Forecasting Oman’s Blue Economy Trade: A Sector-Disaggregated Comparison of Five Small-Sample Forecasting Methods with Structural-Break Adjustment

Resumen

This study forecasts export and import values across four sectors of Oman’s blue economy, fisheries and seafood processing, coastal and maritime tourism, maritime freight transport, and port services and logistics, using UNCTADstat trade data from 2005 to 2024. Two forecasting approaches, autoregressive integrated moving average (ARIMA) and Bayesian structural time series (BSTS), are compared against each other, against a simple equal-weight combination of both, and against two standard small-sample benchmarks, a random walk with drift and exponential smoothing (ETS), with accuracy measured throughout by genuine out-of-sample rolling-origin backtesting. Structural-break dummies for the COVID-19 disruption and the 2023 to 2024 Red Sea shipping crisis are entered as regressors where the
affected sector and years overlap, and the fisheries sector is additionally cross-checked against raw UN Comtrade goods trade (HS codes 03 and 16), which runs on average 15 per cent below UNCTADstat’s own ocean-adjusted figure, a gap attributable to UNCTAD’s own coefficient adjustment for ocean-based activity. That same Comtrade data extends back to 2000, twelve years before UNCTADstat’s own fisheries table, and a robustness check on this longer series confirms the fisheries sector’s preferred method with five times as many backtest origins as the headline result alone provides. Method preference varies by sector: ARIMA wins one series, BSTS wins two, ETS wins three, and the random walk with drift wins one, with the equal-weight combination winning none outright and no method dominating across Oman’s blue economy. A Holm-corrected Diebold-Mariano test was applied to all nine method rankings, the seven headline series plus the two extended fisheries checks, comparing each winner against its runner-up. None of the nine rankings reached statistical significance. A short annual backtest therefore offers limited power to distinguish between competing forecasting methods.