Reverse mode differentiation for DSGE models

This paper provides a reverse mode derivative for DSGE models. Reverse mode differentiation enables the efficient computation of gradients from the model likelihood to the model parameters. These gradients can then be used by derivative based sampling algorithms including the No U-Turn Sampler. Benchmarks are provided using a small scale New Keynesian model. Our benchmarks demonstrate that MCMC chains generated using the No U-turn Sampler converge much more quickly than those generated using Metropolis Hastings.