Forecasting US bond yields at weekly frequency

Forecasting models for bond yields often use macro data to improve their properties. Unfortunately, macro data are not available at frequencies higher than monthly. In order to mitigate this problem, we propose a nonlinear VEC model with conditional heteroskedasticity (NECH) and find that such model has superior in-sample performance than models which fail to encompass nonlinearities and/or GARCH-type effects. Out-of-sample forecasts by our model are marginally superior to competing models; however, the data points we used for evaluating forecasts refer to a period of relative tranquillity on the financial markets, whereas we argue that our model should display superior performance under "unusual" circumstances.