An expansion in the model space in the context of utility maximization (Q1709603)

In general settings, the numerical computation of the investor's value function and corresponding optimal trading strategy remains a challenging problem. The paper, in this regard, presents an alternative method to overcome this problem. In particular, in an incomplete market where the underlying stochastic driver is a continuous semimartingale and the investor's preferences conform to a negative power utility function, a second-order Taylor expansion of the investor's value function with respect to perturbations in the underlying market price of risk (MPR) process is provided. As a by-product, first-order approximations of both the primal and the dual optimizers are also derived. In a nutshell, the paper can be seen as a nonlinear version of the primal-dual second-order estimation techniques used in the context of mathematical finance. The challenge here is, when the investor's value function is considered as a function of the underlying MPR, it is neither convex nor concave. This distinguishes the paper from the previous literature in that the latter mostly relies on these useful characteristics. In Section 1, \textit{Introduction}, the paper surveys main numerical approaches to find an optimal trading strategy. The paper, however, distinguishes itself from most of the previous literature by relaxing the Markovian assumption and dealing with a market which is more general and not necessarily complete. As a result, the paper introduces both a novel ``approximation tool'' that employs a highly tractable model to approximate highly intractable ones and a \textit{stability result} for identifying not only the paths along which the MPR process remains continuous but also the features of the MPR process whose effects on the utility maximisation problem are most significant. Through its Section 2, \textit{A family of utility maximization problems}, the paper introduces the primal and dual value functions for an investor with constant relative risk aversion (CRRA) and defines the utility maximization problem for each of these functions (cf. Eq. (2.2) and (2.3)). An explicit expression for the equivalent martingale measure in terms of both utility and value functions is also presented. Section 3, \textit{The problem and the main results}, provides first-order expansions and error estimates of the primal and dual value functions (cf. Theorem 3.1 and 3.2). An infinitesimal growth rate of the certainty equivalent of the initial wealth with respect to the changes in the MPR is also derived. The CRRA assumption, indeed, brings many stochastic interest rate models into the scope of this paper. The primal optimiser -- in terms of the fraction of the initial wealth invested in the risk asset -- is verified to be an optimal-control for the model under consideration -- with an error term \(O(\varepsilon^2)\) -- when the change in the MPR process is small. In addition, a correction term to the primal optimiser is provided such that the converge becomes even stronger (i.e., an \(o(\varepsilon^2)\)). In the Brownian setting, a second-order expansion and an \(O(\varepsilon^3)\)-optimal-control is available. Accordingly, Section 4 is devoted to the detailed \textit{Proofs of the main theorems} introduced in Section 3. In Section 5, \textit{Examples}, the performance of the approximation model is tested, first, through a short list of ``trivial and extreme cases'' that can be handled either numerically or explicitly, namely, when the \textit{base} MPR is close to zero or it is a non-zero constant process (i.e., Black-Scholes setting), or when the deviations from the \textit{base} MPR are constant. The scenarios considered suggest the same qualitative behaviour as in Section 3: the percentage changes in the investor's utility are proportional to the \textit{base} MPR process as well as the size of the perturbations from this process. The latter result seems to have become particularly simple to attain as the investor's value function can be approximated, at least locally, through an exponential function. Secondly, the numerical approximation results are compared against those of the well-known Kim-Omberg (KO) model for the MPR process. Appendix A, in this sense, is useful for getting to know the details of the KO model which indeed allows for many explicit calculations. Thirdly, the present model is applied to single- as well as multi-dimension extended affine models for the MPR process and, similar to the case of the Kim-Omberg model above, the results point to the significant superiority of the second-order approximations over their first-order counterparts. Appendix B further details the extended affine models used in the paper. Overall, the paper is a valuable attempt for enriching the portfolio of available methods to answer the question of what the optimal trading strategy would be in an incomplete market with no Markovian assumptions, and therefore is an interesting reading. Inspired by this work, further analytic and computational refinements and generalisations, as well as strong results and tools, can be expected in the related quantitative finance literature.