We use a dynamic panel Tobit model with heteroskedasticity to generate forecasts for a large cross‐section of short time series of censored observations. Our fully Bayesian approach allows us to flexibly estimate the cross‐sectional distribution of heterogeneous coefficients and then...