A class of globally solvable Markovian quadratic BSDE systems and applications (Q1747757)

The authors consider the following backward stochastic differential equation (BSDE) \[ Y_t = G + \int_t^T f(s,Y_s,Z_s) ds - \int_t^T Z_s dW_s, \] where \(W\) is a \(d\)-dimensional \(\mathcal F_t\) Brownian motion, \(G\) is \(\mathcal F_T\) measurable and an \(N\)-dimensional random vector and \(f\) a random function. By a solution here we mean a pair \((Y,Z)\) consisting of an \(N\)-dimensional semimartingale \(Y\) and an \(N \times d\)-dimensional adapted process \(Z\) which satisfy the above equation a.s. The authors assume that \(G\) is of the form \(G=g(X_T)\) and the generator \(f(t,X_t,Y_t,Z_t)\) grows at most quadratically in \(Z\). They work with a novel structural condition on \(f\) which requires the existence of a Lyapunov function. The main result of this paper states that whenever a Lyapunov function exists and an additional a priori local-boundedness condition holds then the above BSDE admits a Markovian solution if in addition \(g\) belongs to a local Hölder space.