Stochastic invariance and consistency of financial models (Q5926045)

Considering a connection between interest rates dynamics and consistent forward rate curves, the pioneering work of \textit{T. Bjørk} and \textit{B. J. Christensen} [Math. Finance 9, No. 4, 323-348 (1999)] has proposed to use for this purpose the invariance question for nonlinear Itô equation on a separable Hilbert space \(H\) \[ dX(t)=(AX(t)+F(X(t))) dt+B(X(t)) dW(t),\quad X(0)=x\in H.\tag{1} \] We recall that the set \(K\subset H\) is said to be invariant for (1) if and only if \(P(X^x(t)\in K)=1\) for all solutions \(X^x\) to (1) with \(x\in K\), \(t>0\), and the Wiener process \(W(t)\) in (1) takes values in a separable Hilbert space \(U\) and has covariance \(Q\), \(F\) is a mapping from \(H\) to \(H\) and \(B\) is a mapping from \(H\) to the space \(L(U,H)\) of all linear transformations from \(U\) to \(H\). The author concentrates on the approach which uses a connection between stochastic equations and deterministic control theory. First of all the invariance question for (1) with \(B(\cdot) \equiv B\) is reduced to the invariance problem for a controlled deterministic equation \[ dy(t)/dt=Ay(t)+F(y(t))+BQ^{1/2}u(t),\quad y(0)=x\in H, \] and necessary and sufficient conditions for stochastic invariance in the Nagumo form are formulated. Invariant, finite-dimensional manifolds \(K\) are of special interest, which are precisely those, that should be used for curve fitting. Then, if-and-only-if conditions for the existence of an invariant finite-dimensional linear subspace for Ornstein-Uhlenbeck processes are established, and Theorem 4 gives the first complete answer to the invariance question even in the specific case of Heath-Jarrow-Morton (HJM) models. The results are then applied to the HJM model in the Musiela parametrization both on infinite and finite intervals and to a recent model elaborated by El Karoui and collaborators. The final application concerns the model for LIBOR rates introduced by Brace, Gatarek and Musiela and something else.