A numerical modelling apparatus and method of performing numerical modelling are described. An input unit receives signals giving information relating to a set of assets. A processor unit is arranged to provide a set of Risk Relation Matrices Vτ for set of investment horizons indicated by τ. Each of the Risk Relation Matrices Vτ comprises a plurality of elements, wherein each of the elements represents a relationship of risk related to a respective pair of the assets and each element is given by a scalar product of two risk vectors, such that each of the assets has an associated risk vector according to the elements of the risk relation matrix. The processor unit is arranged to decompose each of the Risk Relation Matrices Vτ into eigenvectors and eigenvalues according to Vτ=Eτ·Λτ·E′τ, where, at each tenor τ, Eτ is a set of eigenvectors of the risk matrix Vτ in columns, Λτ is the corresponding diagonal eigenvalue matrix, and E′τ is the transpose of Eτ. Components of each of the risk vectors are derived at each tenor τ in the basis of unit independent risks by the corresponding row of the matrix product Eτ·Λτ1/2 relating to each of the assets. An output unit is arranged to output the components of each of the risk vectors as a risk vector data set.