Developing algorithms for estimating high frequency principal components (PCs) for monthly macroeconomic data, the author finds that when suitably aggregated those PCs can closely approximate those calculated from monthly data even during a government shutdown: