In a co-fractional vector autoregressive (VAR) model two more parameters are estimated, compared
to the traditional cointegrated VAR model. The increased number of parameters that needs to be
estimated leads to identification problems; there is no unique formulation of a co-fractional system,
though usually one formulation is preferred. This paper has the following contributions: (i) it
discusses different kinds of identification problems in co-fractional VAR models; (ii) it proposes a
specification test for higher order fractional processes; (iii) it presents an Ox program that can be
used for estimating and testing co-fractional systems; and (iv) it uses the above mentioned
contributions to analyse a system of Government Bonds in the US and Norway where the results
indicates that the level and trend in the yield curve have a longer memory than the curvature (i.e., a
linear combination of the yields of the Government Bonds that corresponds to representing the
curvature of the yield curve is a co-fractional relationship). |

working papers >