A1820
Title: Inference for sparse duration models
Authors: Jonas Striaukas - Copenhagen Business School (Denmark) [presenting]
Jan Roth - Copenhagen Business School (Denmark)
Abstract: Inference in sparse duration models based on a log-linearized autoregressive conditional duration (ACD) framework with high-dimensional covariates is studied through a three-step approach: first, a log-linear regression with serially correlated errors is estimated using l1-penalization; second, the ACD model is fitted using the first-stage estimates; and third, a decorrelated score combining both stages is constructed. The resulting statistic enables joint hypothesis testing for the sparse high-dimensional coefficient vector while accounting for serial dependence and model selection. Under suitable sparsity conditions, the test is asymptotically valid, and simulation results demonstrate good size control and power in finite samples.
