If long maturity bonds are traded frequently and traders have non-nested information sets, speculative behavior in the sense of Harrison and Kreps (1978) arises. Using a term structure model displaying such speculative behavior, this paper proposes an empirically plausible re-interpretation of predictable excess returns that is not based on the value traders attach to a marginal increase of wealth in different states of the world. It is demonstrated that (i) dispersion of expectations about future short rates is sufficient for individual traders to systematically predict excess returns even in a model with constant risk premia and (ii) the new term structure dynamics driven by speculative trade is orthogonal to public information in real time. The model is estimated using monthly data on US short to medium term Treasuries from 1964 to 2007 and it provides a very good fit of the data. Speculative dynamics are found to be quantitatively important, accounting for a substantial fraction of the variation of bond yields and is more important at long maturities. We also show that a three factor no-arbitrage factor model would find overwhelming but misleading evidence in favor of a time varying risk premia if the world is characterized by the model presented here.