Duration modelling of the after-hours electronic futures market

This thesis explores a class of models for modelling the time between trades, known as trade duration, in the after-hours electronic market for U.S. equity futures. These electronic markets have grown significantly over the last 10 years but little empirical work has been done to describe them. This is particularly so with duration modelling. High frequency trade duration data for the S&P 500 and NASDAQ-100 modelled in this thesis are collected from the GLOBEX electronic trading platform from the Chicago Mercantile Exchange for the period of 2004 to 2008. This thesis first fits standard linear Autoregressive Conditional Duration (ACD) models with Exponential, Weibull and Generalized Gamma error distributional assumptions to the period 2004 to 2006. The Generalized Gamma distribution outperforms the alternatives but still provides unsatisfactory results in the form of serially correlated residuals (volume is used as an additional mark in the model specifications). In order to improve the models, nonlinear forms of ACDmodel are estimated. In particular, the threshold and logarithmic forms are implemented. Although the results improve with these more flexible forms, there remains continued evidence of nonlinearity in the results. As a consequence, and taking into consideration the fact that the sample period of this thesis is much longer than the 3 month samples typically examined in the existing ACD literature, the thesis then examines the S&P 500 data for potential structural changes. Structural breaks are detected using a range of conditional Lagrange Multiplier tests associated with Andrews (1993) and Andrews and Ploberger (1994). FittingWeibull ACD models to the segmented sub-periods identified with the structural break tests significantly improves the model estimation results. Finally, this thesis examines the evidence for structural breaks in ACD models in the global financial crisis period using S&P 500 data from 2006 to 2008. The most significant structural change is found to occur in July, 2007, which is consistent with the onset of the crisis. Many of the structural change points detected in the data can be aligned with economic events during the crisis period, and sub-period estimations reveal the impact of the crisis on the electronic futures market.