Solving Unbounded Quadratic BSDEs by a Domination Method

Abstract

We introduce a domination argument which asserts that: if we can dominate theparameters of a quadratic backward stochastic differential equation (QBSDE) with continuousgenerator from above and from below by those of two BSDEs having ordered solutions, thenalso the original QBSDE admits at least one solution. This result is presented in a generalframework: we do not impose any integrability condition on none of the terminal data of thethree involved BSDEs, we do not require any constraint on the growth nor continuity of thetwo dominating generators. As a consequence, we establish the existence of a maximal anda minimal solution to BSDEs whose coefficient H is continuous and satisfies |H(t, y, z)| $\le$$\alpha$ t + $\beta$ t |y| + $\theta$ t |z| + f (|y|)|z| 2 , where $\alpha$ t , $\beta$ t , $\theta$ t are positive processes and the function f ispositive, continuous and increasing (or even only positive and locally bounded) on R. This isdone with unbounded terminal value. We cover the classical QBSDEs where the function f isconstant ([10], [12], [23], [25]) and when f (y) = y p ([21]) and also the cases where the generator has super linear growth such as y|z|, e |y| |z| p , e e |z| 2 , (k $\ge$ 0, 0 $\le$ p < 2) and so on. Incontrast to the works [10, 12, 21, 23, 25], we get the existence of a a maximal and a minimalsolution and we cover the BSDEs with at most linear growth (take f = 0). In particular,we cover and extend the results of [22] and [24]. Furthermore, we establish the existence anduniqueness of solutions to BSDEs driven by f (y)|z| 2 when f is merely locally integrable on R.