This paper discusses the econometric analysis of ultra-high frequency data, which are complete transaction data that arrive at random times. The author introduces the Autoregressive Conditional Duration (ACD) model developed by Engle and Russell (1995) to analyze IBM transaction data. The model is used to estimate semi-parametric hazard functions and measures of instantaneous conditional variances. The variances are negatively influenced by long durations, as suggested by some market microstructure literature. The paper formulates economic questions in a statistical framework, focusing on the analysis of transaction data. It discusses the use of point processes and the theoretical framework of marked point processes for analyzing such data. The paper also presents econometric models for analyzing the data, including the ACD model, and applies these models to IBM stock transactions. The results show that the ACD model can be used to estimate the hazard function and conditional variances. The paper also discusses the estimation of price volatility using transaction data. It introduces a model that incorporates the relationship between the timing of trades and the volatility of prices. The model uses a GARCH-type process for the prices and incorporates the current duration in the volatility specification. The results show that longer durations are associated with lower volatilities, as predicted by the Easley and O'Hara model. The paper also presents a semiparametric approach to estimating the hazard function and discusses the implications of the results for understanding market microstructure and volatility. The paper concludes that the ACD model is a useful tool for analyzing ultra-high frequency data and that the results support the idea that longer durations are associated with lower volatilities.This paper discusses the econometric analysis of ultra-high frequency data, which are complete transaction data that arrive at random times. The author introduces the Autoregressive Conditional Duration (ACD) model developed by Engle and Russell (1995) to analyze IBM transaction data. The model is used to estimate semi-parametric hazard functions and measures of instantaneous conditional variances. The variances are negatively influenced by long durations, as suggested by some market microstructure literature. The paper formulates economic questions in a statistical framework, focusing on the analysis of transaction data. It discusses the use of point processes and the theoretical framework of marked point processes for analyzing such data. The paper also presents econometric models for analyzing the data, including the ACD model, and applies these models to IBM stock transactions. The results show that the ACD model can be used to estimate the hazard function and conditional variances. The paper also discusses the estimation of price volatility using transaction data. It introduces a model that incorporates the relationship between the timing of trades and the volatility of prices. The model uses a GARCH-type process for the prices and incorporates the current duration in the volatility specification. The results show that longer durations are associated with lower volatilities, as predicted by the Easley and O'Hara model. The paper also presents a semiparametric approach to estimating the hazard function and discusses the implications of the results for understanding market microstructure and volatility. The paper concludes that the ACD model is a useful tool for analyzing ultra-high frequency data and that the results support the idea that longer durations are associated with lower volatilities.