Time series with a single observation per period are commonly studied in the field of time series analysis. However, advancements in data recording technology have enabled the collection of time series with multiple observations per period. In the context of the insurance industry, for instance, we can consider claims within a period and compute loss ratios by dividing each claim by its corresponding premium. This approach results in a time series with multiple observations per period. In this talk, I will discuss methods for analyzing such datasets. Specifically, I will explore the application of both parametric and non-parametric approaches, combined with multivariate singular spectrum analysis, to produce forecasts for future observations.
