Comparative quantitative research into the causes, responses to, and effects of banking crisis uses two series of crisis data: Reinhart and Rogoff (2009, 2010) and Laeven and Valencia (2013, and their predecessors). While these data sets provide broad coverage, the measures they code have several shortcomings. They are constructed post hoc and so tend to be biased towards severe crises with large government responses and away from circumstances where governments effectively calmed emerging trouble. They suffer from clear selection bias. Because they are simple dichotomous indicators of financial crisis, they do not indicate crisis severity. They use often ad hoc methods of determining when a crisis has ended. Our goal in this paper is to create a measure that is accurate, reliable, and comparable across countries, and that includes information about crisis severity. We use a kernel principal component analysis (PCA) of Economist Intelligence Unit (EIU) monthly country reports to develop a new real-time and continuous measure of perceived banking system stress. We refer to this measure as the EIU Perceptions of Financial Market Stress (FinStress) Index. We not only develop a novel indicator of financial market stress, but we also make a contribution to the wider political science and finance literatures on measurement by demonstrating how kernel PCA can be used to efficiently summarise vast quantities of qualitative texts into useful continuous cross-sectional time-series indicators. Finally, we provide an application of our measure to the political business cycle literature, with a focus on changes in debt. We demonstrate that governments reveal more of the debt created by responding to financial market stress when they are electorally safe.