Autoregressive models for panel data causal inference with application to state-level opioid policies