Autoregressive conditional duration and liquidity

This thesis provides an in-depth Autoregressive Conditional Duration (ACD) application and examines the relationship between the consecutive transaction, price, and volume durations in Borsa Istanbul (BIST) stock exchange and investigates the explanatory power of models' coefficients on return and liquidity measures. ACD models enable the usage of intraday high-frequency order book data to model the duration between consecutive transactions and predict the next meaningful duration. Durations are modeled in two parts, past and conditional, and the power of dependencies between successive trades is investigated in this way. In this study, by utilizing a subset of the ten most traded stocks in BIST and applying a framework for investigating the most suitable error term specification, various ACD models and extensions are employed to understand the intraday duration behaviors of different stocks. First, a brief explanation is specified about why the widely used low-frequency liquidity measures are inadequate at capturing the appropriate intraday liquidity. Afterwards, the durations between the transactions, prices, and volumes are modeled by using various types of ACD models, and a framework is established. Finally, new coefficients taken from the ACD applications are comparatively studied using regressions across different trading windows. Building on the framework of in-depth applications and analysis of ACD, a panel regression is used to understand the explanatory power of the new model. The findings of this study demonstrate the effectiveness of using ACD applications for modeling intraday duration effects on equity returns and liquidity.