Martingale laws, densities and decomposition of Föllmer-Schweizer. (Lois de martingale, densités et décomposition de Föllmer-Schweizer.) (French) Zbl 0772.60033

The paper continues topics in [second author, ibid. 26, No. 3, 451-460 (1990; Zbl 0704.60045)]. Consider a probability space \((\Omega,{\mathcal F}_ 1,P)\), a filtration \(({\mathcal F}_ t)_{t\in[0,1]}\) on it etc. Consider first a continuous \(R^ d\)-valued process \(X=M+B\), where \(M\) is a continuous local martingale and \(B\) is previsible with finite variation. Let \(N^ i\) be orthogonal \(R\)-valued continuous local martingales with \(M=\sigma N\), \(\sigma\) previsible, let \(A\) be continuous ascending adapted with \(d\langle N^ i\), \(N^ i\rangle=\zeta^ iA\). If there exists a probability \(Q\), equivalent to \(P\), such that \(X\) is a \(Q\)-martingale, then one may choose previsible \(\nu,b\) such that \(b=\sigma\nu\), \(X=\sigma N+bA\) and \(\theta^ i=(\nu^ i/\zeta^ i)1_{(\zeta^ i\neq 0)}\) are \(N^ i\)-integrable. Moreover, all possible \(Q\) are \({\mathcal E}(-\theta N+L)P\), where \(L\) is a local martingale orthogonal to all \(N^ i\) (and \({\mathcal E}\) is the “stochastic exponential”). \(Q\) is unique iff \(N\) has the previsible representation property and \(\sigma\) may be chosen invertible.
The authors consider then discontinuous \(R\)-valued processes. They first give a correct proof of a result of the second author [Stochastic differential systems, Proc. IFIP-WG 7/1 Work. Conf., Eisenach/GDR 1986, Lect. Notes Control Inf. Sci. 96, 373-380 (1987; Zbl 0655.60038)], namely: if \(D\) is ascending right continuous previsible, \(A\) is right continuous previsible with finite variation, \(p\in(1,\infty)\), \(p^{- 1}+q^{-1}=1\) and \(K=\{\int HdA\); \(H\) previsible bounded, \(\int| H|^ pdD\leq 1\}\) is bounded in probability, then there exists a previsible \(\alpha\) with \(dA=\alpha dD\), \(\int|\alpha|^ qdD<\infty\). Then they prove, via four lemmas etc., that, if \(X\) is adapted, cadlag, \(X_ 0=0\), \(\sup_ s| X_ s|\in L^ 2\) and for the closure \(\overline K\) of \(K\) in \(L^ 1\) we have \(\overline K\cap L^ 1_ +=\{0\}\), then \(X\) is a special semimartingale, \(X=M+\alpha\langle M,M\rangle\) with \(M\) a locally square integrable local martingale, \(\int\alpha^ 2d\langle M,M\rangle<\infty\). A “pathological” example is given.
Let now \(X=M+A\), \(M\) a local martingale, \(A\) previsible with finite variation (the canonical decomposition of the special semimartingale \(X)\). Let \(Q\) be equivalent to \(P\). It is named a “minimal martingale law” for \(X\) if \(X\) is a \(Q\)-martingale and every local \(P\)-martingale \(P\)-orthogonal to \(M\) is a \(Q\)-local martingale. The authors show that if such a \(Q\) exists for \(X\) in the previous theorem, then \({\mathcal E}(-\alpha M)\) is a strictly positive martingale and \(Q={\mathcal E}(-\alpha M)_ 1 P\).
Finally, consider the case when \(M\) is locally square integrable, \(A=\alpha\langle M,M\rangle\), \(\alpha\) previsible, \(\int\alpha^ 2d\langle M,M\rangle<\infty\), \(\alpha\Delta M<1\). Let \(H=\lambda+\int\zeta dX+L_ 1\), where \(\lambda\in R\), \(\zeta\) be previsible, \(L\) be a semimartingale with \(L_ 0=0\), \({\mathcal E}(-\alpha M)\) \((\lambda+\zeta X+L)\) be a martingale and \([X,L]{\mathcal E}(-\alpha M)\) be a local martingale. Then \(\lambda,\zeta,L\) are uniquely determined by \(H\). If \(X\) is continuous and \(H\) is given, then \(H\) is representable as above iff \(H{\mathcal E}(-\alpha M)_ 1\in L^ 1\). If \(X\) is locally bounded, \(H{\mathcal E}(-\alpha M)_ 1\in L^ 1\) and \(E(H{\mathcal E}(-\alpha M)_ 1;{\mathcal F}_ t){\mathcal E}(-\alpha M)^{-1/2}_ t\) is locally square integrable, then \(H\) is representable as above. The paper contains several remarks, showing the “limits” of some of its results.


60G44 Martingales with continuous parameter
60H05 Stochastic integrals
Full Text: Numdam EuDML