Analysis · Finance
A Narrower Forecast Range Can Hide More Misses
By AWEI · AI-compiled · Published · Analysis prepared · 1 source · doi.org
A forecast range can look reassuringly precise while missing most of the outcomes it is meant to cover. In a study of Thailand’s seasoning exports, comparing the published tables reveals that only four of 24 monthly global export values fell inside the reported 95% bootstrap prediction intervals.
The 2026 paper by Nitinai Rungjindarat and Nalinee Phansaita compared forecasting methods using monthly export values in Thai baht. Models were fitted to January 2013–December 2022, with January 2023–December 2024 reserved for testing. The analysis covered worldwide exports and a regional grouping called ASEAN+8. These were monetary values, not quantities shipped or individual companies’ sales.
The researchers used SARIMA, which models patterns over time, including seasonality. They compared a systematic search through model configurations with an automated selection procedure. Their accuracy table reports lower average errors for the systematically selected models across six-, twelve- and twenty-four-month evaluation windows. That is evidence from a historical forecasting comparison, not an experiment establishing what causes export growth.
For uncertainty, the authors generated 1,000 bootstrap replications by resampling past model errors and adding them to forecasts. The resulting ranges were generally narrower than the standard intervals. The paper interprets this as more realistic uncertainty, but narrowness alone cannot establish reliability. The four-of-24 result comes from comparing its actual-value table with its global bootstrap interval table.
That comparison involves just one testing period, so it cannot establish performance across other market conditions. Nevertheless, it directly challenges the paper’s confident interpretation of those intervals. Reusing historical errors also does not demonstrate protection against future structural changes; adding exchange rates and trade-policy information remains a proposed extension.
For readers assessing market forecasts, the useful distinction is between an accurate central estimate and a trustworthy uncertainty range. Check both the size of forecast errors and how often actual outcomes fall within the advertised bounds.
Sources used for this article (1)
Direct links to the publisher reports used to prepare this article.
- Source 1
- Enhancing Forecast Accuracy of Thailand’s Seasoning Exports Using SARIMA Models with Grid Search and Bootstrap Prediction Intervals Nalinee Phansaita
