Utility Preferences Over Time and Wealth in Portfolio Management

Computational cognitive modelling is a central area within cognitive science and psychology that involves developing formal, quantitative models of human cognition and behavior. A critical challenge in current research is how to appropriately model individual differences. Traditional modeling approaches often assume all individuals operate under the same parameter values, which can obscure meaningful variability and lead to artifacts. In contrast, hierarchical models [1] (also known as multilevel models) allow researchers to model both shared group-level patterns and individual-specific deviations. This structure enables more accurate parameter estimation, especially when data per individual is limited, and opens up new possibilities for real-world applications-such as tailoring cognitive training interventions to an individual's learning profile or understanding variability in clinical populations. To date, most hierarchical implementations of cognitive models have relied on Bayesian approaches. Bayesian frameworks are flexible and theoretically well-suited for hierarchical modeling. However, they are typically computationally expensive, inefficient, susceptible to bias and may fail to converge, particularly if models have complex likelihood function and/or data are limited. These limitations greatly limit the suitability of such models for real-world applications, where efficiency and scalability are often critical.