Testing Affine Term Structure Models in Case of Transaction Costs.

In this paper we empirically analyze the impact of transaction costs on the performance of affine interest rate models. We test the implied (no arbitrage) Euler restrictions, and we calculate the specification error bound of Hansen and Jagannathan to measure the extent to which a model is misspecified. Using data on T-bill and bond returns we find, under the assumption of frictionless markets, strong evidence of misspecification of one- and two-factor affine interest rate models. This is in line with earlier research. However, we show that the pricing errors of these models are reduced considerably, if relatively small transaction costs are taken into account. The average transaction costs for T-bills, due to the bid-ask spread, are around 1.5 basis points. At this size of transaction costs and for monthly holding periods, the misspecification of one- and two-factor affine interest rate models becomes statistically insignificant and economically very small. For quarterly holding periods, higher transaction costs of around 3 basis points are required to avoid misspecification.