A Hybrid Soft Computing Approach to Inflation Forecasting: HybridSutte Versus Exponential Smoothing Benchmarks in an Emerging Economy

Abstract

Central banks in commodity-dependent emerging economies face a structural forecasting challenge: exponential smoothing methods calibrated on supply-shock training windows systematically overproject the downward trend into post-shock stabilisation phases, producing compounding errors that undermine monetary policy communication. This paper proposes HybridSutte, a soft computing model that fuses four-point alpha-Sutte recurrence with exponential smoothing correction, as an alternative to conventional exponential smoothing benchmarks. Monthly year-on-year Consumer Price Index data published by Bank Indonesia cover January 2021 through December 2025 (n = 60 observations), capturing Indonesia's complete monetary policy cycle: COVID-19 demand recovery, Russia-Ukraine commodity supply shock (peak: 5.95%, September 2022), Bank Indonesia's 250 basis-point rate-hike disinflation campaign, and the subsequent 2025 post-shock stabilisation within the 2.5% ± 1% target band. The 51/9 in-sample/out-of-sample partition places the evaluation window (April–December 2025) entirely within the structurally distinct post-shock stabilised regime. HybridSutte achieves out-of-sample RMSE of 0.606% and MAPE of 21.25%, compared with Holt's double exponential smoothing (ETS) RMSE of 3.069% and MAPE of 121.60%, yielding reductions of 80.2% and 82.5%, respectively. The performance advantage grows monotonically with forecast horizon h, reaching a 451.1% cumulative absolute error differential by  = 9. This is the first application of HybridSutte to central bank inflation data in an emerging market and the first to evaluate a soft computing hybrid model across a complete five-year monetary policy cycle. Findings support regime-aware model selection for central bank forecasting departments.