Anomalous PDEs in Markov chains: domains of validity and numerical solutions (Q2488493)

The conditional expected values in Markov chains are solutions to a set of backward differential equations, which may be ordinary or partial depending on the number of relevant state variables. The second case is more difficult because the equations are not quite standard as they involve the unknown functions and their derivatives at different states. Moreover, the analytic properties of the unknown functions cannot be fully explored by mere inspection of the differential equations. Typically, the functions are not continuously differentiable everywhere, and they may even be discontinuous. The paper presents methods for selecting the points of non-smoothness before numerical methods can be devised. A numerical method is constructed that accounts for the non-smoothness problem and that works with the controlled global error. The method is akin to the Lax-Wendroff scheme, but does not involve any approximation of the second order derivatives. Two cases leading to first order PDEs in two variables are considered: the option pricing in a Markov chain driven financial market, and probability distributions of discounted cash flows generated by multi-state life insurance contracts.