Dimension-wise integration of high-dimensional functions with applications to finance (Q708311): Difference between revisions

The authors derive a new class of methods for the so-called ``dimension-wise'' numerical integration of a function \(f\) of \(d\) variables, \[ I_d(f) = \int_{[0,1]^d} f({\mathbf{x}}) d{\mathbf{x}}. \] A highly readable survey of theoretical and practical aspects of the computation of high-dimensional integrals is presented. The anchored-ANOVA (analysis of variance) decomposition where only a finite number of function values is required for its computation is employed. The methods then result from a truncation of the anchored-ANOVA decomposition and from integration of the remaining terms using appropriate low-dimensional quadrature rules. It is demonstrated that the new methods can be applied in locally adaptive and dimension-adaptive ways. Their efficiency is demonstrated based on numerical experiments from finance (price to collateralized mortgage obligation, price zero coupon bounds, and test problem to determine the fair value of an Asian option with geometric average).