82

We use an M-SETAR (Momentum–Self-Exciting Threshold Auto-Regressive) model to analyze U.S. real short-term interest rates over the last five decades. To separate non-linearity cases from non-stationarity cases, we use threshold integration tests against a stationary but non-linear alternative hypothesis. One innovation consists in introducing a structural break in the deterministic component of the process. This enables our model to take account of shifting regimes both in the deterministic part (mean shift) and in the stochastic part (threshold effects). The empirical application concerns the gap between the ex post real interest rate and its natural level, which changes after the break date. We find evidence that the real interest-rate gap follows a two-regime threshold process. Furthermore, the process seems to behave like a martingale in one of the regimes, highlighting the “reactive” characteristics of monetary policy in the corresponding periods.