MODELS AND ALGORITHMS OF AN INTELLIGENT INFORMATION TECHNOLOGY FOR PERSONALIZED INVESTMENT PORTFOLIO OPTIMIZATION BASED ON ADAPTIVE RISK PROFILING

Abstract

The paper develops an intelligent information technology for personalized investment portfolio optimization based on adaptive risk profiling. The relevance of the study is determined by the growing demand for digital decision-support tools that can account for individual investor characteristics and operate with heterogeneous financial instruments within a unified portfolio construction process. In many practical solutions, personalization is reduced either to selecting one of several predefined portfolio templates or to assigning an investor to a discrete risk category. Such approaches are insufficiently flexible and do not provide a formal connection between the estimated risk profile and the optimization constraints that determine the final portfolio structure. The purpose of the study is to develop and formally describe an intelligent information technology for personalized portfolio optimization and to justify its algorithmic support and validation approach. The proposed technology is represented as the tuple IT = 〈D, M, Algo, C, V 〉, where D denotes market and user data with preprocessing procedures, M denotes intelligent models, Algo denotes algorithmic portfolio-construction components, C denotes a parameterized system of optimization constraints, and V denotes visualization and validation modules. Adaptive questionnaire responses are transformed into a user feature vector and then mapped by a machine-learning model to a continuous risk profile. Personalization is implemented through a formal dependence between the user risk value and the feasible portfolio set. Two key algorithmic modules are proposed. The first forms a candidate asset universe by reducing redundancy in the initial instrument set on the basis of correlation or covariance dependencies, which decreases dimensionality and improves the stability of optimization results. The second parameterizes optimization constraints as a function of the continuous risk profile, including bounds on asset-class weights, the share of anchor instruments, and concentration restrictions. Portfolio construction is formulated as an optimization problem in which the feasible set is determined by personalized constraints and the reduced candidate asset universe. The paper also considers computational complexity in offline and online contours, portfolio-quality metrics, stability metrics with respect to small risk-profile changes, and an ablation-based validation procedure. The practical value of the proposed approach lies in combining explainability, reproducibility, controllable personalization, and computational feasibility within one integrated technology. The obtained results show that the proposed formalization provides a coherent foundation for building applied portfolio recommendation systems that are more stable, interpretable, and adaptable to individual investor characteristics.