ARIMA model for forecasting using EViews
EViews from IHS Markit offers academic researchers, corporations, government agencies and students access to powerful statistical forecasting and modeling tools through an easy-to-use object-oriented interface.
EViews has established a reputation as a worldwide leader in Windows-based econometric and forecasting software. Originally developed and distributed by Quantitative Micro Software (QMS), now part of IHS Markit, the popular MicroTSP software from QMS, was one of the first forecasting and analytical packages available for the personal computer. EViews, a Window-based software, replaced MicroTSP in 1994. (www.eviews.com)
Time series analysis is the analysis of a set of data in a given time period past which is useful for knowing or predicting future conditions (Soejoeti, 1987).
forecasting forms an integral part of any decision-making process. It attempts to decrease the dependence on chance and provide some way to predict future events (DeLurgio, 1998)
Forecasting can help us determine our future steps, where after forecasting we can see the results of the strategies we apply for the future. so that if there is a risk then we will be prepared to deal with it and handle it well.
Open the EViews application, here we use EViews 8. Once opened, the following display will appear:
Set the desired workfile. For date specifications, the column frequency is the type of data whether annual, monthly or daily. Then, set the start date and end date. In this case used 23 observations. So its start with 1 and end with 23.
By clicking object, you can create a new object and select series and change its name, then it will appear in the workfile section. This step can also be done in another way, namely using the quick menu then selecting an empty group and directly copying the data you would analyse.
STASIONARITY
It can be seen the stationarity of the data visually with the graph. Select the view menu then select the graph, the display will appear as follows. After that to save the freeze menu can be used then save.
To perform the formal stationarity test, use the view menu and select the unit root test. after that do to differencing levels. After the results come out, look at the probability section, if it is less than the specified alpha (0.05) then it meets the stationary test (stationary data) if not then retest it and change the difference in coloumn Test for unit root in.
The results of the stationarity test for the differencing level show that the probability is more than alpha (0.05), therefore it can be concluded that the data is not stationary.
It is necessary to test with the 1st difference and the probability value is less than alpha so that the stationarity test is fulfilled.
For level, d = 0
For 1st difference, d =1
For 2nd difference, d=2
Autocorellation
Autocorrelation testing can be done with the view menu then select correlogram. After that select the difference according to the previous test (in this case the previous test shows that it meets stationary at the 1st difference).
After the results come out it can be concluded that the lag in ACF (Autocorrelation) and PACF (Partial Autocorrelation) intersects at lag-2, the possible models are AR (2), MA (2), or ARIMA (2,1,2). ACF will shows us about MA models and PACF for AR models and arima model is ARIMA (p,d,f).
From the possible models, it is necessary to estimate the best model. Select the quick menu and select Estimated Equation. Enter the formula listed above where d (ihsg) indicates that the data uses 1st difference. Repeatedly for each model.
All models meet the F test (determine whether the model is suitable) and the T test (determine whether the variable is significant), but in the comparison of AIC and SC values there is a difference where the lowest value is the ARIMA model (2,1,2) So the best model is ARIMA (2 , 1,2).
The next step is to forcast what the IHSG value is for the next 7 days. So the first thing to do is edit the structure of the workfile and add observations by changing the end date. On the proc menu, select structure / resize current structure.
In testing the selected model, do a forcast by selecting the forecasting menu, then selecting dynamic forcast, the results of the forecast will be stored in ihsgf
A candle loses nothing by lighting another candle -James Kaller
