(Un)naturally Low? Likelihood inference for a DSGE model