Year: 2026
Type: article
Abstract: This study thoroughly evaluates a Quantum Bayesian Time Series Forecasting model, focusing on how well it can measure uncertainty in different types of noise and with varying numbers of data points. The paper compared the model‘s performance using several key metrics” Mean Squared Error (MSE), Mean Absolute Error (MAE), Coverage at 95%, Aleatoric and Epistemic uncertainty” against both synthetic noise models (heteroscedastic, heavy-tailed, homoscedastic, and regime-dependent) and real financial data. The results highlighted quantum advantages in breaking down uncertainty, especially in situations where data is volatile or not stable over time. Using a Bayesian time series model, the study obtained average levels of aleatoric and epistemic uncertainty for each assumption at different forecast sample sizes ”60-, 120-, 180-, and 240- time units (which can be years). The heavy-tailed error model performed best, with the lowest MSE and MAE across many sample sizes, such as MSE of 1.71 at sample size 60 and a high coverage of 0. The regime-dependent model performed well in one situation (lowest MSE of 1). The paper showed that including heavy-tailed errors and accounting for regime changes improved the accuracy and reliability of Bayesian forecasts. These results suggested that both aleatoric and epistemic uncertainty should be modelled in time series data, as done in full predictive density approaches, and have important implications for forecasting in financial, economic, or environmental contexts where volatility and sudden changes are common.Keywords: Bayesian; Quantum Computing, Simulation; Time Series; Forecasting
DOI: 10.66721/3s6jbr20
Author: Isiaka Oloyede
Institution: University of Ilorin
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7165563925