Journal article: This paper highlights two drawbacks with respect to the current use of rational expectations in new-Keynesian dynamic stochastic general equilibrium models. The first drawback is that a rational expectation model unobservable autoregressive shocks has the same predictions for given observations than an adaptive expectations model with white noise shocks. The second drawback is related to interest rate rules responding to inflation where both inflation and interest rate are non-predetermined variables. In this case, a unique rational expectations solution implies that the parameter in the interest rate rules destabilises inflation with a positive-feedback effects. However, these inflation or deflation spirals do not show off because the autocorrelation of inflation is the one of unobservable cost-push shocks. However this destabilizing stabilization paradox disappears if one assumes that the interest rate is a predetermined variable when inflation is a non- predetermined variable. Then, the rational expectations unique solution implies a policy rule parameter which stabilizes inflation.