The Best Forecasting Model For Cassava Price

This study aims to analyze and select the most accurate forecasting for predicting cassava prices in Indonesia. The data used is monthly data during the period of 2009 to 2017. This predicting uses the forecasting model, such as Moving Average, Exponential Smoothing, and Decomposition. Selecting the models found by comparing the smallest values of MAPE, MAD, and MSD. Therefore, it concluded that the Moving Average model is the most appropriate to Forecasting the price of cassava.


INTRODUCTION
Indonesia is a country that has various natural resources compared to other countries in the world. Abundant natural remedies can improve the economy to sustain food security in a region. Cassava is one food that can use as a substitute for rice or corn.
In Indonesia, Cassava has a significant economic value than other types of the tuber. In the arid regions in Indonesia, the function of cassava as a staple food because it is rich in carbohydrates. Humans can use almost all parts of the cassava plant, for example, as a vegetable and old leaves used as fodder, the stem is used as firewood. Products processed from cassava, among others: noodles, crackers, pluntiran, tiwul instant, bidaran, stick, tiwul gatot, and layer cake, Demand for cassava throughout the entire region of Indonesia has growth of about 3.16% and productivity of 228.16 KU / Ha over the past five years (Pusat Data dan Sistem Informasi Pertanian, 2016). Cassava requests in a region so vary that it affects the difference in the price of cassava in each area. We can see cassava price development in Table 1.
Based on Table 1 shows that the volume of the price of cassava has increased from 2009 to 2017. The development of the cassava price at rural consumers not only increases every year but also every month and It rose to the highest rate in December 2017 about Rp. 276,160/100 Kg. Price stability in the future happens through a price forecasting approach. The purpose of this study was to find the best method of forecasting of the cassava price.

MATERIALS AND METHODS
This study using the data of rural consumer price of cassava development in Indonesia from 2009 to 2017. Total 98 observation model used in this study.

Method of Moving Averages (moving average)
Methods Moving Average (MA) is an indicator often used in technical analysis that shows the average value of the data during the period specified. Data averaged time-dependent data (time series). Moving averages are used widely in stock/forex technical analysis, prices to measure momentum, and determine areas of support and resistance that are possible. Simple Moving Average (SMA) used to create a smooth or smooth Stock/forex price curve and filter noise data so that it is easier to see the trend data (Irfan Abbas, 2016). The Formulas For Moving Averages Are : = ( + −1 + −2 + ⋯ .+ − +1) / Where: = data series N = Total number of average periods = Prediction in period t + 1

Decomposition Model
In the decomposition method, there are additive and multiplicative decomposition models. Additive and multiplicative decomposition models can be used to predict a trend, seasonal, and cycle factors. The simple average decomposition method assumes the additive model, while the ratio decomposition method on the moving average (classical decomposition) and the Census II method assume a multiplicative model.

Exponential Smoothing Model
Exponential smoothing is a method that describes repetition procedures in continuous calculations using new data. The weighting system can be symbolized by a. A symbol can be freely determined to reduce forecast error. Smoothing constant values, a value of 0 can be chosen because it applies: 0 < a < 1 (Garspersz, 2004).

Size of forecasting results
According to Wardah and Iskandar (2017), Measurement of forecasting is a measure of error about the slight difference between the results of predicting and actual demand. To calculate forecast errors are usually used Mean Absolute Percentage Error (MAPE), Mean Absolute Deviation (MAD), and Mean Square Deviation (MSD) (Sidik, 2010).

RESULTS AND DISCUSSION
The article on national price data throughout the territory of Indonesia is used to analyze data. This price data uses the last nine years period from January 2009 to December 2017. The cassava commodity price data is obtained from the secondary data of the Ministry of Agriculture Price Information System. The data collected is then analyzed with the following results:

Moving Average Plot for Harga Ubi Kayu
MSD have the smallest value, then the model is the best forecasting model.

Moving Average (MA) Method
Moving Average model estimation is presented to see the forecasting of cassava prices in the next period is order two movements in the -110 period. The model estimation results are shown in Figure 1. Visually the model estimation results are done five times from the three images. It can be seen that the line shows the forecasting value almost coincident or closest to the line representing the actual value is the forecasting value produced by MA (3). While for the forecasting value produced by MA (2) and MA (4), it is not too coincide with the line that represents the actual value. Thus MA (3)

Exponential Smoothing Method
Exponential smoothing forecasting in this article uses a single exponential smoothing method. The exponential way is a weighting forecasting technique where data is weighting by an exponential function (Render and Heizer, 2005). Exponential smoothing has a more accurately level of accuracy compared to the moving average forecasting method even though it has similarities. The estimation results of the model using various levels of α (0.1 -0.9) presented in Table 5.  Table 5, it is known that the best criteria are the modelwith the smallest error value that value of α = 0.9, the MSE value 4344500, MAD value 1438, and MAPE value 1. Based on this method, the price of cassava forecasts for the period of February 2018 is Rp 275,913. From this forecasting method, it can be seen that there is a decrease in the price of cassava by Rp 247 in February 2018.
The decomposition method tries to identify ways that are separate from basic patterns that tend to characterize data series, especially economic and business data, to see stationary data. The following are the results of an analysis of cassava data using the Decomposition Method, either the Additive Decomposition Method or the Multiplicative Decomposition Method.

Decomposition Method
Multiplicative and additive decomposition models are methods that are often used to generate predictions by regarding various factors such as trends (cycles), cycles, and seasonally. In figure (a) is a multiplicative decomposition model while in figure (b)

Accurate Model Selection
Forecasting the price of cassava to predicting cassava prices has not yet occurred to forecast the cost of cassava in the future by using data on cassava prices from the past. Forecasting the price of cassava in this article uses the forecasting Moving Average method, the Exponential Smoothing model, and the Decomposition model. Of the three models, the most accurate model will be chosen to determine the best forecasting of cassava prices. Model selection is made by comparing the MAPE (Mean Absolute Percentage Error), MAD (Mean Absolute Deviation), and MSD (Mean Square Deviation) values of each model have done before. The results of the three models show in Table 7 as follows: