Portfolio Optimization via Credal Probabilistic Circuits
David Ricardo Montalván Hernández, Cassio Polpo de Campos · 2023
Portfolio optimization is a crucial part of many investment approaches and is arguably employed by almost all traders in a way or another. We introduce novel approaches for determining optimal weights for portfolios using a class of robust probabilistic generative models. Specifically, we utilize credal probabilistic circuits, a type of generative model known for their ability to perform efficient exact probabilistic inferences and to handle uncertainty in a sound manner. To account for model or epistemic uncertainty, these models use the theory of imprecise probability. Sets of parameter values represent perturbations of the probabilistic circuit model and can be interpreted as an uncertainty-aware correction of the parameters of an underlying portfolio. We call the result as credal portfolio. We propose a method for determining the amount of perturbation that well-captures the uncertainty of the problem, which is employed for the analysis of investments with real-world daily stock market data, showing promising results when compared to usual approaches.