Pembuatan Sistem Peramalan Nilai Tukar Petani Berbasis Model SARIMAX dan Hybrid SARIMAX-LSTM

Abstract

Nilai Tukar Petani (NTP) is a key indicator of farmers’ welfare in Indonesia, reflecting the ratio between prices received by farmers and prices paid for consumption and production inputs. Despite its strategic role in policy formulation, responses to NTP fluctuations tend to be reactive, while disparities across agricultural subsectors and regions remain evident. This study aims to develop an adaptive forecasting system for NTP by incorporating relevant exogenous variables, namely the IT and IB. Monthly NTP data from 2010 to 2024 published by Badan Pusat Statistik (BPS) Indonesia were analyzed across 166 datasets representing province- subsector combinations. The proposed methodology integrates SARIMAX and a hybrid SARIMAX-LSTM model in a sequential and conditional framework. Ecogenous variables were forecasted using a baseline comparison between SARIMA and Seasonal Naive, with safreguard mechanism to handle flat forecast and low-accuracu cases. SARIMAX captures linear, seasonal, and exogenous effects, while residual diagnostics using Ljung-Box and BDS tests determine the presence of remaining nonlinear dependencies. When nonlinear structures are detected, an LSTM model is activated to model SARIMAX residuals. The hybrid SARIMAX-LSTM model was applied to all 166 datasets. The results indicate that most NTP series exhibit non-stationarity behavior, heterogenous trends, and generally weak seasonal patterns. Evaluation results show that the adaptive approach achieves good forecasting accuracy for the entire of datasets, with an average MPAE 0.93%, average RMSE of 1.32, and average MAE of 1.03. These metrics indicate that the combination of hybrid modelling and high-quality exogenous inputs significantly improves forecasting accuracy. Overall, this study demonstrates that an adaptive hybrid modelling approach improves NTP forecasting accuracy and provides a robust data-driven foundation for anticipatory agricultural policy analysis.