Modeling persistent interest rates with volatility-induced stationarity

It is well-known that interest rates are extremely persistent, yet they are best modeled and understood as stationary processes. These properties are contradictory in the workhorse Gaussian affine term structure model in which persistent data often result in unit roots that imply non-stationarity. We resolve this puzzle by proposing a term structure model with volatilityinduced stationarity. Our model employs a leveldependent conditional volatility that maintains stationarity despite the presence of unit roots in the characteristic polynomial corresponding to the conditional mean. The model is consistent with key characteristics of U.S. Treasury data and obtains term premia that are economically plausible and consistent with survey data. Compared to the Gaussian affine term structure model, we improve out-of-sample forecasting of the yield curve.