| Abstrakt: |
Asuransi Usaha Tani Padi (AUTP) is a Multiperil Crop Insurance (MPCI). However, in practice AUTP has a weakness in which they have a lengthy claim process. Nevertheless, this weakness can be overcome by implementing weather index-based agriculture insurance. To create a weather index based agricultural insurance, a forecasting model is needed. The two most crop damaging disasters in Indonesia are flood and drought, and the climate factor that is responsible for it is precipitation. There are two data commonly used in climate research, the weather station and satellite data. In this research, satellite data is selected because the insured crops are not always near a weather station. It is known that the variance of precipitation is not constant, hence the GARCH method will be used in this research in order to overcome the assumption of constant variance in a conventional forecasting method. The Sangkapura, Banyuwangi and Karangkates district is chosen to represent districts with high, low and medium rainfall among districts in East Java as the province with the widest crop land in Indonesia. In this research, SARIMA is first used to create the initial model, then heteroskedasticity is not found on the residual model, hence the GARCH model is not needed and only the SARIMA model will be used. The MAPE for the Sangkapura and Banyuwangi model is 3.62% and 3.48%, however the Karangkates model MAPE is at 10.7%. This happens because the SARIMA model is too heavily affected by the last yearly pattern of the training dataset. With a little shift on the training cut-off, the MAPE became 4.8%. The only models that are able to predict flood and drought with decent accuracy, precision and recall are the Sangkapura and Banyuwangi flood models. This prediction can later be used to calculate the premium using the indemnity function. The auto ARIMA function in R has a lower BIC, MAPE, precision and recall on the auto ARIMA forecasting results, and some of the models created fail to pass the white noise assumption in ARIMA. In conclusion, heteroskedasticity are not present on precipitation data in East Java, SARIMA is able to create a good model for precipitation, however it is important to properly choose the cut-off for train testing data since SARIMA is greatly affected by the last yearly pattern of the training dataset, model still needs improvement on predicting disasters and auto ARIMA is not yet able to be a substitute to the conventional way of deciding ARIMA order. [ABSTRACT FROM AUTHOR] |
