Stochastic analysis based on deterministic Brownian motion (Q5951517)

The leading idea of this original article is to provide (in one dimension) an alternative model to the Brownian one, sharing many properties with Brownian motion, but having \(0\)-entropy and being consequently well adapted to prediction. To proceed, a deterministic simple sequence of piecewise linear functions \(f_n\) from \([0,1]\) onto \([0,1]\) is constructed by induction and trichotomy; the basic deterministic process is then the limiting function \(f_\infty\) of this procedure. The so-called ``deterministic Brownian motion'' \(N_t\) is then introduced by randomly shifting and scaling the deterministic \(f_\infty\). This gives a centered, time reversible, square integrable, self-similar with order \(1/2\) process, which has stationary, uncorrelated, strictly ergodic increments with \(0\)-entropy. The stochastic integrals \(\int^b_a H_x'(N_t,t) dN_t\) are well defined for real functions \(H(x,t)\) which are of class \(C^2\), \(C^1\) with respect to \(x\), \(t\), respectively, and they satisfy Itô's formula. Finally, given \(a< b\) and assuming \(H>0\), the stochastic process \(Y_t:= H(N_t, t)\) is well predictible in the following sense: assuming \(Y_{[a,b]}\) known, an estimator \(\widehat Y_c\) of \(Y_c\) exists for \(c> b\), such that for small \((c-b)\): \(\|\widehat Y_c- Y_c\|_2= O(c-b)\).