Structural break detection in financial durations

High‐frequency financial data are more readily available and provide a deeper understanding of market infrastructure, market dynamics, and structural instability. The class of autoregressive conditional duration models is useful for statistical analysis of intra‐event durations in asset prices. Often, time series of durations exhibit structural breaks that may be due to change in level, or change in model order, or change in parameter values. This article studies the structural break detection problem in univariate time series of durations using penalized estimating functions. A retrospective algorithm based on the entire data allows simultaneous detection of the number of structural breaks, locations of the structural breaks, as well as the model order. This paper also provides a simulation study to compare the proposed algorithm with a Group LASSO method discussed in the literature for piecewise autoregressive models. Our method based on the penalized estimating function is attractive because it provides insights for understanding the stochastic dynamics of high‐frequency financial data.