We study treatment-effect estimation using panel data. The treatment may be nonbinary, non-absorbing, and the outcome may be affected by the treatment lags. We make parallel-trends assumptions, but do ...