Generalized stochastic differential utility and preference for information (Q1769427)

This paper studies the preference for information for a specific class of intertemporal utilities. For a fixed consumption horizon \(T> 0\), a utility function is a function mapping the set of objects of choice, that is, the pairs of state contingent consumption process \(c= \{c_t,\,0\leq t\leq T\}\) and information filtration \({\mathcal A}= \{{\mathcal A}_t,\,0\leq t\leq T\}\) satisfying the usual conditions, into \(\mathbb{R}\). The question of preference for information consists in analyzing the dependency of a utility function on its filtration information argument. This specific question has been greatly simplified within the familiar context of the von Neumann-Morgenstern expected utility. Specifically, an expected utility function is defined by \[ U^{{\mathcal A}}_t(c)= E\Biggl[\int^T_t e^{-\beta(s- t)}v(c_s)\, ds\,\left|\,{\mathcal A}_t\Biggr]\right.\tag{\(*\)} \] for time \(t< T\), \(v(\cdot)\) is the felicity function and the expectation \(E\) is conditioned by the time \(t\) available information \({\mathcal A}_t\). In fact, the expected utility model \((*)\) does not allow for preference for information in the sense that if \({\mathcal A}\subsetneq{\mathcal B}\) are two filtrations such that both \({\mathcal A}_0\) and \({\mathcal B}_0\) are trivial \(U^{{\mathcal A}}_0(c)= U^{{\mathcal B}}_0(c)\) for any \({\mathcal A}\)-adapted consumption process \(c\).