64th ISI World Statistics Congress - Ottawa, Canada

64th ISI World Statistics Congress - Ottawa, Canada

Assessing the Accuracy of Paddy Harvested Area Prediction from the Area Sampling Frame Survey Results and the Alternative Forecasting Methods


Octavia Rizky Prasetyo


  • K


64th ISI World Statistics Congress - Ottawa, Canada

Format: CPS Abstract

Keywords: forecasting, hierarchical, paddy


Since 2018, Statistics Indonesia (BPS) has implemented the Area Sampling Frame (ASF) method to estimate the paddy harvested area in Indonesia. This method upgraded the traditional so-called eye-estimate method, which allegedly produced overestimated figures of paddy harvested area. One of the advantages of the ASF method is its ability to yield the upcoming three months prediction of the paddy harvested area by observing the state of the paddy growing phase at the current month, which can give valuable information for policy formulation. This study aims to evaluate the accuracy of those predictions and proposes an alternative method for forecasting. We use Mean Absolute Percentage Error (MAPE) for the evaluation and apply a Hierarchical forecasting method under various reconciliation and forecast methods for the latter. The data used are from monthly ASF survey results for the period of January 2018 to September 2022 produced by BPS. The results reveal that the upcoming one-month prediction of paddy harvested area has good accuracy, indicated by a MAPE value of around 11 percent while a lower accuracy is found in the upcoming two-month and three-month predictions with a MAPE value of around 16 percent. However, there are systematic error patterns of the ASF prediction based on the state of the paddy growing phase, which tends to be higher than the harvested area estimation results. Hence, we examined the alternative forecasting method of Hierarchical forecasting using ETS and ARIMA to obtain forecasts of monthly paddy harvested area from January to September 2022 (test set). In doing so, we applied a rolling window method under three reconciliation scenarios: bottom-up forecast, top-down forecast based on the average historical proportion, and optimal combination forecast. Our findings point out that the out-of-sample forecasts of the upcoming three consecutive months of the harvested area using either ETS or ARIMA for the test set yield more accurate results than the ASF prediction for the same period. The smallest MAPE of below 8 percent is obtained from the combination of the bottom-up reconciliation method and the ARIMA model. It aligns with the nature of the calculation for the national figures of paddy harvested area based on the ASF method, which aggregates all provinces’ estimations. Our results underline that the ARIMA with a bottom-up reconciliation can be an alternative method to obtain the upcoming three-month paddy harvested area estimation instead of using the current prediction based on the growing phase state from the ASF survey results.