In this paper we give a mathematical frame for interest rate series and an evaluation procedure for forecasting interest rate series as autoprojection. The mathematical frame is a random system with complete connections and the forecasting procedure a combination of Fourier polynomials and and an AR(3) target function, whose parameters are estimated by the method of conditional least squares, as described in the paper by Klimko and Nelson (1978). The precision of our results are very promising, compared with traditional approaches