisticated model using nonlinear functions was proposed in order to manage the problems arising in the classical models. More over, the model enables different dynamics of intervals between trades, depending on last price change. Comparing with the classical methods, the proposed model much better describes nonlinear dependence existing in the data. It enhances a goodness of fit, and is more applicable for practical use. Theoretical features such as an existence of equilibrium, stationarity and ergodicity where considered under specific functional form. But the idea can be extended for more general classes of models. |
