A financial stress index for the United States is introduced – an index that was used in real time by the staff of the Federal Reserve Board to monitor the financial crisis of 2008-9 – and the interaction with real activity, inflation and monetary policy is demonstrated using a richly parameterized Markov-switching VAR model, estimated using Bayesian methods. A "stress event" is defined as a period where the latent Markov states for both shock variances and model coefficients are adverse. Results show that allowing for time variation is economically and statistically important, with solid (quasi) real-time properties. Stress events line up well with financial events in history. A shift to a stress event is highly detrimental to the outlook for the real economy, and conventional monetary policy is relatively weak during such periods.