Three essays on high frequency financial data and their use for risk management

This dissertation contains three essays investigating the modeling and use of financial tick-by-tick data. High-frequency finance has become a very active field of research over the last two decades. Research making use of irregularly time-spaced transaction data has its roots in the seminal article of Engle and Russell (1998) that introduced the Autoregressive Conditional Duration (ACD) model for the analysis of arrival times between events based on all past information.;The first essay provides an up-to-date survey of the main theoretical developments in ACD modeling and empirical studies using financial data. First, we discuss the properties of the standard ACD specification and its extensions, existing diagnostic tests, and joint models for the arrival times of events and some market characteristics. Then, we present the empirical applications of ACD models to different types of events, and identify possible directions for future research.;The second essay proposes two classes of test statistics for duration clustering and one class of test statistics for the adequacy of ACD models, using a spectral approach. The tests for ACD effects of the first class are obtained by comparing a kernel-based normalized spectral density estimator and the normalized spectral density under the null hypothesis of no ACD effects, using a norm. The second class of test statistics for ACD effects exploits the one-sided nature of the alternative hypothesis. The class of tests for the adequacy of an ACD model is obtained by comparing a kernel-based spectral density estimator of the estimated standardized residuals and the null hypothesis of adequacy using a norm. With the L2 norm and the truncated uniform kernel, we retrieve generalized versions of the classical Box-Pierce/Ljung-Box test statistics. However, using non-uniform kernels, we obtain more powerful test procedures in many situations. The proposed test statistics possess a convenient asymptotic normal distribution under the null hypothesis. We present a simulation experiment and an application on IBM transaction data.;The third essay investigates the use of tick-by-tick data for market risk measurement. We propose an Intraday Value at Risk (IVaR) at different horizons based on irregularly time-spaced high-frequency data by using an intraday Monte Carlo simulation. An UHF-GARCH model extending the framework of Engle (2000) is used to specify the joint density of the marked point process of durations and high-frequency returns. We apply our methodology to transaction data for the Royal Bank and the Placer Dome stocks traded on the Toronto Stock Exchange. Results show that our approach constitutes reliable means of measuring intraday risk for traders who are very active on the market. The UHF-GARCH model performs well out-of-sample for almost all the time horizons and the confidence levels considered even when normality is assumed for the distribution of the error term, provided that, intraday seasonality has been accounted for prior to the estimation.;Keywords. tick-by-tick data, Autoregressive Conditional Duration model, duration clustering, model adequacy, spectral density, marked point process, Intraday Value at Risk, intraday market risk, UHF-GARCH models, intraday Monte Carlo simulation.