A1951
Title: Robust prediction intervals for time series forecasting: A bootstrap and Bayesian approach
Authors: Betty X-Y Chu - National Chung Hsing University (Taiwan) [presenting]
Cathy W-S Chen - Feng Chia University (Taiwan)
Abstract: A new bootstrap strategy, the standardized skewed Student-t block multiplier bootstrap (skew-tBMM), is proposed to capture heteroskedasticity and heavy-tailed behavior in time series forecasting with machine learning models. Prediction intervals are constructed by integrating bootstrap resampling with machine learning methods, including support vector regression, random forest, XGBoost, and long short-term memory networks, while Bayesian structural time series (BSTS) and autoregressive integrated moving average models serve as probabilistic benchmarks. To enable fair comparisons across models and data-generating mechanisms, a scale-adjusted relative interval score is introduced for evaluating interval forecasts. Simulation studies under nonlinear dynamics, deterministic seasonality, transfer-function effects, and conditional heteroskedasticity show that skew-tBMM provides well-calibrated intervals with competitive sharpness. An empirical application to monthly suicide counts in Japan illustrates the practical applicability of the framework. Results indicate that machine learning models combined with skew-tBMM, particularly long short-term memory networks, produce reliable forecast distributions in complex and heavy-tailed settings, while BSTS performs well for probabilistic forecasting.
