Vector autoregressions (VARs) are popular for forecasting, but ill-suited to handle occasionally binding constraints, like the effective lower bound on nominal interest rates. We examine reduced-form "shadow rate VARs" that model interest rates as censored observations of a latent shadow rate...