Decomposing the yield curve with linear regressions and survey information

The decomposition of bond yields into term premiums and average expected future short rates is impaired by the limited availability of information about the dynamics of the expectations component. Therefore, many studies require the model-implied average expected future short rates to be close to short rate expectations from surveys. In this paper, I restrict the variance of changes in model-implied average expected future short rates to match the variance of changes in short rate expectations from surveys. The variance of changes in survey expectations is relatively similar across markets and thus provides a reliable source of additional information about the expectation formation of investors. Technically, I impose a nonlinear restriction to the term structure model of Adrian, Crump, and Moench (2013). I show that typical small sample problems of term structure estimations can be mitigated if the restriction on the variance of changes is imposed. However, the analysis also makes a case for unrestricted estimations if they are based on a dataset with a typical sample length in macro finance.