Hello everybody,I'm trying to calibrate AFNS model (christensen, Diebold 2011)http://www.ssc.upenn.edu/~fdiebold/pape ... easurement equation:Rt = A + B.Xt + Et Et~ N(0, H) (Xt vector: Level, Slope, Curvature)such as : B matrixe of loadings (constant in time), A matrix of the adjustement term (calculated in Christensen 2011)Transition Equation: Xt = (I-exp(-K*dt)).Theta + exp(-K*dt).Xt-1 +Nt Nt~ N(0, Q) My error covariance matrix F seems to be "quasi-" singular or ill-conditioned (scilab message), so the algorithm is having some problems inverting it when computing the the kalman gain matrix ( KalmanGain = VPrediction(:,:,t)*B'*inv(F))Can anyone help plz ?!Thanks in advance.

Hope this helps...http://econpapers.repec.org/article/eee ... -308.htmIn general, VCV of kalman filter model will vanish when the design matrices are ill-defined. Above paper will show you the exact form of design matrices.

Thanks Weare, i'll look at it