This paper addresses the issue of forecasting term structure. We provide a unified state-space modeling framework that encompasses different existing discrete-time yield curve models. Within such a framework we analyze the impact of two modeling choices, namely the imposition of no-arbitrage restrictions and the size of the information set used to extract factors, on forecasting performance. Using US yield curve data, we find that both no-arbitrage and large information sets help in forecasting but no model uniformly dominates the other. No-arbitrage models are more useful at shorter horizons for shorter maturities. Large information sets are more useful at longer horizons and longer maturities. We also find evidence for a significant feedback from yield curve models to macroeconomic variables that could be exploited for macroeconomic forecasting. Copyright © 2010 John Wiley & Sons, Ltd.