Ansari Saleh Ahmar
This study presents BetaSutte — a novel hybrid forecasting model applying the four-lag α-Sutte Indicator to OLS-detrended residuals rather than to the raw level series — and evaluates it on Indonesia’s monthly Nilai Tukar Petani (NTP, Farmer Exchange Rate), a farmer exchange rate for over 40 million farming households. Using 84 monthly observations from January 2019 to December 2025 (Badan Pusat Statistik), the model separates NTP into a linear trend component Tt = a + b · t and a residual Rt = Xt − Tt, applies the α-Sutte formula AS(Rt) to the stationary residual domain, and generates forecasts as X̂t = Tt + β · AS(Rt) with β optimised by grid search. Calibrating on the first 60 observations (January 2019–December 2023), the OLS trend explains 82.75% of NTP variance (Tt = 98.05 + 0.24 · t, R2 = 0.8275), and the optimal β = 0.30 yields in-sample RMSE = 1.6887, MAE = 1.3695, and MAPE = 1.2881% — an 11.5% RMSE reduction versus the trend-only baseline. Crucially, two full years of genuinely out-of-sample validation (January 2024–December 2025, n = 24) confirm BetaSutte’s operational superiority: RMSE = 5.4841 versus 6.0782 for trend-only, a 9.8% improvement representing 112 months of independently collected data never seen during calibration. Residuals are normally distributed (Shapiro-Wilk p = 0.130), confirming well-conditioned model inputs. The full-sample retrained model (n = 84) estimates Tt = 96.02 + 0.33 · t (R2 = 0.9169), forecasting January 2026 NTP at 123.91. This study constitutes the first BetaSutte application to a farmer exchange rate with two-year prospective out-of-sample validation. © 2026, PT Mattawang Mediatama Solution. All rights reserved.
Department of Statistics, Faculty of Mathematics and Natural Sciences, Universitas Negeri Makassar, Makassar, 90223, Indonesia
