Abstract

We propose a block time series framework for count data observed over disjoint time windows. We represent irregular time series as a sequence of observed blocks, each with its own local temporal dynamics. At the same time, the blocks are not assumed to be independent. Instead, we introduce a structured dependence across blocks to investigate if nearby blocks are more correlated than those observed with large gaps. This yields a flexible compromise between complete independence of the blocks and the too restrictive global stationarity assumptions. The proposed framework is especially appealing for irregular count time series in which long periods without observations arise from external constraints rather than from the underlying data-generating mechanism. We propose an application to astronomical X-ray binary (XRB) time series.