Inconsistencies in bond market quotes: Is it the wrong model or the wrong data?

We use the linear programming approach to quantify quote inconsistencies in risk-free bond markets. We present an algorithm to identify whether an inconsistency is probably due to the insufficient framework flexibility, the insufficient data quality, or the non-homogeneity of the dataset. In the latter case we study the problem of filtering out some instruments so that the remaining dataset be homogeneous. We show that the traditional filtering approach performs unacceptably poor and propose new algorithms. We find that the bonds, which get mispriced the most by a fitting algorithm, surprisingly are not the bonds, which cause the inconsistencies.